Split (3)

train · 4.35k rows

article stringlengths 1.98k 169k	summary stringlengths 1.01k 4.15k	section_headings listlengths 2 38	keywords listlengths 0 12	year stringclasses 11 values	title stringlengths 30 189
Imaginal progenitors in Drosophila are known to arrest in G2 during larval stages and proliferate thereafter . Here we investigate the mechanism and implications of G2 arrest in progenitors of the adult thoracic tracheal epithelium ( tracheoblasts ) . We report that tracheoblasts pause in G2 for ~48–56 h and grow in size over this period . Surprisingly , tracheoblasts arrested in G2 express drivers of G2-M like Cdc25/String ( Stg ) . We find that mechanisms that prevent G2-M are also in place in this interval . Tracheoblasts activate Checkpoint Kinase 1/Grapes ( Chk1/Grp ) in an ATR/mei-41-dependent manner . Loss of ATR/Chk1 led to precocious mitotic entry ~24–32 h earlier . These divisions were apparently normal as there was no evidence of increased DNA damage or cell death . However , induction of precocious mitoses impaired growth of tracheoblasts and the tracheae they comprise . We propose that ATR/Chk1 negatively regulate G2-M in developing tracheoblasts and that G2 arrest facilitates cellular and hypertrophic organ growth . The precise regulation of cell division in space and time is essential for normal development . This is achieved by the integration of developmental signals and the machinery that drives cell cycle progression . Non-cycling cells are known to pause in G1 ( or G0 ) ( Cheung and Rando , 2013 ) or , less commonly , in the G2 phase ( Bouldin and Kimelman , 2014 ) of the cell cycle and rekindle a mitotic program at the appropriate time and location . G2-arrested cells have been observed in developing Drosophila ( Johnston and Edgar , 1998; Ayeni et al . , 2016 ) , Ciona ( Ogura et al . , 2011 ) , zebrafish ( Nguyen et al . , 2017 ) , chick ( Boije et al . , 2009; Stone et al . , 1999 ) and mice ( Seki et al . , 2007 ) . In this study we investigate the cell-intrinsic mechanisms underlying developmental G2 arrest and the significance of this mode of arrest in the context of the Drosophila tracheal ( respiratory ) system . Fruit flies undergo a complete metamorphosis during their life cycle and have distinct larval and adult body forms . Progenitors of the adult animal ( hereafter referred to as imaginal progenitors ) are set aside during embryonic development , nourished in the larva , and activated during the larval-adult transition ( Cohen et al . , 1993 ) . These progenitors generate adult organs de novo ( sensory organs , wings , legs ) or by remodeling larval organs ( gut , abdominal epidermis , tracheal system ) . Imaginal progenitors of different tissues may either proliferate throughout larval life or in a punctuated manner . The progenitors of the abdominal epidermis ( histoblasts ) ( Ninov et al . , 2009 ) , and the thoracic tracheal ( respiratory ) system ( Djabrayan et al . , 2014; Guha et al . , 2008; Sato et al . , 2008; Weaver and Krasnow , 2008; Pitsouli and Perrimon , 2010 ) exhibit the latter pattern . These cells remain mitotically arrested in G2 through most of larval life and initiate a program of rapid cell division and morphogenesis thereafter . The mechanism for G2-M is highly conserved among eukaryotic cells ( Bouldin and Kimelman , 2014 ) . The G2-M transition is triggered by the dephosphorylation of the Cyclin-dependent Kinase Cdc2/Cdk1 by the phosphatase Cdc25 . Dephosphorylation of Cdc2/Cdk1 by Cdc25 results in the activation of the Cdk-Cyclin B complex and in turn to the phosphorylation of substrates in the cytoplasm and nucleus and to mitotic entry . G2-arrested progenitors of sensory organs ( SOPs ) in Drosophila have been shown to repress Cdc25/String ( Stg ) expression ( Johnston and Edgar , 1998 ) . A recent study showed that SOPs forced to divide precociously undergo normal neuronal differentiation but generate supernumerary non-neuronal sensory organ support cells . Taken together , studies on SOPs show that G2 arrest is mediated by the transcriptional repression of Stg and that the arrest ensures that a proper balance of cell types within each sensory organ is achieved ( Ayeni et al . , 2016 ) . Whether other G2-arrested progenitors in Drosophila and in other organisms are regulated like SOPs is an open question . Histoblasts appear to be regulated in a similar manner . Non-cycling histoblasts lack Stg expression and upregulate Stg in an ecdysteroid signaling-dependent manner during G2-M ( Ninov et al . , 2009 ) . In the sea squirt Ciona , progenitors undergoing neurulation are known to arrest in G2 ( Ogura et al . , 2011 ) . The overexpression of Stg in these cells triggers precocious G2 exit and perturbs the formation of the neural tube ( Ogura et al . , 2011 ) . This suggests that G2-arrested neural progenitors in Ciona may be regulated in a manner akin to SOPs . A recent study on the muscle stem cells that contribute to the growth of the zebrafish myotome has shown that these cells arrest in G2 but utilize a different mechanism for arrest . G2-arrest in this context is mediated by Meox1-dependent repression of Cyclin B expression ( Nguyen et al . , 2017 ) . Proliferating cells in the G2 phase that are subjected to DNA damaging agents stall cell cycle progression , initiate DNA repair , and rekindle the cell cycle after repair is completed ( Branzei and Foiani , 2008 ) . Cells with DNA damage utilize other mechanisms for G2 arrest . Genotoxic stress leads to the activation of the phosphoinositide-3-kinase-related kinases ATR and ATM that phosphorylate and activate Checkpoint Kinases 1 and 2 ( Chk1 , 2 ) respectively ( Kumagai et al . , 2004; Chaturvedi et al . , 1999 ) . Chk1 and Chk2 , in turn , inhibit Cdc25 and arrest cell cycle progression ( Xiao et al . , 2003; Chaturvedi et al . , 1999 ) . In addition , Chk1 can also stabilize the Wee/Myt kinases that phosphorylate and inhibit Cdc2/Cdk1 ( O'Connell et al . , 1997 ) . While there is evidence that ATR/Chk1 can regulate cell cycle progression in cultured cells in the absence of induced DNA damage ( Sørensen et al . , 2004; Tang et al . , 2006 ) and during early embryonic development ( Sibon et al . , 1999; Sibon et al . , 1997; Su et al . , 1999; Liu et al . , 2000 ) their roles in developmental G2 arrest have not been fully explored . In this study we focus on the progenitors of the adult thoracic tracheal system ( tracheoblasts ) . It has been reported that tracheoblasts arrest in G2 during early larval life and rekindle a mitotic program at the onset of the pupal period . Our efforts to determine precisely when tracheoblasts enter and exit G2 showed that the cells enter larval life in G1 , transition from G1 to G2 in the first larval instar ( L1 ) and remain in G2 till the mid third larval instar ( L3 ) , a period of ~48–56 h , whereupon they enter mitosis . We investigated the status of Stg expression in G2-arrested tracheoblasts to find that these cells express Stg throughout . Moreover , we observed that paused tracheoblasts also expressed Cdc2/Cdk1 and Cyclin B . These findings alerted us to the possibility that G2 arrest is not mediated via transcriptional repression of Stg , or of the essential drivers for G2-M , and led us to investigate alternate mechanisms . The findings presented here show that the G2-M transition in tracheoblasts is negatively regulated by ATR ( mei-41 , hereafter ATR ) /Chk1 ( Grapes ( Grp ) , hereafter Chk1 ) , that the transition involves the coordination of several processes including the downregulation of Chk1 , and that arrest in G2 facilitates cellular and organ growth . The tracheal system of the Drosophila larva originates in the embryo from a pair of placodes in each segment . Each placode undergoes branching morphogenesis to generate tracheal tubes that anastomose at stereotyped locations and generate a connected network . The larval tracheal system is largely comprised of cells that undergo repeated cycles of endoreplication and are post-mitotic . In addition to serving the oxygen demands of the larva , the larval tracheal network also serves as a scaffold for the development of the tracheal system of the adult fruit fly . Embedded in larval tracheae , at stereotyped locations , are imaginal tracheal progenitors . These cells remain mitotically quiescent through larval life and proliferate and replace larval cells during metamorphosis ( Pitsouli and Perrimon , 2010; Weaver and Krasnow , 2008 ) . The progenitors that contribute to the development of the thoracic tracheal system of the adult animal , specifically progenitors of the second thoracic metamere ( Tr2 ) , are unusual . To the best of our knowledge , majority of the cells that populate Tr2 tracheae during larval stages are also imaginal progenitors ( Figure 1A , shown in green ) ( Guha et al . , 2008; Sato et al . , 2008 ) . Unlike post-mitotic larval tracheal cells , the cells that comprise Tr2 tracheae are mitotically competent , remain paused in G2 during larval life , and rekindle mitoses in the third larval instar ( L3 ) ( Guha et al . , 2008; Sato et al . , 2008; Djabrayan et al . , 2014 ) . For the analysis of the mechanism for G2 arrest in Tr2 we focused our attention on the cells that comprise the Dorsal Trunk ( DT ) in this segment ( Figure 1A , demarcated with dashed lines ) . Tr2 DT constitutes a developmental compartment consisting of 16–18 cells that become mitotically active mid-L3 ( Rao et al . , 2015 ) . To characterize when precisely cells in Tr2 DT enter and exit G2 , we counted the numbers of cells at different larval stages ( first instar ( L1 ) , second instar ( L2 ) , third instar ( L3 ) ( early L3 , 32–40 h L3 , wandering L3 ( WL3 ) ) and analyzed the timecourse of BrdU incorporation ( S phase ) and phospho-Histone H3 ( pH3 , M phase ) labeling at these respective stages ( Figure 1I , Figure 1—figure supplement 1 ) . Counts of cell numbers in Tr2 DT showed that there are 16–18 cells in L1 , that this number remains unchanged through 16–24 h L3 , that there are 25–35 cells by 32–40 h L3 and ~250 cells at wandering L3 ( WL3 ) ( Figure 1I , n ≥ 5 tracheae per timepoint here and in all subsequent figures showing cell frequencies ) . BrdU incorporation in Tr2 was observed in the first larval instar ( L1 ) and again at 32–40 h L3 but not in the interim period . Phospho-histone H3 labeling was detected from 32- 40 h L3 onwards ( representative images from different stages and a quantitation of pH3+ cells at these stages is shown in Figure 1—figure supplement 1 ) . Based on these analyses we surmised that cells in Tr2 DT complete S phase in L1 and remain in G2 from late L1/L2 till 24–32 h L3 . The FUCCI ( Fluorescence Ubiquitination-based Cell Cycle Indicator ) system facilitates precise cell cycle staging based on the levels of expression of fluorescent reporters ( Zielke et al . , 2014 ) . To characterize cell cycle phasing of cells in Tr2 DT , we expressed fluorescently tagged degrons from E2F1 ( E2F1-GFP , degraded at the onset of S phase ) and Cyclin B ( Cyclin B-RFP , degraded in mitosis ) in the tracheal system and analyzed expression of these reporters in Tr2 DT at different larval stages ( Figure 1B–F , n ≥ 6 tracheae per timepoint analyzed ) . In this system , cells in G1 are GFP+ , cells in S RFP+ and cells in G2 GFP+RFP+ . We found that cells in Tr2 DT are heterogeneous with respect to reporter expression in L1 ( GFP+ , RFP+ and GFP+RFP+ ) ( Figure 1B ) , homogeneous from L2 till early L3 ( GFP+RFP+ ) ( L2 shown Figures 1C , 0-8 h L3 not shown , 16–24 h L3 shown in Figure 1D ) and heterogeneous from 32-40 hL3 onwards ( GFP+ , RFP+ and GFP+RFP+ ) ( 32–40 h L3 shown in Figure 1E , WL3 shown in Figure 1F ) . Analysis of freshly hatched L1s showed that the cells in Tr2 DT are all GFP+ ( Figure 1—figure supplement 1 ) . Taken together , FUCCI analysis show that cells in Tr2 enter larval life in G1 , transition from G1 to G2 in the first larval instar ( L1 ) and remain paused in G2 from the second larval instar ( L2 ) till 24–32 h L3 ( ~48–56 h ) . In addition to these analyses , we also compared the DNA content of paused cells in L2 and early L3 with the DNA content of mitotically active cells in the G1 or G2 phase ( G1/G2 identified by FUCCI at WL3 ) ( Figure 1G ) . We noted that the DNA content of cells in L2 and early L3 was comparable to the DNA content of cells in G2 . Thus , FUCCI and DNA content analyses corroborated our initial assessment of cell cycle phasing of cells in Tr2 DT during larval life ( summarized in the diagram in Figure 1H ) . Studies in SOPs and other imaginal progenitors have shown that G2 arrest is due to the absence of Stg expression ( Johnston and Edgar , 1998 ) . To test whether this model is applicable to Tr2 tracheoblasts , we compared levels of Stg mRNA in Tr2 DT prior to and post mitotic entry using real-time PCR ( qPCR ) and immunohistochemistry . For qPCR analysis we micro-dissected Tr2 DT fragments from different stages and isolated RNA from these fragments ( ≥15 tracheal fragments per timepoint per experiment , n = 3 experiments ) . Levels of Stg mRNA at L2 , 0–8 h L3 , 16–24 h L3 and 32–40 h L3 were comparable ( Figure 1I ) and the level at WL3 was significantly higher than earlier stages ( Figure 1I ) . Stg immunostaining revealed low levels of Stg , in both nucleus and cytosol , at L2 , 0–8 h L3 and 32–40 h L3 ( Figure 1J , n = 9 tracheae per condition per experiment , n = 3 experiments ) . We detected higher levels , with the same spatial distribution as earlier stages , at WL3 ( Figure 1J ) . To confirm that the anti-Stg antibody was specific , we stained trachea from animals expressing StgRNAi in the tracheal system under the control of Btl-Gal4 ( Btl-StgRNAi ) . Staining in these animals was comparable to the staining in specimens incubated with the secondary antibody alone indicating that the Stg staining is specific ( Figure 1J ) . These experiments suggest that unlike SOPs , tracheoblasts paused in G2 express both Stg mRNA and protein . Next we investigated whether Cdc2/Cdk1 and Cyclin B , the other drivers of G2-M , are also expressed in tracheoblasts paused in G2 . Both mRNA and protein were detected in paused cells and exhibited a timecourse of expression like Stg/Stg ( qPCR data and immunostaining shown in Figure 1—figure supplement 2 , qPCR as above n = 3 , immunohistochemistry n = 6 tracheae per condition , n = 2 experiments ) . Together , the findings suggest that G2-arrested cells in Tr2 DT express all the cell cycle regulators necessary for G2-M . The expression of drivers of G2-M in paused cells was unexpected . To investigate whether mitotic arrest was due to insufficient expression , we co-overexpressed Stg , Cdc2/Cdk1 and Cyclin B in L2 and early L3 via heat shock and counted cells in Tr2 DT post heat shock . The co-overexpression of Stg , Cdc2/Cdk1 and Cyclin B did not result in increased numbers of cells ( Figure 1K ) . This suggestedthat G2 arrest in tracheoblasts could be due to the expression of negative regulators of G2-M and/or the paucity of other positive regulators . As mentioned earlier , downregulation of Cdc2/Cdk1 activity by the ATR-Chk1 or the ATM-Chk2 kinase cascades can lead to arrest in G2 . Next we investigated whether ATR and ATM have any role in the regulation of G2 arrest in Tr2 . We knocked down ATR or ATM ( Telomere Fusion ( Tefu ) , hereafter referred to as ATM ) levels in tracheae by RNAi ( Btl-ATRRNAi , Btl-ATMRNAi ) and counted the numbers of cells in Tr2 DT at L2 , early L3 , 32–40 h L3 and WL3 . The numbers of cells in Tr2 DT in Btl-ATRRNAi larvae were significantly higher than wild type by 16–24 h in L3 ( 21–26 cells compared to 16–18 cells , Figure 2A ) while there was no increase in cell number in Btl-ATMRNAi expressing animals . This finding implicated the ATR-Chk1 axis and not the ATM-Chk2 axis in the regulation of G2 arrest in Tr2 DT . To investigate the role of ATR-Chk1 further , we stained larvae with antisera against an activated ( phosphorylated ) form of Chk1 ( Sørensen et al . , 2004 ) . Anti-phospho-Chk1 ( pChk1 ) immunostaining was high in L2 and early L3 and low at 32–40 h L3 ( Figure 2B , n = 6–8 tracheae per condition per experiment , n = 3 ) and subsequent stages ( WL3 , data not shown ) . We stained for pChk1 in Btl-ATRRNAi animals and found no detectable signal indicating that Chk1 phosphorylation is indeed ATR-dependent ( Figure 2C ) . To confirm that pChk1 staining was specific , we stained Btl-Chk1RNAi expressing animals with the same antisera . pChk1 staining in Btl-Chk1RNAi was comparable to specimens stained with the secondary antibody alone ( Figure 2C , n = 3 experiments ) . Next we determined whether the loss of Chk1 also perturbed the timecourse of cell proliferation in Tr2 DT . Cell counts of Btl-Chk1RNAi larvae showed that the number of cells in in Tr2 DT is higher by 16–24 h in L3 ( 40–50 cells , Figure 2D ) . In addition to counting numbers of cells , we assayed mitotic activity in Btl-ATRRNAi and Btl-Chk1RNAi by anti-pH3 immunostaining at L2 , 0–8 h L3 and 16–24 h L3 . As indicated in the previous section there are no pH3+ cells in wild type Tr2 tracheae at these stages ( Figure 1—figure supplement 1 ) . We observed pH3+ mitotic figures in Btl-ATRRNAi at 16–24 h and in Btl-Chk1RNAi expressing animals at 0–8 h L3 and 16–24 h L3 ( Figure 2E ) . Whether the observed differences between Btl-ATRRNAi and Btl-Chk1RNAi are due to differences in the efficiencies of the respective RNAi lines or due to the presence of other negative regulators of Chk1 is currently unclear . Next we examined how Chk1 overexpression impacted the timing of mitotic entry in Tr2 DT . The overexpression of Chk1 under Btl control ( Btl-Chk1 ) inhibited mitotic entry at 32–40 h L3 ( Figure 2F ) . Based on these findings we conclude that ATR/Chk1 are negative regulators of G2-M in tracheoblasts . Despite precocious mitotic entry in ATR/Chk1-deficient animals , the numbers of cells in Tr2 DT at WL3 in these animals were significantly lower than in wild type ( Figure 2A , D ) . To investigate whether Chk1 has a role post-mitotic entry , we expressed Chk1RNAi in the tracheal system in a timed ( conditional ) manner using the Gal4-UAS-Gal-80ts system ( Figure 3 ) . The expression of Chk1RNAi from embryonic stages ( animals raised at 29°C till WL3 to inactivate Gal-80 ) led to a reduction in the numbers of cells at WL3 ( Figure 3A ) . However , expression of Chk1RNAi from embryonic stages till 24 h L3 ( animals grown at 29°C from embryonic stages to 24 h L3 and shifted to 18°C from 24 h L3 to WL3 ) did not lead to a reduction in the numbers of cells in Tr2 DT at WL3 ( Figure 3B ) . This suggests that in addition to regulating G2-M , Chk1 has a role in tracheoblasts post mitotic entry . Next we investigated whether reduced numbers of cells in Btl-Chk1RNAi at WL3 was due to decreased rate of proliferation , increased apoptosis or both . We measured the frequencies of mitotically active cells ( BrdU incorporation , anti-pH3 immunostaining ) and of apoptotic cells ( activated Caspase3 immunostaining ) in Btl-Chk1RNAi and wild type in late L3 . The frequencies of BrdU+ cells were comparable in Btl-Chk1RNAi and wild type but the frequency of pH3+ cells was significantly higher in Btl-Chk1RNAi ( Figure 3C ) . The frequencies of activated Caspase3+ cells were comparable in Btl-Chk1RNAi and wild type ( Figure 3D ) . Together , reduced numbers of cells in Btl-Chk1RNAi in late L3 and increased frequency of pH3+ cells in these animals at the same stage suggests that cells lacking Chk1 divide more slowly on account of a prolonged M phase . These data are consistent with a previously described role for ATR-Chk1 in the regulation of the mitotic spindle during cytokinesis ( Gruber et al . , 2011; Tang et al . , 2006 ) . Cells in Tr2 DT enter G2 late in L1/early L2 and remain in G2 till 24–32 h L3 . The loss of ATR/Chk1 results in mitotic entry in early L3 , ~24–32 h earlier than normal . The reason cells in Btl-ATRRNAi/Btl-Chk1RNAi initiated mitoses in L3 but not earlier was unclear and led us to further probe the dependence on Chk1 . Chk1 is thought to mediate G2 arrest by inhibition of Cdc2/Cdk1 ( Xiao et al . , 2003; O'Connell et al . , 1997 ) . The activation of Cdc2/Cdk1 involves dephosphorylation of residues Threonine-14 and Tyrosine-15 . Chk1 may prevent activation by inhibiting the phosphatase Stg and/or by stabilizing the kinases Wee/Myt . Next we examined how expression of an ‘activated’ form of Cdc2/Cdk1 ( Cdc2AF ) in which Threonine-14 and Tyrosine-15 have been mutated to Alanine and Phenylalanine respectively ( Chow et al . , 2003; Edgar and O'Farrell , 1990 ) , impacted proliferation . Cdc2AFwas induced via heat shock in L2 , 0–8 h L3 and 32–40 h L3 and numbers of cells in Tr2 DT were counted . We observed that Cdc2AF induction in L2 did not lead to an increase in cell number but induction at 0–8 h L3 and 32–40 h L3 did ( Figure 4A ) . We examined levels of Cdc2 expression at these stages , both prior to and post heat shock , using an anti-cdc2 antibody ( n = 6 tracheae per condition per experiment , n = 3 experiments ) . The staining showed that Cdc2 was expressed in L2 and L3 and upregulated post heat shock ( Figure 4B ) . We conclude that the induction of Cdc2AF is unable to induce proliferation in L2 but is able to at subsequent stages . We also co-overexpressed Cdc2AF and Cyclin B in L2 and found that this combination was also unable to induce division in L2 ( Figure 4C ) . The inability of Cdc2AF to induce mitoses in L2 was consistent with the lack of proliferation in ATRRNAi/Chk1RNAi at this stage and implicated an ATR-Chk1-independent process in the regulation of G2 arrest in L2 ( see Discussion ) . Our analysis of Tr2 DT has shown that Chk1 is phosphorylated in an ATR-dependent manner ( Figure 2 ) and that levels of pChk1 are high in L2 and early L3 and diminished at 32–40 hr L3 upon mitotic entry ( Figure 2 ) . ATR is known to be recruited to double-strand breaks in DNA that result from exposure to DNA damaging agents , collapsed/stalled replication forks during DNA synthesis or DNA recombination ( Branzei and Foiani , 2008 ) . The recruitment of ATR to DNA results in the phosphorylation of the histone variant H2AX ( gamma-H2AX/gamma-H2AV in Drosophila ) and other proteins including Chk1 . Antisera against gamma-H2AX are able to detect DNA damage in Drosophila ( Bayer et al . , 2017; Rogakou et al . , 1999 ) . Larvae subject to X-ray irradiation-induced DNA damage showed increased levels gamma-H2AX staining in larval tissues ( Bayer et al . , 2017 ) . To investigate whether the activation of ATR/Chk1 in Tr2 DT is associated with double-strand breaks in DNA in these cells , we examined the distribution of gamma-H2AX in L2 , early L3 and WL3 ( Figure 4 , n = 6–8 tracheae per condition per experiment , n = 3 experiments ) . We detected no gamma-H2AX staining in Tr2 DT in L2 and sporadic staining in WL3 ( Figure 5A–B , early L3 not shown ) . To validate the gamma-H2AX antisera used in our experiments we stained imaginal discs from larvae irradiated with X-rays at 20 Gy and 40 Gy , 1 h post irrradiation ( n = 5–6 imaginal discs per condition per experiment , n = 2 experiments ) . The frequency of gamma H2AX+ cells was higher in significantly higher in treated discs than controls and higher in discs exposed to 40 Gy than 20 Gy ( data not shown ) . Thus , we found no evidence for double-strand breaks in DNA in cells paused in G2 . We also examined gamma-H2AX levels in Btl-Chk1RNAi animals to see if the levels were elevated . Here again , we detected no gamma-H2AX in L2 and some labeling , at levels comparable to wild type , at WL3 ( Figure 5A–B ) . The sporadic gamma-H2AX staining observed in wild type and Btl-Chk1RNAi animals is intriguing and the underlying reason is currently unclear . It has been reported that the Ser-139 residue on H2AX is also phosphorylated in cells undergoing apoptosis ( Rogakou et al . , 2000 ) . Since the frequencies of gamma-H2AX+ nuclei/trachea and activated-Caspase3+ cells/trachea in control and Btl-Chk1RNAi at WL3 are comparable ( compare Figure 5B and Figure 3D ) , it is plausible that some of the gamma-H2AX+ nuclei in control and Btl-Chk1RNAi tracheae at WL3 are apoptotic cells . In an independent set of experiments , we probed the expression of ATR/Chk1 mRNA in Tr2 DT at different larval stages ( L2 , 0–8 h L3 , and 32–40 h L3 ) by qPCR ( ≥15 tracheal fragments per timepoint per experiment , n = 3 experiments ) . With respect to expression at L2 , levels of ATR expression were slightly increased at 0–8 h L3 ( Figure 5C ) . Interestingly , the levels of Chk1 mRNA were ~5 fold higher at both L2 and 0–8 h L3 in comparison to 32–40 h L3 ( Figure 5D ) . This suggests that the high levels of pChk1 in Tr2 DT in L2 and early L3 could be regulated at a transcriptional level via the regulation of Chk1 expression . The larval tracheal system consists of a network of epithelial tubes that grow in length and in circumference as the larva grows . As indicated earlier ( Figure 1 ) , Tr2 DT and the cells that comprise this segment grow in size during larval stages . Measurements of the length and width of Tr2 DT at different larval stages showed that most of the growth of this segment occurs from the time the animals enter L2 till mid-L3 ( Figure 6A–B ) . During this period , Tr2 DT grows in length by 268 ± 32% ( n = 6 tracheae ) and in width by 247 ± 19% ( n = 12 tracheae ) . We also estimated the sizes of tracheoblasts in Tr2 DT ( 2D area ) over the L2-L3 interval . For this , tracheae were stained with Phalloidin to delineate margins of cells and cellular areas were measured ( see Figure 6—figure supplement 1 ) . Between L2 and 24–32 h L3 , cells virtually doubled in size ( cellular area increased by 112 ± 37% ( n = 15 ) ) ( Figure 6C ) . Thus , analysis of the growth trajectory of Tr2 DT , and the tracheoblasts that comprise it , showed that maximal growth occurs in the period when the tracheoblasts are paused in G2 . This raised the possibility that the arrest in G2 facilitates the growth of cells and the tracheal branches they comprise . Next we investigated how abrogating ( Figure 2D ) or prolonging G2 arrest ( Figure 2F ) impacted growth of Tr2 DT . Measurements of length and width of Tr2 DT in Btl- Chk1RNAi animals at 32–40 hr L3 showed that the growth was significantly reduced in comparison to wild type ( Figure 6D , both average length and width are statistically different p<0 . 05 , see Source data file 1 ) . Conversely , the measurements of length and width of Tr2 DT in animals overexpressing Chk1 showed thatTr2 DT segments were significantly larger in this background ( Figure 6D , p<0 . 05 , both average length and width are statistically different p<0 . 05 , see Source data file 1 ) . We then compared the sizes of trachea in wild type , Btl- Chk1RNAi and Btl-Chk1 animals at 0–8 h L3 and found that they are comparable ( Figure 6—figure supplement 1 ) . Based on these findings , we conclude that ATR/Chk1-dependent G2 arrest facilitates cellular and organ growth in Tr2 . Here we investigate the mechanism for and implication of G2 arrest in progenitors of adult thoracic tracheal system in Drosophila . We show that Tr2 tracheoblasts remain arrested in G2 for ~48–56 h during larval life during which the cells and the tracheae they comprise grow in size . Our findings are that tracheoblasts paused in G2 express both the essential drivers for G2-M like Cdc2/Cdk1 , Cyclin B and Stg , and negative regulators of G2-M like ATR and Chk1 , and that the G2-M transition in these cells involves the coordination of several genetically distinguishable processes including the downregulation of Chk1 . Our analysis also reveals that arrest in G2 is necessary for growth of tracheoblasts and the tracheae they comprise . In the sections that follow we discuss the processes underlying G2 arrest , the relationship between G2 arrest and cellular growth and the broader implications of the developmental program described here . ATR/Chk1 have been implicated in the negative regulation of Cdc2/Cdk1 activity leading to a slowdown of the S/G2 phases of the cell cycle ( Branzei and Foiani , 2008; Su et al . , 1999; Blythe and Wieschaus , 2015 ) . In these contexts , ATR is recruited to stalled/collapsed replication forks during DNA replication and to double-strand DNA breaks that occur upon exposure DNA damaging agents or during recombination ( Kumagai et al . , 2004; Blythe and Wieschaus , 2015 ) . Recruitment of ATR leads in turn to the phosphorylation and activation of Chk1 and to the inhibition of Cdc2/Cdk1 activity . The inhibition of ATR/Chk1 has been shown to hinder completion of DNA replication and DNA repair and result in aberrant mitoses; ATR/Chk1 mutant animals are embryonic lethal ( Liu et al . , 2000; Sibon et al . , 1999 ) . In the context of tracheoblasts , we find that ATR/Chk1 act to arrest cells in G2 . But unlike in other contexts , we find no evidence for any DNA damage in cells in which Chk1 is active nor any increase in DNA damage upon reduction of Chk1 levels and the induction of precocious mitoses . We independently measured the frequency of apoptotic nuclei in Chk1 mutants to find that it is indistinguishable from wild type . Thus , it appears unlikely that the activation of ATR/Chk1 in Tr2 DT is the outcome of the activation of the canonical DNA damage checkpoint . The mechanisms for the activation of ATR/Chk1 in Tr2 DT merit further investigation . Analysis of ATR/Chk1 expression by qPCR suggests that levels of Chk1 mRNA are significantly higher in arrested cells . Increased expression of Chk1 mRNA could contribute toward increased levels of phosphorylated Chk1 and to G2 arrest . Activated Chk1 is thought inhibit Cdc2/Cdk1 by either inhibiting Stg and/or stabilizing Wee/Myt ( O'Connell et al . , 1997; Xiao et al . , 2003 ) . We find that the expression of WeeRNAi and Myt1RNAi does not recapitulate the Chk1RNAi phenotype ( data not shown ) . This suggests that Wee/Myt1 do not contribute toward G2 arrest in tracheoblasts and that Chk1 acts via the inhibition of Stg . We have examined if the precocious proliferation in Btl-Chk1RNAi animals is dependent on Stg . Counts of cell numbers in Tr2 DT in Btl-Chk1RNAiStgRNAi animals showed that there was no increase at 16–24 h L3 or at 32–40 h L3 ( Figure 2—figure supplement 1 ) . This is consistent with the possibility that Chk1 inhibits mitotic activity by inhibiting Stg . Tracheoblasts lacking ATR/Chk1 pause in G2 for 24 h ( L2 ) and proliferate ~24–32 h prior to the normal time for mitotic entry ( early L3 ) . We find that expression of a constitutively active form of Cdc2/Cdk1 , that is insensitive to ATR/Chk1 , is unable to induce precocious mitotic exit in cells in L2 . This shows that the mechanism for arrest in L2 is likely to be independent of ATR/Chk1 . Mitochondrial fragmentation that occurs in G2 is necessary for the segregation of these organelles into daughter cells during mitosis . Several studies have shown that the inhibition of mitochondrial fission in actively proliferating cells leads to arrest in G2 ( Westrate et al . , 2014; Lee et al . , 2014 ) . Importantly , a recent study has shown that cells in which mitochondrial fission was inhibited were also unresponsive to expression of Cdc2AF ( Lee et al . , 2014 ) . Whether the timecourse of mitochondrial fission in Tr2 DT impacts mitotic competence remains to be determined . Arrest in G2 correlates with increase in size of tracheoblasts and hypertrophic growth of the tracheal branch they comprise ( DT ) . Based on the quantitation of tracheal size ( Figure 6 ) , we estimate that the tracheal epithelium , and at least some of the tracheoblasts that comprise it , grow in volume ~13 fold in the L2-32-40 h L3 interval . The association between G2 arrest and enormous cellular growth has been reported in other developmental contexts . Histoblasts arrested in G2 grow ~60 fold in volume ( Ninov et al . , 2009 ) and Drosophila spermatocytes arrested in pre-meiotic G2 arrest grow ~25 fold in volume ( Ueishi et al . , 2009 ) . We find that abrogating G2 arrest in developing tracheoblasts via knockdown of ATR/Chk1 diminishes cellular and organ growth . Conversely , prolonging G2 arrest leads to increased growth . It is plausible that arrest in G2 facilitates cellular growth . This cellular growth may drive hypertrophic organ growth as we show here and facilitate rapid proliferation subsequently . Analysis of the mechanisms underlying hypertrophic growth in the vertebrate kidney ( Shankland and Wolf , 2000 ) and heart ( Braun-Dullaeus et al . , 1999 ) has shown that it is associated with cells in G1 arrest . The findings presented here suggest that G2-arrested cells may also be relevant to hypertrophic organ growth and pathogenesis . Interestingly , hypertrophic cellular growth in G1-arrested cells referred to above has been shown to be dependent on the juxtaposition of cell cycle activators and inhibitors . In the context of glomerular cell hypertrophy associated with diabetic nephropathy , elevated glucose levels induced quiescent cells to enter G1 while Angiotensin II and TGF-Beta signaling elevated levels of CDK inhibitors that prevented entry into S phase ( Fujita et al . , 2004 ) . The negative regulation of cell cycle progression in cells that are mitotically active may be necessary for the hypertrophic growth of G2-arrested cells as well . The following strains were obtained from repositories: UAS FUCCI , TubGAL80ts;TM2/TM6b , Tb , UAS-ATMRNAi ( Bloomington Drosophila Stock Center ) , UAS-Chk1RNAi , UAS-StgRNAi , and UAS-ATRRNAi ( Vienna Drosophila Resource Center ) , UAS-Chk1 ( In-house fly facility ) . The following strains were received as gifts: Btl-Gal4 , hs-String , hs-Cdc2 , hs-Cdc2AF , hs-CyclinB . Strains were raised on a diet of cornmeal-agar and maintained at 25°C except hs and GAL80ts strainsthat were maintained at 22°C and 18°C respectively . For experiments involving GAL80tsstrains , the animals were moved to 29°C at indicated stages for indicated time periods . For experiments involving hs strains , the animals were heat-shocked at 37°C for 30 min at indicated stages , transferred to 22°C for 3 h and then sacrificed . Larval staging was based on the morphology of the anterior spiracles as previously described ( Guha and Kornberg , 2005 ) . For timepoints in L3 , L2 larvae were collected and examined at 8 h intervals to identify animals that had undergone the L2-L3 molt ( 0–8 h L3 ) . 0–8 h L3 cohorts isolated in this manner were staged for subsequent timepoints . Larvae were fed 5-Bromo-2’-deoxyuridine ( 1 mg/ml final concentration , Sigma , in cornmeal-agar ) for 2 hr and sacrificed for analysis . Animals were dissected in PBS and fixed with 4% ( wt/vol ) Paraformaldehyde in PBS for 30 min . Immunohistochemical analysis utilized the following antisera: Rabbit anti-gamma-H2AX ( Novus Biologics , 1:300 ) , Rabbit anti-cleaved-Caspase3 ( Cell signaling technology , 1:300 ) , Rabbit anti-pH3 ( Millipore , 1:500 ) , Guinea pig anti-stg ( gift from Dr . Yukiko Yamashita , 1:500 [Inaba et al . , 2011] ) , Rabbit anti-Cdk1/Cdc2 ( PSTAIR ) ( Millipore , 1:500 ) , Mouse anti-CyclinB ( DSHB , 1:300 ) , Rabbit anti-phospho Chk1 ( Abcam , 1:200 ) , Mouse anti-BrdU ( Sigma , 1:100 ) and Alexa 488/568/647-conjugated Donkey/Goat anti-mouse/rabbit/guinea pig secondary antibodies ( Invitrogen , 1:300 ) . Tyramide signal amplification was used as per manufacturer recommendations for pChk1 detection . As part of this protocol the following reagents were used: Tyramide amplification buffer and Tyramide reagent ( Invitrogen ) , Vectastain A and B and Biotinylated goat anti Rabbit IgG ( 1:200 , Vector Labs ) . Tracheal preparations were flat-mounted in ProLong Diamond AntifadeMountant with DAPI ( Molecular Probes ) and imaged on Leica TCS SP5 or Zeiss LSM-780 laser-scanning confocal microscopes . Images were processed using Image J and Adobe Photoshop . For quantification of cell number , fixed specimens were mounted in ProLong Diamond AntifadeMountant with DAPI and the number of nuclei were counted on a Zeiss Axio Scope A1 microscope . The DT of the second thoracic metamere was identified morphologically based on the cuticular banding pattern at anterior and posterior junctions . Length of DT was measured as the distance between the cuticular bands of respective metameres . Area of a cell was measured by creating a mask over the cell boundaries as stained by Alexa 568 Phalloidin ( Molecular Probes , 1:500 ) using ImageJ . Btl-FUCCI-expressing larvae were dissected in PBS , incubated with 5 μM Draq5 ( Abcam ) for 5 min ( PBS ) , fixed in 5% paraformaldehyde ( PBS ) for 10 min , and rinsed in PBS + 0 . 3% Triton X-100 . Trachea were flat-mounted in 50% ( vol/vol ) glycerol and compressed gently to expel air from the tubes to eliminate light scattering by air-filled tracheal tubes . Nuclear regions were selected using ImageJ and intensity of DRAQ5 fluorescence was estimated . Intensities were corrected for background by subtracting an average intensity value of five selected regions devoid of nuclei from all pixels in the image . The DT of the second thoracic segment was identified morphologically as described in the previous section , micro-dissected with forceps in PBS and transferred to Trizol reagent ( Ambion ) on ice for RNA extraction as per instructions provided by the manufacturer . RNA was then precipitated using Isopropanol/4M Lithium chloride . cDNA was synthesized ( 100 ng RNA ) using Maxima First Strand cDNA Synthesis Kit ( Thermo Fisher ) . 1 μL of cDNA was then used to perform qPCR using the SYBR Green ( Maxima Probe/ROX qPCR Master Mix , Thermo Fisher ) protocol . Primer sequences for candidate genes and GAPDH ( internal control ) are provided below . Relative mRNA levels were quantified using the formula RE = 2- ∆∆Ctmethod . The following primer sets were used: GAPDH forward5' CGTTCATGCCACCACCGCTA 3'GAPDH Reverse5' CACGTCCATCACGCCACAA 3'String Forward5' CAGCATGGATTGCAATATCAGTAAT 3'String Reverse5' AGACCCATCAGCTCCGGACT 3'Chk1 Forward5' AACAACAGTAAAACGCGCTGG 3'Chk1 Reverse5' TGCATATCTTTCGGCAGCTC 3'ATR Forward5' CCAGATAGCAGCGAGTGCAT 3'ATR Reverse5' CGAGGTCCAGGGAACTTAGC 3'Cdc2 Forward5' CCATCAACCGCGATCAGAGAAAT 3'Cdc2 Reverse5' CTCTCCATGTGCTTATCAACTGGC 3'Cyclin B Forward5' TTACAGGCCATCGGAGATTGC 3'Cyclin B Reverse5' TTCGCGATCAGCCGGGTAAT 3'	Every organism begins as a single cell . That cell , and all the other cells it generates over time , need to divide at the right time and in the right place to develop into an adult . As they do so , they pass through the stages of the cell cycle . As cells prepare to divide they enter into the first growth phase , G1 , ramping up their metabolic activity . They then enter S phase , duplicate their DNA , and subsequently a second growth phase G2 . Finally , during the mitotic phase , the chromosome separate and cells undergo cytokinesis to form new cells . Dividing cells can pause at certain stages of the cell cycle to assess whether the conditions are suitable to proceed . The length of the pause depends on the stage of development and the cell type . Signals around the cell provide the cues that it needs to make the decision . The fruit fly Drosophila melanogaster , for example , undergoes metamorphosis during development , meaning it transforms from a larva into an adult . The larva contains small patches of ‘progenitor’ cells that form the adult tissue . These remain paused for various intervals during larval life and restart their cell cycle as the animal develops . A key challenge in biology is to understand how these progenitors pause and what makes them start dividing again . Here , Kizhedathu , Bagul and Guha uncover a new mechanism that pauses the cell cycle in developing animal cells . Progenitors of the respiratory system in the adult fruit fly pause at the G2 stage of the cell cycle during larval life . Some of these progenitors , from a part of the larva called the dorsal trunk , go on to form the structures of the adult respiratory system . By counting the cells and tracking their dynamics with fluorescent labels , Kizhedathu et al . revealed that the progenitor cells pause for between 48 and to 56 hours . Previous research suggested that this pause happens because the cells lack a protein essential for mitosis called Cdc25/String . However , these progenitors were producing Cdc25/String . They stopped dividing because they also made another protein , known as Checkpoint Kinase 1/Grapes ( Chk1/Grp ) . Chk1 is known to add a chemical modification to Cdc25 , which dampens its activity and stops the cell cycle from progressing . This is likely what allow the flies to co-ordinate their development and give the cells more time to grow . When Chk1 was experimentally removed , it reactivated the paused cells sooner , resulting in smaller cells and a smaller respiratory organ . This work extends our understanding of stem cell dynamics and growth during development . Previous work has shown that cells that give rise to muscles and the neural tube ( the precursor of the central nervous system ) also pause their cell cycle in G2 . Understanding more about how this happens could open new avenues for research into developmental disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2018	Negative regulation of G2-M by ATR (mei-41)/Chk1(Grapes) facilitates tracheoblast growth and tracheal hypertrophy in Drosophila
Bacterial populations vary in their stress tolerance and population structure depending upon whether growth occurs in well-mixed or structured environments . We hypothesized that evolution in biofilms would generate greater genetic diversity than well-mixed environments and lead to different pathways of antibiotic resistance . We used experimental evolution and whole genome sequencing to test how the biofilm lifestyle influenced the rate , genetic mechanisms , and pleiotropic effects of resistance to ciprofloxacin in Acinetobacter baumannii populations . Both evolutionary dynamics and the identities of mutations differed between lifestyle . Planktonic populations experienced selective sweeps of mutations including the primary topoisomerase drug targets , whereas biofilm-adapted populations acquired mutations in regulators of efflux pumps . An overall trade-off between fitness and resistance level emerged , wherein biofilm-adapted clones were less resistant than planktonic but more fit in the absence of drug . However , biofilm populations developed collateral sensitivity to cephalosporins , demonstrating the clinical relevance of lifestyle on the evolution of resistance . Antimicrobial resistance ( AMR ) is one of the main challenges facing modern medicine . The emergence and rapid dissemination of resistant bacteria is decreasing the effectiveness of antibiotics and some estimates suggest as many as 700 , 000 people die per year due to AMR-related problems ( O'Neill et al . , 2016 ) . AMR , like all phenotypes , is an evolved property , either the ancient product of living amidst other microbial producers of antimicrobials ( Martínez , 2008 ) , or the recent product of strong selection by human activities for novel resistance-generating mutations ( Ventola , 2015 ) . The dominant mode of growth for most microbes is on surfaces , and this biofilm lifestyle is central to AMR ( Høiby et al . , 2010; Olsen , 2015; Ahmed et al . , 2018 ) , especially in chronic infections ( Wolcott et al . , 2010; Wolcott , 2017 ) . However , with few exceptions ( Ridenhour et al . , 2017; Ahmed et al . , 2018; France et al . , 2019 ) , most of the research on the evolution of AMR has been conducted in well-mixed populations ( reviewed in Hughes and Andersson , 2017 ) or on agar plates ( Baym et al . , 2016a ) , conditions that cannot simulate the effects of biofilms on the evolution of AMR . Consequently , our understanding of how this lifestyle influences the evolution of AMR , whether by different population-genetic dynamics or molecular mechanisms , is limited . One example is that the close proximity of cells in biofilms may facilitate the horizontal transfer and persistence of resistance genes in bacterial populations ( Stalder and Top , 2016; Ridenhour et al . , 2017 ) . Less appreciated is the potential for the biofilm lifestyle to influence the evolution of AMR by de novo chromosomal mutations . This emergence of AMR in biofilms is important because: i ) the environmental structure of biofilms can increase clonal interference , rendering selection less effective and enhancing genetic diversity ( Habets et al . , 2006; Traverse et al . , 2013; Cooper et al . , 2014; Ellis et al . , 2015; France et al . , 2019 ) , ii ) distinct ecological conditions within the biofilm can favor functionally distinct adaptations to different niches ( Poltak and Cooper , 2011 ) , iii ) the biofilm itself can protect its residents from being exposed to external stresses like antibiotics or host immunity , and weaken selection ( Geisinger and Isberg , 2015; Eze et al . , 2018 ) , and iv ) slower growth within biofilms can reduce the efficacy of antibiotics that preferentially attack fast-growing cells ( Walters et al . , 2003; Kirby et al . , 2012 ) . The first two hypotheses would predict more complex evolutionary dynamics within biofilms than in well-mixed environments ( Steenackers et al . , 2016 ) , while the second two predict different rates of evolution , targets of selection , and likely less potent mechanisms of AMR ( Andersson and Hughes , 2014 ) . Together , these potential factors call into question the conventional wisdom of a tradeoff between fitness and antimicrobial resistance , a relationship that remains to be clearly defined . Here , we study the evolutionary dynamics and effects of new resistance mutations in the opportunistic nosocomial pathogen Acinetobacter baumannii , which is often intrinsically resistant to antibiotics or has been reported to rapidly evolve resistance to them ( Doi et al . , 2015 ) . This pathogen is categorized as one of the highest threats to patient safety ( Asif et al . , 2018 ) , partly due to its ability to live on inanimate surfaces in biofilms ( Eze et al . , 2018 ) . We experimentally propagated populations of A . baumannii exposed either to subinhibitory or increasing concentrations of ciprofloxacin ( CIP ) over 80 generations in biofilm or planktonic conditions to ascertain whether these lifestyles select for different mechanisms of AMR . Rather than focusing on the genotypes of single isolates , which can limit the scope of an analysis , we conducted whole-population genomic sequencing over time to define the dynamics of adaptation and the fitness of certain resistance alleles compared to others in the experiment . We then identified clones with specific genotypes that we linked to fitness and resistance phenotypes . This approach sheds new light on the ways that pathogens adapt to antibiotics while growing in biofilms and has implications for treatment decisions . Replicate cultures of the susceptible A . baumannii strain ATCC 17978 ( Piechaud and Second , 1951; Baumann et al . , 1968 ) were established under planktonic or biofilm conditions in one of three treatments: i ) no antibiotics , ii ) sub-inhibitory concentration of the antibiotic ciprofloxacin ( CIP ) and iii ) evolutionary rescue ( Bell and Gonzalez , 2009 ) in which CIP concentrations were increased every 72 hr from subinhibitory concentrations to four times the minimum inhibitory concentration ( MIC ) ( Figure 1A ) . Before the start of the antibiotic evolution experiment , we propagated the ATCC strain for ten days in planktonic conditions to reduce the influence of adaptation to the laboratory conditions on subsequent comparisons . CIP was chosen because of its clinical importance in treating A . baumannii ( Lopes and Amyes , 2013; Ardebili et al . , 2014; Doi et al . , 2015 ) , its ability to penetrate the biofilm matrix ( Tseng et al . , 2013 ) allowing similar efficacy in well mixed and structured populations ( Kirby et al . , 2012 ) , and because it is not known to stimulate biofilm formation in A . baumannii ( Aka and Haji , 2015 ) . Planktonic populations were serially passaged by daily 1:100 dilution while biofilm populations were propagated using a bead model simulating the biofilm life cycle ( Poltak and Cooper , 2011; Traverse et al . , 2013; Turner et al . , 2018 ) . This model selects for bacteria that attach to a 7 mm polystyrene bead , form a biofilm , and then disperse to colonize a new bead each day . ( A video tutorial for this protocol is available at http://evolvingstem . org/see-it-in-action ) . The transfer population size in biofilm and in planktonic cultures was set to be nearly equivalent at the beginning of the experiment ( approximately 1 × 107 CFU/mL ) , because population size influences mutation availability and the response to selection ( Salverda et al . , 2017; Cooper , 2018 ) . Each day , the population size increases 100-fold during regrowth , generating approximately 106 new mutations per day using a conservative but experimentally justified estimate of the mutation rate ( Lynch et al . , 2016; Dillon et al . , 2017 ) . Effects of fluoroquinolones like CIP have been shown to increase the mutation rate by an order of magnitude , and some studies suggest biofilm growth may also increase the mutation rate ( Boles and Singh , 2008; Long et al . , 2016; Pribis et al . , 2019 ) . Thus any differences in mutated genes reaching high frequency between treatments are almost certainly the product of selection and not a lack of mutation availability , though early-arising or more probable beneficial mutations could sweep and limit invasion of selectively equivalent mutations in different genes ( Khan et al . , 2011; Flynn et al . , 2013; Kryazhimskiy et al . , 2014 ) . The mutational dynamics of three lineages from each treatment were tracked by whole-population genomic sequencing ( Figure 1A ) . We also sequenced 49 single clones isolated from 22 populations at the end of the 12 day experiment to determine mutation linkage . The rate and extent of evolved resistance depends on the strength of antibiotic selection ( Andersson and Hughes , 2014; Oz et al . , 2014 ) , the distribution of fitness effects of mutations that increase resistance to the drug ( Maclean et al . , 2010 ) , and the population size of replicating bacteria ( Salverda et al . , 2017; Cooper , 2018 ) . The mode of bacterial growth can in principle alter each of these three variables and generate different dynamics and magnitudes of AMR . In the populations exposed to the increasing concentrations of CIP ( the evolutionary rescue ) , the magnitude of evolved CIP resistance differed between planktonic and biofilm populations . Planktonic populations became approximately 160x more resistant on average than the ancestral clone while the biofilm populations became only 6x more resistant ( Figure 1B and Table 1 ) . Planktonic populations also evolved resistance much more rapidly , becoming 10x more resistant after only 24 hr of growth in sub-inhibitory CIP . This level of resistance would have been sufficient for surviving the remainder of the experiment , but MICs continued to increase at each sampling ( Figure 1B ) . The evolution of resistance far beyond the selective requirement indicates that mutations conferring higher resistance also increased fitness in planktonic populations exposed to CIP . In contrast , biofilm-evolved populations evolved under the evolutionary rescue regime acquired much lower levels of resistance ( ca . 3– 7x the ancestral MIC ) and primarily in a single step between days 3 and 4 ( Figure 1B ) . In one notable exception , the MIC of biofilm population B2 increased ~50 x after 3 days of selection in subinhibitory concentrations of CIP ( Figure 1B ) , but then the resistance of this population declined to only 6x higher than the ancestral strain . This dynamic suggested that a mutant conferring high-level resistance rose to intermediate frequency but was replaced by a more fit , yet less resistant , mutant ( this possibility is evaluated below ) . Lower levels of resistance were observed in populations selected at constant subinhibitory concentrations of CIP . Biofilm populations were 4x more resistant than the ancestor and planktonic populations were 20x more resistant ( Table 1 ) . We can infer that biofilm growth does not select for the high-level resistance seen in planktonic populations , instead favoring mutants with low levels of resistance and better adapted to life in a biofilm . It is important to note that these MIC measurements were made in planktonic conditions according to the clinical standards ( CLSI , 2019 ) and that these values increased when measured in biofilm ( Table 2 ) . Our results correspond with studies of clinical isolates in which those producing more biofilm ( and likely having adapted in biofilm conditions ) were less resistant than non-biofilm-forming isolates ( Wang et al . , 2018 ) . Nevertheless , antibiotic resistance levels are context-dependent ( Borriello et al . , 2004; Hill et al . , 2005; Kirby et al . , 2012 ) , and because the biofilm environment at least partially protects cells from antibiotic exposure ( Table 2 ) , it can be argued that differences in MICs are due to the fact that evolved biofilm populations experienced lower CIP concentrations than planktonic populations . However , we selected CIP because it can penetrate the biofilm barrier ( Tseng et al . , 2013 ) , and furthermore , cells growing in the bead model must disperse from one bead to colonize the next one in a less protected state . Overall , the fact that the planktonic populations exposed to subinhibitory concentrations of CIP increased their resistance level approximately 20-fold ( Table 1 ) demonstrates that exposing bacteria to low levels of antibiotic risks selection for high levels of resistance that can make future treatment more difficult ( Wistrand-Yuen et al . , 2018 ) . In large bacterial populations ( >105 cells ) growing under strong selection , adaptive mutations conferring beneficial traits such as antibiotic resistance will dominate population dynamics ( Barrick and Lenski , 2013; Cooper , 2018 ) . Therefore , if a single mutation renders the antibiotic ineffective and provides the highest fitness gain , it would be expected to outcompete all other less fit mutations . Further , the stronger the selection for resistance , the greater the probability of genetic parallelism among replicate populations ( Bolnick et al . , 2018 ) . Under the population-genetic conditions of these experiments described above , a conservative estimate of 106 mutations occur in the first growth cycle and at least 107 mutations arise over the 12 days of selection , leading to a probability of 0 . 98 that every site in the 4Mbp A . baumannii genome experiences a mutation at least once over the course of the 12 day experiment ( see Supplementary file 1 for details of these calculations ) . Further , as stated above , fluoroquinolones like CIP or biofilm growth may increase the mutation rate so the probability that every site is mutated may be higher than estimated by this simplistic model ( Long et al . , 2016; Geisinger et al . , 2018; Pribis et al . , 2019 ) . However , these studies do not indicate that fluoroquinolones like CIP alter the mutation spectra or particular mutation targets ( Long et al . , 2016 ) , so a lifestyle-dependent difference in CIP exposure seems unlikely to alter the availability of resistance mutations under selection in these experiments . Rather , the dramatic differences in the evolved resistance levels of planktonic and biofilm populations suggested distinct genetic causes of resistance produced by different selective forces that appear incongruent with mutation availability . We also predicted greater genetic diversity in the biofilm treatments , owing to spatial structure and/or niche differentiation ( Traverse et al . , 2013 ) , than in the planktonic cultures , in which we expected selective sweeps ( Barrick et al . , 2009 ) . A signature of spatial structure alone might be different mutations in the same gene with predicted similar function coexisting over time , which is a form of clonal interference ( de Visser and Rozen , 2006 ) . A signature of niche differentiation might be the coexistence of mutations in different genes with unique functions , which is a form of adaptive radiation ( Kassen , 2009 ) . We conducted whole-population genomic sequencing of three replicates per treatment to identify all contending mutations above a detection threshold of 5% ( see Materials and methods ) . The spectrum of mutations from CIP-treated populations are consistent with expectations from strong positive selection on altered or disrupted coding sequences ( see Table 3 for day 12 results and Supplementary file 2 for dynamics across the experiment ) . High nonsynonymous to synonymous mutation ratios were observed in both lifestyles ( 8 . 5 in planktonic and 9 . 7 in biofilm ) . 43% of the total mutations in planktonic and 34% in biofilm were insertions or deletions , which is vastly enriched over typical mutation rates of ~10 SNPs/indel under neutral conditions ( Lynch et al . , 2016; Dillon et al . , 2017 ) . Roughly 30% of the mutations in CIP-treated populations of either lifestyle occurred in intergenic regions , which is statistically enriched over the approximately 13 . 5% of intergenic regions of the ancestral strain ( X-squared = 8 . 2237 , df = 2 , p-value=0 . 01638 ) . Of the intergenic mutations , 72% of the planktonic mutations and 18% of the biofilm mutations occurred in promoters , 5’ untranslated regions , sRNAs or in putative terminators ( Kröger et al . , 2018 ) indicating that , as in Pseudomonas , intergenic mutations can be adaptive by regulating the transcription of different genes , while avoiding possible pleiotropic effects of mutations in the coding regions ( Khademi et al . , 2019 ) . As expected from theory , in CIP-selected planktonic populations where selection is most efficient , one or two mutations rapidly outcompeted others and fixed ( Figure 2 ) . Selection in biofilms , however , produced fewer selective sweeps and maintained more contending mutations , especially at lower antibiotic concentrations . In one population , multiple mutations in the same locus ( adeL ) rose to high frequency and persisted , which is consistent with the effect of population structure producing clonal interference . In the other two populations , mutations in different efflux pumps ( adeL , adeS , adeN ) contended during the experiment , which could be explained by population structure or ecological diversification , if these mutations produced different phenotypes . Overall , across all treatments and timepoints , biofilm-adapted populations were significantly more diverse than the planktonic-adapted populations ( Shannon index; Kruskal Wallis , chi-squared = 7 . 723 , p=0 . 005 ) , particularly at subinhibitory concentrations of CIP ( Figure 2—figure supplement 1a ) . Notably , increasing drug concentrations eliminated the differences in diversity between treatments ( Figure 2—figure supplement 1b ) , but the greater diversity in biofilms treated with lower doses generated more diversity for selection to act upon in a changing environment . This higher standing diversity is important when considering dosing and when antibiotic exposure may be low ( e . g . in the external environment or when bound to tissues ) ( Baquero et al . , 1998; Khan et al . , 2013 ) because biofilms with more allelic diversity have a greater chance of survival to drug and immune attack ( Fux et al . , 2005 ) . In contrast with the data observed in the populations evolving under CIP pressure , drug-free control populations contained no mutations that achieved high frequency during the experiment ( Figure 2C and D ) . These results suggest that the ancestral starting clone was already well adapted both to planktonic and biofilm lifestyles , likely because we had previously propagated the A . baumannii ATCC 17978 clone under identical drug-free conditions for 10 days leading to the fixation of mutations in three genes ( Supplementary file 2 ) . The absence of mutations specific to lifestyle in the absence of antibiotics and the acquisition of mutations specific to the growth mode under antibiotic pressure highlight different evolutionary responses to combined selective pressures that were not observed with each selective pressure alone ( Harrison et al . , 2017 ) . A . baumannii clinical samples acquire resistance to CIP by two principal mechanisms: modification of the direct antibiotic targets — gyrase A or B and topoisomerase IV — or by the overexpression of efflux pumps reducing the intracellular concentrations of the antibiotic ( Doi et al . , 2015 ) . To directly associate genotypes with resistance phenotypes , we sequenced 49 clones isolated at the end of the experiment , the majority of which were selected to delineate genotypes in the evolutionary rescue populations ( Figure 2E and Figure 2—figure supplement 2 ) . Both the genetic targets and mutational dynamics of selection in planktonic and biofilm environments differed . Mutations disrupting three negative regulators of efflux pumps evolved in parallel across populations exposed to CIP , but mutations in two of these ( adeL and adeS ) were nearly exclusive to biofilm clones ( Figure 2E ) . The most common and highest frequency mutations observed in the biofilm populations were in the repressor gene adeL ( Figure 2E , Figure 2—figure supplement 2 , and Table 4 ) , which regulates AdeFGH , one of three resistance-nodulation-division ( RND ) efflux pump systems in A . baumannii ( Coyne et al . , 2010; Fernando et al . , 2013; Pournaras et al . , 2016 ) . The overexpression of the AdeFGH is predicted to enhance transport of acylated homoserine lactones , which can increase both biofilm and antibiotic resistance ( He et al . , 2015; Alav et al . , 2018 ) . In the planktonic lines , the predominant mutations were found in adeN , which is a negative regulator of AdeIJK and were mainly insertions of IS701 that disrupted the gene ( Li et al . , 2016 ) . AdeIJK contributes to resistance to biocides , hospital disinfectants , and to both intrinsic and acquired antibiotic resistance in A . baumannii ( Damier-Piolle et al . , 2008; Rosenfeld et al . , 2012 ) and may decrease biofilm formation , which could explain its prevalence in planktonic populations here ( Yoon et al . , 2015 ) . In biofilm lines , different contending adeL mutations were detected in each replicate after 24 hr then eventually fixed as CIP concentrations increased ( green lines in Figure 2B ) , sometimes along with a secondary adeL mutation . This pattern suggests that altering efflux via adeL generates adaptations to the combination of CIP and biofilm which is supported by the increase in biofilm formation by the adeL mutants ( Figure 2—figure supplement 3 ) . Further , mutants with higher resistance than necessary appear to be maladaptive in the biofilm treatment . For example , adeN ( found more often in planktonic culture ) and adeS mutations found simultaneously on day three in population B2 ( Figure 2 ) led to a spike in resistance at that timepoint ( Figure 1 ) , but these alleles were subsequently outcompeted by adeL mutants that were evidently more fit despite lower planktonic resistance . In contrast to the biofilm populations , all planktonic populations with increasing concentrations of CIP eventually underwent selective sweeps of a single high frequency mutation in gyrA ( S81L ) , the canonical ciprofloxacin-resistant mutation in DNA gyrase . These gyrA mutations evolved in genetic backgrounds containing either an adeN mutant or a pgpB mutant . pgpB is a gene that encodes a putative membrane associated lipid phosphatase and is co-regulated by adeN ( Hua et al . , 2014 ) . Other mutations associated with high levels of resistance affected parC , encoding topoisomerase IV , and regulatory regions of two putative transporters , ACX60_RS15145 and ACX60_RS1613 , the latter being co-transcribed with the multidrug efflux pump abeM ( Su et al . , 2005 ) . Few other mutations exceeded the 10% of the total population filter in the planktonic lines . The repeated , rapid fixation of only adeN and adeN-regulated alleles in the planktonic CIP-exposed populations indicate that adeN conferred higher fitness than other CIP-resistant mutations at low drug concentrations or that these mutations were more accessible than others conferring resistance , though their mutation types do not support the latter interpretation ( Supplementary file 2 ) . Subsequently , at increased concentrations of CIP , on-target mutations in gyrA were favored in each line . Together , our results demonstrate that bacterial lifestyle influences the evolutionary dynamics and targets of selection of AMR . Multiple selective pressures , particularly in the biofilm life cycle , may affect evolutionary dynamics and constrain the evolution of AMR if negative genetic correlations exist ( Harrison et al . , 2017 ) . For instance , adeN mutations decrease biofilm formation and increase resistance by altering the adeN-controlled adeIJK efflux pump ( Yoon et al . , 2015 ) , which could explain their prevalence in planktonic populations but not biofilm populations . In contrast , loss-of-function mutations in regulators of the adeFGH and adeABC RND efflux pumps were selected in CIP-treated biofilm populations and increased resistance ~4 fold , but these were not selected in planktonic populations perhaps because of this low resistance phenotype . Subsequent adaptation by planktonic populations exposed to CIP then selected mutations in the targets of the fluoroquinolone , gyrA and parC , leading to much higher levels of resistance . A longstanding hypothesis is that de novo acquired antibiotic resistance is associated with a fitness cost in the absence of antibiotics ( reviewed in Vogwill and MacLean , 2015 ) . The extent of this cost and the ability to compensate for it by secondary mutations ( compensatory evolution ) is a key attribute determining the spread and maintenance of the resistance mechanism ( Moore et al . , 2000; Zhao and Drlica , 2002; Maclean et al . , 2010; Vogwill and MacLean , 2015 ) . A negative correlation between CIP resistance and fitness of resistant genotypes in the absence of antibiotics has been previously described in Escherichia coli , suggesting a trade-off between these traits ( Marcusson et al . , 2009; Huseby et al . , 2017; Basra et al . , 2018 ) . To determine the relationship between resistance and fitness in the absence of antibiotics in our experiment , we chose 10 clones ( five each from biofilm and planktonic populations , Figure 2F and Figure 2—figure supplement 2 ) with different genotypes and putative resistance mechanisms and measured their resistance and fitness phenotypes in both planktonic and biofilm conditions ( Figure 3 ) . As expected from the population mean values ( Figure 1B ) , the biofilm clones much were less resistant in planktonic conditions than those evolved planktonically [MIC = 0 . 58 mg/L ( SEM = 0 . 13 ) vs . MIC = 8 . 53 mg/L ( SEM = 1 . 96 ) , two-tailed t-test: p<0 . 05 , t = 4 . 048 , df = 80] . However , biofilm-evolved clones were more fit relative to the ancestral strain than the planktonic-evolved clones in the absence of antibiotic ( two-tailed t-test: p=0 . 008 , t = 2 . 984 df=18 ) ( Figure 3 ) . Importantly , these fitness measurements were made in both planktonic and biofilm conditions , demonstrating that even in the conditions they evolved in , and even following the preadaptation phase conducted in planktonic cultures , planktonic-selected clones were less fit as a result of fitness trade-offs of antibiotic resistance . However , one planktonic-evolved clone with mutations in both gyrA and parC exhibited no significant fitness cost and high levels of resistance . This suggests that , as in Pseudomonas aeruginosa , the parC mutation may compensate for the cost imposed by the gyrA mutation ( Kugelberg et al . , 2005 ) , an example of sign epistasis ( Sackman and Rokyta , 2018 ) . Overall , mutants selected in biofilm-evolved populations were less resistant than mutants selected in planktonic populations ( Figure 1B ) but produced more biofilm ( Figure 2—figure supplement 3 ) and paid little or no fitness cost in the absence of antibiotics ( Figure 3 ) . This cost-free resistance implies that these subpopulations could persist in the absence of drug , limiting the treatment options and demanding new approaches to treat high fitness , resistant pathogens ( Baym et al . , 2016b ) . When a bacterium acquires resistance to one antibiotic , the mechanism of resistance can also confer resistance to other antibiotics ( cross-resistance ) or increase the susceptibility to other antibiotics ( collateral sensitivity ) ( Pál et al . , 2015 ) . We tested the MIC of the evolved populations to 23 different antibiotics in planktonic conditions and reported quantitative changes in susceptibility by two-fold dilution , but not necessarily clinical breakpoints in resistance . Changes in susceptibilities were observed in 13 antibiotics that depended upon the growth mode of prior selection ( Figure 4 ) . For example , planktonic-evolved populations exhibited cross resistance to cefpodoxime and ceftazidime , but biofilm-evolved populations evolved collateral sensitivity to these cephalosporins . Cross-resistance was associated genetically with adeN , adeS , gyrA or pgpB mutations , and collateral sensitivity was associated with adeL mutations . Selection in these environments evidently favors the activation of different efflux pumps or modified targets that have different pleiotropic consequences for multidrug resistance ( Podnecky et al . , 2018 ) . The mechanisms leading to collateral sensitivity are still poorly understood but they depend on the genetic background of the strain , the nature of the resistance mechanisms ( Barbosa et al . , 2017; Yen and Papin , 2017 ) , and the specific physiological context of the cells ( Leus et al . , 2018 ) . In A . baumannii , each RND efflux pump is suggested to be specific for certain classes of antibiotics ( Table 4 ) ( Coyne et al . , 2011; Li et al . , 2016; Leus et al . , 2018 ) . Similar to our results ( Figure 4 ) , Yoon and collaborators demonstrated that efflux pumps AdeABC and AdeIJK , regulated by adeS and adeN respectively , increased the resistance level to some β-lactams when overexpressed ( Yoon et al . , 2015 ) . However , production of AdeFGH , the efflux pump regulated by adeL , decreased resistance to some β-lactams and other families of antibiotics or detergents by an unknown mechanism ( Yoon et al . , 2015; Leus et al . , 2018 ) . Increased sensitivity to β-lactams with efflux overexpression has also been reported in P . aeruginosa ( Azimi and Rastegar Lari , 2017 ) , which demonstrates the urgency of understanding the physiological basis of collateral sensitivity to control AMR evolution . Exploiting collateral sensitivity has been proposed to counteract the evolution of resistant populations both in bacteria ( Imamovic and Sommer , 2013; Kim et al . , 2014; Nichol et al . , 2019 ) and in cancer ( Dhawan et al . , 2017 ) . Remarkably , our results show that biofilm growth , commonly thought to broaden resistance , may actually generate collateral sensitivity during treatment with CIP and potentially other fluoroquinolones . We used experimental evolution of the opportunistic pathogen A . baumannii in both well-mixed and biofilm conditions to examine how lifestyle influences the dynamics , diversity , identity of genetic mechanisms , and direct and pleiotropic effects of resistance to a common antibiotic ( Figure 5 ) . Experimental evolution is a powerful method of screening naturally arising genetic variation for mutants that are the best fit in a defined condition ( Elena and Lenski , 2003; Cooper , 2018; Van den Bergh et al . , 2018 ) . When population sizes are large and reproductive rates are rapid , as they were here , the probability that all possible single-step mutations that can increase both resistance and fitness occurred in each population is very likely . The few mutations selected here as well as their repeated order with increasing CIP concentrations may indicate that these are the most fit mutations in this A . baumannii strain and set of environmental conditions . The prevalence of some of these mutations in clinical samples suggests that they too may have been exposed to selection in similar conditions . For instance , S81L in gyrA and S80L in parC have been reported worldwide as the primary mechanism conferring high levels of resistance to fluroquinolones in clinical isolates ( Adams-Haduch et al . , 2008; Warner et al . , 2016; Dahdouh et al . , 2017 ) , and mutations in RND efflux pumps have been associated with multidrug resistant phenotypes in clinical samples isolated worldwide ( Damier-Piolle et al . , 2008; Coyne et al . , 2010; Rosenfeld et al . , 2012; Fernando et al . , 2013; Pournaras et al . , 2016; Leus et al . , 2018 ) . Likewise , the absence of other mutations reported in shotgun mutant screens of resistance in A . baumannii ( Geisinger et al . , 2018 ) means that these mutants produced less resistance , lower fitness , or both . Evolution experiments hold promise for ultimately forecasting mutations selected by different antimicrobials and anticipating treatment outcomes , including the diversification of the pathogen population and the likelihood of collateral sensitivity or cross-resistance ( Brockhurst et al . , 2019 ) . Furthermore , knowledge of the prevailing lifestyle of the pathogen population may be critically important for treatment design . As predicted by our experiment , biofilm-forming clinical isolates were more susceptible to CIP and other antibiotics than non-biofilm forming clinical isolates even when the resistance levels were measured in biofilms ( Rodríguez-Baño et al . , 2008; Qi et al . , 2016; Wang et al . , 2018 ) . Most infections are likely caused by surface-attached populations ( Wolcott et al . , 2010; Wolcott , 2017 ) , and some treatments include cycling antibiotics that promote biofilm as a primary response . For example , tobramycin is used for treating P . aeruginosa in cystic fibrosis patients ( Hamed and Debonnett , 2017 ) and promotes biofilm formation ( Hoffman et al . , 2005; Linares et al . , 2006 ) , wherein the evolution of antibiotic resistance without a detectable fitness cost may arise during treatment . As in our experiment , the overexpression of RND efflux pumps in both P . aeruginosa and Neisseria gonorrhoeae may produce little or no fitness cost ( Warner et al . , 2007; Olivares Pacheco et al . , 2017 ) . But the more diverse biofilm-adapted lineages in our experiments revealed a striking vulnerability to cephalosporins , which could provide a new strategy for treatment . Broader still , conventional wisdom has long held that the relationship between resistance and fitness is antagonistic , and that the efficacy of many antimicrobials is aided by a severe fitness cost of resistance ( Vogwill and MacLean , 2015; Baym et al . , 2016b; Hughes and Andersson , 2017 ) . This study demonstrates that the form of the relationship between fitness and resistance can be altered by the mode of growth , whereby biofilms can align resistance and fitness traits . Continued efforts to determine how the fitness landscape of various resistance pathways depends on the environment and its structure , including growth mode , could produce a valuable forecasting tool to stem the rising AMR tide . Before the start of the antibiotic evolution experiment , we propagated well mixed tubes founded by one clone of the susceptible A . baumannii strain ATCC 17978-mff ( Piechaud and Second , 1951; Baumann et al . , 1968 ) in a modified M9 medium ( referred to as M9+ ) containing 0 . 1 mM CaCl2 , 1 mM MgSO4 , 42 . 2 mM Na2HPO4 , 22 mM KH2PO4 , 21 . 7 mM NaCl , 18 . 7 mM NH4Cl and 11 . 1 mM glucose and supplemented with 20 mL/L MEM essential amino acids ( Gibco 11130051 ) , 10 mL/L MEM nonessential amino acids ( Gibco 11140050 ) , and 1 mL each of trace mineral solutions A , B , and C ( Corning 25021–3 Cl ) . This preadaptation phase was conducted in the absence of antibiotics for 10 days ( ca . 66 generations ) with a dilution factor of 100 per day . After the ten days of preadaptation to M9+ medium , we selected a single clone and propagated for 24 hr in M9+ in the absence of antibiotic . We then subcultured this population into twenty replicate populations . Ten of the populations ( 5 planktonic and five biofilm ) were propagated every 24 hr in constant subinhibitory concentrations of CIP , 0 . 0625 mg/L , which corresponds to 0 . 5x the minimum inhibitory concentration ( MIC ) . After 72 hr under subinhibitory concentrations of CIP , the populations were exposed to two different antibiotic regimes for nine more days , either constant subinhibitory concentrations of CIP or increasing concentrations of CIP ( called the evolutionary rescue ) . For the latter , we doubled the CIP concentrations every 72 hr until 4x MIC . As a control , the 10 remaining populations were propagated in the absence of CIP ( Figure 1 ) . We propagated the populations into fresh media every 24 hr as described by Turner et al . ( 2018 ) . For planktonic populations , we transferred a 1:100 ( 50 µl into 5 mL of M9+ ) dilution , which corresponded to 6 . 64 generations per day . For biofilm populations , we transferred a polystyrene bead ( Polysciences , Inc , Warrington , PA ) to fresh media containing three sterile beads . We rinsed each bead in PBS before the transfer , therefore reducing the transfer of planktonic cells . Each day we alternated between black and white marked beads , ensuring that the bacteria were growing on the bead for 24 hr , which corresponds to approximately 6 to 7 . 5 generations/day ( Traverse et al . , 2013; Turner et al . , 2018 ) . For the experiment with increasing concentrations of antibiotics , we froze a sample of each bacterial population on days 1 , 3 , 4 , 6 , 7 , 9 , 10 and 12 . In the experiment with constant exposure to subinhibitory concentrations of antibiotics , we froze the populations on days 1 , 3 , 4 , 9 , and 12 . We froze the control populations at days 1 , 4 , 9 , and 12 . For planktonic populations , we froze 1 mL of culture with 9% of DMSO . For freezing the biofilm populations , we sonicated the beads in 1 mL of PBS with a probe sonicator and subsequently froze with 9% DMSO . We determined the MIC of CIP by broth microdilution in M9+ according to the Clinical and Laboratory Standards Institute guidelines ( CLSI , 2019 ) , in which each bacterial sample was tested to 2-fold-increasing concentration of CIP from 0 . 0625 to 64 mg/L . To obtain a general picture of the resistance profiles we determined the MIC to 23 antibiotics ( amikacin , ampicillin , ampicillin/sulbactam , aztreonam , cefazolin , cefepime , cephalothin , meropenem , ertapenem , cefuroxime , gentamicin , CIP , piperacillin/tazobactam , cefoxitin , trimethoprim/sulfamethoxazole , cefpodoxime , ceftazidime , tobramycin , tigecycline , ticarcillin/clavulanic acid , ceftriaxone and tetracycline ) by broth microdilution in commercial microtiter plates following the instructions provided by the manufacturers ( Sensititre GN3F , Trek Diagnostics Inc , Westlake , OH ) . We tested the MIC at days 1 , 3 , 4 , 6 , 7 , 9 , 10 and 12 for the populations propagated under increasing concentrations of antibiotic , and at days 1 and 12 for the subinhibitory and non-antibiotic treatments . For the CIP-MICs , we used Pseudomonas aeruginosa PAO1 in Mueller Hinton broth as a control . No differences in the MICs were found between Mueller Hinton and M9+ or if measuring the MIC in 96 well-plate or in 5 mL tubes , which are the experimental conditions . Each MIC was performed in triplicate . The CIP was provided by Alfa Aesar ( Alfa Aesar , Wardhill , MA ) . We also determined the MIC of CIP in biofilm conditions adapting the method described by Diez-Aguilar to the bead model ( Díez-Aguilar et al . , 2018 ) . We resuspended each clone into fresh M9+ containing sterile beads ( as in the experimental evolution conditions , each tube used contained three sterile beads and 5 mL of M9+ ) . After 24 hr growing at 37° , each bead was propagated into new fresh M9+ containing different CIP concentrations ( from 4 to 128 mg/L in 2-fold-increasing manner ) . After 24 growing at 37° , we rinsed each bead in PBS and sonicate them individually as explained before . 10 μl of the sonicated liquid were transferred to 100 μL of M9+ . The MIC was calculated after measuring the optical density at 650 nm before and after 24 hr incubation . The inhibition of growth was defined as the lowest antibiotic that resulted in an OD difference at or below 0 . 05 after 6 hr of incubation . We estimated the biofilm formation of the selected clones using a modification of the previously described protocol ( O'Toole and Kolter , 1998 ) . We resurrected each clone in 5 mL of M9+ containing 0 . 5 mg/L of CIP and grew them for 24 hr . For each strain , we transferred 50 µl into 15 mL of M9+ . We tested 200 µl of the previous dilution of each clone to four different subinhibitory CIP concentrations ( 0 mg/L , 0 . 01 mg/L , 0 . 03 mg/L and 0 . 0625 mg/L ) . After 24 hr of growing at 37°C , we measured population sizes by optical density ( OD ) at 590 nm ( ODPopulations ) . Then , we added 250 µl of 0 . 1% crystal violet and incubated at room temperature for 15 min . After washing the wells and drying for 24 hr , we added 250 µl 95% EtOH solution ( 95% EtOH , 4 . 95% dH2O , 0 . 05% Triton X-100 ) to each well and incubated for 15 min and biofilm formation was measured by the OD at 590 nm ( ODBiofilm ) . Biofilm formation was corrected by population sizes ( ODBiofilm/ODPopulation ) . Results are the average of three experiments ( Figure 2—figure supplement 3 ) . We selected 5 biofilm and five planktonic clones from the end of the evolutionary rescue experiment with known genotype ( Figure 2—figure supplement 2 ) and measured their fitness by directly competing the ancestral strain and the evolved clone variants both in planktonic and in biofilm conditions in the absence of antibiotic ( Figure 3 ) ( Turner et al . , 2018 ) . We revived each clone from a freezer stock in M9+ for 24 hr . We maintained the same evolutionary conditions to revive the clones , adding three beads and/or CIP to the broth when required . After 24 hr , we added equal volume of the clones and the ancestors in M9+ in the absence of antibiotics . For planktonic populations , we mixed 25 µl of each competitor in 5 mL of M9+ . For biofilm competitions , we sonicated one bead per competitor in 1 mL of PBS and mixed in 5 mL of M9+ containing three beads . The mix was cultured at 37°C for 24 hr . We plated at time zero and after 24 hr . For each competition , we plated aliquots onto nonselective tryptic soy agar and tryptic soy agar containing CIP . Selection rate ( r ) was calculated as the difference of the Malthusian parameters for the two competitors: r = ( ln ( CIP resistantd=1/CIP resistantd=0 ) ) / ( ln ( CIP susceptibled=1/CIP susceptibled=0 ) ) /day ( Lenski , 1991 ) . Susceptible populations were calculated as the difference between the total population ( number of colonies/mL growing on the nonselective plates ) and the resistant fraction ( number of colonies/mL growing on the plates containing CIP ) . As a control for calculating the correct ratio of susceptible vs . resistant populations , we replica-plated 50 to 100 colonies from the nonselective plates onto plates containing CIP as previously described ( Santos-Lopez et al . , 2017 ) . Results are the average of three to five independent experiments . We sequenced whole populations of three evolving replicates per treatment . We sequenced the populations at days 1 , 3 , 4 , 6 , 7 , 9 , 10 , and 12 of the populations under increasing concentrations of CIP ( populations P1 , P2 , P3 and B1 , B2 , B3 for planktonic and biofilm populations ) and at days 1 , 4 , 9 , and 12 of the populations under subinhibitory concentration and no antibiotic treatments . In addition , we selected 49 clones for sequencing at the end of the experiment ( Figure 2F ) . 12 of the clones were recovered from the populations propagated in the absence of the antibiotic , 12 clones from the subinhibitory concentrations of CIP treatment and 25 were isolated from the increasing concentrations of antibiotic . We revived each population or clone from a freezer stock in the growth conditions under which they were isolated ( i . e . the same CIP concentration which they were exposed to during the experiment ) and grew for 24 hr . DNA was extracted using the Qiagen DNAeasy Blood and Tissue kit ( Qiagen , Hiden , Germany ) . The sequencing library was prepared as described by Turner and colleagues ( Turner et al . , 2018 ) according to the protocol of Baym et al . ( 2015 ) , using the Illumina Nextera kit ( Illumina Inc , San Diego , CA ) and sequenced using an Illumina NextSeq500 at the Microbial Genome Sequencing center ( http://migs . pitt . edu ) . All sequences were first quality filtered and trimmed with the Trimmomatic software v0 . 36 ( Bolger et al . , 2014 ) using the criteria: LEADING:20 TRAILING:20 SLIDINGWINDOW:4:20 MINLEN:70 . Variants were called with the breseq software v0 . 31 . 0 ( Deatherage and Barrick , 2014 ) using the default parameters and the -p flag when required for identifying polymorphisms in populations . This option calls a mutation if it is observed in two reads from each strand and reaches 5% in the population . The average depth of sequencing for populations was 219 ± 51 x and average genome coverage was 98 . 7 ± 0 . 128% . The reference genome used for variant calling was downloaded from the NCBI RefSeq database using the 17-Mar-2017 version of A . baumannii ATCC 17978-mff complete genome ( GCF_001077675 . 1 ) . In addition to the chromosome NZ_CP012004 and plasmid NZ_CP012005 sequences , we added two additional plasmid sequences to the reference genome that are known to be present in our working strain of A . baumannii ATCC 17978-mff: NC009083 , NC_009084 . Mutations were then manually curated and filtered to remove false positives . Mutations were filtered if the gene was found to contain a mutation when the ancestor sequence was compared to the reference genome or if a mutation never reached a cumulative frequency of 10% across all replicate populations . Diversity measurements were made in R using the Shannon index considering the presence , absence , and frequency of alleles . Significant differences between biofilm and planktonic populations were determined by the Kruskal Wallis test . Filtering , mutational dynamics , and plotting were done in R v3 . 4 . 4 ( www . r-project . org ) with the packages ggplot2 v2 . 2 . 1 ( Wickham , 2016 ) , dplyr v0 . 7 . 4 ( Wickham et al . , 2018 ) , vegan v2 . 5–1 ( Oksanen et al . , 2018 ) , and reshape2 ( Wickham , 2007 ) . R code for filtering and data processing can be found here: https://github . com/sirmicrobe/U01_allele_freq_code ( Santos-Lopez , 2019; copy archived at https://github . com/elifesciences-publications/U01_allele_freq_code ) . All sequences were deposited into NCBI under the Biosample accession numbers SAMN09783599-SAMN09783682 .	A bacterium known as Acinetobacter baumannii causes serious lung infections in people with weakened immune systems . These illnesses are becoming more common largely because A . baumannii is increasingly developing resistance to antibiotics . Inside the airways , individual A . baumannii cells can stick together and coat themselves in a slimy substance to form a structure called biofilm , which physically protects bacteria from antibiotics . This may be one of the reasons why it is often harder to treat bacterial infections associated with biofilms . Another possibility is that bacteria may evolve differently in biofilms compared with cells living independently . For example , A . baumannii may colonize several regions of the lungs during an infection , leading to distinct groups of bacteria that experience different conditions and evolve separately . Each population may therefore respond differently to an antibiotic . In contrast , bacteria living independently in a well-mixed population – such as in the bloodstream of their host – would be more likely to all evolve in the same way . Santos-Lopez , Marshall et al . tested this theory by exposing populations of A . baumannii that lived either independently or in biofilms to increasing levels of an antibiotic called ciprofloxacin . The genetic information of these cells was examined as the populations were evolving , and the bacteria were also put in contact with other types of antibiotics . The analyses revealed that bacteria in well-mixed populations shared the same limited number of mutations: these gave the bacteria high levels of resistance to the antibiotic’s primary target , an enzyme involved in DNA processes . The bacteria had also become resistant to other classes of antibiotics . In contrast , the bacteria in biofilm populations evolved to be more genetically diverse , exhibiting different types of mutations that helped the cells to pump out the drug . These bacteria were less resistant to ciprofloxacin and more sensitive to other types of antibiotics . Further experiments looked into the fitness of the bacteria – their ability to survive , reproduce and compete with each other . High levels of antibiotic resistance came with lower fitness: biofilm bacteria had evolved to become being fitter than those from well-mixed population . Even in the absence of drugs , these populations were in fact fitter than the original cells . Overall , understanding how the lifestyles of bacteria affect the way they respond to drugs may help researchers to develop new approaches that limit the spread of antibiotic resistance and improve treatment .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "microbiology", "and", "infectious", "disease" ]	2019	Evolutionary pathways to antibiotic resistance are dependent upon environmental structure and bacterial lifestyle
Hitherto , membralin has been a protein of unknown function . Here , we show that membralin mutant mice manifest a severe and early-onset motor neuron disease in an autosomal recessive manner , dying by postnatal day 5–6 . Selective death of lower motor neurons , including those innervating the limbs , intercostal muscles , and diaphragm , is predominantly responsible for this fatal phenotype . Neural expression of a membralin transgene completely rescues membralin mutant mice . Mechanistically , we show that membralin interacts with Erlin2 , an endoplasmic reticulum ( ER ) membrane protein that is located in lipid rafts and known to be important in ER-associated protein degradation ( ERAD ) . Accordingly , the degradation rate of ERAD substrates is attenuated in cells lacking membralin . Membralin mutations or deficiency in mouse models induces ER stress , rendering neurons more vulnerable to cell death . Our study reveals a critical role of membralin in motor neuron survival and suggests a novel mechanism for early-onset motor neuron disease . The endoplasmic reticulum ( ER ) is a membrane-enclosed cellular organelle that plays an essential role in the folding of membrane-bound and secreted proteins , synthesis of lipids and sterols , and storage of free Ca2+ . ER stress is often triggered by the accumulation of unfolded proteins due to pathological conditions , such as DNA sequence mutations , transcriptional and translational errors , or protein folding failure ( Kim et al . , 2008; Lin et al . , 2008 ) . Mammalian cells have evolved an intricate system with multiple signaling pathways to respond to ER stress , collectively termed the unfolded protein response ( UPR ) ( Kim et al . , 2008; Lin et al . , 2008; Walter and Ron , 2011 ) . If unchecked , ER stress eventually causes cell death , including neuronal death in neurodegenerative diseases ( Lindholm et al . , 2006; Scheper and Hoozemans , 2009; Hetz and Mollereau , 2014 ) . Increased ER stress is thought to play an early role in motor neuron diseases , including amyotrophic lateral sclerosis ( ALS ) and Charcot-Marie-Tooth ( CMT ) ( Atkin et al . , 2006; Nagata et al . , 2007; Nishitoh et al . , 2008; Kanekura et al . , 2009; Saxena et al . , 2009 ) . Manifestations and consequences of ER stress have been studied in the pathogenesis of SOD1 mutant mice , a commonly used model of ALS that expresses mutant human SOD1 protein as found in some inherited forms of ALS ( Atkin et al . , 2006; Kanekura et al . , 2009; Saxena et al . , 2009 ) . Many ER stress-related molecules are upregulated in SOD1 mice at an early stage of the disease , and some are specific to motor neurons ( Tobisawa et al . , 2003; Atkin et al . , 2006; Kikuchi et al . , 2006; Nagata et al . , 2007; Ito et al . , 2009; Saxena et al . , 2009 ) . In addition , reduction in the ER co-chaperone SIL1 has been found to be associated specifically with ER stress-prone , fast-fatigable motor neurons ( Filezac de L'Etang et al . , 2015 ) . ER upregulation of the UPR and aberrant modification of protein disulfide isomerase ( PDI ) also occur in human tissues from sporadic ALS ( Atkin et al . , 2008; Walker et al . , 2010 ) . Our studies on the activation of the type I interferon signaling in astrocytes at pre-symptomatic stages of SOD1 mice also suggest that ER stress in motor neurons may play a crucial role in disease onset ( Wang et al . , 2011 ) . Moreover , genetic interruption of UPR signaling molecules , such as XBP-1 or apoptosis signal-regulating kinase1 ( ASK1 ) , protects motor neurons in SOD1 mutant mice ( Nishitoh et al . , 2008; Hetz and Mollereau , 2014 ) . It has been suggested that the unusually heavy metabolic demand of motor neurons may make them particularly susceptible to ER stress ( Vinay et al . , 2000; Carrascal et al . , 2005; Li et al . , 2005 ) . Importantly , ER stress inhibitors , such as salubrinal , guanabenz , and sphin1 , delay disease onset and prolong the survival of these mutant mice ( Saxena et al . , 2009; Jiang et al . , 2014; Wang et al . , 2014; Das et al . , 2015 ) . This finding suggests that ER stress-related molecules may represent rational drug targets for motor neuron diseases . Previously , membralin had been predicted by genome sequencing to encompass an open reading frame ( orf61 in mouse and c19orf6 in human ) , but the function of the encoded protein has , heretofore , remained unknown . Orf61 is highly conserved and found in the genome of most species except yeast and bacteria . cDNAs for membralin have been cloned from human and mouse tissues ( Andersson and von Euler , 2002; Chen et al . , 2005 ) , and the protein encoded by orf61 was named membralin because it was predicted to be a membrane protein ( Andersson and von Euler , 2002 ) . Given the absence of known protein domains , membralin may represent a novel class of proteins . Here , we discovered during our cloning and characterization of the glutamate receptor subunit 3B gene ( GluN3B , formerly designated NR3B , Nishi et al . , 2001; Chatterton et al . , 2002; Matsuda et al . , 2002 ) that the 3′ end of the GluN3B gene overlaps with the 3′ end of the membralin gene on the opposite strand . Thus , our GluN3B knockout ( KO ) mice also carry a C-terminal truncation of membralin . We observed that GluN3B/membralin C-terminal double-knockout ( DKO ) mice die at postnatal day 5–6 of paresis due to severe motor neuron degeneration . Transgenic expression of membralin , but not GluN3B , rescued the phenotypes of DKO mice . Additionally , we generated membralin-specific KO mice and found that they phenocopied the DKO mice . These data suggest that membralin plays a critical role in motor neuron survival . We also demonstrate that membralin interacts with Erlin2 , a protein that is enriched in ER lipid rafts and important for ER-associated protein degradation ( ERAD ) ( Ikegawa et al . , 1999; Browman et al . , 2006 ) . Additionally , neurons from membralin mutant mice display increased basal ER stress and increased vulnerability to ER stress-inducing agents . Our discovery that membralin mutations result in motor neuron disease provides mechanistic insight into pathophysiology and may offer a novel target for therapy . We generated GluN3B-deficienct mice by deletion of the 8-kb region encoding the entire GluN3B gene and found that it also resulted in partial deletion of membralin ( Figure 1—figure supplement 1 ) . Sequence analysis confirmed that the translation frame of the altered membralin transcript in null mice terminates 147 nucleotides ( nts ) beyond the end of partial exon XI , resulting in a slightly truncated membralin protein with 49 unrelated amino-acid ( aa ) residues replacing the wild-type ( WT ) 89 C-terminal aa residues . Thus , our targeting strategy generated a GluN3B/membralin C-terminal DKO mouse line . DKO mice appeared normal at birth but unexpectedly died at postnatal day 5–6 ( P5-6 ) . The motor function of these mice was indistinguishable from WT or heterozygous littermates during the first three postnatal days , but motor strength was severely impaired , thereafter , preceding death . DKO mice were also significantly smaller than WT and heterozygous littermates ( Figure 1A ) but displayed normal gross anatomy and histology in the brain ( Figure 1—figure supplement 2 ) , visceral organs , and skeletal muscle . 10 . 7554/eLife . 06500 . 003Figure 1 . Motor neuron death in GluN3B/Membralin DKO mice . ( A ) DKO and WT littermate mice are shown at P5 . The DKO mouse was unable to maintain correct posture . ( B ) Immunostaining with anti-Hb9 antibody revealed a dramatic decrease in the number of motor neurons in the lumbar spinal cord of DKO mice relative to WT littermate mice at P5 but not at P0 . ( C ) The number of motor neurons in the lumbar spinal cord of KO and WT mice was not significantly different at P0 and P3 . However , severe motor neuron loss was observed by P5 in DKO mice ( n = 3 ) compared with WT mice ( n = 3 , p < 0 . 05 by Student's t-test ) . ( D ) Glial fibrillary-associated protein ( GFAP ) and ionized calcium-binding adapter molecule 1 ( IBA-1 ) immunostaining in the spinal cord of WT and DKO mice . GFAP-positive astrocytes and IBA-1-positive microglial cells surrounding motor neurons ( counterstained with Giemsa ) of DKO mice showed swelling cell bodies and processes , a typical sign of astrogliosis . ( E , F ) Electron microscopic analysis of spinal roots showed that DKO mice at age P5 had severe damage of motor neuron axons in the ventral root ( E ) but not of sensory neuron axons in the dorsal root ( F ) . Bar , 2 µm . ( G , H ) Damaged axons in the ventral root showed either vacuoles or loss of axoplasm ( G ) , as an early sign of axonal damage . At a later stage of axonal damage , dark disintegrated myelin sheaths and amorphous lipid were present ( H ) . Bar , 1 µm . ( I ) Percentage of injured axons in the ventral and dorsal roots of DKO and WT mice ( p < 0 . 05 by Student's t-test , n = 3 ) . Data are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 00310 . 7554/eLife . 06500 . 004Figure 1—figure supplement 1 . Generation of GluN3B/membralin DKO mice . ( A ) GluN3B and membralin genes arranged in tail-to-tail fashion with overlapping 3′ coding regions of both genes . Blue and brown boxes indicate predicted exons of GluN3B and membralin genes , respectively . The entire GluN3B coding region and part of the 3′ coding region of membralin between the 5′- and 3′-arm ( green arrows ) in the WT allele were replaced by a Neo/PGK cassette in the DKO allele by homologous recombination . Spe I restriction sites , Southern blot and Northern blot probes , and primers used for PCR genotyping are indicated . ( B ) Genotyping by Southern blot analysis of SpeI-digested genomic DNA and by PCR . Mouse genomic DNA from 12 littermates from a single breeding pair of heterozygotes was subjected to Southern blot ( top panel ) and PCR analysis ( bottom panel ) . Hybridizing bands or PCR fragments corresponding to WT and DKO allele are indicated . Both methods identified 4 WT ( #6 , 7 , 8 , and 10 ) , 6 heterozygote ( #1 , 2 , 3 , 5 , 9 , and 12 ) , and 2 homozygote ( #4 and 11 ) mice . ( C ) Northern blot analysis of total RNA from null ( KO ) and WT mice using probes 1 and 2 , derived from the last exon ( XI ) or exon VII-X of membralin , respectively ( see A ) . Probe 1 generated a hybridization band ( ∼2 . 8 kb ) in WT but not in DKO mice . Probe 2 generated a ∼2 . 8 kb band in WT and a ∼3 . 4 . kb band in DKO mice . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 00410 . 7554/eLife . 06500 . 005Figure 1—figure supplement 2 . GluN3B/membralin DKO and WT mice show similar gross anatomy of the brain . Left: whole brains dissected from DKO and WT mice at P5 . Right: Nissl-stained sagittal brain sections from DKO and WT mice at P5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 00510 . 7554/eLife . 06500 . 006Figure 1—figure supplement 3 . Expression of the motor neuron injury marker ISG15 in the spinal cord and brainstem of membralin KO mice . ( A ) ISG15 immunosignal was barely detectable in spinal cord of P3 WT littermate mice but decorated the ventral grey matter ( arrows ) of cervical and lumbar ( but not thoracic ) spinal cord of membralin KO mice . By P5 , stronger ISG15 immunostaining appeared in the ventral grey matter ( arrows ) at all levels of the spinal cord in KO mice . ( B ) Similarly , prominent ISG15 staining was observed at P5 in the facial nucleus ( dotted circle ) of KO but not WT littermate mice . Large motor neurons are indicated by Nissl staining of contiguous sections . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 00610 . 7554/eLife . 06500 . 007Figure 1—figure supplement 4 . Degeneration of motor neuron fibers in phrenic nerve of GluN3B/membralin DKO mice . ( A ) Primary afferent fibers , stained with anti-calcitonin gene-related peptide ( CGRP , red ) , remained intact in DKOK compared to WT mice . Axons counterstained with anti-parvalbumin ( Parv , green ) . ( B ) Axonal degeneration of phrenic motor neurons at P5 in DKO mice , as evidenced by paucity of fibers compared to WT . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 007 As DKO mice showed signs of paresis , we conducted histological examination of motor neurons in the lumbar spinal cord using the motor neuron-specific marker , Hb9 . The number of motor neurons in DKO mice was unchanged at P0 and P3 , but reduced by 40% at P5 , compared to littermate WT controls ( Figure 1B , C ) . The remaining motor neurons appeared smaller than those of WT mice . Motor neuron loss and surrounding astrogliosis were also apparent in the cervical spinal cords of P5 DKO mice ( Figure 1D ) . We next examined Isg15 expression ( interferon-induced 17 kDa protein ) , which is thought to play a role in innate immunity and is upregulated at both pre-symptomatic stages and post-symptomatic stages in a mouse model of ALS ( Wang et al . , 2011 ) . WT mice manifested minimal Isg15 expression throughout the brain and spinal cord . In contrast , by P3 , DKO mice had increased Isg15 expression in the ventral horn of the cervical and lumbar enlargement of the spinal cord ( Figure 1—figure supplement 3A ) . By P5 , DKO mice showed increased Isg15 expression in additional areas containing primary motor neurons , with highest expression in the ventral horn of all segments of the spinal cord and moderate expression in the facial nucleus of the brainstem ( Figure 1—figure supplement 3A , B ) . No Isg15 expression was detectable in other brain areas , even in late-stage DKO mice . These results suggest that the motor neurons are progressively injured in DKO mice , starting with those in the spinal cord . Since the normal process of programmed cell death in embryonic motor neurons is complete at birth ( Lance-Jones , 1982 ) , the loss of motor neurons in DKO mice was most likely due to cellular events occurring perinatally . To confirm specific loss of motor neurons in DKO mice , we examined the spinal roots of the lumbar spinal cord by electron microscopy . Motor neuron axons in the ventral root of DKO mice showed advanced degeneration ( Figure 1E ) , whereas sensory neuron axons in the dorsal root were intact ( Figure 1F ) . Under high magnification of motor neuron axons , we saw intra-axonal vacuoles and the loss of axoplasm ( Figure 1G ) , as well as disintegrated myelin and amorphous lipids ( Figure 1H ) , as typically seen in Wallerian degeneration . The percentage of degenerated axons in the ventral root of DKO mice reached 48 . 0 ± 4 . 1% by P5 , which was significantly greater ( p < 0 . 01 ) than in WT mice ( 8 . 6 ± 1 . 5% ) ( Figure 1I ) . In contrast , the percentage of degenerated axons in the dorsal root of DKO mice ( 10 . 5 ± 1 . 9% ) was similar to that of the WT ventral root , consistent with the normal background level of axonal degeneration that occurs during development . Moreover , the sensory afferent terminals in the spinal cord of DKO mice appeared normal ( Figure 1—figure supplement 4A ) . In contrast , we found degeneration of phrenic motor nerve fibers and terminals in P5 DKO mice ( Figure 1—figure supplement 4B ) . These results further confirm the selective death of motor neurons in DKO mice , potentially leading to respiratory failure due to loss of motor neuron innervation of the intercostal muscles and diaphragm . Taken together , our DKO mice present a phenotype of early-onset and apparently selective motor neuron degeneration . Since both membralin and GluN3B are mutated in DKO mice , we needed to determine which gene was responsible for the motor neuron degeneration . Therefore , we tested whether transgenically expressed intact membralin or GluN3B could rescue DKO mice . We generated a transgenic mouse line carrying full-length membralin under the control of the murine prion promoter , followed by internal ribosome entry site ( IRES ) and enhanced green fluorescent protein ( EGFP ) cDNA sequences ( Figure 2A ) . The membralin transgenic mice [Tg ( membralin ) ] were fertile , normal in size , and did not display any gross physical or behavioral abnormalities compared with their littermate controls . When Tg ( membralin ) mice were crossbred with DKO mice , the DKO/Tg ( membralin ) mice were apparently normal , manifesting neither paresis nor premature death ( Figure 2B ) . Thus , the membralin transgene rescued DKO mice , even though membralin mRNA levels remained somewhat lower than in WT mice ( Figure 2C ) . GFP immunostaining in the ventral horn of the spinal cord showed that membralin/GFP transgene expression was predominantly neuronal , although we could not rule out relatively weak expression in glial cells ( Figure 2—figure supplement 1 ) . The DKO/Tg ( membralin ) mice showed no sign of motor defects or motor neuron loss , despite the absence of GluN3B expression ( Figure 2D ) . In contrast , the GluN3B transgene did not rescue DKO mice ( Figure 2—figure supplement 2 ) , and GluN3B KO mice generated by deleting the first exon did not show any defect in motor neurons ( Niemann et al . , 2007 ) . Taken together , these findings are consistent with the hypothesis that the loss of intact membralin in DKO mice causes motor neuron degeneration . 10 . 7554/eLife . 06500 . 008Figure 2 . Membralin transgene [Tg ( membralin ) ] rescues GluN3B/membralin ( DKO ) mice . ( A ) Generation of the transgenic constructs expressing membralin . The murine prion promoter was cloned with full-length mouse membralin cDNA followed by the IRES and the EGFP sequence . The entire transgenic sequence was isolated by enzymatic digestion and used to generate Tg ( membralin ) mice . ( B ) DKO/Tg ( membralin ) mice were viable and fertile . Left: genotyping by PCR showed that 3 littermates ( #3 , 4 and 5 ) from a breeding of DKO and Tg ( membralin ) mice were positive to DKO primers and negative to WT primers ( same primers used as in Figure 1 ) . Of the three DKO mice , the two receiving membralin transgene ( #3 and 5 , as positively identified by GFP primers ) were rescued . Right: Weight of the 7 littermates was monitored after birth . Only DKO mouse #4 , which was negative for the membralin transgene , showed weight loss after P3 and died at P5 . 5 . Data shown here are from one representative litter ( 5 independent litters were analyzed with similar results ) . ( C ) Gene expression analysis by RT-PCR for mice #1 , 3 , 4 of the same litter as in B . Membralin transgene expression is seen in the brains of littermates #1 and 3 , but not 4 , whereas GluN3B is only expressed in #1 . Membralin and GluN3B are not expressed in muscle samples . GAPDH is expressed in both brain and muscles of all mice and served as a positive control . ( D ) Rescued DKO/Tg ( membralin ) mice lived to adulthood without any sign of paresis . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 00810 . 7554/eLife . 06500 . 009Figure 2—figure supplement 1 . Expression of GFP-membralin transgene in spinal cord of membralin transgenic mice . GFP immunostaining shows expression of membralin-GFP transgene in all spinal cord layers of membralin ( MEM ) KO and transgenic ( Tg ) mice ( left ) , but not WT mice ( right ) . GFP immunostaining was located predominantly in neurons , including motor neurons ( arrows ) . Images at low ( top ) and high ( bottom ) magnification are shown from the lumbar spinal cord at P5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 00910 . 7554/eLife . 06500 . 010Figure 2—figure supplement 2 . GluN3B transgene cannot rescue GluN3B/membralin DKO mice . ( A ) DNA construct used for generating GluN3B transgenic [Tg ( GluN3B ) ] mice containing the 7 kb murine GluN3B promoter followed by murine full-length coding region for GluN3B , IRES , and GFP . ( B ) Stronger immunocytochemical signal for GluN3B signal in motor neurons of lumbar spinal cord of Tg ( GluN3B ) mice compared to WT mice . ( C ) Tg ( GluN3B ) cannot rescue DKO mice . Left: PCR genotyping shows one mouse ( #3 ) among five from a DKO x Tg ( GluN3B ) litter was positive for DKO and GFP primers , but negative for WT primers ( same primers as used in Figure 1 ) . Right: weight of the 5 littermates was monitored after birth , and DKO mouse #3 , which contained the GluN3B transgene , lost weight , and died at P6 . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 010 Since heterozygous DKO mice did not manifest motor defects , and expression of the truncated form of membralin appears to lack any deleterious effect . This conclusion is also supported by results from a second membralin KO mouse line we generated using a gene trapping strategy ( Figure 3A ) . RT-PCR experiments using primers from exon 1 and 2 further confirmed that homozygous mice carrying membralin/trapping vector alleles did not express membralin mRNAs with sequences beyond exon 2 . Membralin protein was also reduced in heterozygotes and absent in homozygotes . These membralin KO mice phenocopied the DKO mice in motor defects , with ∼50% motor neuron loss preceding death around P5 ( Figure 3B , C ) . 10 . 7554/eLife . 06500 . 011Figure 3 . Membralin KO mice die of motor neuron degeneration and consequent paresis . ( A ) Gene trapping was used to generate membralin KO mice by inserting a trapping vector that contained a splicing acceptor sequence between exon 1 and 2 to disrupt normal RNA splicing . Primers ( P1 , P2 , P3 , red arrows ) were designed to detect normal and trapped membralin transcripts . RT-PCR experiments showed that PCR products using the P1 and P3 primer pair were only detected in WT mouse brain ( lane 1 ) , liver ( lane 2 ) , and kidney ( lane 3 ) , whereas PCR products using the P1 and P2 primers were only detected in these tissues of KO mice . ( B ) Membralin KO mice phenocopied GluN3B/membralin DKO mice and died of paresis around P5 . ( C ) Lumbar motor neurons , identified by anti-Hb9 staining ( top panels ) , were significantly reduced ( lower panel ) in membralin KO mice compared to WT mice ( n = 3 for each group of mice , p < 0 . 05 , Student's t-test ) . Data are mean +s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 011 Membralin is a novel protein with no conserved domains . RT-PCR analysis showed that membralin mRNA was expressed in multiple tissues , including brain and spinal cord , but not heart , skeletal muscle , or spleen ( Figure 4A ) . Western blot analysis of spinal cord samples from WT and KO mice showed specific expression of membralin in the ER but not in the cytosol or mitochondria ( Figure 4B ) . Since our anti-membralin antibody proved ineffective for immunostaining , we made a fusion protein by adding a myc tag to the C-terminus of membralin ( membralin-myc ) . Immunocytochemistry demonstrated that membralin-myc co-localized with an ER marker PDI but not with markers for Golgi or mitochondria ( Figure 4C ) . Membralin-myc was not expressed on the cell surface , since it was not detected in transfected cells without first permeabilizing the membrane ( data not shown ) . Software for topology prediction from the Center for Biological Sequences ( CBS , TMHMM ) indicated that the membralin is most likely a transmembrane protein with 4–6 membrane-spanning regions , containing an ER-lumen domain between membrane-spanning regions 1 and 2 . Additionally , the protein has multiple glycosylation sites and cytosolic regions at both the N-terminal ends and C-terminal ends ( Figure 4—figure supplement 1A , B ) . These data suggest that membralin is a membrane protein most likely localized to the ER and/or outer nuclear envelope . 10 . 7554/eLife . 06500 . 012Figure 4 . Cellular expression and localization of membralin . ( A ) Semi-quantitative RT-PCR analysis showed the expression of membralin mRNA in brain ( 1 ) , spinal cord ( 2 ) , lung ( 4 ) , liver ( 5 ) , and kidney ( 6 ) , but not in heart ( 3 ) , spleen ( 7 ) , and muscle ( 8 ) tissues of WT mice . ( B ) Western blot analysis showed the expression of membralin protein in the ER fraction , but not in the mitochondrial fraction ( Mit ) or the cytosolic fraction ( Cyt ) , of the WT mouse brains . ( C ) Subcellular localization of Myc-tagged membralin in HEK 293 cells . Cells were transiently transfected with Myc-tagged membralin . At 48 hr after transfection , localization of membralin was detected by confocal microscopy ( Zeiss 710 ) . Organelles were labeled with specific markers ( anti-PDI for ER , anti-mitochondria for mitochondria , and anti-GMP130 for Golgi ) . Region of interest shown at higher magnification in insets ( bottom right corner ) . Merged pictures show membralin co-localized with the ER marker , PDI . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 01210 . 7554/eLife . 06500 . 013Figure 4—figure supplement 1 . Structural prediction of membralin . ( A ) Membralin's predicted structure has four transmembrane regions with termini on the same side of membrane . ( B ) The predicted glycosylation sites are also indicated by brown tree signs , which suggest the topology of the membralin on ER membrane ( lumen and cytosolic side is indicated as ‘in’ and ‘out’ , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 013 A recent study defining human ERAD networks through an integrative mapping strategy identified a potential interaction of membralin ( C19orf6 ) with Erlin2 ( Christianson et al . , 2012 ) . Erlin2 is a protein that is enriched in ER lipid rafts ( Ikegawa et al . , 1999; Browman et al . , 2006 ) , and mutations in Erlin2 have been linked to several human diseases with motor dysfunction ( Al-Yahyaee et al . , 2006; Alazami et al . , 2011; Yildirim et al . , 2011; Al-Saif et al . , 2012; Wakil et al . , 2013 ) . We designed experiments to verify the membralin/Erlin2 interaction . We found that GFP-tagged membralin not only co-localized with but also manifested a strong FRET signal with mCherry-tagged Erlin2 in the ER ( Figure 5A ) . The FRET efficiency was significantly higher for GFP-tagged membralin and mCherry-tagged Erlin2 ( 18 . 00 ± 0 . 70% ) than the negative control ( GFP and mCherry , 0 . 15 ± 0 . 05% ) , consistent with a direct interaction of the two molecules ( Figure 5B ) . Additionally , Myc-tagged full-length membralin co-immunoprecipitated with HA-tagged Erlin2 ( Figure 5C ) . This interaction of membralin with Erlin2 further supports the notion that membralin is an ER membrane protein and also suggests that loss of membralin could potentially increase ER stress by interrupting ERAD . Indeed , we found that the ER membrane protein , CD3-δ , was cleared more slowly in mouse embryonic fibroblasts ( MEFs ) prepared from membralin KO mice than from WT mice ( Figure 5D ) , whereas there was no change in the clearance of the ER lumenal protein , NHK ( Figure 5—figure supplement 1 ) . These results suggest that membralin deficiency affects degradation of ER membrane proteins , which could potentially increase ER stress . Thus , we next analyzed ER stress in membralin KO mice . 10 . 7554/eLife . 06500 . 014Figure 5 . Membralin interacts with Erlin2 and regulates protein degradation . ( A ) HEK 293T cells were transfected with control ( mCherry and EGFP ) or target ( membralin-mCherry and Erlin2-EGFP ) molecules . Individual or merged fluorescence images were shown in pre-bleaching and post-bleaching conditions . ( B ) FRET efficiency was quantified for control and target groups . All data shown are mean ± s . e . m . n = 10; *p < 0 . 001 by Student's t-test . ( C ) The interaction of membralin with Erlin2 was confirmed by co-immunoprecipitation experiments using HA-tagged Erlin2 and Myc-tagged membralin fusion proteins co-transfected into HEK 293T cells . Whole-cell lysates were immunoprecipitated with anti-Myc or IgG ( negative control ) and immunoblotted using anti-Erlin2 . ( D ) MEFs from both WT and membralin KO mouse were transfected with HA tagged CD3-δ and subjected to pulse-chase analysis of CD3-δ degradation after exposing with cycloheximide ( Chx ) . Proteins collected at the indicated time points were subjected to immunoblotting with antibodies to HA . Image shown represents the example of immunoblotting and graphs , the quantification of three experiments . ( n = 3 , p < 0 . 05 by Student's t-test ) . All data shown are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 01410 . 7554/eLife . 06500 . 015Figure 5—figure supplement 1 . Deletion of membralin did not prolong the half life of ERAD substrate NHK . MEFs from both WT and membralin KO mouse were transfected with HA tagged NHK and subjected to pulse-chase analysis of degradation after exposed to cycloheximide ( Chx ) . Proteins collected at the indicated time points were subjected to immunoblotting with antibodies to HA . Image shown represent example of immunoblotting and graphs , the quantification of three experiments . All data shown are mean ± s . d . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 015 Increased ER stress is thought to be involved in some forms of motor neuron disease , including ALS ( Atkin et al . , 2006; Nagata et al . , 2007; Nishitoh et al . , 2008; Kanekura et al . , 2009; Saxena et al . , 2009 ) . We , therefore , tested whether the absence of membralin impacted ER stress ( Walter and Ron , 2011 ) . Western blot analysis showed that the levels of GRP78 and ATF4 , two molecules induced during ER stress via activation of the PERK signaling pathway , were consistently higher in the spinal cord of P3 membralin KO mice than WT mice ( 154 ± 35% and 412 ± 39% of WT , respectively; KO , n = 4; WT , n = 3; Figure 6A ) . In contrast , levels of spliced XBP-1 , indicating activation of the IRE1 signaling pathway during ER stress , were not altered ( Figure 6—figure supplement 1A ) . Furthermore , tunicamycin , an inducer of ER stress , produced a greater degree of cell death in MEFs from KO mice than from WT mice ( Figure 6B ) . Tunicamycin exposure also increased CHOP and ATF4 levels earlier and to a greater extent in MEFs from KO mice; GRP78 levels were also higher at 24 hr post-exposure ( Figure 6C ) . Basal levels of ATF4 , but not CHOP , were higher in membralin KO mice than WT mice ( Figure 6C ) , but XBP-1 splicing was not altered ( Figure 6—figure supplement 1B ) . Collectively , these data suggest that loss of membralin increases basal ER stress and makes cells more vulnerable to additional ER stress-induced injury . 10 . 7554/eLife . 06500 . 016Figure 6 . Elevated ER stress in membralin KO mice . ( A ) Upregulation of GRP78 and ATF4 in spinal cord of membralin KO mice . Left: immunoblot analysis shows higher levels of GRP78 and ATF4 in the spinal cord of P3 membralin KO mice compared to that of littermate WT mice . Right: increased expression level of GRP78 and ATF4 normalized to actin in membralin KO mice over WT mice . ( B ) Survival of MEFs after 24-hr exposure to tunicamycin or thapsigargin determined by a cytotoxicity assay . ( C ) The expression level of GRP78 , CHOP , and ATF4 in MEFs from KO and WT mice is shown after exposure to tunicamycin at different time points . Levels of CHOP and ATF4 are elevated earlier and higher in MEFs from KO than that from WT mice . A representative immunoblot is shown; graphs include data from three experiments . For each panel , data are mean ± s . e . m . ; n = 3 for each bar; *p < 0 . 05 by Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 01610 . 7554/eLife . 06500 . 017Figure 6—figure supplement 1 . Membralin deletion does not alter Xbp-1 splicing . ( A ) RT-PCR assay of Xbp-1 mRNA from brain and spinal cord from membralin WT and KO mouse . ( B ) RT-PCR assay of spliced and unspliced Xbp1 mRNA from membralin WT and KO MEF cells exposed to the ER stress inducer , tunicamycin ( 300 nM ) , for the indicated times . DOI: http://dx . doi . org/10 . 7554/eLife . 06500 . 017 Hereditary motor neuropathy ( HMN ) is usually subdivided into two groups: proximal HMN , i . e . , classical spinal muscular atrophy ( SMA ) syndromes , and distal HMN , which clinically resemble CMT syndromes without obvious sensory abnormalities . In contrast to proximal HMN , distal HMN represents a heterogeneous group of peripheral neuropathies affecting mainly distal muscles with both autosomal dominant and recessive inheritance ( Irobi et al . , 2004 , 2006; Irobi-Devolder , 2008 ) . Seven subgroups of distal HMN were initially proposed based on age of onset , mode of inheritance , and clinical features of a limited number of affected families ( Harding , 1993 ) . Additional forms of distal HMN were subsequently reported , adding more genetic complexity . A dozen of causal gene loci and other associated genes have been identified for distal HMN ( exclusive of type I ) ; these genes encode for a functionally heterogeneous group of proteins , as summarized in recent reviews ( Irobi et al . , 2004 , 2006; Irobi-Devolder , 2008 ) . The biological functions of the affected proteins include stress responses ( small heat shock proteins HSP22 , HSP27 ) , housekeeping ( GARS , glycyl tRNA synthetase ) , protein glycosylation in the ER ( seipin ) , RNA processing ( immunoglobulin μ-binding protein 2 [LGHMBP2] and senataxin ) , and axonal transport ( dynactin ) . It is not clear , however , how these mutated proteins of diverse cellular function converge to affect motor neuron survival selectively . In fact , their expression is not even restricted to motor neurons . A similar case is found for membralin in this study . Therefore , further studies on molecular mechanisms underlying the selectivity of motor neuron death in distal HMN will be critical to understanding the disease . A hint towards pathogenesis may be found in prior work in conjunction with our new findings and involves an unusual susceptibility of motor neurons to various forms of cell stress , particularly ER stress . Accordingly , the membralin KO mouse generated in our laboratory provides a good model to study selective motor neuron vulnerability . These mice display severe motor neuron loss and muscle wasting , leading to paresis and death . Although the early disease onset in membralin KO mice is similar to that seen in an SMA mouse model ( Hsieh-Li et al . , 2000; Monani et al . , 2000 ) , the pattern of motor neuron injury in membralin KO mice is reminiscent of human distal HMN rather than proximal HMN , as observed in SMA . Of the dozen known causal genes for distal HMN , only a few show autosomal recessive inheritance , and no mouse model has , heretofore , been generated ( Irobi et al . , 2004 , 2006; Irobi-Devolder , 2008 ) . Therefore , the membralin null mouse represents a novel model for human distal HMN . Additionally , our studies demonstrate the previously unknown function of membralin in motor neuron survival . Our findings from endogenous and heterologous expression studies suggest that membralin is mainly located in the ER . Relatively low expression of membralin seems sufficient for normal function , as heterozygotes of the membralin KO survived , and transgene rescue of the membralin KO did not require high expression levels . Importantly , in our rescue experiments , membralin transgene expression was primarily in brain , not muscle , consistent with the notion that the pathogenic process is predominantly neural in origin . The exact mechanisms underlying motor neuron death in membralin KO mice are not yet clear , although death occurs very rapidly at an early and well-defined postnatal stage . Thus , membralin KO mice can be used not only as an early-onset model of motor neuron disease but also to determine if there are features in common with late-onset motor neuron disease , such as motor neuron-specific vulnerability to ER stress ( Saxena et al . , 2009; Roselli and Caroni , 2015 ) , dying back axonopathy ( Fischer et al . , 2004; Coleman , 2005 ) , or non-cell autonomous death ( Boillee et al . , 2006; Yamanaka et al . , 2008; Kang et al . , 2013 ) . Previously , membralin was predicted to encode a transmembrane protein , but it lacked homology with any known protein domains and was not known to interact with other membrane proteins ( Andersson and von Euler , 2002 ) . In the present study , we found that membralin interacts with Erlin2 , an ER membrane protein potentially involved in ERAD ( Christianson et al . , 2012 ) . The exact cellular functions of Erlin2 are not clear , although it has been reported to interact with several ER resident E3 ligases such as GP78 and Hrd1 that are important for ERAD ( Christianson et al . , 2012 ) . Additionally , Erlin2 has been shown to regulate ER membrane proteins such as Inositol 1 , 4 , 5-trisphosphate receptors ( Pearce et al . , 2007 ) . It is conceivable that membralin assists Erlin2 in a complex that retrotranslocates unfolded proteins from the ER lumen to the cytosol , thus facilitating their ubiquitination for degradation ( Schulze et al . , 2005; Carvalho et al . , 2006; Denic et al . , 2006; Vembar and Brodsky , 2008; Carvalho et al . , 2010; Smith et al . , 2011; Brodsky , 2012 ) . Consequently , membralin deficiency might increase both basal ER stress and vulnerability to ER stress-induced cell death , and this is exactly what we observed . We also found differences in Isg15 levels in membralin KO mice , leading to ISGylation , a process that represents a type I interferon-dependent , ER stress-triggered event . This observation is interesting in light of prior findings showing that ISGylation is a pre-symptomatic event observed in a mouse model of ALS ( Wang et al . , 2011 ) . Collectively , our data suggest that the loss of membralin may contribute to degeneration by increasing ER stress in motor neurons , which are especially vulnerable to such stress due to their large metabolic demand . This demand is particularly prevalent postnatally when maturation of motor neurons requires synthesis of proteins for production of dendritic trees , synaptic connections , and ionic conductances ( Vinay et al . , 2000; Carrascal et al . , 2005; Li et al . , 2005 ) . Additionally , disruption of ERAD in SOD1 mutant mice is known to induce ER stress , activation of ASK1 , and motor neuron cell death ( Nishitoh et al . , 2008 ) . Mutations in Erlin2 have been linked to human disease , including intellectual disability , motor dysfunction , hereditary spastic paraplegia , and juvenile primary lateral sclerosis ( Al-Saif et al . , 2012; Al-Yahyaee et al . , 2006; Alazami et al . , 2011; Wakil et al . , 2013; Yildirim et al . , 2011 ) . Moreover , disturbance in various other components of ERAD , including OS-9 , erasin , ubiquilin2 , torsinA , and Derlin1 , causes ER stress ( Nishitoh et al . , 2008; Alcock and Swanton , 2009; Lim et al . , 2009; Deng et al . , 2011; Nery et al . , 2011 ) , and some of these genes have been linked to ALS ( Nishitoh et al . , 2008; Alcock and Swanton , 2009; Lim et al . , 2009; Alazami et al . , 2011; Nery et al . , 2011; Yildirim et al . , 2011; Al-Saif et al . , 2012; Wakil et al . , 2013 ) . Therefore , future elucidation of the exact role of membralin in ERAD will undoubtedly be important for understanding the contribution of ER stress to motor neuron diseases . Interestingly , there are more than 50 single nucleotide polymorphisms ( SNPs ) in the coding region of human membralin that result in missense mutations ( NBCI SNP database ) . The minor allele frequency of most SNPs is either low or has not been determined , suggesting that these SNPs are not common in the human population . Thus , our studies point to a specific role of membralin in motor neuron survival and potentially open a new avenue for research in the field of human motor neuron diseases . We used the 3 . 6 kb ( Apa I/Spe I ) and 3 . 2 kb ( BamH I/Kpn I ) DNA fragments flanking the target region as the 5′- and 3′-arms , respectively , for the ∼11 . 4 kb pGTN29/GluN3B/membralin targeting vector ( see Figure 1—figure supplement 1A ) . Homologous recombination generated a ∼5 kb size difference between the original and recombinant genes , facilitating their analysis . During construction of the targeting vector , one SpeI site at the junction of the 5′-arm and the target region were eliminated , and another SpeI site was introduced at the junction of the target region and 3′-arm . These restriction site variations allowed us to designed a 3′ probe , a 436 bp fragment from Xho I digestion , that hybridized to the 14 kb and 6 kb fragments generated by SpeI digestion or 9 . 1 kb and 8 . 4 kb fragments generated by EcoR I digestion ( see Figure 1—figure supplement 1A , B ) from the WT and recombinant alleles , respectively . All described procedures for animal were approved by the Institutional Animal Care and Use Committee of Sanford–Burnham Medical Research Institute and conducted in compliance with the Guide for the Care and Use of Laboratory Animals . Both sexes of mice were used for experiments and maintained in an institute facility accredited by the Association for Assessment and Accreditation of Laboratory Animal Care ( AAALAC ) . The C1 ES cells were cultured in the presence of leukemia inhibitor factor on primary embryo fibroblast feeder cells and were transfected with the linearized pGTN29/GluN3B/membralin target vector DNA by electroporation at 400 V , 25 mF . Thirty-six hours after transfection , 0 . 48 mg/ml G418 was added to select for cells that have acquired the Neor gene by homologous recombination with the targeting vector . The ES cells with the correct recombination were confirmed by standard Southern blot analysis using the probe described above . GluN3B/membralin-targeted ES cell lines were used for blastocyst injection . Then , 6 to 10 embryos , including 2 uninjected blastocysts as carriers , were transferred into one uterine horn of each pseudopregnant foster mother ( F1 of DBA X C57BL/J6j ) . Chimeric mice were identified by eye pigmentation . Further breeding was made to test for germ line transmission of the injected ES cells . Mouse tail DNA was used to identify homozygotes and heterozygotes by the same procedure used in testing ES cells . Once the DKO line was established , we also designed a PCR method to simplify genotyping using a pair of primers for the GluN3B sequence in WT alleles and a pair of primers for the Neo sequence in the DKO allele ( see Supplementary file 1 ) . We have obtained the mouse full-length GluN3B cDNA clone from Dr Yuzaki as a gift ( Matsuda et al . , 2002 ) . A full-length mouse membralin cDNA clone ( ID# 3813678 ) was purchased from distributors of the Integrated Molecular Analysis of Genomes and their Expression ( I . M . A . G . E ) consortium ( http://www . imageconsortium . org/ ) . We used the mouse GluN3B promoter that drives gene expression mainly in motor neurons from late embryonic stages to adult ( database from NINDS Gensat Bac Transgenic Project , http://www . gensat . org/index . html ) . In addition , we used the mouse prion promoter , which drives high transgene expression in mouse neurons ( Baybutt and Manson , 1997; Loftus et al . , 2002; Gispert et al . , 2003 ) , to express the transgene in a broader brain area . The selected full-length transgene was subcloned downstream of either promoter , followed by an IRES or a sequence coding the GFP . The insert containing promoter , transgene , and IRES/GFP sequences were isolated by restriction digest followed by gel purification . The purified DNA fragments were injected into fertilized eggs to generate transgenic mice ( C57BL/J6j ) in our institute's Transgenic Facility according to established protocol . We used PCR methods to genotype transgenic mice using a pair of primers ( see Supplementary file 1 ) . Hemizygotes of transgenic mice were bred with DKO mice to obtain double heterozygotes [transgene+/−/ ( GluN3B/membralin ) +/−] . The double heterozygotes were further bred to obtain mice that express the transgene in the GluN3B/membralin null background [transgene+/−/ ( GluN3B/membralin ) −/−] , and tested for GFP and transgene expression using RT-PCR . The crossed transgene/DKO mouse line was subjected to histological tests for examining motor neuron integrity . The survival time of the [transgene+/−/ ( GluN3B/membralin ) −/−] mice were monitored and compared with that of DKO mice to determine the ability of the transgene to rescue the DKO mice phenotype . We generated membralin KO mice by using ES cells ( gift of Sanger Institute ) with a disrupted membralin gene by a gene trapping method ( Brennan and Skarnes , 1999; Skarnes et al . , 2004 ) . Briefly , a trapping vector containing an RNA splicing acceptor sequence was inserted between exon 1 and 2 of the membralin gene in ES cells to disrupt normal RNA splicing . The positive ES cell clone was confirmed by sequencing and used for blastocyst injection to generate membralin KO mice ( C57BL/J6j ) . RT-PCR experiments using primers from exon 1 and 2 further confirmed that homozygote mice carrying membralin/trapping vector alleles did not express membralin mRNAs with sequences beyond exon 2 . Previously established protocols were used for staining spinal cord sections or cultured cells ( Xing et al . , 2006 ) , with the following antibodies: mouse anti-GFAP ( Sigma-Aldrich , St . Louis , MO ) , rabbit anti-IBA-1 ( Wako Chemicals , Inc . , Richmond , VA ) , and anti-Hb9 ( a gift from Dr Pfaff ) , followed by incubation with either a fluorophore-attached ( Life Technologies , Grand Island , NY ) or a biotinylated secondary antibody ( Vector Laboratories , Burlingame , CA ) . Immunosignals were detected either directly under epifluorescence microscopy or after using a Vectastain Elite ABC kit ( Vector Laboratories ) with 3 , 3′-diaminobenzidine visualization ( Roche Applied Science , Indianapolis , IN ) . The number of motor neurons in layers VIII and IX of the lumbar spinal cord was counted using stereological methods in WT , DKO , and membralin KO mice ( Coggeshall and Lekan , 1996 ) . Briefly , consecutive sections ( 12 μm in thickness ) of the lumber enlargement ( L1–L5 ) were collected for immunostaining with anti-Hb9 antibody to identify motor neurons . The number of motor neurons in each section was counted stereologically using adjacent sections for reference and look-up ( physical disector ) . The total number of motor neurons was obtained for each mouse , and the mean value for 3 mice at each age group was calculated and normalized to that of WT mice . The general appearance of motor neurons was examined by conventional staining methods used in analysis of SOD1 mutant mice , including hematoxylin/eosin and cresyl violet staining . HEK 293 cells were used for investigating the subcellular localization of membralin . A membralin-myc fusion protein was constructed by tagging the C-terminal of membralin with a Myc sequence using the pcDNA3 . 1 ( − ) Myc/his vector ( Life Technologies ) . HEK 293 cells were plated onto coverslips , transfected with the membralin-Myc overnight , and fixed 48 hr after transfection with 4% PFA with 0 . 5% Triton X-100 in PBS . After blocking with 10% normal goat serum in PBS for 60 min , the cells were double stained by anti-Myc rabbit polyclonal antibody ( Sigma ) for membralin-Myc detection , followed by Alexa fluor-conjugated secondary antibodies ( Life Technologies ) . Subcellular organelles were stained by either anti-PDI antibody ( Enzo Life Science , Farmingdale , NY ) for ER labeling , anti-GMP130 antibody ( BD bioscience , Franklin Lakes , NJ ) for Golgi labeling , or anti-mitochondria antibody ( 113-1 , Abcam , Cambridge , MA ) for mitochondrial labeling , followed by Alexa fluor-conjugated secondary antibodies ( Life Technologies ) . DAPI was used to stain nuclei . Images were captured under confocal microscopy ( Zeiss 710 ) . Dorsal and ventral roots of lumbar segments ( L3–L5 ) of the spinal cord were dissected out from P5 WT and DKO mice . Samples were immediately placed in 4% paraformaldehyde plus 1% glutaraldehyde in 0 . 1 M phosphate buffer and simultaneously processed for electron microscopy as described previously ( Liu et al . , 1998 ) . Briefly , samples were osmicated in 1% OsO4 for 10–20 min followed by washing in 0 . 1 M phosphate buffer and then dehydrated in graded ethanol and 100% acetone . Each sample was oriented and placed in a Flat Embedding Mold ( Ted Pella , Redding , CA ) filled with Araldite ( EMS , Fort Washington , PA ) . Embedded dorsal or ventral roots of different genotypes were dissected , and the cross sections of the roots were re-cut on an ultramicrotome ( Ultracut , Leica ) at 70–80 nm . Serial thin sections were collected on Formva-coated single slot nickel grids and stained with uranyl acetate and lead citrate . Ultrathin sections were examined in a Philips CM120 electron microscope at 80 KV . Digitized images were acquired by a high-resolution ( 2K × 2K ) CCD camera ( Gatan , Inc . , Pleasanton , CA ) , processed using software provided by the manufacturer ( DigitalMicrograph ) , and displayed with Photoshop CS ( Adobe Systems , San Jose , CA ) . Two DNA fragments were purified by enzymatic digestion of membralin cDNA using BamH I/Bgl II and Not I/BamH I , respectively , to generate two probes specific to regions of exon VII-X and the last exon ( XI ) of membralin . Northern blot analysis was performed using these two probes labeled with 32P by random priming . 10 μg of total mouse RNA per lane were analyzed by electrophoresis on a 1 . 1% denaturing gel and subsequently transferred to a nylon membrane . The blot was hybridized at 42°C in a solution containing 50% formamide , 6x SSC , 5x Denhardt's reagent , and 0 . 5% SDS . The blots were dried and exposed to autoradiographic films for analysis . cDNAs for mouse membralin and Erlin2 were purchased from distributors of the Integrated Molecular Analysis of Genomes and their Expression ( I . M . A . G . E ) consortium ( http://www . imageconsortium . org/ ) and subcloned into a pcDNA3 . 1 vector for mammalian cell expression . C-terminal tagged membralin ( membralin-myc or membralin-mCherry ) and N-terminal or C-terminal tagged Erlin2 ( Erlin2-EGFP or Erlin2-HA ) were constructed by introducing corresponding tag fragments generated by the PCR method , and all constructs were confirmed by sequencing . Membralin-myc and Erlin2-HA were co-transfected in HEK 293 cells and total cell lysates collected in 48 hr after transfection for co-immunoprecipitation-immunoblot assay . The interaction of Erlin2-EGFP and membralin-mCherry was validated by the acceptor photobleaching method for FRET detection ( Karpova and McNally , 2006 ) . Briefly , HEK 293 cells were transfected with Erlin2-EGFP and membralin-mCherry for 24 hr . A Zeiss 710 NLO microscope ( Carl Zeiss Inc . ) was used to record the fluorescence of EGFP and mCherry in transfected cells . Three pre-bleached and five post-bleached images were acquired . Averaged fluorescence intensities of the donor were calculated from the measurement of regions of interest for each experimental set before and after bleaching . The efficiency of FRET was calculated by Efret = 1 – ( Ia/Ib ) , where Ia and Ib represent the steady-state donor fluorescence in the presence and the absence of the acceptor , respectively . Seven Ala-linked EGFP-mCherry plasmids were transfected into HEK 293 cells and were used as a positive FRET control ( FRET efficiency = ∼50% , data not shown ) . FRET efficiencies were reported as mean ± s . e . m . We measured levels of GRP78 , CHOP , and ATF4 , components of three canonical branches of UPR , by Western blot analysis using the spinal cord tissues and MEFs from membralin KO mice and WT littermates . The antibodies against these proteins were purchased from commercial sources: GRP78 ( H-129 , 1:1000; Santa Cruz Biotechnology , Santa Cruz , CA ) , CHOP ( 2895S , 1:1000; Cell Signaling , Denvers , MA ) , and ATF4 ( SC200 , 1:500; Santa Cruz Biotechnology ) . Total RNA was isolated from the spinal cord of membralin mutant mice and WT littermate using TRIzol reagent ( Life Technologies ) . The following set of primers was used to detect the expression of mouse Xbp-1 ( GATCCTGACGAGGGTCCAAGA and ACAGGGTCCAACTTGTCCAG ) . MEFs from membralin KO mice and WT littermates were isolated from day 12 . 5 embryos and cultured by conventional methods . In the toxicity assay , the percentage of cell death was measured by counting cells or using the CellTiter 96 Aqueous Non-Radioactive Cell Proliferation Assay kit ( Promega , Madison , WI ) after MEF cells were exposed to the ER stress inducer , tunicamycin , for 24 hr . The sample size was 3–10 per genotype for animal histology and 3–6 for protein or transcript expression as well as for cell viability studies , as determined by Power Analyses of previous data . The experiments were not randomized . Although the initial investigator performing the experiments was not blinded , the samples and animal results were then examined by a group of the investigators masked to experimental identity in order to evaluate the results . All data points demonstrated a normal distribution and were all included in the analysis . Data are presented as mean ± s . e . m . and analyzed by a Student's t-test for pairwise comparisons . Statistical analyses were conducted using GraphPad Prism software ( version 6 ) . A p value <0 . 05 was considered statistically significant .	As new proteins are built inside a cell , many will pass into a structure called the endoplasmic reticulum for processing . There , the proteins are folded into the specific three-dimensional shapes that allow them to carry out their respective jobs . Sometimes the folding process goes awry , leading to a build-up of unfolded proteins that stress the endoplasmic reticulum and can kill the cell . Brain cells are particularly vulnerable to death from endoplasmic reticulum stress . To combat a deadly build-up of unfolded proteins , each cell has systems that respond when the endoplasmic reticulum is under stress . Unchecked stress on the endoplasmic reticulum has been linked to diseases like amyotrophic lateral sclerosis ( called ALS for short ) . In diseases like ALS , the nerve cells that control muscle movements gradually die off , causing a loss of muscle control and eventually death . Scientists suspect that these nerve cells ( called motor neurons ) are particularly sensitive to endoplasmic reticulum stress because they are highly active . Drugs that help counteract stress on the endoplasmic reticulum extend the lives of mice with motor neuron disease , suggesting this may be a useful strategy for treating such diseases in humans . Now , Yang , Qu et al . identify a new protein that appears necessary for a healthy endoplasmic reticulum . Mice that lack the gene for a protein called membralin die within five or six days after birth because their motor neurons die off . Further experiments showed that re-introducing membralin in their nervous system can rescue these membralin-deficient mice . Yang , Qu et al . found that membralin interacts with another protein that helps eliminate poorly folded or unfolded proteins in the endoplasmic reticulum , and thus relieves stress on the cell . Mutations in this endoplasmic reticulum stress response protein have previously been linked to motor neuron diseases . The motor neurons in membralin-deficient mice show signs of endoplasmic reticulum stress and are extra vulnerable to chemicals that induce protein misfolding . Together , the experiments show membralin plays an important role in mitigating stress on the endoplasmic reticulum . More studies of mice lacking membralin may help explain why the endoplasmic reticulum stress increases in motor neuron diseases and may point to possible treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	The critical role of membralin in postnatal motor neuron survival and disease
Insect herbivores use different cues to locate host plants . The importance of CO2 in this context is not well understood . We manipulated CO2 perception in western corn rootworm ( WCR ) larvae through RNAi and studied how CO2 perception impacts their interaction with their host plant . The expression of a carbon dioxide receptor , DvvGr2 , is specifically required for dose-dependent larval responses to CO2 . Silencing CO2 perception or scrubbing plant-associated CO2 has no effect on the ability of WCR larvae to locate host plants at short distances ( <9 cm ) , but impairs host location at greater distances . WCR larvae preferentially orient and prefer plants that grow in well-fertilized soils compared to plants that grow in nutrient-poor soils , a behaviour that has direct consequences for larval growth and depends on the ability of the larvae to perceive root-emitted CO2 . This study unravels how CO2 can mediate plant–herbivore interactions by serving as a distance-dependent host location cue . Insect herbivores can use different cues to locate suitable host plants from a distance . Volatile cues , in particular , can convey information about the identity and physiological status of a host plant and are integrated by herbivores to locate host plants for oviposition and feeding ( Visser and Avé , 1978 ) . Over the years , many attractive and repellent plant volatiles were identified ( Bruce et al . , 2005; Späthe et al . , 2013; Webster and Cardé , 2017 ) , and the importance of individual compounds and volatile blends was documented using synthetic chemicals ( Bruce and Pickett , 2011; Carrasco et al . , 2015; Fraenkel , 1959; Dorn et al . , 2003; Visser and Avé , 1978 ) . More recently , molecular manipulative approaches were used to manipulate plant volatile production and herbivore perception in vivo ( Fandino et al . , 2019; Halitschke et al . , 2008; Robert et al . , 2013 ) , thus confirming the important role of plant volatiles in plant–herbivore interactions . While the role of plant volatiles such as green-leaf volatiles , aromatic compounds , and terpenes is well understood , much less is known about the role of plant-associated carbon dioxide ( CO2 ) in plant–herbivore interactions . As many plant organs and their associated microbial communities release CO2 , it may be integrated into herbivore foraging as a marker of metabolic activity . Datura wrightii flowers , for instance , emit the highest levels of CO2 during times of high nectar availability; as hawkmoth pollinators are attracted to CO2 , they may thus use this cue to locate rewarding flowers ( Goyret et al . , 2008; Guerenstein et al . , 2004; Guerenstein and Hildebrand , 2008; Stange , 1996; Stange , 1999; Stange and Stowe , 1999; Thom et al . , 2004 ) . Similarly , lesions in apples result in high CO2 release and attract Bactrocera tryoni fruit flies . As CO2 at corresponding concentrations is attractive to the flies , it has been suggested that they may use plant-associated CO2 to locate suitable oviposition sites ( Stange , 1999 ) . Root-feeding insects are highly attracted to CO2 in vitro ( Bernklau and Bjostad , 1998a; Bernklau and Bjostad , 1998b; Eilers et al . , 2012; Hibbard and Bjostad , 1988; Jones and Coaker , 1978; Klingler , 1966; Nicolas and Sillans , 1989; Rogers et al . , 2013; Strnad et al . , 1986; Strnad and Dunn , 1990 ) . Given that CO2 is produced and released by plant roots and diffuses relatively well through the soil , a likely explanation for this phenomenon is that root herbivores use CO2 as a host location cue ( Bernklau and Bjostad , 1998a; Bernklau and Bjostad , 1998b; Doane et al . , 1975; Erb et al . , 2013; Johnson and Gregory , 2006; Johnson and Nielsen , 2012 ) , However , the reliability of CO2 as a host location cue for root feeders has been questioned due to a number of reasons: ( i ) CO2 can be emitted by many other sources apart from host plant roots , including decaying organic matter , microorganisms , and non-host plants; ( ii ) there is a strong diurnal fluctuation in plant CO2 emissions that does not necessarily match with insect foraging habits; and ( iii ) other plant-released chemicals can be used by root herbivores for host location within a CO2 background ( Agus et al . , 2010; Eilers et al . , 2012; Erb et al . , 2013; Hansen , 1977; Hibbard and Bjostad , 1988; Hiltpold and Turlings , 2012; Johnson and Nielsen , 2012; Reinecke et al . , 2008; Weissteiner et al . , 2012 ) . A model that may reconcile these different views is that CO2 may be used as an initial cue at long distances , while other , more host-specific volatiles may be used at shorter distances ( Erb et al . , 2013; Johnson et al . , 2006; Johnson and Nielsen , 2012 ) . So far , this model has not been experimentally validated , and the precise role of plant-associated CO2 as a host location cue by herbivores , in general , and root herbivores , in particular , remains unclear ( Eilers et al . , 2016 ) . To the best of our knowledge , no studies so far have investigated the role of plant-associated CO2 in plant–herbivore interactions in vivo using molecular manipulative approaches . The larvae of Diabrotica virgifera virgifera ( the western corn rootworm [WCR] ) feed almost exclusively on maize roots in agricultural settings and cause major yield losses in the US and Eastern Europe ( Ciosi et al . , 2008; Gray et al . , 2009; Meinke et al . , 2009; Toepfer et al . , 2015 ) . The larvae rely on a number of volatile and non-volatile chemicals to identify and locate host plants , and distinguish between suitable and less-suitable maize plants and forage within the maize root system ( Hiltpold et al . , 2013; Johnson and Gregory , 2006; Johnson and Nielsen , 2012; Robert et al . , 2012c; Schumann et al . , 2018 ) . Non-volatile primary metabolites such as sugars and fatty acids as well as secondary metabolites such as benzoxazinoids and phenolic acid conjugates modulate larval behaviour ( Bernklau et al . , 2011; Bernklau et al . , 2015; Bernklau et al . , 2016a; Bernklau et al . , 2018; Erb et al . , 2015; Hu et al . , 2018; Huang et al . , 2017; Machado et al . , 2021; Robert et al . , 2012c ) . Volatiles including ( E ) -β-caryophyllene , ethylene , and CO2 attract the larvae ( Bernklau and Bjostad , 1998a; Bernklau and Bjostad , 1998b; Robert et al . , 2012b; Robert et al . , 2012a ) , while methyl anthranilate repels them ( Bernklau et al . , 2016b ) . Based on the finding that high CO2 levels can outweigh the attractive effects of other maize volatiles , it was suggested that CO2 may be the only relevant volatile attractant for WCR larvae ( Bernklau and Bjostad , 1998b ) . However , under conditions where CO2 levels are similar , WCR larvae reliably choose between host plants of different suitability using other volatile cues ( Huang et al . , 2017; Lu et al . , 2016; Robert et al . , 2012b; Robert et al . , 2012a ) . The demonstrated ability of WCR larvae to respond to different volatile cues and the recent identification of putative CO2 receptors from transcriptomic data ( Rodrigues et al . , 2016 ) make this species a suitable model system to investigate the role of CO2 in plant–herbivore interactions . Ongoing efforts to use CO2 as a bait to control WCR in the field ( Bernklau et al . , 2004; Schumann et al . , 2014a; Schumann et al . , 2014b ) provide further motivation to assess the importance of this volatile for WCR foraging . To understand the importance of CO2 for WCR foraging in the soil , we manipulated the insect’s capacity to perceive CO2 . We reduced the expression levels of three putative WCR CO2 receptor-encoding genes through RNA interference ( RNAi ) , resulting in the identification of DvvGr2 as an essential gene for CO2 perception . Using DvvGr2-silenced larvae in combination with CO2 removal , we then assessed the importance of CO2 perception for WCR behaviour and foraging in olfactometers and soil arenas . Our experiments reveal how root-associated CO2 modulates the interaction between maize and its economically most damaging root pest and expand the current repertoire of potential adaptive explanations for the attraction of insect herbivores to CO2 . Plant-emitted CO2 may be used as a host location cue by root herbivores . To understand whether the presence of plant roots is associated with higher CO2 levels , we measured CO2 levels in the soil at different distances from young maize seedlings . We observed a significant CO2 gradient in the soil , with concentrations of 548–554 ppm in the rhizosphere , 506–515 ppm at distances between 8 and 16 cm from the plant , and 460–484 ppm at distances between 16 and 32 cm ( Figure 1A ) . At distances of more than 40 cm , CO2 levels levelled off at 425–439 ppm . We then removed the plants and remeasured CO2 levels 1 hr afterwards . In the absence of the plants , no CO2 gradient was observed , and CO2 concentrations in the soil were around 430 ppm ( Figure 1B ) . When surrounding soil was removed and seedling roots were washed , we observed 542 ± 6 . 74 ppm CO2 around the roots ( n = 3 ) . Thus , the release of CO2 from maize roots can account for the CO2 difference between soil trays with and without plants . This experiment shows that elevated CO2 levels derived from roots and probably from root-associated microorganisms are temporally and spatially associated with the presence of maize roots , and may thus be used as a host location cue by the WCR . To test this hypothesis , we identified CO2 receptors in WCR larvae , genetically impair their expression , and conducted a series of behavioural experiments as described below . To identify genes encoding putative CO2 receptors in WCR , we used known CO2 receptor-encoding gene sequences as queries against the WCR genome ( available from the National Center for Biotechnology Information [NCBI] ) . Three putative carbon dioxide receptor candidates , DvvGr1 , DvvGr2 , and DvvGr3 , were identified , matching three candidate genes that were found in previous transcriptome analyses ( Rodrigues et al . , 2016 ) . Phylogenetic reconstruction based on in silico-predicted protein sequences revealed orthologous relationships for the three WCR candidate receptors and the receptors of several other insects ( Figure 2A ) . Consistent with their taxonomy , we observed close homology between the protein sequences of the CO2 receptors of WCR and the protein sequences of other coleopteran insects such as Tribolium castaneum ( Figure 2A ) . Expression levels of DvvGr1 and DvvGr2 were found to be significantly higher in the head than in the rest of the body ( thorax and abdomen ) of second instar WCR larvae ( Figure 2B , C ) . No significant difference in expression was observed for DvvGr3 ( Figure 2D ) . Protein tertiary structure and topology models indicated that all three genes encode for 7-transmembrane domain proteins , which is consistent with their roles as receptors ( Figure 2B–D ) . To determine the importance of DvvGr1 , DvvGr2 , and DvvGr3 for the responsiveness of WCR larvae to CO2 , we knocked down the expression of each gene individually through double-stranded RNA ( dsRNA ) -mediated RNAi and conducted initial behavioural experiments with carbonated water as a CO2 source ( Figure 3 ) . Oral administration of dsRNA targeting either DvvGr1 , DvvGr2 , or DvvGr3 reduced the expression levels of these genes by 80% , 83% , and 66% compared to WCR larvae fed with dsRNA of the green fluorescent protein ( GFP ) gene ( herein referred to as wild type [WT] ) ( Figure 3A ) . All RNAi constructs were confirmed to be gene specific ( Figure 3A ) . Measurements within the olfactometers showed that CO2 levels were approximately 100 ppm higher in the arms of the L-shaped pots that contained plastic cups filled with carbonated water than in the arms of L-shaped pots that contained plastic cups filled with distilled water ( Figure 3B , Figure 3—figure supplement 1 ) . A higher proportion of WT larvae moved towards olfactometer arms with higher CO2 levels ( Figure 3C ) . Silencing DvvGr1 or DvvGr3 expression did not alter this preference . In contrast , DvvGr2-silenced larvae did not show preference for any olfactometer arm ( Figure 3C ) . To explore the role of DvvGr2 in different aspects of WCR behaviour , we conducted a series of additional experiments . First , we assessed the impact of silencing DvvGr2 on the capacity of WCR larvae to respond to other volatile and non-volatile host cues ( Figure 3D–G ) . DvvGr2-silenced larvae responded similarly to the repellent volatile methyl anthranilate as WT larvae ( Figure 3E ) . Responsiveness to non-volatile compounds such as Fe ( III ) ( DIMBOA ) 3 and a blend of glucose , fructose , and sucrose was also unaltered in DvvGr2-silenced larvae ( Figure 3F , G ) , demonstrating that knocking down DvvGr2 expression does not alter the capacity of WCR larvae to respond to other important chemical cues . Second , we assessed the contribution of DvvGr2 to CO2 responsiveness using synthetic CO2 at different concentrations ( Figure 4 , Figure 4—figure supplement 1 ) . WT larvae showed characteristic dose-dependent behavioural responses to CO2 . While they did not respond to 22 ppm CO2 above ambient CO2 levels , they were attracted to CO2 concentrations between 59 and 258 ppm above ambient and repelled by CO2-enriched air at 950 ppm above ambient CO2 levels and above ( Figure 4 ) . In contrast , DvvGr2-silenced larvae did not respond to CO2 enrichment at any of the tested concentrations ( Figure 4 ) . These experiments show that WCR larvae are attracted to CO2-enriched environments within the physiological range of the maize rhizosphere and that DvvGr2 silencing fully and specifically suppresses CO2 responsiveness in WCR larvae . To assess the impact of DvvGr2 on larval motility , we followed the trajectories of individual larvae in humid filter paper-lined Petri plates that were outfitted with a CO2 point releaser ( Figure 5 ) . WT larvae made frequent turns , but consistently oriented themselves towards the CO2 release point . Once they reached the CO2 release point , they stopped moving ( Figure 5A ) . DvvGr2-silenced larvae exhibited similar turning behaviour as WT larvae , but did not move towards the CO2 release point ( Figure 5B ) . WT larvae spend more time on CO2 release point than DvvGr2-silenced larvae ( Figure 5C ) . During the movement phase , the mean speed of WT larvae and DvvGr2-silenced larvae was similar ( Figure 5C ) , but the distance covered by DvvGr2-silenced larvae was higher , as they did not stop at the CO2 release point . In a second experiment , we followed the trajectories of individual larvae in Petri plates with maize roots ( Figure 5D , E ) . Speed and distance covered were similar between WT and DvvGr2-silenced larvae ( Figure 5F ) . Surprisingly , both WT and DvvGr2-silenced larvae oriented themselves towards the maize roots and reached the maize roots after a similar amount of time ( Figure 5F ) . This result shows that DvvGr2 expression is required for the location and detection of CO2 , but does not influence WCR motility nor its ability to locate maize roots over short distances ( i . e . , <9 cm ) . To further explore the role of plant-associated CO2 and DvvGr2 in volatile-mediated host location , we performed a series of olfactometer experiments with maize plants grown in sand on one side and sand only on the other side . We tested attraction at two distances , 9 and 18 cm , from the volatile sources and the release points of the larvae ( Figure 6 ) . We also manipulated the diffusion of CO2 into the arms of a subset of olfactometers by adding a layer of CO2-absorbing soda lime into the olfactometer arms . CO2 measurements revealed that the presence of a host root system increased CO2 concentrations by approximately 100 ppm above ambient CO2 levels in the corresponding olfactometer arm ( Figure 6 , Figure 6—figure supplement 1 ) . The soda lime reduced ambient CO2 concentrations in the olfactometer arms by approximately 100 ppm and equalized CO2 concentrations between arms with and without a host plant ( Figure 6 ) . The diffusion of other maize root volatiles was not affected by the soda lime ( Figure 6—figure supplement 2 ) , thus validating the CO2 scrubbing approach . Larvae did not have direct access to the plant , the plant growth medium , or the soda lime , and received no visual cues , and thus had to rely on host plant volatiles for orientation . When released at distance of 9 cm from the volatile sources , both WT and DvvGr2-silenced larvae showed a clear preference for the olfactometer arms leading to host plants ( Figure 6A ) . This preference was still intact in olfactometers outfitted with soda lime , showing that volatiles other than CO2 are sufficient for volatile-mediated host location at a short distance . At a distance of 18 cm from the volatile sources , WT larvae showed a similarly strong preference for arms leading to host plants ( Figure 6B ) . By contrast , DvvGr2-silenced larvae did not exhibit any preference ( Figure 6B ) . In the presence of soda lime , neither WT nor DvvGr2-silenced larvae were attracted to arms with a host plant ( Figure 6B ) . Taken together , these experiments provide strong support for the hypothesis that WCR larvae use plant-associated CO2 to locate host plants over distances greater than 9 cm in a DvvGr2-dependent manner . WCR larvae can move up to 1 m in the soil . Second and third instar larvae in particular are known to move between maize plants across rows in maize fields ( Hibbard et al . , 2003 ) . To test whether DvvGr2-mediated CO2 responsiveness mediates host location over longer distances in a soil context , we planted maize plants in soil-filled plastic trays , released WCR larvae at distances of 16 , 32 , 48 , or 64 cm from the maize plants , and evaluated larval positions after 8 hr ( Figure 7 ) . This time point was chosen based on preliminary observations showing that larvae take approximately 8 hr to cross the soil arenas . Direct access to the roots was impeded by using volatile-permeable fabrics , referred to hereby as root barriers . The CO2 emitted by maize roots formed a gradient in the soil , starting at about 506 ppm in the rhizosphere ( zone 1 ) and 430 ppm at distances of 16–32 cm from the plant ( zone 2 ) ( Figure 7B , Figure 7—figure supplement 1 ) . At distances of more than 32 cm from the plant , the CO2 levels were around 400 ppm and statistically indistinguishable from soil without plants or ambient air ( Figure 7—figure supplement 1 ) . To confirm that larval motility is not altered by DvvGr2 silencing in a soil context , we first released WT and DvvGr2-silenced larvae into the middle of a set of arenas without a host plant and evaluated larval positions after 8 hr . We found that the larvae dispersed equally across the arenas , without any difference between WT and DvvGr2-silenced larvae ( Figure 7A ) . Eight hours after releasing the larvae into arenas that included host plants on one side , 53% of WT larvae that were released at 64 cm from the plant were retrieved close to the maize rhizosphere , that is , in zone 1 ( Figure 7C ) . In contrast , only 33% of the DvvGr2-silenced larvae that were released at the same distance were recovered from the maize rhizosphere ( Figure 7C ) . Significantly more DvvGr2-silenced larvae were recovered further away from the plants , in zones 3 and 4 ( Figure 7C ) . The number of WT and DvvGr2-silenced WCR larvae found close to the host plant increased with decreasing release distance , as did the difference between WT and DvvGr2-silenced larvae ( Figure 7C–F ) . At a release distance of 16 cm , only slightly more WT than DvvGr2-silenced larvae were found close to the plant roots ( Figure 7F ) . To further confirm the role of DvvGr2 in mediating host plant location over long distances in the soil , we performed a time-course experiment where we released WT and DvvGr2-silenced larvae in zone 5 ( 64 cm away from the host plant ) and then recorded how rapidly they reached zone 1 containing host plants ( Figure 7—figure supplement 2 ) . The capacity of the larvae to directly feed on the host roots was impeded using a volatile-permeable root barrier ( Figure 7—figure supplement 2 ) . Within 10 hr , 36% of the released WT larvae were found in zone 1 , and within 32 hr , this number had increased to 90% ( Figure 7—figure supplement 2 ) . By contrast , only 24% of the released DvvGr2-silenced larvae were found in zone 1 after 10 hr , and after 32 hr , this value had only increased to 56% ( Figure 7—figure supplement 2 ) . Thus , the capacity to detect CO2 gradients contributes to successful host location by WCR larvae in a distance-specific manner in the soil . While larvae released at a distance equal to or below 32 cm from the host plant ( zones 2–3 ) can use CO2 directly as a host location cue , larvae released at greater distances likely move randomly before reaching zones with plant-associated CO2 gradients . Plant nutritional status determines plant growth and defence , and can thus modulate plant–herbivore interactions ( Wetzel et al . , 2016 ) . To test for a possible connection between plant nutritional status , host suitability , and CO2-dependent herbivore attraction , we varied the nutrient supply of maize plants and then carried out CO2 measurements , and behavioural and insect performance experiments ( Figure 8 ) . To exclude direct or soil-mediated effects of fertilization , plants were first grown under different fertilization regimes and then , prior to experiments , harvested , washed , and replanted . Higher CO2 levels were observed close to the roots of plants that were well fertilized compared to the levels that were observed close to the roots of plants that received medium ( 50% of optimally fertilized plants ) or low ( 10% of optimally fertilized plants ) fertilizer doses ( Figure 8A , B , Figure 8—figure supplement 1A , B ) . As observed before , soil CO2 levels decreased with increasing distance from the plants and were lowest in the middle of the experimental trays ( Figure 8A , B ) . In choice experiments with maize plants planted approximately 50 cm apart , which corresponds to row spacing used for high planting densities in maize cultivation , WT larvae showed a significant preference for well-fertilized over medium- or low-fertilized plants ( Figure 8C , D ) . DvvGr2-silenced larvae did not show any preference ( Figure 8C , D ) . In no-choice experiments , WCR larvae gained most weight on washed roots of well-fertilized maize plants than on washed roots of plants treated with medium or low doses of fertilizer ( Figure 8E ) . Hence , intact CO2 perception allows WCR larvae to locate suitable host plants at agriculturally relevant distances , which may result in specific insect distribution patterns in heterogeneous environments . In this study , we conducted gene sequence similarity analyses , phylogenetic relationship reconstructions , RNA interference , and behavioural experiments to explore the biological relevance of root-associated CO2 for plant–herbivore interactions . We found that the WCR genome contains at least three putative CO2 receptor-encoding genes: DvvGr1 , DvvGr2 , and DvvGr3 , which is consistent with previous transcriptomic-based studies ( Rodrigues et al . , 2016 ) . Protein tertiary structure and topology prediction models show that the identified genes code for proteins that contain seven transmembrane domains , which is consistent with the protein topology of gustatory and olfactory receptors ( Dahanukar et al . , 2005; Hallem et al . , 2006 ) . Larval behaviour and gene silencing based-functional characterization of the three identified WCR putative CO2 receptor genes revealed that the intact expression of DvvGr2 is essential for the attractive effects of CO2 to WCR larvae . Knocking down DvvGr2 rendered larvae fully unresponsive to synthetic and plant-associated CO2 without impairing responses to other stimuli or affecting search behaviour and motility . In Aedes aegypti , Helicoverpa armigera , and Drosophila melanogaster , both carbon dioxide receptors Gr1 and Gr3 are required for CO2 detection ( Erdelyan et al . , 2012; Jones et al . , 2007; Kwon et al . , 2007; McMeniman et al . , 2014; Ning et al . , 2016; Suh et al . , 2004 ) . In Culex quinquefasciatus , both Gr2 and Gr3 carbon dioxide receptors are required , while Gr1 acts as a modulator ( Xu et al . , 2020 ) . In A . aegypti , the involvement of Gr2 in carbon dioxide responsiveness is still under debate ( Erdelyan et al . , 2012; Kumar et al . , 2019 ) . Taken together , the molecular elements required for carbon dioxide perception may be species-specific . Our results support this notion as DvvGr2 , but not DvvGr1 and DvvGr3 , are crucial for CO2 responsiveness . The role of DvvGr1 and DvvGr3 for WCR remains to be determined , but their presence and expression may hint at additional complexity in developmental and/or tissue-specific patterns of CO2 responsiveness in this species . Despite the inability of DvvGr2-silenced WCR larvae to respond to differences in CO2 levels , the larvae were still able to orient towards maize roots at short distances of 8–10 cm . Olfactometer experiments in combination with CO2 removal demonstrate that other volatile cues can be used by WCR larvae to locate maize plants at distances shorter than 9 cm . Earlier studies found that ( E ) -β-caryophyllene , which is emitted from the roots of certain maize genotypes when they are attacked by root herbivores , attracts second and third instar WCR larvae and allows them to aggregate on maize plants and thereby enhance their fitness ( Robert et al . , 2012b ) , while neonate larvae are not attracted to this volatile ( Hiltpold and Hibbard , 2016 ) . Ethylene has also been shown to attract WCR larvae ( Robert et al . , 2012a ) , and MBOA or its breakdown products have also been proposed as volatile attractants ( Bjostad and Hibbard , 1992 ) . Methyl anthranilate , on the other hand , has been shown to repel WCR larvae ( Bernklau et al . , 2016b; Bernklau et al . , 2019 ) . Many other leaf- and root-feeding herbivores are known to respond to plant volatiles other than CO2 ( Bruce et al . , 2005 ) . Given the low reliability of CO2 as a host-specific cue , it is probably not surprising that WCR , as a highly specialized maize feeder , can use other volatile cues to locate host plants . Integrating other volatile cues likely allows WCR larvae to locate maize plants even in the absence of reliable CO2 gradients in the soil , thus increasing the robustness of its foraging behaviour at short distances . An intriguing result in this context is the fact that WCR larvae show the same efficiency in locating maize roots at short distances in the absence of a CO2 gradient , suggesting that this volatile may not play a role as a cue at close range . Although intact CO2 perception was not required for host location at short distances , it had a strong impact on the capacity of WCR larvae to reach the maize rhizosphere at long distances . A gradient of plant-associated CO2 was detected at distances of up to 32 cm from the plant . When WCR larvae were released at distances greater than 32 cm , they still managed to locate plants in a DvvGr2-dependent manner . This result can be explained by random movement , where the larvae move randomly until they encounter a CO2 gradient , or by localized CO2 gradients along preferential gas-phase pathways that may extend beyond 32 cm , or a combination of both . The advantages of CO2 as a host location cue are that it is abundantly produced through respiration by most organisms , is relatively stable ( Jones and Coaker , 1977; Li et al . , 2016 ) , and diffuses rapidly in air , water , and soil ( Hashimoto and Suzuki , 2002; Ma et al . , 2013 ) . CO2 may thus be a suitable long-range cue to locate organisms with high respiratory rates , such as mammals and heterotroph plant parts , including roots and their associated microbial communities ( Johnson and Nielsen , 2012 ) . Aboveground insects can be attracted to CO2 traps located as far away as 10 m , and it is estimated that this distance could even be as long as 60 m under optimal environmental circumstances ( Guerenstein and Hildebrand , 2008; Zollner et al . , 2004 ) . For belowground insects , this distance is hypothesized to be within the lower centimetre range as CO2 diffusion is substantially decreased within the soil matrix compared to CO2 diffusion in air ( Bernklau et al . , 2005; Doane et al . , 1975; Doane and Klingler , 1978; Klingler , 1966 ) . Other volatiles that are less abundant and diffuse even less well through the soil such as ( E ) -β-caryophyllene are unlikely to be detectable at distances of more than 10 cm ( Chiriboga M . et al . , 2017; Hiltpold and Turlings , 2008 ) . These volatiles are thus likely useful host location cues at short , but not long , distances in the soil . The finding that WCR integrates CO2 perception with other environmental cues and that attraction to CO2 is context dependent is in line with patterns reported for other insects such as mosquitoes , whose response to stimuli such as colour , temperature , and human body odours is enhanced by CO2 ( McMeniman et al . , 2014; van Breugel et al . , 2015 ) , and pollinating hawkmoths , which use CO2 as a redundant volatile distance stimulus in a sex-specific manner ( Goyret et al . , 2008 ) . A recent study shows that a CO2 receptor in Drosophila flies is also involved in the detection and behavioural responses to other volatiles ( MacWilliam et al . , 2018 ) . We observed that DvvGr2-silenced larvae were repelled by methyl anthranilate , a potent maize root repellent , to a similar extent as WT larvae , suggesting that their sensitivity to this plant volatile is unchanged ( Bernklau et al . , 2016b ) . In Drosophila flies , the CO2 receptor Gr63a is required for spermidine attractiveness over short time spans , that is , less than 1 min , but not over longer time spans ( hours ) , when other receptors likely become more important ( MacWilliam et al . , 2018 ) . In the present experiments , WCR behaviour was evaluated after one or more hours . The CO2 scrubber experiment provides further evidence that the foraging patterns observed in this study are not due to different sensitivity of DvvGr2-silenced larvae to other root volatiles . Apart from acting as a long-distance host location cue , CO2 also links plant fertilization to herbivore behaviour by guiding WCR to well-fertilized plants . As WCR larvae are resistant to root defences of maize ( Robert et al . , 2012b ) , it is likely to benefit from increased fertilization , independently of the plant’s defensive status . As the plant nutritional status and host quality for WCR larvae are associated with higher CO2 release from the roots , following the highest concentrations of CO2 in the soil may be adaptive for the herbivore as it may increase its chance not only to find a maize plant per se but also to identify a plant that has the resources to grow vigorously and that is a better host . More experiments are needed to confirm this hypothesis as in the current set-up the larvae may have followed the only available CO2 gradient close to their release point rather than having made a choice between two gradients . However , given the dose-dependent responses of WCR , preferential orientation towards plants surrounded by higher CO2 levels appears likely . Well-fertilized maize plants increase photosynthesis and biomass production , which results in higher CO2 release from the roots ( Zhu and Lynch , 2004 ) . WCR larvae are specialized maize pests that have evolved with intense maize cultivation in the corn belt of the US ( Gray et al . , 2009 ) and are resistant to maize defence metabolites ( Robert et al . , 2012b ) . Following the strongest CO2 gradient in an equally spaced maize monoculture may indeed be a useful strategy for this root feeder to locate suitable food sources . An association between CO2 emission and food-source profitability was also suggested for Datura flowers , which emit the highest level of CO2 in times when nectar is most abundant ( Guerenstein et al . , 2004; Thom et al . , 2004 ) . These findings support the general hypothesis that CO2 is a marker of metabolic activity that allows for an assessment of the vigour and profitability of a wide variety of hosts . The impact of CO2 for the distribution of root herbivores such as the WCR in heterogeneous environments remains to be determined . Based on our results , we expect plant-associated CO2 to contribute to uneven herbivore distribution and to aggregation on plants with a good nutritional status within monocultures . In summary , this work demonstrates how a herbivore uses its capacity to perceive CO2 to locate host plants . Volatiles other than CO2 are also integrated into host-finding behaviour in the soil , but their effects are more important at short than at long distances . Random movement in the soil may help this root herbivore to increase its capacity to find host cues at even greater distances . Thus , evidence is now accumulating that CO2 acts as an important host location cue in different insects , likely because of its unique role as a highly conserved long-range marker of metabolic activity within complex sensory landscapes . Maize seeds ( Zea mays L . , var . Akku ) were provided by Delley Semences et Plantes SA ( Delley , Switzerland ) . Seedlings were grown under greenhouse conditions ( 23 ± 2°C , 60% relative humidity , 16:8 h L/D , and 250 mmol/m2/s1 additional light supplied by sodium lamps ) . Plantaaktiv 16+6+26 Typ K fertilizer ( Hauert HBG Dünger AG , Grossaffoltern , Switzerland ) was added twice a week after plant emergence following the manufacturer’s recommendations . The composition of the fertilizer is: total nitrogen ( N ) 16% , nitrate 11% , ammonium 5% , phosphate ( P2O5 ) 6% , potassium oxide ( K2O ) 26% , magnesium oxide ( MgO ) 3 . 3% , boron ( B ) 0 . 02% , copper ( Cu , EDTA-chelated ) 0 . 04% , iron ( Fe , EDTA-chelated ) 0 . 1% , manganese ( Mn , EDTA-chelated ) 0 . 05% , molybdenum ( Mo ) 0 . 01% , and zinc ( Zn , EDTA-chelated ) 0 . 01% . When plants were used as insect food , seedlings were germinated in vermiculite ( particle size: 2–4 mm; tabaksamen , Switzerland ) and used within 4 days after germination . Diabrotica virgifera virgifera ( WCR ) insects used in this study were derived from a non-diapausing colony reared at the University of Neuchâtel . The eggs used to establish the colony were supplied by USDA-ARS-NCARL , Brookings , SD . New insects of the same origin are introduced into the colony every 3–6 months . Upon hatching , insects were maintained in organic soil ( Selmaterra , Bigler Samen AG , Thun , Switzerland ) and fed freshly germinated maize seedlings ( var . Akku ) . To identify CO2 receptor orthologues in WCR , we used CO2 receptor-encoding gene sequences of T . castaneum and several sequences from other insects as queries against publicly available WCR genome sequences ( NCBI accession: PXJM00000000 . 2 ) using the National Center for Biotechnology Information Basic Local Alignment Search Tool ( NCBI BLAST ) ( Robertson and Kent , 2009; Wang et al . , 2013; Xu and Anderson , 2015 ) . The full gene sequences can be retrieved from the NCBI databank using the following accession numbers: XM_028276483 . 1 ( DvvGr1 ) , XM_028280521 . 1 ( DvvGr2 ) , and XM_028272033 . 1 ( DvvGr3 ) . These gene sequences were translated to obtain protein sequences . The obtained protein sequences and the protein sequences of CO2 receptors from different insects were used to infer evolutionary relationships using the neighbor-joining method in MEGA 7 ( Kumar et al . , 2016; Robertson and Kent , 2009; Rodrigues et al . , 2016; Saitou and Nei , 1987 ) . The optimal tree with the sum of branch length = 4 . 44068889 is provided in Figure 2A . The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test ( 100 replicates ) are shown next to the branches ( Felsenstein , 1985 ) . The tree is drawn to scale , with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic tree . The evolutionary distances were computed using the Poisson correction method ( Zuckerkandl and Pauling , 1965 ) and are in the units of the number of amino acid substitutions per site . A total of 242 amino acid positions were included in the final data set . Graphical representation and edition of the phylogenetic tree were performed with the Interactive Tree of Life ( version 3 . 5 . 1 ) ( Letunic and Bork , 2016 ) . Protein tertiary structures and topologies were predicted using Phyre2 ( Kelley et al . , 2015 ) . Escherichia coli HT115 were transformed with recombinant L4440 plasmids that contained a 211–240 bp long gene fragment targeting one of the three CO2 receptors . Cloned nucleotide sequences were synthetized de novo ( Eurofins , Germany ) . To induce the production of dsRNA , an overnight bacterial culture was used to inoculate fresh Luria–Berthani broth ( 25 g/L , Luria/Miller , Carl Roth GmbH , Karlsruhe , Germany ) . Once the bacterial culture reached an OD600 of 0 . 6–0 . 8 , it was supplemented with isopropyl β-D-1-thiogalactopyranoside ( Sigma-Aldrich , Switzerland ) at a final concentration of 2 mM . Bacterial cultures were incubated at 37°C in an orbital shaker ( Ecotron , Infors HT , Bottmingen , Switzerland ) at 130 rpm for 16 additional hours . Bacteria were harvested by centrifugation ( 2000 rpm , 10 min ) using a top bench centrifuge ( IEC Centra GP6R , Thermo Fisher Scientific , Waltham , MA , USA ) and stored at −20°C in a freezer ( Bosch , Gerlingen , Germany ) for further use ( Kim et al . , 2015 ) . To induce gene silencing in WCR , 6–10 second instar WCR larvae were released in solo cups ( 30 ml , Frontier Scientific Services , Inc , Germany ) containing approximately 2 g of autoclaved soil ( Selmaterra , Bigler Samen AG , Thun , Switzerland ) and 2–3 freshly germinated maize seedlings . Maize seedlings were coated with 1 ml of bacterial solution containing approximately 200–500 ng of dsRNA targeting the different CO2 receptor genes . As controls , larvae were fed with bacteria-producing dsRNA-targeting GFP genes , which are absent in the WCR genome ( Rodrigues et al . , 2016 ) . dsRNA was produced as described above . Fresh bacteria and seedlings were added to solo cups every other day for three consecutive times . Two days after the last dsRNA/bacteria application , larvae were collected and used for experiments . Total RNA was isolated from approximately 10 mg of frozen , ground , and homogenized WCR larval tissue ( 3–7 larvae per biological replicate , n = 8 ) using the GenElute Universal Total RNA Purification Kit ( Sigma-Aldrich , St . Louis , MO , USA ) . A NanoDrop spectrophotometer ( ND-1000 , Thermo Fisher Scientific , Waltham , MA , USA ) was used to estimate RNA purity and quantity . DNase-treated RNA was used as template for reverse transcription and first-strand cDNA synthesis with PrimeScript Reverse Transcriptase ( Takara Bio Inc , Kusatsu , Japan ) . DNase treatment was carried out using the gDNA Eraser ( Perfect Real Time ) following manufacturer’s instructions ( Takara Bio Inc ) . For gene expression analysis , 2 μl of undiluted cDNA ( i . e . , the equivalent of 100 ng total RNA ) served as template in a 20 μl qRT-PCR using the TB Green Premix Ex Taq II ( Tli RNaseH Plus ) kit ( Takara Bio Inc ) and the Roche LightCycler 96 system ( Roche , Basel , Switzerland ) , according to manufacturer’s instructions . Transcript abundances of the following WCR genes were analysed: DvvGr1 , DvvGr2 , and DvvGr3 ( Rodrigues et al . , 2016 ) . Actin was used as reference gene to normalize expression data across samples . Relative gene expression levels were calculated by the 2-ΔΔCt method ( Livak and Schmittgen , 2001 ) . The following primers were used: DvvGr1-F CGTTAATTTAGCTGCTGTGG , DvvGr1-R GTTTTCTGTTGCTAGAGTTGC , DvvGr2-F GAACTAAGCGAGCTCCTCCA , DvvGr2-R CAGAAGCACCATGCAATACG , DvvGr3-F GCAACGCTTTCAGCTTTACC , DvvGr3-R GTGCATCGTCATTCATCCAG , DvvActin-F TCCAGGCTGTACTCTCCTTG , and DvvActin-R CAAGTCCAAACGAAGGATTG . CO2 was quantified by an infrared CO2 gas analyser or by gas chromatography coupled to a flame ionization detector ( GC-FID ) . In the first case , air samples were collected by a micropump ( Intelligent Subsampler TR-SS3 , Sable Systems International , Las Vegas , NV , USA ) connected to an airstream selector ( RM8 Intelligent Multiplexer , V5 , Sable Systems International , Las Vegas , NV , USA ) controlled by a computer via a Universal Interface ( UI-2 ) and the Expedata software version 1 . 2 . 6 ( Sable Systems International , Las Vegas , NV , USA ) . The sampled air passed through an infrared CO2 gas analyser ( LI 7000 , Li-Cor Inc , Lincoln , NE , USA ) ( Frei et al . , 2017 ) . In the second case , air samples were collected by using a syringe equipped with a Luer connector , a stopcock valve , and a needle . Also , 3 ml of air samples were immediately injected and CO2 was analysed by a GC-FID ( Shimadzu GC-8 ) equipped with a methanizer ( VWR , Radnor , PA , USA ) and a Poropack N column . Nitrogen was used as carrier gas ( 400 kPa ) . After injection , the column temperature was maintained at 100°C for 1 . 3 min , and then increased to 130°C at a rate of 30°C/min and maintained at this temperature for 4 min . The FID temperature was set at 250°C . To measure root volatiles , three 4-day-old maize seedling were transplanted into moist white sand ( Migros , Switzerland ) in spherical glass pots ( 7 cm diameter , Verre and Quartz Technique SA , Neuchâtel , Switzerland ) . To boost volatile release , the roots were damaged mechanically before transplantation by briefly twisting them . The pots were wrapped in aluminium foil . Clean humidified air was pushed through the pots at a rate of 1 l·min−1 and pulled through Porapak filters ( 25 mg of Porapak adsorbent , 80–100 mesh; Alltech Assoc . , Deerfield , IL , USA ) at a rate of 0 . 6 L·min−1 . Root volatiles were collected over 6 hr . After this period , the filters were eluted with 150 μl of dichloromethane , and N-octane and nonyl-acetate ( Sigma , Buchs , Switzerland ) were further added as internal standards ( 200 ng in 10 μl dichloromethane ) . The root volatiles were analysed by gas chromatography coupled to mass spectrometry ( Agilent 7820A GC coupled to an Agilent 5977E MS , Agilent Technologies , Santa Clara , CA , USA ) . The aliquot was injected in the injector port ( 230°C ) and pulsed in a spitless mode onto an apolar column ( HP-5MS 5% Phenyl Methyl Silox , 30 m × 250 μm internal diameter × 0 . 25 μm film thickness , J&W Scientific , Agilent Technologies SA , Basel , Switzerland ) . Helium at a constant flow of 1 ml·min−1 ( constant pressure 8 . 2317 psi ) was used as carrier gas . After injection , the column temperature was maintained at 40°C for 3 . 5 min , and then increased to 100°C at a rate of 8°C/min and subsequently at 5°C/min to 230°C , followed by a post run of 3 min at 250°C . Volatile identification was obtained by comparing mass spectra with those of the NIST17 Mass Spectra Library , and relative quantities for the major compounds were calculated based on the peak areas of the internal standards . To determine whether silencing putative CO2 receptor genes impairs the ability of WCR larvae to behaviourally respond to CO2 , we silenced DvvGr1 , DvvGr2 , and DvvGr3 as described above and evaluated larvae responses to CO2 in dual-choice experiments using belowground olfactometers . Larvae that were fed bacteria that express dsRNA that targets GFP were used as controls ( herein referred to as wild type larvae; WT ) . The belowground olfactometers consist of two L-shaped glass pots ( 5 cm diameter , 11 cm deep ) connected to a detachable central glass tube ( 24–29 mm diameter , 8 cm in length ) by detachable Teflon connectors ( Verre and Quartz Technique SA , Neuchâtel , Switzerland ) ( Figure 2B ) . The Teflon connectors contained wire mesh screens ( 2300 mesh , Small Parts Inc , Miami Lakes , FL , USA ) to restrain the larvae from moving into the plants . The central glass tubes remained empty to only allow volatile compounds to diffuse through the central glass tubes . The central glass tubes have an access port in the middle to allow the release of insects ( Figure 2B ) . Insects can freely move inside the central glass tube ( 8 cm in length ) and reach the metal wire screens . For further technical specifications regarding the belowground olfactometers , refer to Rasmann et al . , 2005 and Robert et al . , 2012a . To increase CO2 levels in one side of the olfactometers , we used carbonated water as a CO2 source ( Bernklau and Bjostad , 1998a; Huang et al . , 2017; Jewett and Bjostad , 1996 ) . For this , a plastic cup containing 50 ml of carbonated water ( Valais , Aproz Sources Minérales , Aproz , Switzerland ) was placed in one L-shaped glass pot and a plastic cup containing 50 ml of distilled water was placed in the opposite pot . L-shaped glass pots did not contain any substrate to allow CO2 to freely diffuse into the central glass tubes ( Figure 2B ) . Ten minutes after placing the cups into the L-shaped glass pots , L-shaped glass pots were connected to the central glass tubes and six larvae were released in the middle of the olfactometer ( red arrow , Figure 2B ) . Larval positions were recorded 1 hr after their release . Insect preference for a given treatment was considered when the larvae were found at a distance of 1 cm or less from the odour source; this is 1 cm apart from the wire mesh . Seven olfactometers per larval type and experiment were assessed . Olfactometers were covered with aluminium foil to reduce light disturbance to the larvae . Aluminium foil was removed shortly before evaluating larval positions . The experiment was repeated twice . CO2 levels on each arm of the olfactometer were measured by GC-FID as described above . Air samples to determine CO2 concentrations were collected by using a syringe equipped with a Luer connector , stopcock valve , and needle . Prior to sampling , we first connected the Teflon connectors to the L-shaped glass pots and closeed them with parafilm . Then , we placed a plastic cup containing either 50 ml of carbonated water or 50 ml of distilled water . Ten minutes after , we pierced the parafilm with the needle and collected 3 ml of air samples using the syringe and immediately injected the samples to the GC-FID for CO2 measurements . To determine whether silencing the DvvGr2 carbon dioxide receptor affects larval responses to a plant volatile other than CO2 , we evaluated larval responses to methyl anthranilate . For this , we evaluated insect preferences for seedling roots or for seedling roots placed next to a filter paper disc treated with methyl anthranilate following a similar experimental procedure as described by Bernklau et al . , 2019 . To this end , either five 2nd–3rd instar WT WCR larvae or five 2nd–3rd instar DvvGr2-silenced WCR larvae were released in the middle of a moist sand-filled Petri plate ( 9 cm diameter , Greiner Bio-One GmbH , Frickenhausen , DE ) where they encountered two 3-day-old maize seedling roots in one side or two 3-day-old maize seedlings roots and a filter paper disc ( 0 . 5 cm diameter , Whatman no . 1 , GE Healthcare Life Sciences , UK ) treated with 10 µl of methyl anthranilate solution ( 10 mg/ml of water ) in the opposite side . Methyl anthranilate was purchased from Sigma ( CAS: 134-20-3; Sigma Aldrich Chemie , Switzerland ) . Sand layers were 3–4 mm high and allowed the larvae to move freely in and on the substrate . Ten Petri plates per larval type with five larvae each were evaluated ( n = 10 ) . Petri plates were covered with black plastic sheets to avoid light disturbance to the insects . Larval positions were recorded 1 hr after releasing the larvae . Insect preference for a given treatment was considered when the larvae were found on the roots or in contact with the filter paper discs . To determine whether silencing the DvvGr2 carbon dioxide receptor affects larval responses to plant metabolites other than CO2 , we evaluated larval responses to Fe ( III ) ( DIMBOA ) 3 . For this , we evaluated insect preferences for filter paper discs impregnated with Fe ( III ) ( DIMBOA ) 3 or for filter paper discs treated with water following the procedure described by Hu et al . , 2018 with minor modifications . To this end , we released either six 2nd–3rd instar WT WCR larvae or six 2nd–3rd instar DvvGr2-silenced WCR larvae in the middle of a moist sand-filled Petri plate ( 6 cm diameter , Greiner Bio-One GmbH , Frickenhausen , DE ) where they encountered a filter paper disc ( 0 . 5 cm diameter , Whatman no . 1 , GE Healthcare Life Sciences , UK ) treated with 10 μl of Fe ( III ) ( DIMBOA ) 3 ( 1 µg/ml of water ) or , on the opposite side , a filter paper disc treated with water only . Twenty Petri plates with six larvae each were evaluated ( n = 20 ) . Fe ( III ) ( DIMBOA ) 3 was prepared fresh by mixing FeCl3 and DIMBOA at a 1:2 ratio as described by Hu et al . , 2018 . Petri plates were covered with black plastic sheets to avoid light disturbance to the insects . Larval preferences were recorded 1 hr after releasing the larvae . Insect preference for a given treatment was considered when the larvae were found in contact with the filter paper discs . To determine whether silencing the DvvGr2 carbon dioxide receptor affects larval responses to plant metabolites other than CO2 , we evaluated larval responses to soluble sugars . For this , we evaluated insect preferences for a mixture of glucose , fructose , and sucrose following the procedure described by Bernklau et al . , 2018 with minor modifications . Briefly , we released either six 2nd–3 instar WT WCR larvae or six 2nd–3rd instar DvvGr2-silenced WCR larvae in the middle of a moist sand-filled Petri plate ( 6 cm diameter , Greiner Bio-One GmbH , Frickenhausen , DE ) where they encountered a filter paper disc ( 0 . 5 cm diameter , Whatman no . 1 , GE Healthcare Life Sciences , UK ) treated with 10 μl of a mixture of glucose , fructose , and sucrose ( 30 mg/ml of each sugar ) , or , on the opposite side , a filter paper treated with water only . Twenty Petri plates with six larvae each were assayed ( n = 20 ) . Petri plates were covered with black plastic sheets to avoid light disturbance to the insects . Larval preferences were recorded 3 hr after release . Insect preference for a given treatment was considered when the larvae were found in contact with the filter paper discs . To determine whether silencing the DvvGr2 carbon dioxide receptor affects larval responses to CO2 and to determine the range of behaviourally active CO2 concentrations , we evaluated larval responses to different concentrations of CO2 in dual-choice experiments using belowground olfactometers . CO2 levels were increased in one side of the olfactometer by delivering CO2-enriched synthetic air ( 1% CO2 , Carbagas , Switzerland ) . For this , the L-shaped glass pots were closed on top using parafilm during CO2 delivery . A manometer connected to the synthetic air bottle allowed to fine-tune CO2 delivery rates and concentrations . CO2 levels were measured using a gas analyser ( Li7000 , Li-Cor Inc , Lincoln , NE , USA ) as described above . CO2 levels were increased at different levels from 22 to 1832 ppm above ambient CO2 levels . Once the desired CO2 concentrations were reached , CO2 delivery was terminated , and the olfactometers were assembled by connecting two L-shaped glass pots to a central glass tube as described above . Immediately after this , either six 2nd–3rd instar WT WCR larvae or six 2nd–3rd instar DvvGr2-silenced WCR larvae were released in the middle of the central glass tubes . Larval positions were evaluated within 10 min of release . Experiments were conducted in a dark room to reduce light disturbance to the larvae . Red-light headlamps were used by the experimenters during the experiment . Insect preference for a given treatment was considered when the larvae were found at a distance of 1 cm or less from the odour source; this is 1 cm apart from the wire mesh . Three olfactometers per larval type and six larvae per olfactometer were assayed ( n = 3 ) . To determine whether silencing the DvvGr2 carbon dioxide receptor affects larval motility and speed , larval behaviour and the trajectories followed by individual larvae in open Petri plates that contained either maize root pieces or that were outfitted with a CO2 point releaser in the middle were evaluated . In the first experiment , two root pieces ( 3–4 cm long ) of 4-day-old maize seedlings were placed at the rim of a Petri plate lined with moist filter paper . Then , on the opposite rim ( i . e . , 9 cm apart ) , either one 2nd–3rd instar WT WCR larvae or one 2nd–3rd instar DvvGr2-silenced WCR larvae was released . The trajectories followed by the insects and the time required to reach the roots were evaluated . The trajectories followed by the insects were drawn on circular pieces of papers of 9 cm diameter . Six Petri plates with one larva each were assayed ( n = 6 ) . In the second experiment , CO2 point releasers were installed in the centre of Petri plates ( 9 cm diameter , Greiner Bio-One , Austria ) . For this , Petri plates were pierced with a hot metal needle . Then , another needle that released CO2 at 581 ppm , resulting in CO2 concentrations 60 ppm above ambient CO2 levels , was inserted in the resulting whole . CO2 levels were adjusted using a manometer connected to the synthetic air bottle . CO2 levels were measured using a gas analyser ( Li7000 , Li-Cor Inc , Lincoln , NE , USA ) as described above . Once the desired CO2 concentration was reached , either one 2nd–3rd instar WT WCR larva or one 2nd–3rd instar DvvGr2-silenced WCR larva was released at the rim of the Petri plates ( i . e . , 4 . 5 cm apart from the CO2 point releaser ) . Petri plates were lined with moist filter paper ( 9 cm diameter , GE Healthcare , UK ) . Insect behaviour was observed for 3 min , the trajectories followed by the insects during this time were drawn on circular pieces of papers of 9 cm diameter , and the time spent in close contact with the CO2 point releaser was quantified . Six Petri plates with one larva each were assayed ( n = 6 ) . Both experiments were conducted in a dark room to reduce light disturbance to the larvae . Red-light headlamps were used by the experimenters during the experiment . The pieces of paper with the drawings of the insect trajectories were scanned . The resulting images were analysed in ImageJ 1 . 53a to determine the distances crawled by the insects . To determine the importance of plant-associated CO2 for host location by WCR larvae and to test for distance-specific effects , we evaluated host location ability of WCR larvae in dual-choice experiments using belowground olfactometers ( Figure 5 ) . To specifically investigate the importance of plant-associated CO2 for host location , preferences of CO2-sensitive and CO2-insensitive insects for intact plant odours and for plant odours without CO2 were evaluated . CO2 was experimentally removed using soda lime ( Carl Roth , Karlsruhe , Germany ) . For this , layers of 5 g of soda lime granules ( 2–4 mm ) were placed between the metal wire screens of the Teflon connectors and the sand contained in the L-shaped glass pots . To test for distance-specific effects , we used two olfactometer types that were otherwise the same but differed in the length of their arms ( Figure 5 ) . General specifications of the olfactometers are described above . One set of olfactometers has short arms that allowed for the release of larvae at 9 cm ( Figure 5A ) from plant volatile sources , and the other one has long arms that allow to release the larvae at 18 cm ( Figure 5B ) from plant volatiles sources . Insect larvae can move freely into the central glass tube and reach the metal wire screens located at both ends of the central glass tube . L-shaped glass pots were covered with aluminium foil , filled with sand , and one 3-week-old maize plant was transplanted 48 hr before the experiments . L-shaped glass pots without plants were treated similarly . Olfactometers remained detached until shortly before the experiments . Thirty minutes before the experiments , soda lime layers were applied and one L-shaped glass pot with a plant and one L-shaped glass pot without a plant were connected to the central glass tubes . Soda lime layers were used in both sides of the olfactometers . Olfactometers without soda lime served as controls . Then , either six 2nd–3rd instar WT WCR larvae or six 2nd–3rd instar DvvGr2-silenced WCR larvae were released . Olfactometers were covered with aluminium foil to reduce light disturbance to the larvae . Aluminium foil was removed shortly before evaluating larval positions . Larval positions were recorded 1 hr after their release . Insect preference for a given treatment was considered when the larvae were found at least 1 cm from the odour source; this is 1 cm apart from the wire mesh . Four olfactometers with six larvae each were assayed ( n = 4 ) . CO2 levels were measured using a gas analyser ( Li7000 , Li-Cor Inc , Lincoln , NE , USA ) as described above . To determine the importance of plant-associated CO2 for host location by WCR larvae and to test for distance-specific effects , host location by CO2-sensitive and CO2-insensitive insects was evaluated in soil-filled trays ( Figure 7 ) . For this , four or five 2-to 3-week-old maize plants were transplanted into custom-made fabric pockets ( 12 × 3 × 5 cm ) made out of volatile-permeable fabrics ( Trenn-Vlies , GeoTex Windhager , Switzerland ) filled with soil ( Selmaterra , Bigler Samen AG , Thun , Switzerland ) . The plants were transplanted into the fabric pockets , and the fabric pockets were placed in a corner of plastic trays ( 80 cm × 15 cm × 4 . 5 cm ) ( Migros Do it + Garden , Switzerland ) containing soil 24 hr before the experiments . Then , 20 second instar WT WCR larvae or 20 second instar DvvGr2-silenced WCR larvae were released at 16 , 32 , 48 , or 64 cm from the plants . Larval release points are indicated by arrows ( Figure 7B ) . Eight hours after releasing the larvae , their positions were recorded by carefully removing the soil at each tray zone and inspecting it in search for the insects . Six trays per larval type and distance with 20 larvae each were assayed ( n = 6 ) . CO2 levels in the soil gas phase of each tray zone were measured by GC-FID as described above . Optimally fertilized plants emit higher levels of respiratory CO2 than suboptimally fertilized plants ( Zhu and Lynch , 2004 ) . To test whether WCR larvae orient towards and prefer optimally fertilized maize plants and to evaluate the importance of CO2 in this context , host location by CO2-sensitive and CO2-insensitive insects and preference for plants that were differentially fertilized was evaluated . To this end , maize plants were fertilized with three doses of fertilizer: 0 . 1% ( optimally fertilized ) , 0 . 05% ( medium fertilized ) , or 0 . 01% ( low fertilized ) . For this , Plantaaktiv 16+6+26 Typ K fertilizer ( Hauert HBG Dünger AG , Grossaffoltern , Switzerland ) was dissolved to the abovementioned concentrations and applied following manufacturer’s indications . Fertilizer macro- and micronutrient composition are described above ( see ‘Plants and planting conditions’ ) . For the choice experiments , plants were grown under the different fertilizer regimes . Twenty-four hours before the choice experiment , five 15-day-old , optimally fertilized plants were re-planted into fabric pockets ( described above ) and transferred to one corner of an 80 cm long plastic tray ( Migros Do it + Garden , Switzerland ) filled with soil ( Figure 8A , B ) . At the opposite side , plants grown either under low or medium fertilizer regimes were equally re-planted and transferred . Then , 20 second instar WCR larvae were released in the middle of the tray ( zone 3 , red arrows ) . Six independent trays per larval type and fertilizer regime pair were evaluated ( n = 6 ) . Eight hours after releasing the larvae , larval positions were recorded . To test whether larval preference for optimally fertilized plants is reflected in their growth , we measured larval weights of larvae feeding on roots of plants that were grown under the different fertilizer regimes . Fertilizer regimes are described above . For this , seven 1st instar larvae were released into solo cups ( 30 ml , Frontier Scientific Services , Inc , DE ) containing approximately 2 g of organic soil ( Selmaterra , Bigler Samen AG , Thun , Switzerland ) . Fresh roots of 15-day-old plants were provided everyday ad libitum . Roots were washed thoroughly to remove fertilizer traces from the root surface . Twenty solo cups per treatment were included ( n = 20 ) . Eight days after the beginning of the experiment , larvae were weighed using a micro balance . Four to seven larvae were recovered per experimental unit at the end of the experiment . Differences in gene expression levels , larval performance , and carbon dioxide concentrations were analysed by either Student’s t tests or by one-way ANOVA using Sigma Plot 12 . 0 ( SystatSoftware Inc , San Jose , CA , USA ) . Normality and equality of variance were verified using Shapiro–Wilk , Levene's , and Brown–Forsythe tests . Holm–Sidak post hoc tests were used for multiple comparisons . Data sets from experiments that did not fulfil the assumptions for ANOVA were natural log‐ , root square‐ , or rank‐transformed before analysis . Carbon dioxide concentrations and differences in larval preference were assessed using GLM under binomial distribution and corrected for overdispersion with quasi-binomial function when necessary followed by analysis of deviance and FDR-corrected post hoc tests . All analyses were followed by residual analysis to verify the suitability of the error distribution and model fitting . All The above analyses were conducted using R 3 . 2 . 2 ( 43 ) using the packages ‘lme4’ , ‘car’ , ‘lsmeans’ , and ‘RVAideMemoire’ ( Bates et al . , 2014; Fox and Weisberg , 2011; Herve , 2015; Lenth , 2016; R Development Core Team , 2014 ) .	Living deep in the ground and surrounded by darkness , soil insects must rely on the chemicals released by plants to find the roots they feed on . Carbon dioxide , for example , is a by-product of plant respiration , which , above ground , is thought to attract moths to flowers and flies to apples; underground , however , its role is still unclear . This gaseous compound can travel through soil and potentially act as a compass for root-eating insects . Yet , it is also produced by decaying plants or animals , which are not edible . It is therefore possible that insects use this signal as a long-range cue to orient themselves , but then switch to another chemical when closer to their target to narrow in on an actual food source . To test this idea , Arce et al . investigated whether carbon dioxide guides the larvae of Western corn rootworm to maize roots . First , the rootworm genes responsible for sensing carbon dioxide were identified and switched off , making the larvae unable to detect this gas . When the genetically engineered rootworms were further than 9cm from maize roots , they were less able to locate that food source; closer to the roots , however , the insects could orient themselves towards the plant . This suggests that the insects use carbon dioxide at long distances but rely on another chemicals to narrow down their search at close range . To confirm this finding , Arce et al . tried absorbing the carbon dioxide using soda lime , leading to similar effects: carbon dioxide sensitive insects stopped detecting the roots at long but not short distances . Additional experiments then revealed that the compound could help insects find the best roots to feed on . Indeed , eating plants that grow on rich terrain – for instance , fertilized soils – helps insects to grow bigger and faster . These roots also release more carbon dioxide , in turn attracting rootworms more frequently . In the United States and Eastern Europe , Western corn rootworms inflict major damage to crops , highlighting the need to understand and manage the link between fertilization regimes , carbon dioxide release and how these pests find their food .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2021	Plant-associated CO2 mediates long-distance host location and foraging behaviour of a root herbivore
During vertebrate embryogenesis , dorsal-ventral patterning is controlled by the BMP/Chordin activator/inhibitor system . BMP induces ventral fates , whereas Chordin inhibits BMP signaling on the dorsal side . Several theories can explain how the distributions of BMP and Chordin are regulated to achieve patterning , but the assumptions regarding activator/inhibitor diffusion and stability differ between models . Notably , ‘shuttling’ models in which the BMP distribution is modulated by a Chordin-mediated increase in BMP diffusivity have gained recent prominence . Here , we directly test five major models by measuring the biophysical properties of fluorescently tagged BMP2b and Chordin in zebrafish embryos . We found that BMP2b and Chordin diffuse and rapidly form extracellular protein gradients , Chordin does not modulate the diffusivity or distribution of BMP2b , and Chordin is not required to establish peak levels of BMP signaling . Our findings challenge current self-regulating reaction-diffusion and shuttling models and provide support for a graded source-sink mechanism underlying zebrafish dorsal-ventral patterning . The dorsal-ventral axis is one of the earliest coordinate systems established during animal development and divides the embryo into dorsal ( back ) and ventral ( belly ) territories . This axis forms under the influence of the BMP/Chordin patterning system . The activator BMP induces the formation of ventral tissues , and BMP signaling is antagonized on the dorsal side by the inhibitor Chordin . There are currently several disparate models that can explain how BMP signaling is restricted to the ventral side ( Ben-Zvi et al . , 2008; Barkai and Ben-Zvi , 2009; Francois et al . , 2009; Ben-Zvi et al . , 2011b; Inomata et al . , 2013; Ramel and Hill , 2013; Ben-Zvi et al . , 2014 ) , but the underlying biophysical assumptions have not been fully tested . In the ‘Graded source-sink + mobile BMP model’ ( Model 1 ) , BMP is produced in a graded , ventrally biased source , and signaling from diffusing BMP is antagonized by binding to its inhibitor Chordin ( Figure 1—figure supplement 1 , Table 1 ) . Chordin ( Chd ) diffuses from a localized source on the opposing dorsal side and therefore provides a ‘sink’ that inactivates BMP molecules diffusing through the embryo , helping to shape the signaling distribution into a gradient that peaks ventrally . The distributions of bmp and chd mRNA in developing embryos are consistent with this idea – initially nearly uniform bmp expression refines to a ventrally biased gradient over time ( Ramel and Hill , 2013; Zinski et al . , 2017 ) , and chd expression is restricted to the dorsal region ( Miller-Bertoglio et al . , 1997 ) . Similar to Model 1 , BMP signaling activity in the ‘Graded source-sink + immobile BMP model’ ( Model 2 , Figure 1—figure supplement 1 , Table 1 ) is also restricted by the inhibitor Chordin diffusing from the dorsal side . However , Model 2 assumes that BMP does not diffuse ( Ramel and Hill , 2013 ) and that it binds to Chordin with weaker affinity than in Model 1 ( see Materials and methods ) . Proponents have argued that the similarities between the graded bmp mRNA distribution , signaling gradient , and target gene expression indicate negligible BMP diffusion during patterning ( Ramel and Hill , 2013 ) . Consistent with this , BMP4 was unable to induce long-range signaling in Xenopus experiments ( Jones et al . , 1996 ) , although BMP target genes are induced outside of BMP-expressing clones in zebrafish ( Xu et al . , 2014 ) . However , measuring the diffusivity of BMP in vivo is the most direct way to determine whether BMP is mobile ( Kicheva et al . , 2007; Zinski et al . , 2017 ) . Although these two relatively simple models are generally supported by biological observations , they do not take into account other regulators known to be crucial for dorsal-ventral patterning , such as the BMP-like ligand ADMP , and Sizzled , an inhibitor of the Chordin protease Tolloid/Xlr . Three models described below include these important dorsal-ventral regulators in addition to BMP and Chordin and have also been shown to explain scale-invariant patterning , a phenomenon in which embryos adjust their tissue proportions to differently sized patterning fields . The recent ‘Long-range accumulation and feedback model’ ( Model 3 , Figure 1—figure supplement 1 , Table 1 ) postulates that BMP and Chordin have equally high mobility , but that dorsal-ventral patterning is controlled by differences in BMP and Chordin protein stability ( Inomata et al . , 2013 ) . In this model , BMP and ADMP induce the secreted , highly diffusible and stable Chordin protease inhibitor Sizzled . This protects Chordin from proteolysis and promotes its expansion towards the ventral side . Over time the resulting inhibition of BMP signaling leads to decreased Sizzled production , destabilizing Chordin and relieving inhibition of BMP . In this way , an appropriate balance between ventral BMP and dorsal Chordin levels can be established even in differently sized embryos . In the ‘Self-regulating reaction-diffusion model’ ( Model 4 , Figure 1—figure supplement 1 , Table 1 ) , BMP and Chordin both have low diffusivities and equivalent protein stabilities . Interactions with highly mobile ADMP and Sizzled in two coupled reaction-diffusion networks eventually result in the restriction of BMP signaling activity on the ventral side , assuming an initial dorsal Chordin or ventral BMP bias ( Francois et al . , 2009 ) . Such a system self-regulates even with noisy initial conditions and could provide robustness during embryogenesis – e . g . , the ability of developing organisms to withstand noise in gene expression or fluctuating environmental conditions – that can be difficult to explain with other models . Finally , the prominent ‘Shuttling model’ ( Model 5 , Figure 1—figure supplement 1 , Table 1 ) postulates that Chordin not only acts as an inhibitor of BMP , but also modulates the mobility and distribution of BMP protein ( Ben-Zvi et al . , 2008; Barkai and Ben-Zvi , 2009; Ben-Zvi et al . , 2011b; Ben-Zvi et al . , 2014 ) . In this model , BMP is poorly diffusive , Chordin is highly diffusive , and BMP mobility increases when bound to Chordin . Cleavage of the BMP/Chordin complex by the uniformly distributed protease Tolloid/Xlr combined with a flux of Chordin from the dorsal side is thought to ‘shuttle’ BMP towards the ventral side by facilitated diffusion over time . In this way , Chordin is responsible for the accumulation of BMP protein on the ventral side , and actively helps establish the subsequent ventral BMP signaling peak . These five conflicting models postulate different diffusion ( no diffusion , equal diffusion , differential diffusion , facilitated diffusion ) and stability properties of BMP and Chordin proteins ( Table 1 , Figure 1—figure supplement 1 ) . However , these biophysical properties have not been fully measured experimentally , in part due to the lack of reagents and techniques to detect active BMP and Chordin in living vertebrate embryos . To test the biophysical tenets of these models , we developed active BMP and Chordin fluorescent fusion proteins , and used a combination of mathematical modeling and quantitative experiments to determine how BMP2b and Chordin gradients form . Additionally , we tested the distinct predictions that the five models make about how BMP signaling changes in the absence of Chordin . We found that ( i ) BMP2b and Chordin proteins have similar stabilities , ( ii ) both BMP2b and Chordin diffuse and form gradients in the extracellular space , and ( iii ) Chordin does not significantly facilitate BMP2b diffusion or play an active role in establishing peak ventral BMP signaling levels . Together , our results are most consistent with dorsal-ventral patterning mediated by Model 1 , the ‘Graded source-sink + mobile BMP’ model . BMP signaling induces phosphorylation and nuclear localization of the transcriptional effectors Smad1/5/9 ( Schier and Talbot , 2005 ) . To quantitatively measure BMP signaling activity during early dorsal-ventral patterning , we imaged pSmad1/5/9-immunostained zebrafish embryos fixed at different developmental stages using in toto light sheet microscopy , converted pSmad1/5/9 signaling activities into information-compressed two-dimensional maps ( Schmid et al . , 2013 ) , and quantified pSmad1/5/9 intensities along the ventral-dorsal axis ( Figure 1A , Materials and methods ) . Over the course of approximately 3 hr during early zebrafish development , BMP signaling rapidly shifts from a low-level near-uniform distribution to a gradient with peak levels on the ventral side ( Figure 1A+B , Videos 1–5 ) ( Tucker et al . , 2008 ) , similar to changes in the distribution of bmp2b mRNA over time ( Ramel and Hill , 2013; Zinski et al . , 2017 ) . We simulated pSmad1/5/9 gradient formation kinetics predicted by each of the five models over a similar time period ( Figure 1C–G ) . Our measurements are consistent with the gradient kinetics predicted by Models 1 , 2 , 4 , and 5 , whereas the dynamics predicted by Model 3 do not resemble the experimentally observed distributions . All five major models of BMP/Chordin-mediated dorsal-ventral patterning qualitatively explain the formation of a ventral signaling peak , but they assign different roles to the inhibitor Chordin ( Figure 2A–E , Table 1 , and Figure 1—figure supplement 1 ) . Models 1 and 2 assume that a flux of the inhibitor Chordin from the dorsal side restricts the range of BMP signaling activity throughout the embryo . They thus predict that in the absence of Chordin , BMP signaling should be expanded throughout the embryo with a small increase in the peak levels on the ventral side ( Figure 2A+B ) . Model 3 adds an additional regulatory layer: Here , the abundance of Chordin is regulated by feedback interactions that modify its stability and affect ventral BMP signaling levels ( Figure 1—figure supplement 1 ) . Similar to Models 1 and 2 , Model 3 also predicts that in the absence of Chordin , BMP signaling should be expanded throughout the embryo ( Figure 2C ) . In Model 4 , two reaction-diffusion systems involving BMP/Sizzled and Chordin/ADMP are coupled . In a completely homogenous field of cells with no initial expression biases , this self-organizing system would give rise to both ventral and dorsal BMP peaks ( Francois et al . , 2009 ) . To achieve a single ventral BMP peak , an initial dorsal Chordin or ventral BMP bias is required ( see Materials and methods ) . Under these conditions , the initial advantage in BMP signaling on the ventral side is amplified by autoregulation of BMP production . Since Chordin inhibits the autoregulation of BMP production , the absence of Chordin leads to a more pronounced ventral BMP peak but has no effect in the rest of the embryo ( Figure 2D ) . Model 4 thus predicts that in the absence of Chordin , pSmad1/5/9 levels would be increased on the ventral but not the dorsal side . In contrast to Models 1–4 , Model 5 assigns a more active role to Chordin in promoting the ventral BMP signaling peak . This model proposes that Chordin activity results in increased BMP signaling ventrally: Chordin increases ventral BMP levels by binding to and physically moving BMP protein towards the ventral side . This model therefore predicts that in embryos lacking Chordin , BMP signaling should be lower on the ventral side compared to wild type embryos ( Figure 2E ) . To experimentally test these predictions , we quantitatively measured BMP signaling activity in fixed chordin−/− zebrafish embryos ( Video 6 ) and their wild type siblings using pSmad1/5/9 immunostaining and in toto light sheet microscopy . Strikingly , BMP signaling was increased in dorso-lateral domains in chordin−/− mutants compared to wild type embryos , but BMP signaling on the ventral side was not significantly affected ( Figure 2F–H ) , consistent with the predictions from Models 1–3 and observations in Xenopus and zebrafish embryos ( Plouhinec et al . , 2013; Zinski et al . , 2017 ) , but not with the BMP signaling distributions predicted by Models 4 and 5 ( Table 1 ) . In order to understand the underlying basis of BMP/Chordin distribution and directly test the biophysical assumptions of the five dorsal-ventral patterning models , we developed fluorescent fusion proteins . We fused superfolder-GFP ( sfGFP [Pédelacq et al . , 2006] ) and the photoconvertible protein Dendra2 ( Gurskaya et al . , 2006 ) to zebrafish Chordin and BMP2b , the major BMP ligand regulating zebrafish dorsal-ventral patterning ( Kishimoto et al . , 1997; Xu et al . , 2014 ) . Basing our design on previously established fusions with small peptide tags ( Cui et al . , 1998; Degnin et al . , 2004; Sopory et al . , 2006 ) , we inserted fluorescent proteins to label the mature signaling domains , and obtained fusion proteins that are processed similarly and have similar biological activity as untagged versions or constructs fused to small FLAG tags ( Figure 3A–E , Figure 3—figure supplement 1 ) . Indeed , BMP2b mutants ( swr−/− , which are normally severely dorsalized [Kishimoto et al . , 1997] ) can be rescued by injection of mRNA encoding BMP2b-Dendra2 or BMP2b-sfGFP at levels equivalent to untagged BMP2b ( Figure 3C ) . In these experiments , the injected mRNA should be uniformly distributed , highlighting the important role of Chordin or other antagonists in shaping the graded BMP signaling distribution . To measure the kinetics of BMP and Chordin protein gradient formation , we expressed BMP2b-sfGFP and Chordin-sfGFP from local sources in wild type zebrafish embryos ( Müller et al . , 2012 ) and imaged the distribution profiles over time using light sheet microscopy ( Figure 3F–I ) . Importantly , in previous experiments it has been demonstrated that BMP2b clones generated in a similar manner can recapitulate BMP signaling comparable to that observed along the dorsal-ventral axis ( Xu et al . , 2014 ) . Strikingly , both BMP2b-sfGFP and Chordin-sfGFP are secreted and diffuse in the extracellular space ( Figure 3F+G , Videos 7+8 ) , in contrast to the proposal of Model 2 that only Chordin – but not BMP – diffuses ( Ramel and Hill , 2013 ) ( Table 1 ) and the absence of long-range BMP4 signaling in Xenopus ( Jones et al . , 1996 ) . Both BMP2b-sfGFP and Chordin-sfGFP rapidly establish concentration gradients over the course of one hour ( Figure 3H+I ) , consistent with the rapid patterning of the dorsal-ventral axis during zebrafish development . The gradient formed by Chordin-sfGFP has a moderately longer range than the one formed by BMP2b-sfGFP . For example , 60 min post-transplantation the BMP2b-sfGFP signal drops to 50% of the maximal concentration at a distance of 30–40 µm , whereas the gradient formed by Chordin-sfGFP reaches 50% of its maximal concentration at a distance of 50–60 µm from the source boundary at this time point ( Figure 3H+I ) . This suggests that stability or diffusivity might differ between these proteins ( Müller and Schier , 2011; Müller et al . , 2013 ) . Importantly , Models 3 and 5 assume that BMP is more stable than Chordin , whereas the other models assume either similar or unconstrained stabilities ( Table 1 ) . To distinguish between these possibilities , we first determined protein stability in living zebrafish embryos using a Fluorescence Decrease After Photoconversion ( FDAP ) assay ( Müller et al . , 2012; Bläßle and Müller , 2015; Rogers et al . , 2015 ) . We expressed BMP2b and Chordin fused to the green-to-red photoconvertible protein Dendra2 uniformly in zebrafish embryos , used brief UV exposure to convert the signal from green to red to generate a pulsed protein pool , and monitored the decrease in extracellular red fluorescence over time ( Figure 4A+B ) . For BMP2b-Dendra2 , we found a clearance rate constant of k1 = ( 8 . 9 ± 0 . 1 ) × 10−5/s ( half-life 130 min , Figure 4A ) . For Chordin-Dendra2 , we measured a similar clearance rate constant of k1 = ( 9 . 6 ± 0 . 3 ) × 10−5/s ( half-life 120 min , Figure 4B ) . The similar clearance rate constants suggest that differential protein stabilities cannot account for the different protein distributions of BMP2b and Chordin . Importantly , these results are inconsistent with the differential protein stabilities predicted by Models 3 and 5 ( Table 1 ) . Our finding that BMP2b- and Chordin-Dendra2 fusions have similar stabilities ( Figure 4A+B ) suggests that differences in diffusivity could account for the slight differences in gradient formation kinetics . Indeed , when we fitted a gradient formation model based on local production , uniform diffusion , and clearance constrained with our measured protein half-lives in a realistic three-dimensional zebrafish embryo-like geometry ( Müller et al . , 2012 ) to the measured protein distributions , we obtained the best agreement between model and data with lower diffusivity of BMP2b ( 4 µm2/s ) compared to Chordin ( 6 µm2/s ) ( Figure 3—figure supplement 2A+B ) . Importantly , the five models assume distinct BMP and Chordin diffusion properties ( Table 1 , Figure 1—figure supplement 1 ) , from no BMP diffusion ( Model 2 ) to substantially higher Chordin mobility compared to BMP ( Model 5 ) . To directly test these predictions , we determined the effective diffusivities of fluorescently tagged BMP2b and Chordin moving through developing zebrafish embryos . We used a Fluorescence Recovery After Photobleaching ( FRAP ) assay ( Müller et al . , 2012 ) that measures the dynamics of re-appearance of fluorescence in a bleached region in embryos uniformly expressing fluorescent fusion proteins ( Figure 4C–E ) . We found effective diffusion coefficients of 2–3 µm2/s for BMPs ( BMP2b-Dendra2: 2 . 0 ± 0 . 4 µm2/s; BMP2b-sfGFP: 2 . 6 ± 0 . 7 µm2/s ( similar to [Zinski et al . , 2017] ) and of 6–7 µm2/s for Chordin ( Chordin-Dendra2: 6 . 0 ± 0 . 7 µm2/s; Chordin-sfGFP: 7 . 3 ± 3 . 9 µm2/s ) , indicating that slight differences in diffusivities could underlie the differences in protein distributions . This idea is further supported by the agreement between gradients simulated with the measured diffusivities and clearance rate constants and our experimentally determined protein gradients ( Figure 3—figure supplement 2E–H ) . The measured diffusion coefficients are most consistent with Models 1 and 4 , which assume either similarly low diffusivities ( Model 4 ) or that BMP has a moderately lower diffusion coefficient than Chordin ( Model 1 , Table 1 ) . As observed in the BMP2b-sfGFP gradient formation experiment ( Figure 3F–I ) , our FRAP data demonstrate that BMP2b-sfGFP is mobile in vivo , inconsistent with Model 2 . Strikingly , local diffusion measurements in very small extracellular volumes far away from cell surfaces using Fluorescence Correlation Spectroscopy ( FCS ) assays showed that BMP2b-sfGFP ( free diffusion coefficient: Df = 46 ± 1 µm2/s ) and Chordin-sfGFP ( free diffusion coefficient: Df = 59 ± 2 µm2/s ) are highly mobile over short spatial and temporal scales ( Figure 4F ) , whereas their diffusivities are reduced at the global scale when they move across a field of cells ( Figure 4E ) . We hypothesize that the difference between effective diffusivities ( measured by FRAP ) and local diffusivities ( measured by FCS ) is due to binding to immobile extracellular molecules , which could serve as diffusion regulators that hinder the mobility of BMP2b and Chordin , similar to what has been proposed for other developmental signals such as Nodal and FGF ( Müller et al . , 2012; Müller et al . , 2013 ) . Models 3 and 4 assign important roles to the secreted proteins ADMP and Sizzled in regulating BMP signaling and distribution . Model 3 postulates diffusivities of ADMP and Sizzled equivalent to BMP and Chordin , whereas Model 4 requires approximately 25-fold higher diffusivities of ADMP and Sizzled compared to BMP and Chordin ( Table 1 ) . To measure the diffusivities of ADMP and Sizzled and test these assumptions , we developed fluorescent ADMP and Sizzled fusion proteins ( see Materials and methods ) . Whereas Sizzled fusion proteins had activity comparable to untagged Sizzled ( Figure 4—figure supplement 1A–C ) , ADMP fusions with sfGFP or FLAG tags inserted 2 , 5 , or 11 amino acids after the Furin cleavage site were much less active than untagged ADMP ( data not shown ) , and could therefore not be used for diffusion measurements . Using FRAP , we measured an effective diffusion coefficient of 9 . 7 ± 3 . 2 µm2/s for Sizzled-sfGFP ( Figure 4E , Figure 4—figure supplement 1D ) . This measurement is consistent with Model 3 , but not Model 4 , the latter of which requires much higher Sizzled mobility ( Table 1 ) . When parameterized with these measured diffusion coefficients and over a ~100-fold range of ADMP diffusion coefficients , Model 3 can form ventral-dorsal gradients over relevant time scales ( Figure 4—figure supplement 1F–J ) , but the kinetics of gradient formation do not reflect the measurements of pSmad1/5/9 distribution profiles in Figure 1A+B . Moreover , the relatively minor difference between BMP/Chordin and Sizzled diffusivity is not compatible with the 25-fold differential required for Model 4 ( Figure 4—figure supplement 1K–P ) . Model 5 ( Shuttling ) postulates that highly diffusive Chordin enhances the mobility of poorly diffusive BMPs ( Ben-Zvi et al . , 2008 ) . In this model , Chordin is secreted dorsally , binds to relatively immobile BMP , and creates a highly mobile BMP/Chordin complex . This complex then diffuses until Chordin is cleaved by a protease ( Xlr ) , rendering BMP immobile again ( Figure 1—figure supplement 1 ) . To investigate whether Chordin is not only an inhibitor of BMP , but also enhances BMP diffusivity , we increased Chordin levels and measured the effective diffusivity of fluorescent BMP2b . In embryos overexpressing Chordin , we did not observe a significant change in the effective diffusivity of fluorescently tagged BMP2b compared to embryos that did not overexpress Chordin ( BMP2b-Dendra2 + Chordin: 2 . 2 ± 0 . 2 µm2/s; BMP2b-sfGFP + Chordin: 2 . 8 ± 0 . 7 µm2/s; Figure 4G ) . The ability of Chordin to enhance the diffusivity of BMP , a major tenet of Model 5 , is therefore not supported by FRAP data . Model 5 also predicts that Chordin alters the distribution of BMP protein . Over time , the shuttling of BMP by Chordin causes BMP to accumulate away from the Chordin source , resulting in an opposing peak of BMP . Our observation that Chordin does not affect the diffusivity of BMP challenges this view ( Figure 4G ) . However , to directly test whether a Chordin source can alter BMP distribution ( Figure 5A+B ) , we juxtaposed clones of BMP2b-sfGFP-producing cells with clones of cells secreting untagged Chordin and imaged the formation of the BMP2b-sfGFP gradient over time using light sheet fluorescence microscopy ( Figure 5C+D , Videos 9–10 ) . Model 5 predicts a steeper BMP2b-sfGFP gradient in the presence of an adjacent Chordin-producing clone compared to a wild type clone ( Figure 5A+B ) . Although BMP2b-sfGFP gradients tend to be slightly steeper in the presence of a neighboring Chordin-expressing clone compared to a non-Chordin-expressing clone ( Figure 5D ) , this minor change is unlikely to account for the formation of a ventral peak in BMP signaling during the short time ( hours ) required to complete dorsal-ventral patterning ( Figure 1A+B ) . We also failed to observe significant redistribution of BMP in simulations of adjacent BMP and Chordin clones using our measured diffusion coefficients and half-lives ( Figure 5E+F ) . This suggests that shuttling of BMP2b by Chordin is not relevant for the early aspects of dorsal-ventral patterning in zebrafish embryos . The BMP signaling gradient patterns the dorsal-ventral axis during animal development . Five major models can explain how a ventral peak of BMP signaling forms , but the biophysical assumptions underlying these models differ widely ( Table 1 ) . After experimentally examining these assumptions , our findings lead to four main conclusions . First , Chordin does not play an active role in generating BMP signaling peaks , but only globally inhibits BMP ( Figure 2 ) . This is consistent with graded source-sink-type models ( e . g . Models 1 and 2 ) and Model 3 , but inconsistent with Models 4 and 5 ( Table 1 ) . Interestingly , BMP signaling in the absence of Chordin is not raised on the extreme dorsal side , indicating that other extracellular inhibitors such as Follistatin or Noggin ( Umulis et al . , 2009 ) or inhibitors of bmp expression ( Koos and Ho , 1999; Leung et al . , 2003; Ramel and Hill , 2013 ) that were not included in the tested models might further restrict BMP signaling in these regions . Second , BMP2b and Chordin both diffuse in the extracellular space ( Figure 3F–I ) , challenging models involving immobile BMP ( Model 2 ) . Third , fluorescently tagged BMP2b and Chordin have similarly high local diffusivities ( Figure 4F ) , but on a global scale they move much more slowly through the embryo ( Figure 4E ) . These findings rule out Models 2 , 3 , and 5 , but are consistent with Models 1 and 4 . Fourth , Chordin does not significantly affect BMP2b diffusion or protein distribution in zebrafish embryos ( Figure 4G , Figure 5 ) , undermining shuttling models in this developmental context . Instead , our data are most consistent with Model 1 , the graded source-sink model of BMP/Chordin-mediated dorsal-ventral patterning during early zebrafish development . Our conclusions are also consistent with a recent complementary study ( Zinski et al . , 2017 ) . Notably , shuttling models ( e . g . Model 5 ) have gained prominence in many developmental contexts including scale-invariant patterning ( Ben-Zvi et al . , 2008; Barkai and Ben-Zvi , 2009; Francois et al . , 2009; Plouhinec and De Robertis , 2009; Ben-Zvi and Barkai , 2010; Ben-Zvi et al . , 2011a; Haskel-Ittah et al . , 2012 ) , but the fundamental tenet , that is , whether putative shuttles such as Chordin change the diffusivity and distribution of signals such as BMP , has not been directly examined . Alternative models that do not invoke Chordin-dependent facilitated BMP diffusion ( Model 4 ) ( Francois et al . , 2009 ) or that postulate differential protein stability ( Model 3 ) ( Inomata et al . , 2013 ) can also explain scale-invariant patterning . Our data do not provide strong evidence for shuttling of BMP2b at time scales relevant for dorsal-ventral patterning during early zebrafish embryogenesis: We failed to observe a significant modulation of BMP2b-sfGFP or BMP2b-Dendra2 diffusivity or distribution by Chordin ( Figure 4G , Figure 5 ) . It is , however , possible that other BMPs ( e . g . BMP4 , BMP7 , ADMP ) are shuttled by interactions with Chordin and its protease Tolloid/Xlr . Indeed , tolloid mutants display a mild patterning defect of the ventral tail fin ( Connors et al . , 1999 ) that might reflect a requirement for the ventral accumulation of a weakly active , dorsally expressed BMP ligand such as ADMP ( Dickmeis et al . , 2001; Lele et al . , 2001 ) . The graded source-sink model ( Model 1 ) that is best supported by our data describes a system in which the graded , ventrally biased distribution of bmp mRNA and the dorsally localized chd mRNA distribution produce opposing sources of extracellular , diffusing BMP and Chordin protein , which together generate the BMP signaling gradient required for proper dorsal-ventral patterning . Notably , this model fails to take other known dorsal-ventral regulators into account ( e . g . , ADMP , Sizzled , Follistatin , Noggin ) . Furthermore , approximately one third of bmp2b and chordin mutant embryos can be rescued by apparently uniform bmp and chordin expression , respectively ( Kishimoto et al . , 1997; Fisher and Halpern , 1999 ) ( Figure 3C ) , arguing against a strong requirement for concurrent opposing BMP and Chordin sources as long as one component of the system is biased ( i . e . ventrally biased bmp2b expression with uniform Chordin , or dorsally biased chordin expression with uniform BMP ) . Thus , further adjustments to the basic Model 1 will be required to fully describe dorsal-ventral patterning . Although our results support a role for BMP diffusion in dorsal-ventral patterning , the necessity of signal diffusion for developmental patterning has recently been challenged by several studies ( Brankatschk and Dickson , 2006; Roy and Kornberg , 2011; Alexandre et al . , 2014; Dominici et al . , 2017; Varadarajan et al . , 2017 ) . It will be interesting to determine whether BMP diffusion is indeed required for proper patterning using emerging nanobody-mediated diffusion perturbations ( Harmansa et al . , 2015 ) or optogenetics-based cell-autonomous modulation of signaling range ( Sako et al . , 2016 ) . To visualize pSmad1/5/9 , wild type TE embryos were dechorionated at the one-cell stage using 1 mg/ml of Pronase ( Roche , Cat . No . 11 459 643 001 ) . Dechorionated embryos were incubated at 28°C and fixed at different developmental stages in 4% formaldehyde ( Roth ) in PBS overnight at 4°C on a shaker . Embryos were then stored in 100% methanol at −20°C for at least 2 hr . All subsequent steps were carried out at room temperature . Embryos were re-hydrated with 70% , 50% , and 30% methanol in PBS for 10 min each . The embryos were then washed eight times with PBST ( 0 . 1% Tween ) for 15 min and blocked twice with blocking solution ( 10% fetal bovine serum and 1% DMSO in PBST ) for 1 hr , and incubated with 1:100 anti-pSmad1/5/9 antibody ( Cell Signaling Technology , Cat . No . 9511 ) for 4 hr . Embryos were washed with blocking solution for 15 min , washed seven times with PBST , blocked with blocking solution for 1 hr , incubated with 1:500 Alexa 488-coupled goat anti-rabbit secondary antibody ( Life Technologies , Cat . No . A11008 ) for 4 hr , and washed similarly to the procedure after primary antibody application . Embryos were then counterstained with DAPI solution ( 0 . 2 µg/ml in PBST ) for 1 hr and washed with PBST . Immunostainings were performed using an In situ Pro hybridization robot ( Abimed/Intavis ) . To analyze pSmad1/5/9 distributions in the absence of Chordin , embryos from one pair of chordintt250 ( Hammerschmidt et al . , 1996 ) heterozygous parents were collected , fixed , immunostained with anti-pSmad1/5/9 antibody ( Cell Signaling Technology , Cat . No . 13820S ) as above , and imaged simultaneously to minimize differences between samples . Embryos were treated as described above , except that progeny from chordin+/- incrosses were first permeabilized with ice-cold acetone at −20°C for 7 min before the re-hydration step . After imaging and DNA extraction ( Meeker et al . , 2007 ) , progeny from the chordintt250 heterozygote incross were identified as wild type , heterozygous , or homozygous mutant embryos by PCR amplification using the forward primer 5’-TTCGTTTGGAGGACAACTCG-3’ and the reverse primer 5’-AACTCAGCAGCAGAAGTCAATTC-3’ with an initial denaturation step of 94°C for 3 min; 39 cycles of 94°C for 30 s , 55°C for 40 s , and 72°C for 30 s; and a final extension at 72°C for 5 min with subsequent digestion with MspI ( New England Biolabs , Cat . No . R0106 ) for 2 hr . The genotyping assay for the chordintt250 line was designed by the Zebrafish International Resource Center ( ZIRC ) staff and downloaded from the ZIRC website at http://zebrafish . org . All constructs were generated by PCR-based methods ( Horton et al . , 1990 ) , contain the consensus Kozak sequence gccacc 5’ of the start codon , and were inserted into the EcoRI and XhoI sites of the pCS2 ( + ) vector . To generate BMP2b-sfGFP and BMP2b-Dendra2 , sequences encoding sfGFP or Dendra2 flanked by LGDPPVAT linkers were inserted two amino acids downstream of the BMP2b Furin cleavage site . Sequences encoding the FLAG tag DYKDDDDK were inserted between the first linker and sfGFP or Dendra2 to generate BMP2b-sfGFP-FLAG and BMP2b-Dendra2-FLAG . To generate BMP2b-FLAG , the FLAG tag was inserted between two LGDPPVAT linkers two amino acids downstream of the BMP2b Furin cleavage site . All constructs were generated by PCR-based methods ( Horton et al . , 1990 ) and contain the consensus Kozak sequence gccacc 5’ of the start codon . Chordin was inserted into the ClaI site of pCS2 ( + ) . All other Chordin-containing constructs were inserted into the EcoRI and XbaI sites of the pCS2 ( + ) vector . To generate Chordin-sfGFP and BMP2b-Dendra2 , sequences encoding sfGFP or Dendra2 flanked by LGDPPVAT linkers were inserted immediately 5’ of the Tolloid cleavage site 2 . To generate Chordin-FLAG , sequences encoding the FLAG tag DYKDDDDK were inserted immediately 5’ of the Tolloid cleavage site 2 without additional linkers . To generate Chordin-sfGFP-FLAG and Chordin-Dendra2-FLAG , sequences encoding the FLAG tag were inserted between the first linker and sfGFP or Dendra2 of Chordin-sfGFP and Chordin-Dendra2 constructs . All Sizzled constructs were generated by PCR-based methods ( Horton et al . , 1990 ) , contain the consensus Kozak sequence gccacc 5’ of the start codon , and were inserted into the EcoRI and XbaI sites of the pCS2 ( + ) vector . To generate Sizzled-sfGFP , sequences encoding sfGFP with an N-terminal LGLG linker were fused to the C-terminus of Sizzled . Sequences encoding the FLAG tag DYKDDDDK were inserted between the LGLG linker and sfGFP to generate Sizzled-sfGFP-FLAG . To generate Sizzled-FLAG , the FLAG tag was fused to the C-terminus of Sizzled separated by an LGLG linker . mRNA was generated using SP6 mMessage mMachine kits ( Thermo Fisher ) after vector linearization with NotI-HF ( New England Biolabs , Cat . No . R3189 ) . mRNA was purified using LiCl precipitation or Qiagen RNeasy kits following the manufacturers’ instructions . Scoring of ventralization and dorsalization was executed as previously described ( Mullins et al . , 1996; Kishimoto et al . , 1997 ) . Embryos were injected at the one- to two-cell stage with equimolar amounts of BMP2b ( 1 pg ) , BMP2b-sfGFP ( 1 . 49 pg ) , and BMP2b-Dendra2 ( 1 . 47 pg ) mRNA to assess ventralizing activity . At 1 day post-fertilization , BMP2b-injected embryos were classified as weakly ventralized ( V1 ) to strongly ventralized ( V4 ) . V1 embryos have reduced eyes but a prominent head . V2 embryos have no eyes , reduction of the head , and expansion of posterior structures such as somites and tail . V3 embryos completely lack head structures and exhibit a further expanded tail and enlarged blood islands . Finally , V4 embryos lack most structures except for a short , protruding , and expanded tail . To assess dorsalizing activity of the Chordin constructs , embryos were injected with equimolar amounts of Chordin ( 30 pg ) , Chordin-sfGFP ( 37 pg ) , Chordin-Dendra2 ( 37 pg ) , and Chordin-FLAG mRNA ( 30 pg ) . Embryos were scored at 1 day post-fertilization and classified as weakly dorsalized ( C1 ) to strongly dorsalized ( C5 ) ( Kishimoto et al . , 1997 ) . C1 embryos lack the ventral tail fin . C2 embryos have a further loss of ventral structures , such as the ventral tail vein , and a bent tail . C3 embryos exhibit a tail that is shortened and twisted . C4 embryos have observable head structures and develop eyes with twisting of the posterior structures above the yolk . C5 embryos are fully dorsalized and frequently lyse ( Mullins et al . , 1996; Kishimoto et al . , 1997 ) . Injection of BMP2b mRNA can rescue BMP2b mutants ( Kishimoto et al . , 1997 ) . To investigate whether tagged BMP2b constructs can rescue swrtc300−/− mutants ( Mullins et al . , 1996 ) , the rescuing amount of BMP2b mRNA was first determined ( 1 . 8 pg ) , and equimolar amounts of mRNA encoding fluorescent fusion constructs were subsequently injected into the progeny of heterozygous swr+/- mutant incrosses . Embryos with wild type morphology at 24 hpf were anesthetized and mounted in 2% methylcellulose for imaging with an AxioZoom V16 ( ZEISS ) microscope at 30–33 hpf . To genotype embryos following DNA extraction ( Meeker et al . , 2007 ) , PCR was performed to amplify a BMP2b fragment with the forward primer 5'-AAAAGCCGAGGAGAAAGCAC-3' and the reverse primer 5'-AGTCCTTCATTGGGGAGATTGTTC-3' , and the following thermocycling parameters: An initial denaturation step of 94°C for 3 min; 39 cycles of 94°C for 30 s , 58°C for 40 s , and 72°C for 40 s; and a final extension at 72°C for 5 min . PCR amplicons were subsequently digested with HaeIII ( New England Biolabs , Cat . No . R0108 ) at 37°C for 2 hr . The genotyping assay for the swrtc300 line was designed by the Zebrafish International Resource Center ( ZIRC ) staff and downloaded from the ZIRC website at http://zebrafish . org . Extracellularly enriched and cellular fractions from manually deyolked embryos between sphere and dome stage were obtained as described previously ( Müller et al . , 2012 ) . mRNAs encoding FLAG-tagged constructs were injected at the one- or two-cell stage at equimolar amounts ( BMP2b-FLAG: 444 pg , BMP2b-sfGFP-FLAG: 638 pg , BMP2b-Dendra2-FLAG: 630 pg; and Chordin-FLAG: 500 pg , Chordin-sfGFP-FLAG: 620 pg , Chordin-Dendra2-FLAG: 615 pg ) . For protein samples with BMP2b constructs , fractions from approximately 19 embryos were loaded and resolved by SDS-PAGE using 12% polyacrylamide gels . For protein samples with Chordin constructs , fractions from approximately 17–18 embryos were loaded and resolved in 8% polyacrylamide gels . Proteins were subsequently transferred onto PVDF membranes using a Trans-Blot Turbo Transfer System ( Bio-Rad , Cat . No . 170–4272 ) . Membranes were blocked with 5% non-fat milk ( Roth , Cat . No . T145 . 2 ) in PBST ( 0 . 1% Tween ) and incubated with anti-FLAG antibody ( Sigma , Cat . No . F3165 ) at a concentration of 1:2000 in non-fat milk in PBST at 4°C overnight . HRP-coupled donkey anti-mouse secondary antibody ( Jackson ImmunoResearch , Cat . No . 715-035-150 ) was used at concentration of 1:25 , 000 for 3 hr at room temperature . Chemiluminescence was detected using SuperSignal West Dura Extended Duration Substrate ( Thermo Fisher , Cat . No . 34075 ) and imaged with a chemiluminescence imaging system ( Fusion Solo , Vilber Lourmat ) . To generate clonal sources secreting BMP2b-sfGFP , Chordin-sfGFP , and untagged Chordin ( Figures 3 and 5 ) , approximately 50–75 cells were transplanted from sphere stage wild type TE donor embryos expressing these constructs into uninjected , sphere stage sibling hosts ( similar to [Müller et al . , 2012] ) . Transplantations were carried out in 1 x Ringer’s buffer . Cells were explanted from donors , extruded briefly into the buffer to wash away cellular debris and extracellular fluorescent protein , and then transplanted into host embryos . Donor embryos were dechorionated with 1 mg/ml Pronase ( Roche , Cat . No . 11 459 643 001 ) and injected with 1–2 nl injection mix at the one-cell stage . Sibling host embryos were dechorionated together with donors at the one-cell stage , and all embryos were incubated at 28°C until transplantation . Unfertilized or injured embryos were discarded . For single ( Figure 3 ) and double ( Figure 5 ) transplantation experiments , BMP2b-sfGFP and Chordin-sfGFP donors were injected with 500 pg mRNA ( Figure 3—figure supplement 1F–H ) . For double transplantation experiments ( Figure 5 ) , embryos received one transplantation from a donor expressing BMP2b-sfGFP and a second transplantation from a donor injected at the one-cell stage with either 50 pg Alexa 546-coupled dextran ( 10 kDa , Molecular Probes , Cat . No . D22911 ) or 1000 pg Chordin mRNA + 50 pg Alexa 546-coupled dextran . Alexa 546-coupled dextran was used to mark the location of the second clone . 2–10 min post-transplantation , embryos were mounted in 1% low-melting NuSieve GTG agarose ( Lonza , Cat . No . 50080 ) dissolved in embryo medium ( 250 mg/l Instant Ocean salt dissolved in reverse osmosis water ) . Embryos were immersed in 40°C molten low melting point agarose , pulled into 1 . 5 mm glass capillary tubes ( ZEISS ) , and positioned with the animal pole perpendicular to the capillary using a metal probe . Agarose tubes were then suspended in embryo medium , and imaged at room temperature using a ZEISS Lightsheet Z . 1 microscope ( see Light sheet microscopy section for further imaging details ) . Fluorescence images in Figures 1 , 2 , 3 and 5 , and Figure 3—figure supplement 1 were obtained using a Lightsheet Z . 1 microscope ( ZEISS ) . For fixed , immunostained embryos , samples were mounted into a glass capillary sample holder in 1% low-melting NuSieve GTG agarose ( Lonza , Cat . No . 50080 ) in embryo medium with 0 . 2 µm dark red fluorescent FluoSpheres ( Life Technologies , Cat . No . F8807 ) diluted 1:200 , 000 from a 2% solids stock . Embryos were imaged at 0° , 45° , 180° and 225° angles ( Schmid et al . , 2013 ) using identical imaging conditions . For 3D reconstruction , an interactive bead-based registration algorithm was used to determine the threshold that most accurately selects the beads ( Preibisch et al . , 2010 ) . Reconstructed images were converted to 8-bit format using ImageJ , and Imaris software ( Bitplane ) was used for 3D data visualization and video generation . The videos were cropped using Avidemux 2 . 6 . To visualize the entire embryo in a single image , reconstructed images were first converted to 16-bit files using ImageJ , and equirectangular 2D map projections were then generated ( Schmid et al . , 2013 ) . The 2D maps were re-aligned into Hammer-Aitoff projections using Hugin panorama photo stitcher software ( http://hugin . sourceforge . net ) to orient the peak of pSmad1/5/9 intensity to the ventral pole ( left in Figure 1 panels ) and the trough of pSmad1/5/9 intensity to the dorsal pole ( right in Figure 1 panels ) . For gradient quantifications in Figure 1A+B and Figure 2F–H , the embryo proper was masked using manual polygon selections in Fiji ( Schindelin et al . , 2012 ) in order to exclude signal from the yolk syncytial layer and yolk . The ‘Plot Profile’ function in Fiji was then applied to the entire masked image to determine ventral-to-dorsal gradients . The background signal of immunostained embryos was determined by finding the lowest value in the profiles of sphere stage embryos ( Figure 1A+B ) and the lowest value in the profiles of chordin−/− embryos ( Figure 2F+G ) , respectively . These background values were subtracted from the data sets , and the profiles were normalized to the highest value in each data series . The mean and standard error of the normalized data sets was then calculated piece-wise for every point along the ventral-to-dorsal profile . For transplantation experiments in Figures 3 and 5 , imaging began 5 to 20 min post-transplantation and continued for approximately 1 hr ( see Transplantation section for further details ) . The following imaging conditions were used: Gradients were quantified using maximum intensity projections of 15 z-slices similar to the approach in ( Müller et al . , 2012 ) . A rectangular region of interest abutting the clone with a fixed height of 86 . 34 μm ( corresponding to 189 pixels ) and varying widths depending on embryo length was drawn in Fiji ( Schindelin et al . , 2012 ) , and the average intensity in 0 . 457 μm strips was calculated from the maximum intensity projections . Background intensity resulting from autofluorescence was measured similarly in uninjected embryos ( for single transplantation experiments , n = 4 ) or in uninjected embryos transplanted with a clone of cells containing Alexa 546-coupled dextran ( for double transplantation experiments , n = 2 ) . A single value for background subtraction was determined by calculating the average of the intensity profile values . After subtracting the background value from the experimental intensity profiles , the data was normalized to the value closest to the clonal source boundary . This approach allows for the comparison of the relative gradient range , which is independent of constant production rates . We assume constant production rates over the relatively short time scales of observation ( ≈80 min ) . Embryos with low signal-to-noise ratios were excluded from analysis . FDAP experiments were carried out as described in ( Müller et al . , 2012; Rogers et al . , 2015 ) . Embryos were injected at the one-cell stage with either 60 pg BMP2b-Dendra2 mRNA + 0 . 5 ng Alexa 488-dextran ( 3 kDa , Molecular Probes ) or 150 pg Chordin-Dendra2 mRNA + 0 . 5 ng Alexa 488-dextran . To assess background fluorescence , embryos were injected with 0 . 5 ng Alexa 488-dextran only . Embryos were mounted in 1% low melting point agarose in glass-bottom Petri dishes ( MatTek Corporation ) covered with embryo medium to hydrate the agarose during imaging . FDAP experiments were performed using an LSM 780 ( ZEISS ) confocal microscope . Pre-conversion and post-conversion images were acquired using an LD C-Apochromat 40x/1 . 1 NA water immersion objective . A single pre-photoconversion image was first acquired for each sample followed by photoconversion and multiposition time-lapse imaging with 10 min intervals for approximately 300 min . For photoconversion , embryos were illuminated with a Sola SE II LED lamp at 100% power for 30 s through a C-Apochromat 10x/0 . 45 NA objective and an AHF F36-500 UV filter cube . For both pre- and post-conversion images , Alexa 488 was excited using a 488 nm Argon laser , and a DPSS 561 nm laser was used to excite photoconverted Dendra2 . The emission signal between 494–576 nm ( Alexa 488 ) and 578–696 nm ( photoconverted Dendra2 ) was collected using a 32 channel GaAsP QUASAR detector array . Embryos that produced only low levels of photoconverted Dendra2 signal or whose position shifted significantly over time as well as embryos with non-uniform signal distribution or embryos that died were excluded from analysis . Sample numbers: n = 22 for BMP2b-Dendra2 ( with n = 17 background embryos ) ; n = 6 for Chordin-Dendra2 ( with n = 1 background embryo ) . All experiments were analyzed using PyFDAP ( Bläßle and Müller , 2015; Rogers et al . , 2015 ) , version 1 . 1 . 2 . PyFDAP extracts the extracellular and intracellular photoconverted Dendra2 signal by masking the Alexa 488 signal , and fits the resulting average intensities with a linear decay model . The ordinary differential equation describing linear protein decay is given bydcdt=−k1c where c is the concentration of the protein and k1 is its clearance rate constant . We assume that Dendra2 signal is directly proportional to the protein concentration . The analytical solution of this equation is given byc ( t ) =c0e−k1t+y0 where c0 + y0 is the protein's concentration at t = 0 , and y0 is the protein's concentration at t = ∞ . The half-life τ of the protein can then be calculated asτ=ln ( 2 ) /k PyFDAP estimates a lower bound for y0 by computing the maximum relative effect of photobleaching Fi , r . For each background data set , the strongest influence of photobleaching was computed by taking the minimum over all differences of background intensity Bj , r and background noise Ni , and the difference between pre-conversion background intensity Bpre , i , r and noise level . Here , r denotes the region under consideration , i . e . extracellular , intracellular , or the entire imaging slice; i indicates the ith data set , and j counts the background data sets . The average over all b background data sets was then taken to arrive at the mean effect of photobleaching . The factorFi , r=1b∑j=1bmint ( Bj , r ( t ) −NiBprej , r−Ni ) was used to scale the pre-conversion intensity of the FDAP data set according toy0i , r≥Fi , r ( Iprei , r−Ni ) +Ni This lower bound was then used to constrain a Nelder-Mead simplex algorithm when minimizingSSD=∑n ( I¯ ( tn ) −c ( tn ) ) 2 FRAP experiments and data analysis were carried out as previously described ( Müller et al . , 2012; Müller et al . , 2013 ) using an LSM 780 NLO confocal microscope ( ZEISS ) and an LD LCI Plan-Apochromat 25x water immersion objective . Embryos were injected at the one-cell stage with 30 pg of mRNA encoding BMP2b-sfGFP , 60 pg of mRNA encoding BMP2b-Dendra2 , 60 pg of mRNA encoding Chordin-sfGFP , 120 pg of mRNA encoding Chordin-Dendra2 , or 30 pg of mRNA encoding Sizzled-sfGFP . To analyze the effect of Chordin on BMP2b diffusion , embryos were injected at the one-cell stage with 30 pg of mRNA encoding BMP2b-sfGFP plus 60 or 200 pg of mRNA encoding Chordin , or 60 pg of mRNA encoding BMP2b-Dendra2 plus 200 pg of mRNA encoding Chordin . Embryos were mounted in 1% low-melting point agarose in glass-bottom Petri dishes ( MatTek Corporation ) covered with embryo medium to hydrate the agarose during imaging . Embryos with low or non-uniform fluorescence and embryos that died or whose position shifted significantly over time were excluded from analysis . For FRAP data analysis , the fits of a model with uniform production , diffusion , and clearance were constrained with the clearance rate constants of BMP2b-Dendra2 and Chordin-Dendra2 fusions measured by FDAP in the present study ( BMP2b-Dendra2: k1 = 8 . 9 × 10−5/s; Chordin-Dendra2: k1 = 9 . 6 × 10−5/s ) . Sizzled-sfGFP fits were constrained with the clearance constant measured for BMP2b-Dendra2 assuming similar protein stability . As shown previously , the estimation of diffusion coefficients does not sensitively depend on the values of clearance rate constants if the time scales of observation ( here: 50 min ) and protein stability ( here: approximately 120 min ) are similar ( Müller et al . , 2012 ) . The FCS experiments were done using an LD C-Apochromat 40x/1 . 1 NA water immersion objective on an LSM 780 NLO confocal microscope ( ZEISS ) . Embryos were injected at the one-cell stage with 30 pg of mRNA encoding BMP2b-sfGFP or 60 pg of mRNA encoding Chordin-sfGFP . Embryos were mounted in 1% low-melting point agarose in glass-bottom Petri dishes ( MatTek Corporation ) and covered with embryo medium to hydrate the agarose during imaging . The fluorophores ( sfGFP , Alexa 488 ) were excited using an Argon 488 nm laser , and the emission light between 494 and 542 nm was collected using a 32-channel GaAsP QUASAR detector array . Before each FCS experiment , the pinhole was aligned and set to 1 Airy unit , and the instrument was calibrated using a solution of 40 nM Alexa 488 dye ( Thermo Fisher ) in water . For each FCS sample , fluorescence fluctuations were measured for 10 s with 10 repeats , and any irregularities in the 100 s count trace resulting from cellular movements were excluded from analysis . Auto-correlation curves for Alexa 488 were freely fitted to determine the structural parameter as well as the diffusion time , the triplet state fraction , and the triplet state relaxation time of Alexa 488 for every experiment . The auto-correlation curves for BMP2b-sfGFP and Chordin-sfGFP were fitted with a fixed structural parameter , fixed triplet state fraction , and fixed triplet relaxation time determined from the Alexa 488 calibration measurements . The curves were fitted using ZEISS ZEN Pro software with a one-component ‘free diffusion with triplet state correction’ model . The first 10−6 seconds lag time for the correlation curve was excluded in the fitting ( Yu et al . , 2009; Müller et al . , 2013 ) . The diffusion coefficient was then calculated by comparing the diffusion time of BMP2b-sfGFP and Chordin-sfGFP with Alexa 488 ( reference diffusion coefficient: 435 μm2/s [Petrásek and Schwille , 2008] ) . Since the values of the triplet state fraction and the triplet state relaxation time of sfGFP are unknown and not necessarily identical to those of Alexa 488 , we also freely fitted the autocorrelation curves for BMP2b-sfGFP and Chordin-sfGFP with the experimentally measured structural parameter as the only constraint , and determined free diffusion coefficients of D = 35 ± 2 μm2/s for BMP2b-sfGFP ( n = 17 measurements from 4 embryos ) and D = 50 ± 3 μm2/s for Chordin-sfGFP ( n = 19 measurements from 5 embryos ) , within a deviation of approximately 20–30% compared to the diffusion coefficients determined by constraining the fits with a fixed structural parameter , fixed triplet state fraction , and fixed triplet relaxation time ( D = 46 ± 1 μm2/s for BMP2b-sfGFP , and D = 59 ± 2 μm2/s for Chordin-sfGFP; values reported in Figure 4 ) . The similar diffusion coefficients determined by differently constrained fits indicate that the diffusion time measured in our experiments does not sensitively depend on the values of the triplet state fraction and triplet state relaxation time . The geometry of the zebrafish blastoderm was approximated by the complement of two spheres with a columnar subdomain placed off-center to represent the signal source region with the same parameters as described in Müller et al . ( 2012 ) . Gradient formation was simulated with the source-diffusion-sink model∂c∂t=D∇2c−k1c+δsk2 withδs={1in the source0otherwise For Figure 3—figure supplement 2 , the experimental data were fitted with solutions from a 50 × 50 parameter grid spanning all possible combinations of 50 diffusion coefficients ( logarithmically spaced from 0 . 1 µm2/s to 50 µm2/s ) and 50 clearance rate constants ( logarithmically spaced from 1 × 10−5/s to 5 × 10−4 ) . The finite element method was used for all numerical simulations . All geometries are one-dimensional representations of embryos . The solution at each time step in the discretized geometries was determined using a sparse LU factorization algorithm ( UMFPACK ) , and the time stepping was computed using a backward Euler step method ( Comsol Multiphysics ) . Simulations in Figure 1C–E , G ( Models 1 , 2 , 3 , and 5 ) were executed for a total of 10080 s ( i . e . , for approximately 3 hr from sphere to shield stage during zebrafish embryogenesis [Kimmel et al . , 1995] ) and read out every 2520 s ( i . e . , approximately every 42 min at relevant zebrafish stages ) . The simulation in Figure 1F ( Model 4 ) was executed for a total of 20 time steps near steady state and read out at every fifth time step . The following model descriptions comprise the complete wild type systems . For simulations of chordin mutants , the Chordin flux was set to 0 ( Models 1 , 2 , 3 , and 5 ) , or the Chordin-dependent terms were removed from the equations and the initial concentration of Chordin was set to 0 ( Model 4 ) . To focus on the role of Chordin in regulating BMP signaling and distribution , we did not include other negative regulators of BMP such as Noggin and Follistatin ( Umulis et al . , 2009 ) . For the interpretation of the simulations , we assume that the distribution of free BMPs is correlated with BMP signaling and the distribution of pSmad1/5/9 . To facilitate comparison of the models , the distribution profiles of free BMP are shown as a function of relative embryo length , and the solutions were normalized to the ventral-most free BMP concentration at shield stage ( i . e . , at t = 7560 s for Models 1 , 2 , 3 , and 5 , and at t = 15 for Model 4 ) in wild type simulations . In the graded source-sink model , the BMP source ρBMP ( x ) was modeled after the known distribution of bmp2b mRNA between sphere stage and 30% epiboly ( Ramel and Hill , 2013 ) . The model does not include autoregulation of BMP production since positive feedback only appears to be important for later stages of development ( Ramel and Hill , 2013; Zinski et al . , 2017 ) . Chordin binds BMP irreversibly and acts as a sink . The model was simulated using the following equations:∂[BMP]∂t=DBMP∇2[BMP]−κ[Chd][BMP]−λBMP[BMP]+ρBMP ( x ) ∂[Chd]∂t=DChd∇2[Chd]−κ[Chd][BMP]−λChd[Chd]∂[ChdBMP]∂t=DChdBMP∇2[ChdBMP]+κ[Chd][BMP]−λChd[ChdBMP] As for Model 1 , the graded source-sink model ( immobile BMP ) was modeled without autoregulation of BMP production since positive feedback only appears to be important for later stages of development ( Ramel and Hill , 2013; Zinski et al . , 2017 ) . Here κ , which reflects the binding between Chordin and BMP , is smaller than in Model 1 to obtain a realistic-free BMP distribution; using the same value for κ as in Model 1 creates an unrealistically steep free BMP gradient . The model was simulated using the following equations:∂[BMP]∂t=−κ[Chd][BMP]−λBMP[BMP]+ρBMP ( x ) ∂[Chd]∂t=DChd∇2[Chd]−κ[Chd][BMP]−λChd[Chd]∂[ChdBMP]∂t=DChdBMP∇2[ChdBMP]+κ[Chd][BMP]−λChd[ChdBMP] The model was developed for frog embryogenesis . For the simulations in the present study the equations , geometry , initial conditions , and parameters used were exactly as described in ( Inomata et al . , 2013 ) :∂[BMP]∂t=D∇2[BMP]+vBMP ( [ADMP]+[BMP] ) 10kBMP10+ ( [ADMP]+[BMP] ) 10−λBMP[BMP]+λChd[ChdBMP]1+[Szl]ki+[Chd]+[ChdBMP]+[ChdADMP]km−k[Chd][BMP]∂[Chd]∂t=D∇2[Chd]+vChdkChd10kChd10+ ( [ADMP]+[BMP] ) 10−λChd[Chd]1+[Szl]ki+[Chd]+[ChdBMP]+[ChdADMP]km−k[Chd][BMP ]−k[Chd][ADMP]∂[ADMP]∂t=D∇2[ADMP]+vADMPkADMP10kADMP10+ ( [ADMP]+[BMP] ) 10−λBMP[ADMP]+λChd[ChdADMP]1+[Szl]ki+[Chd]+[ChdBMP]+[ChdADMP]km−k[Chd][ADMP]∂[Szl]∂t=D∇2[Szl]+vSzl ( [ADMP]+[BMP] ) 20kSzl20+ ( [ADMP]+[BMP] ) 20−λSzl[Szl]∂[ChdBMP]∂t=D∇2[ChdBMP]−λChd[ChdBMP]1+[Szl]ki+[Chd]+[ChdBMP]+[ChdADMP]km+k[Chd][BMP]∂[ChdADMP]∂t=D∇2[ChdADMP]−λChd[ChdADMP]1+[Szl]ki+[Chd]+[ChdBMP]+[ChdADMP]km+k[Chd][ADMP] The non-dimensional model , geometry , initial conditions , and parameters used for the simulations were similar to the ones described in [Francois et al . , 2009]:∂[BMP]∂t=DBMP∇2[BMP]+[BMP]2 ( 1+[Chd] ) [Szl]−μBMP[BMP]+ρBMP∂[Chd]∂t=DChd∇2[Chd]+[Chd]2[ADMP]−μChd[Chd]+ρChd∂[ADMP]∂t=DADMP∇2[ADMP]+[Chd]2−μADMP[ADMP]∂[Szl]∂t=DSzl∇2[Szl]+[BMP]2−μSzl[Szl] For Model 5 , a minimal transport model that excludes the effects of downstream patterning circuits was used to illustrate the biophysical aspects of shuttling ( Ben-Zvi et al . , 2008 ) :∂[BMP]∂t=DBMP∇2[BMP]−κ[Chd][BMP]+λ[Xlr][ChdBMP]−λBMP[BMP]+ρBMP ( x ) ∂[Chd]∂t=DChd∇2[Chd]−κ[Chd][BMP]−λChd[Chd]∂[ChdBMP]∂t=DChdBMP∇2[ChdBMP]+κ[Chd][BMP]−λ[Xlr][ChdBMP]−λChd[ChdBMP] The one-dimensional simulations in Figure 5 were executed similarly to the ones described above and solved at 15 and 75 min for comparison to the zebrafish embryo double transplantation experiments . The solutions in Figure 5A and Figure 5E were normalized to the highest free BMP concentration in the simulation without the Chordin source , and the solutions in Figure 5B and Figure 5F were normalized to the free BMP concentration at the BMP source boundary ( at 100 μm ) for each condition to facilitate comparison between the gradient ranges . The double transplantation experiments were modeled using the following equations:∂[BMP]∂t=DBMP∇2[BMP]−λBMP[BMP]−κ[Chd][BMP]+λ[Xlr][ChdBMP]+δBMPηBMP∂[Chd]∂t=DChd∇2[Chd]−κ[Chd][BMP]+δChdηChd∂[ChdBMP]∂t=DChdBMP∇2[ChdBMP]+κ[Chd][BMP]−λ[Xlr][ChdBMP] withδBMP={1 in the BMP source0 otherwise andδChd={1 in the Chordin source0 otherwise	Animals start life as clumps of cells that ultimately give rise to complex structures and organs . Over a century of research has revealed a small number of proteins that are crucial for complex structures to form from these clumps , including one protein called BMP . Different levels of BMP instruct cells to give rise to different tissues . In zebrafish , BMP is more abundant on one side of the embryo than the other . This gradient in BMP levels causes different tissues to form at distinct positions and helps coordinate embryo development . Several theories have been proposed to explain how the BMP gradient is established . They all suggest that a second protein – Chordin – plays an important role in influencing how cells sense the BMP gradient by blocking BMP’s activity . However , the exact role of Chordin in the formation of the BMP gradient is disputed . To address this , Pomreinke , Soh , Rogers et al . directly tested five theories of how BMP and Chordin molecules spread through embryos . The experiments used microscopy to track the movements of fluorescent versions of both molecules in zebrafish embryos . The measurements contradict one theory stating that BMP does not move , and another in which Chordin increases the mobility of BMP . Pomreinke , Soh , Rogers et al . also found that embryos that lack Chordin have increased BMP signaling levels only on the side where Chordin is normally made but not on the opposite side where BMP is made , ruling out several of the theories . The findings are most consistent with the idea that the BMP gradient forms mainly as a result of higher production of BMP on one side of the embryo combined with movement of BMP away from where it is made . Chordin produced at the opposite end of the embryo helps to ensure that only the correct cells receive instructions from BMP . In the future , two approaches could further clarify how the BMP gradient is formed . First , better techniques to directly observe the BMP gradient in normally developing embryos would be useful . Second , new theories that take into account additional players other than BMP and Chordin might help explain some features of development that current theories cannot address . Uncovering the mechanisms that control the formation of BMP gradients will improve our understanding of how clumps of cells can develop into animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "computational", "and", "systems", "biology" ]	2017	Dynamics of BMP signaling and distribution during zebrafish dorsal-ventral patterning
DNA methylation is an essential epigenetic mark whose role in gene regulation and its dependency on genomic sequence and environment are not fully understood . In this study we provide novel insights into the mechanistic relationships between genetic variation , DNA methylation and transcriptome sequencing data in three different cell-types of the GenCord human population cohort . We find that the association between DNA methylation and gene expression variation among individuals are likely due to different mechanisms from those establishing methylation-expression patterns during differentiation . Furthermore , cell-type differential DNA methylation may delineate a platform in which local inter-individual changes may respond to or act in gene regulation . We show that unlike genetic regulatory variation , DNA methylation alone does not significantly drive allele specific expression . Finally , inferred mechanistic relationships using genetic variation as well as correlations with TF abundance reveal both a passive and active role of DNA methylation to regulatory interactions influencing gene expression . DNA methylation is an essential ( Li et al . , 1992 ) epigenetic mark whose role in gene regulation and its dependency on genomic sequence and environment is not yet fully understood ( Jones , 2012; Schubeler , 2012 ) . DNA methylation in vertebrates occurs most commonly in cytosines that are adjacent to guanines ( CpG sites ) . Mammalian DNA methylation levels are generally high , although CpG rich regions , called CpG islands ( CGI ) , appear mostly unmethylated ( Bird , 2002; Weber et al . , 2007; Lister et al . , 2009 ) . The mechanisms of de novo methylation and maintenance of methylation patterns are well known ( Shoemaker et al . , 2011 ) , and alterations of these can cause severe diseases ( Robertson , 2005 ) . Even though it was originally reported to be involved in gene silencing ( Holliday and Pugh , 1975; Riggs , 1975 ) , more recent studies have found that DNA methylation can also be positively correlated to gene transcription when found in gene bodies ( Hellman and Chess , 2007; Lister et al . , 2009 ) . Additionally , its participation in gene expression is proving to be highly variable , ranging from marking alternative intra-genic promoters ( Maunakea et al . , 2010 ) , to being affected by transcription factors ( TFs ) at enhancers ( Stadler et al . , 2011 ) or it-self affecting the binding of TFs such as MYC ( Prendergast and Ziff , 1991 ) . Hence , whether DNA methylation is a consequence of gene regulation , or whether it controls gene expression changes—that is , whether it plays a passive or an active role in gene regulation—still remains a topic of debate ( Schubeler , 2012 ) . Additionally , DNA methylation can be affected by environment ( Kaminsky et al . , 2009 ) but it has also been proven that regions in the genome can autonomously determine DNA methylation states ( Lienert et al . , 2010 ) . Furthermore , studies looking at natural DNA methylation variation in human populations have shown that genetic variation influences DNA methylation levels in different cell-types ( Gibbs et al . , 2010; Zhang et al . , 2010; Bell et al . , 2011 ) , but the mechanisms by which this occurs are far from clear . In these and other studies ( Kulis et al . , 2012 ) , the association between DNA methylation and gene expression in a population context , where the same gene and same methylation site can be compared across multiple individuals , has been reported to be both positive and negative . Overall , the nature of the relationships among genetic variants , DNA methylation and gene expression are still unclear despite some initial efforts ( van Eijk et al . , 2012 ) . In this study we dissect the mechanistic relationships between inter-individual DNA methylation and gene expression variation using DNA sequence variability and TF abundance measured by RNA-Seq . By assaying in a high resolution and genome wide level these three layers of information in three different cell-types originating from the same set of individuals , we are able to study the role of DNA methylation variation in different dimensions . Our results reveal a picture where DNA methylation variable sites are mechanistically associated to gene expression in complex and context dependent ways that can be of passive or active nature . We further highlight some of the mechanisms by which passive DNA methylation may occur and how this role can interplay with genetic variation . We use the GenCord collection ( Dimas et al . , 2009 ) of umbilical cord samples from 204 newborn babies of central European descent , from which we derived three cell-types: fibroblasts ( primary cells ) , T-cells ( primary cells ) and lymphoblastoid cells ( immortalized cell lines , LCLs ) ( Figure 1 ) . We genotyped each individual for 2 . 5 million SNPs , and sequenced the poly-A transcriptome of all three cell-types from the 204 individuals yielding a median of 16 million exonic reads per sample . We subsequently removed samples from 18 to 21 genetic or expression outliers , yielding a final set of 183–186 individuals per cell type ( Figure 1—figure supplements 1 and 2 ) . The assayed SNPs were imputed to the Phase 1 release of the 1000 genomes project ( Abecasis et al . , 2012 ) yielding a set of 6 . 9 million SNPs . We obtained normalized expression levels for 70 , 800–76 , 870 exons belonging to 12 , 265–12 , 863 genes ( Figure 1—figure supplements 3 and 4 ) . DNA methylation levels were measured using bisulfite-conversion and hybridization to a bead chip , assaying 416 , 118 CpG sites in 66–118 samples . Normalized methylation levels of CpG sites range from 0 to 1 , reflecting the percentage of methylation per site ( β-value; Figure 1—figure supplements 5–7 ) . In total , we analyzed 66–186 samples per cell-type and assay , belonging to 195 individuals ( Table 1; see ‘Materials and methods’ ) . 10 . 7554/eLife . 00523 . 003Figure 1 . GenCord project scheme . We collected umbilical cord and cord blood samples from 204 newborn babies , from which we derived three cell-types: fibroblasts , lymphoblastoid cells and T-cells . Genotyping , RNA-sequencing and DNA methylation levels were assayed . The number of samples without genetic and technical outliers is indicated for each assay and each cell-type . We then correlated and utilized different properties of all datasets in order to assess: expression Quantitative Trait Loci ( eQTLs ) , methylation QTLs ( mQTLs ) , positive ( pos ) and negative ( neg ) expression Quantitative Trait Methylation ( eQTMs ) . Green ticks represent Single Nucleotide Polymorphisms ( SNPs ) , purple lollipops represent methylation sites , black boxes represent exons and orange arrows depict associations between two data-types . Shown are the maximum distances between each pair of variables tested . See Figure 1—figure supplement 1–7 for data processing and quality checks . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 00310 . 7554/eLife . 00523 . 004Figure 1—figure supplement 1 . Genetic outliers removed from analyses involving genetic variation . Multidimensional scaling ( MDS ) plot showing the genetic clustering of our 204 GenCord individuals ( in black ) and 30 individuals from each of the following HapMap populations: Western and Northern European in Utah ( CEU , in blue ) , Japanese in Tokyo ( JPT , in green ) , Yoruba in Ibadan , Nigeria ( YRI , in orange ) , Gujarati Indians in Houston ( GIH , in purple ) , Mexican ancestry in Los Angeles ( MXL , in pink ) . The 16 GenCord individuals inside gray circles were considered genotypic outliers and were not included in analyses involving genetic variation . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 00410 . 7554/eLife . 00523 . 005Figure 1—figure supplement 2 . Number of reads per sample , before removing technical outliers . ( A ) Number of total reads per sample . ( B ) Number of total reads mapping uniquely , properly paired and with MAPQ ≥ 10 to exons . ( C ) Proportion of reads mapping to exons . Vertical red lines indicate median . Samples with less than 5M exonic reads were considered technical outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 00510 . 7554/eLife . 00523 . 006Figure 1—figure supplement 3 . Covariates for which expression data was corrected . ( A ) Histogram of mean GC content per library before removing outliers . ( B ) Histogram of insert size mode per library before removing outliers . The two samples in the extreme left were considered technical outliers and were removed . ( C ) – ( F ) p value distributions , with π1 indicated in each plot , of the linear regression effects of the four covariates we later corrected for on scaled exon levels . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 00610 . 7554/eLife . 00523 . 007Figure 1—figure supplement 4 . Pair wise correlations among individuals before and after covariate correction . ( A ) Histograms of scaled exon counts pair wise spearman correlation coefficients between samples ( all libraries scaled to 10M reads ) in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) . ( B ) Histogram of pair wise spearman correlation coefficients of expression levels after covariate correction between individuals in each cell-type . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 00710 . 7554/eLife . 00523 . 008Figure 1—figure supplement 5 . Normalized β-value and variance across individuals . ( A ) Normalized beta-value distribution in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) . ( B ) Distribution of normalized β-value variance per site across individuals . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 00810 . 7554/eLife . 00523 . 009Figure 1—figure supplement 6 . Normalized β-value pair wise correlations between individuals . Distributions of pair wise spearman correlation coefficients between samples in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 00910 . 7554/eLife . 00523 . 010Figure 1—figure supplement 7 . β-value distributions in distinct genomic features for expressed and non-expressed genes . ( A ) Median β-value across individuals is plotted for CpG sites by their position relative to the nearest transcription start site ( TSS ) in each cell-type , for expressed ( blue ) and non-expressed ( red ) genes . Based on the region where expressed genes have lower methylation levels than non-expressed genes , the promoter proximal region of our analyses was defined from -1kb to +2kb relative to the TSS . ( B ) – ( D ) β-value distributions in genes ( B ) , 1kb window upstream of TSS ( C ) and open-chromatin ( D ) in one LCL sample for expressed ( blue ) and non-expressed ( red ) genes . Other cell-types look very similar . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 01010 . 7554/eLife . 00523 . 011Table 1 . Summary of main association analyses in GenCord . See Table 1–source data 1DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 01110 . 7554/eLife . 00523 . 012Table 1—source data 1 . Significant eQTL , mQTL and eQTM associations found in each cell-type . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 012TestSamplesWindow sizeUnitFDR ( % ) Nominal p valueFibroblastsLCLsT-cellseQTLsGenotypes and expression183 ( F ) ; 185 ( L ) ; 186 ( T ) 1 MbGenes102 . 2 × 10−5 ( F ) ; 3 . 2 × 10−5 ( L ) ; 1 . 8 × 10−5 ( T ) 243333722115mQTLsGenotypes and methylation107 ( F ) ; 111 ( L ) ; 66 ( T ) 5 kbMethylation sites104 . 4 × 10−4 ( F ) ; 7 . 9 × 10−4 ( L ) ; 1 . 3 × 10−3 ( T ) 14 , 18922 , 41132 , 318eQTMsMethylation and expression110 ( F ) ; 118 ( L ) ; 66 ( T ) 50 kbGenes107 . 6 × 10−5 ( F ) ; 7 × 10−4 ( L ) ; 6 . 9 × 10−4 ( T ) 59636803838 At 10% FDR , we identify 2115–3372 expression quantitative trait loci ( eQTLs ) using a 1 Mb window to either side of the TSS . We discover 14 , 189–32 , 318 methylation QTLs ( mQTLs ) using a 5-kb window to either side of the CpG site . We find 1541–17 , 267 significant expression to methylation associations ( eQTMs ) using a 50-kb window around the TSS; these pertain to 596–3838 genes and 970–6846 CpG sites ( Table 1 , Figure 1 ) . DNA methylation at promoter regions is widely known to correlate negatively with gene expression levels when looking at comparisons across genes ( Jones , 2012 ) . We observe the same pattern in our study looking at the promoter regions of all genes of each individual separately and for each cell-type ( Figure 2—figure supplement 1 ) . The across individual methylation-gene expression associations ( eQTMs ) however appear to be either positive or negative , even for DNA methylation sites in promoter regions . Hence , we hypothesized that methylation sites in promoter regions from positive ( pos ) and negative ( neg ) eQTMs contribute differently ( positively and negatively , respectively ) to gene expression levels across genes . Contrary to our hypothesis , we find that methylation sites correlate negatively with gene expression across genes independently of whether they correlate positively or negatively with gene expression across individuals ( Figure 2A; Spearman correlation coefficient , rho = −0 . 11 , p=1 . 1 × 10−4 , and rho = −0 . 10 , p=1 . 7 × 10−13 , respectively; see also Figure 2—figure supplement 2 ) . The strength of these negative correlations , despite involving a subset of genes , is comparable to the one found at a genome-wide level , correlating all expressed genes with their promoter DNA methylation status per individual ( Figure 2—figure supplement 1C ) . These results suggest that the mechanisms and processes underlying inter-individual DNA methylation variation associated to gene expression are at least partly independent of the mechanisms involved in the establishment of the repressive mark of promoter DNA methylation across genes during development and differentiation . 10 . 7554/eLife . 00523 . 013Figure 2 . Inter-individual DNA methylation variation in cell-type differentiation and in different genomic contexts . ( A ) The median methylation level of promoter eQTM sites ( x-axis ) correlates negatively with across gene median number of reads per kilobase per million reads ( RPKM ) irrespective of whether they are pos-eQTMs ( yellow , N = 1149 ) or neg-eQTMs ( blue; N = 5112 ) . Spearman correlation coefficient rho is indicated in the plot with p=1 . 1 × 10−4 and p=1 . 7 × 10−13 for pos and neg-eQTMs , respectively . See Figure 2—figure supplements 1 and 2 . ( B ) As the level of cell-type methylation differentiation increases ( x-axis ) , a larger proportion of sites are associated to gene expression ( eQTMs , left y-axis ) and affected by genetic variation ( mQTLs , right y-axis ) . Proportions are plotted by 10 bins each containing 10% of the data ( 0 . 1 quantiles ) . Level of methylation differentiation is measured for each site as the coefficient of variation of the median methylation level per cell-type . See Figure 2—figure supplement 3 . ( C ) Proportion of eQTMs that are positive ( pos-eQTMs , yellow ) or negative ( neg-eQTMs , blue ) overlapping vs non-overlapping ( expected ) distinct genomic features ( promoters , CTCF binding sites , enhancers ) , or overlapping CpG island promoters ( CGI prom ) vs overlapping non-CpG island promoters ( non-CGI prom ) . For T-cells there are no CTCF or chromatin ChIP-seq data available so the data of an LCL were used instead ( see Materials and methods ) . ( D ) For each non-CGI and CGI promoters ( x-axis ) , the proportion ( y-axis ) of overlapping mQTLs was calculated ( red bars ) and was compared to the proportion of overlapping null SNPs ( black bars ) . One star indicates p<0 . 05 , two stars indicate p<1 × 10−6 , Fisher’s exact test . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 01310 . 7554/eLife . 00523 . 014Figure 2—figure supplement 1 . Correlation between promoter DNA methylation and across gene expression . ( A ) Shown is the median methylation level across all sites falling in the promoter region of a gene ( x-axis ) by the gene expression level of the gene ( y-axis ) as the log2 of the median number of reads per kilobase per million reads ( RPKM ) . The background scatter plot represents the data of a single individual , while the fitted lines depict the smoothened mean distribution of values for each individual . Non-expressed genes were artificially plotted at the bottom of the plot . ( B ) Histogram of Spearman rank correlation coefficients for each sample of correlations explained in ( A ) . All p values<2 . 2×10−16 . ( C ) Distribution of Spearman rank correlation coefficients for each sample of correlations explained in ( A ) taking only expressed genes , which were used in our eQTM analyses ( at least 1 exonic read in >90% of individuals ) . In fibroblasts ( F ) all p values<6 . 2×10−8 , in LCLs ( L ) all p values<8 . 2×10−18 , in T-cells ( T ) all p-values<2 . 6×10−44 . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 01410 . 7554/eLife . 00523 . 015Figure 2—figure supplement 2 . pos and neg-eQTMs correlation with across gene expression . Shown are the correlations between the median methylation levels of eQTM sites ( x-axis ) and the median gene expression levels ( y-axis ) for positive-eQTMs ( left panels ) and negative-eQTMs ( right panels ) in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) . The Spearman correlation coefficient rho and p value are indicated in each plot . Fitted lines depict the smoothened mean distribution of values . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 01510 . 7554/eLife . 00523 . 016Figure 2—figure supplement 3 . Tissue-specific methylation is enriched with eQTMs and mQTLs . ( A ) Distribution of the level of differentiation metric which we defined as the coefficient of variation of the median methylation level per cell-type for each site ( i . e . , standard deviation of the medians divided by the mean of medians ) . ( B ) eQTM sites have higher levels of differentiation than non-eQTM sites ( C ) mQTL sites have higher levels of differentiation than non-mQTL sites . Stars between box plots indicate that the difference is significant with p<2 . 2×10−16 , Wilcoxon test . eQTM and mQTL median methylation levels cover a wide range of β-values , so removing the 113 sites that were below the minimum or above the maximum median ( representing effectively unmethylated or fully methylated sites ) in all three cell-types did not alter the results at all . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 016 Differentially methylated regions among cell-types have been shown to play important roles in cell differentiation and tissue-specific regulation ( Meissner et al . , 2008; Schmidl et al . , 2009; Hodges et al . , 2011; Li et al . , 2012 ) . Hence , we asked whether differentially methylated sites across the three cell-types are more likely to be functionally relevant when variable within a population , hence more often associated to gene expression ( eQTMs ) or genetic variation ( mQTLs ) than non-differentiated sites across cell-types . We measured the level of differentiation for each site by calculating its median methylation level in each cell-type and then calculating the coefficient of variation between those three medians , which is a measure of variability that controls for the mean level of methylation ( Figure 2—figure supplement 3A ) . We find that as differentiation per methylation site increases , a higher proportion of eQTMs and mQTLs are observed ( Figure 2B ) . This is statistically supported by the fact that the tissue differentiation level of methylation sites involved in eQTMs and mQTLs is significantly higher than for non-eQTM and non-mQTL sites , respectively ( all p values<2 . 2 × 10−16 , Wilcoxon test , Figure 2—figure supplement 3B , C ) . These results show that the same methylation sites marking or contributing to tissue differentiation are participating in inter-individual variability that is highly determined by genetic variation and associated to gene expression . This could indicate that the establishment of differentially methylated regions during development could be delineating a backbone where local inter-individual changes may occur . In order to further understand the nature of these inter-individual changes we analyzed their participation in different genomic contexts . We find a significant increased presence of negative compared to positive eQTMs in CTCF binding sites , enhancers and promoters , with this increase being higher in non-CpG island ( CGI ) promoters compared to CGI promoters ( Figure 2C ) . Interestingly , we discover a significant depletion of mQTLs in CGI promoters , and a significant enrichment of mQTLs in non-CGI promoters ( Figure 2D ) . This shows that methylation sites in non-CGI promoters are under a stronger genetic control than at CGI promoters , where methylation is in general robustly maintained at low levels . Nevertheless , genetic variation in CpG islands commonly affects gene expression , since we find that eQTLs are significantly enriched at these genomic regions ( all p values<0 . 03 ) . Overall , these results suggest that the role of DNA methylation can be highly dependent on the genomic and functional context . Allele specific expression ( ASE ) , seen as a signal of regulatory difference between two haplotypes of an individual , can in theory be driven either by genetic regulatory variation or epigenetic ( in ) activation of one of the two alleles , for example , by DNA methylation . In order to test whether ASE is caused by genetic variation or differential DNA methylation , we compared the magnitude of allelic imbalance in eQTL genes between individuals who are heterozygous for the eQTL and their respective homozygotes . Similar to other studies ( Dimas et al . , 2009; Pickrell et al . , 2010; Montgomery et al . , 2010 ) , at a genome-wide level a significantly greater allelic imbalance is associated with heterozygote eQTLs in all three cell-types ( p values<3 . 0 × 10−8 , Figure 3A ) . In order to test whether ASE is driven by haplotype differences in DNA methylation alone , potentially represented by semimethylated sites ( partially methylated , β-value >0 . 3 and <0 . 7 ) , we analyzed the genes having eQTM methylation sites that are not associated with SNPs ( filtered out SNP-methylation correlations with nominal p<0 . 01 ) . A comparison of allelic imbalance between individuals with semimethylated eQTMs , and homomethylated ( i . e . , fully methylated , β-value > 0 . 7 , or unmethylated , β-value < 0 . 3 ) eQTMs , revealed no significant differences ( p values>0 . 42 , Figure 3A ) . Additionally , ASE driven by heterozygote eQTLs is significantly higher than that of semimethylated eQTMs ( p values<7 . 5 × 10−3 , Figure 3A ) . Furthermore , as a positive control , we observe that allelic imbalance in imprinted genes , known to be driven by allelic DNA methylation , is significantly higher than in homomethylated eQTMs ( p values<9 . 7 × 10−4 , Figure 3A ) . Thus , while ASE is shown to be significantly driven by genetic variation ( or by DNA methylation in imprinted genes ) , we find no evidence in our data of methylation alone contributing to ASE . This argues that DNA methylation is rarely allele-specific in the absence of DNA sequence variation effects , and while we cannot exclude other possible epigenetic sources of ASE , the results suggest that the widespread ASE across the genome may be primarily driven by common and rare genetic regulatory variants . 10 . 7554/eLife . 00523 . 017Figure 3 . DNA methylation associated to gene expression is not significantly allelic and can interact with genetic variation . ( A ) In green are depicted the distributions of allelic imbalance ( i . e . , absolute distance from the expected 0 . 5 ratio ) of assayable heterozygote sites in eQTL genes of individuals that are homozygous ( HOM ) or heterozygous ( HET ) for the eQTL . The difference between distributions is significant in all cell-types with p<2 . 2 × 10−16 , p=2 . 7 × 10−15 and p=3 . 0 × 10−8 in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) , respectively ( Wilcoxon test ) . This strongly indicates that allele specific expression is significantly driven by genetic variation . In purple are shown the distributions of allelic imbalance of assayable heterozygote sites in eQTM genes ( excluding methylation sites affected by genetic variation ) of individuals that are homomethylated ( HOMOMETH , i . e . , fully methylated , β-value > 0 . 7 , or unmethylated β-value < 0 . 3 ) or semimethylated ( SEMIMETH , i . e . , β-value >0 . 3 and <0 . 7 ) for the eQTM site . The difference between distributions is not significant in any cell-type , with p=0 . 79 , p=0 . 42 and p=0 . 49 in F , L , T , respectively . The difference between distributions of HET eQTLs and SEMIMETH eQTMs is significant in all cell-types with p=7 . 5 × 10−3 , p=3 . 4 × 10−7 , p=1 . 1 × 10−13 , in F , L , T , respectively . The difference between distributions of IMPRINTED genes and HOMOMETH eQTMs is significant in all cell-types with p=6 . 9 × 10−34 , p=9 . 7 × 10−4 , p=8 . 5 × 10−5 in F , L , and T , respectively . This shows that allele specific expression is not significantly driven by DNA methylation that is not affected by genetic variation . ( B ) Using linear regression we tested the interaction term as shown in the illustrated formula for the effects of SNPs ( green tick ) and methylation sites ( purple lollipop ) on gene expression ( black squared arrow and boxes ) . Qqplots illustrate the enrichment of low synergistic interaction observed p values ( black ) , together with the 5th and 95th confidence limits based on expression permutations ( gray ) with respect to the expected uniform distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 017 We then sought to explore the mechanistic relationships among DNA methylation , gene expression and genetic variation . We first tested whether there was a significant enrichment of synergistic interactions between genetic variants and DNA methylation on gene expression using linear regression . We selected for each exon all eQTLs with p<1 × 10−4 that fell in independent recombination intervals and all eQTMs with p<0 . 001 . To avoid artificial inflation of significant interactions , we filtered out any exon-SNP-meth triplets where the SNP and methylation site correlated with p<0 . 05 . To avoid spurious interactions caused by outliers , we further filtered out cases where there were less than four individuals homozygous for the minor allele of a SNP . Finally , to further account for remaining correlation between the SNP and the methylation we permuted the expression values 1000 times to infer the 95% confidence intervals and assess the significance of the enrichment of low p values . Synergistic interactions are enriched in LCLs and T-cells ( Figure 3B ) with π1 being 4 . 3% and 9 . 3% ( all p<0 . 001 ) , which reflects the percent of estimated true positives from the p value distributions ( Storey and Tibshirani , 2003 ) . In fibroblasts , although the p value distribution is not above the 95th confidence limit ( Figure 3B ) , the observed π1 is 12 . 4% ( p<0 . 001 ) , 7 . 2 times larger than the top permuted π1 , which suggests a significant enrichment of interactions . Finally , we find 3 , 91 and 14 individually significant interactions in fibroblasts , LCLs and T-cells , respectively , at 10% false discovery rate ( FDR ) . Overall , these results reflect the interdependency of genetic and epigenetic variation to determine gene expression levels . We further dissected the causative relationships between DNA methylation and gene expression by considering that the SNP triggers the causal network since its state is not modifiable in time . We used Bayesian Network ( BN ) construction and relative likelihood ( see ‘Materials and methods’ ) to test which of the three possible causative models depicted in Figure 4A and described below is the most likely given the data for each set of variables . Under the ‘INDEP’ ( independent ) model , a SNP affects independently gene expression and DNA methylation ( passive role of methylation ) . Under the ‘SME’ ( SNP-methylation-expression ) model , the SNP affects methylation , which then affects gene expression ( active role ) . The ‘SEM’ ( SNP-expression-methylation ) model requires that a SNP affects gene expression , and expression then affects methylation ( passive role ) . We tested the relative likelihood of these models choosing a non-biased approach where at least two of the three pairwise correlations between genetic variation , gene expression and DNA methylation are significant , yielding 831–2928 SNP-methyl-exon triplets tested per cell-type . All three models occur , depending on cell-type and sign of correlation between DNA methylation and gene expression ( Figure 4B , Figure 4—figure supplement 1A ) . The INDEP model , that reflects a passive role for DNA methylation , is the most likely model in fibroblasts and LCLs . However , in T-cells the SME model , where methylation takes an active causative role , is the major contributor . Note that in the SEM model , where methylation is passive being influenced by gene expression levels , a higher likelihood of pos-eQTMs is found compared to the SME model . Overall , these results suggest that DNA methylation can be both active , by being a likely cause of gene expression variation levels , or passive , by being a consequence or an independent mark of gene expression levels . 10 . 7554/eLife . 00523 . 018Figure 4 . Passive and active roles of DNA methylation in gene regulation . ( A ) Illustration of the three possible causative models tested of mechanistic relationships between genetic variation ( SNP ) , DNA methylation ( methyl ) and gene expression ( expr ) . Arrows indicate the causal direction of effects . The name of each model is underlined . ( B ) Mosaicplots illustrate the relative likelihoods of each model ( x-axis ) , partitioned by the relative likelihoods of those involving pos-eQTMs ( yellow ) and neg-eQTMs ( blue; y-axis ) , in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) . The three types of models ( INDEP , SME and SEM ) are present in the three cell-types , suggesting that DNA methylation can have both active and passive roles in gene regulation . See Figure 4 –figure supplement 1 and Figure 4–Source data 1 . ( C ) Heatmap of p value relative frequency distributions of spearman correlations between transcription factors ( TF ) and DNA methylation levels of eQTMs at their binding sites , sorted by π1 . The enrichment of significant associations can be appreciated by the accumulation of reddish colors , reflecting higher relative frequencies , at low p values , and yellowish colors , reflecting lower relative frequencies , at higher p values . These results highlight one of the possible mechanisms of a passive role of DNA methylation regarding gene expression . See Figure 4—figure supplements 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 01810 . 7554/eLife . 00523 . 019Figure 4—source data 1 . High confidence calls for INDEP , SME and SEM models in each cell-type . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 01910 . 7554/eLife . 00523 . 020Figure 4—figure supplement 1 . Passive and active roles of DNA methylation on gene regulation . ( A ) For all exon-SNP-methyl triplets tested the best model was inferred using Bayesian networks ( BN ) and relative likelihood . Mosaicplots illustrate the relative frequencies of each model ( x-axis ) , partitioned by the relative frequencies of pos-eQTMs ( yellow ) and neg-eQTMs ( blue; y-axis ) , in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) . ( B ) Same as in ( A ) only including high confidence calls ( see ‘Materials and methods’ and Figure 4—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 02010 . 7554/eLife . 00523 . 021Figure 4—figure supplement 2 . Associations between TF abundance and DNA methylation at their target binding sites . Observed ( left panels ) and expected ( right panels ) p value distributions of spearman correlations between transcription factor expression levels and DNA methylation levels of eQTMs at their binding sites ( see ‘Materials and methods’ ) in fibroblasts ( F ) , LCLs ( L ) and T-cells ( T ) . π1 statistic , reflecting the fraction of true positives , is indicated on top of each p value distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 02110 . 7554/eLife . 00523 . 022Figure 4—figure supplement 3 . Interactions between SNPs and transcription factor levels on DNA methylation . ( A ) We tested whether genetic variants that could potentially alter TF binding would interact with TF abundance ( transcription level ) on their effect on DNA methylation levels . To do so we took the top mQTL SNPs that fell in TF peaks and whose associated methylation site correlated significantly with the TF expression level of that peak ( 10% FDR for both types of associations ) , and required that the SNP and the TF abundance were not correlated and that the SNP had at least four minor allele homozygotes ( N = 114 ) . ( B ) We find an enrichment of low p values for interactions between genetic variants and TF abundance on DNA methylation levels , with π1 estimating 15% of true positives . ( C ) Plotted is the top interaction with p=1×10−4 involving the TF c-Jun in T-cells . The SNP falls in the binding peak of c-Jun . The methylation site and the SNP are 172 bp away from each other and both fall in an intron of gene SLC9A9 . DOI: http://dx . doi . org/10 . 7554/eLife . 00523 . 022 In order to obtain a set of high confidence calls for each of the models , we have required that the BN calls are confirmed by an independent method called the Causal Inference Test ( Millstein et al . , 2009 ) ( see details in ‘Materials and methods’ ) . From the total number of tests , 61% , 36% and 27% were called as high confidence ( HC ) in fibroblasts , LCLs and T-cells , respectively ( Figure 4—source data 1 ) . The relative frequencies of these HC calls look similar to the general relative likelihood space of the models ( Figure 4—figure supplement 1B ) . As an example of a HC INDEP model , we identified in fibroblasts a case occurring at the promoter of gene DPYSL4 and involving a methylation site associated to age and age rate in blood samples ( Hannum et al . , 2013 ) . In this example , SNP rs12772795 affects independently the DNA methylation status of site cg05652533 and the expression level of gene DPYSL4 , possibly via an effect on the binding of CTCF nearby ( given binding peak reported ) , which is a factor known to alter DNA methylation levels locally ( Stadler et al . , 2011 ) . As an example of an SME model , we have identified in T-cells SNP rs1362125 that could be affecting the binding of SP1 ( overlapping peak reported ) , which is a factor shown to confer methylation protection ( Boumber et al . , 2008 ) . This could then alter the methylation state of site cg24703717 ( 78bp away ) that falls in a YY1 binding peak , a factor whose binding is known to be sensitive to DNA methylation levels ( Kim et al . , 2003 ) . Hence methylation could actively alter binding of YY1 , negatively affecting the expression of the gene HLA-F . Finally , we would like to highlight in LCLs an SEM scenario in which SNP rs3733346 , located in a DNAse hypersensitive site , affects the expression of gene DGKQ , whose transcriptional activity could then be influencing positively DNA methylation levels of site cg00846425 located in its gene body , as has been suggested to be a possible phenomenon for gene body DNA methylation ( Hahn et al . , 2011 ) . To understand potential molecular causes for the passive role of methylation in the INDEP model we postulated that SNPs could be influencing binding levels of DNA-binding factors ( hereon called transcription factors or TFs ) , which in turn affect the methylation status of a site in parallel to their effect on gene expression . It has been shown in mice ( Stadler et al . , 2011 ) that TFs can influence DNA methylation levels near their binding sites ( TFBSs ) . In addition , by correlating expression levels of TFs with DNA methylation at their TFBSs in different cell-types , this pattern has also been observed for some TFs in human tissues ( Thurman et al . , 2012 ) . In order to test whether human inter-individual differences of TF expression levels would affect DNA methylation that is itself associated to gene expression we used the ENCODE dataset of TFBSs based on ChIP-seq assays for 111 TFs ( Bernstein et al . , 2012; Gerstein et al . , 2012 ) , and correlated their expression levels with eQTM methylation levels found at their reported binding sites . We found a strong enrichment of significant TF-methylation associations , with π1 being 18% , 9% and 25% , in fibroblasts , LCL , and T-cells , respectively ( Figure 4—figure supplement 2 ) . At 10% FDR , we find significant associations for 27 , 47 and 99 different TFs in fibroblasts , LCL and T-cells , respectively . A strong enrichment of significant associations can also be appreciated for each TF individually as depicted in Figure 4C . We further tested whether genetic variants that could potentially alter TF binding would interact with TF abundance on their effect on DNA methylation levels ( see ‘Materials and methods’ ) . Despite the limited number of cases that could be tested ( N = 114 ) , we find an enrichment of low p values for such interactions , with π1 estimated at 15% of true positives , and the top interaction involving the TF c-Jun in T-cells ( p=1 × 10−4 , Figure 4—figure supplement 3 ) . These results suggest that TF levels could be influencing DNA methylation and gene expression levels simultaneously . It is possible that in some cases this could occur via the interaction with DNA sequence variation . This could illustrate a realistic model for passive but at the same time correlated levels of DNA methylation with gene expression . In this study , we have shown that inter-individual DNA methylation changes are mechanistically associated to genetic variation and gene expression in complex and context dependent ways that can be of passive or active nature . DNA methylation levels can depend on TF binding ( Stadler et al . , 2011 ) . TF binding levels in turn can be determined by both inter-individual differences of TF abundance and genetic variants at their binding sites . This picture shows how DNA methylation can be linked to genetic variation , and can participate in gene regulation in a passive manner . This scenario is quite different from the often-assumed simple model of genetic variation affecting the epigenome , which in turn affects the transcriptome , and furthermore these relationships can be of different nature depending on tissue and genomic region . This is likely to be true for other epigenetic mechanisms as well , and properly characterizing these relationships will enhance our understanding of genome function and complex disease . Additionally , our results suggest that non-genetically determined DNA methylation is rarely allelic but rather defined at the whole cell level . That is , when we observe a population of cells to be semimethylated the most likely scenario is that only a fraction of the cells are fully methylated rather than one of the alleles in each of the cells is methylated . Nevertheless , given the limitations of our data , further studies addressing this question are needed to confirm our results . Further , we show that methylation sites that represent a repressive state with respect to tissue differentiation can also participate in positive and negative associations with gene expression when looking at inter-individual variability . DNA methylation as a whole emerges both as marker and determinant of cellular identity and may provide a platform by which genetic variation and other variable factors among individuals ( e . g . , TFs ) exert their effect to differentially influence cellular processes . Fifty-six of the 204 newborn umbilical cord and cord blood samples are part of the 85 samples collected previously in Dimas et al . ( 2009 ) . The remaining 148 samples have been newly collected , and primary fibroblasts , EBV-immortalized lymphoblastoid cell lines and primary T-cells have been prepared and grown as stated previously ( Dimas et al . , 2009 ) , with a few exceptions: the preparation of LCLs was done with 1 ml of 20 million re-suspended cells and 1 ml of EBV that were transferred to a 24-well plate . In addition , to confirm that the cord and cord blood acquired from the hospital belonged to the same individual , DNA was extracted from cord tissue and LCLs separately with the Puregene cell kit ( Gentra-Qiagen , Venlo , The Netherlands ) , microsatellite markers were assayed with QF-PCR and compared between the two samples coming from the same individual . We extracted DNA from LCLs with the QIAamp DNA Blood Mini Kit ( Qiagen , Venlo , The Netherlands ) and genotyped the 204 individuals with the Illumina 2 . 5M Omni chip . We applied several filters before imputation with PLINK v1 . 07 ( Purcell et al . , 2007 ) . We filtered out any SNPs with >5% missing genotypes , minor allele frequency ( MAF ) <1% , gene call score <0 . 25 for more than 1% of samples , or Hardy-Weinberg equilibrium p<1 × 10−6 . This left 1 , 535 , 724 SNPs before imputation . We performed the imputation with Beagle v3 . 3 . 2 ( Browning and Browning , 2009 ) , into 21 million European panel SNPs of the 1000 genomes March 2012 release ( Abecasis et al . , 2012 ) . From the final output , we required an R2 ≥ 0 . 9 . This yielded a total of 6 . 9 million SNPs . For association analyses ( eQTLs and mQTLs ) we filtered out genetic outliers based on multidimensional scaling using HapMap populations ( Altshuler et al . , 2010; Figure 1—figure supplement 1 ) , and we further filtered out any SNPs with MAF <5% , yielding sets of 5 , 209 , 348–5 , 278 , 330 SNPs . RNA was extracted from LCLs , fibroblasts and T-cells with RNeasy columns ( Qiagen , Venlo , The Netherlands ) or Trizol ( Invitrogen , Carlsbad , CA ) . RNA samples were quantified with NanoDrop ( Thermo Scientific , Waltham , MA ) and Qubit ( Invitrogen , Carlsbad , CA ) , and analyzed with a 2100 Bioanalyzer ( Agilent , Santa Clara , CA ) . Samples were prepared for sequencing with the Illumina mRNA-seq and TruSeq sample preparation kits ( Illumina , San Diego , CA ) as indicated by manufacturer’s instructions . In these library preparation procedures , poly-A RNA is selected using poly-T oligo-attached magnetic beads; then the RNA is cleaved , converted to first strand cDNA and after RNA digestion and second DNA strand synthesis , the fragments are end repaired and ligated to the adapters containing specific primer indexes . The cDNA libraries are then PCR amplified . Libraries were sequenced in either sets of 6 samples per lane in the Genome Analyzer II machine or 12 samples per lane in the HiSeq2000 machine , randomly pooling in the same lane samples from different cell-types and individuals . Afterwards , the 49-bp sequenced paired-end reads were mapped to the reference genome ( Lander et al . , 2001 ) GRCh37 with BWA v0 . 5 . 9 ( Li and Durbin , 2009 ) . Using SAMtools ( Li et al . , 2009 ) , we kept reads mapping uniquely to the genome , with MAPQ ≥ 10 and properly paired . In order to quantify exons in a non-redundant way , we created a set of merged exons from the GENCODE v10 annotation ( Harrow et al . , 2006 ) . In more detail , we took all protein coding and lincRNA transcripts and merged any overlapping exons into new exon units . We then counted the number of reads mapping to each exon unit ( Figure 1—figure supplement 2 ) . Technical outliers having less than 5M exonic reads or extremely low insert size mode were removed from the study . In order to keep only the set of exons that we considered expressed in each cell-type , we filtered out exons that had no reads mapped in 10% or more of the samples per cell-type . This yielded sets of 70 , 800–76 , 870 exons belonging to 12 , 265–12 , 863 expressed genes depending on the cell-type and analysis . Afterwards , we normalized the raw exon counts by scaling all libraries to 10 million reads , based on the total number of exonic reads per sample . In order to remove technical variance from our samples , we tested the effect of many different covariates on our scaled exon counts in each cell-type by linear regression . We decided to correct for mean GC content per library , run date , primer index and insert size mode , using a random effects model ( Figure 1—figure supplement 3 ) . Thus , our final exon expression levels are the residuals after correcting for those covariates plus the corrected mean ( Figure 1—figure supplement 4 ) . DNA was extracted with the QIAmp DNA blood mini kit ( Qiagen , Venlo , The Netherlands ) and bisulfite converted ( BC ) using the Zymo Research EZ DNA MethylationTM Kit ( using the 50-column or 96-well formats; Zymo Research , Irvine , CA ) . The starting quantity of DNA was 1000 ng and after BC we normalized all our samples to have a final concentration of 40 ng/μl or the closest possible amount . BC-DNA was then processed through the 450K Illumina Infinium HD Methylation Assay ( Illumina , San Diego , CA ) according to manufacturer’s instructions . In brief , this is an assay that involves whole genome amplification , specific hybridization capture , and allele-specific single-base primer extension . Samples from the three cell-types were processed in a randomized manner . The controls available in the methylation assay for verifying the efficiency of our bisulfite conversion indicated that none of our samples had bad efficiency of bisulfite conversion . We removed three technical outliers that didn’t cluster well with any cell-type in a principal component analysis . From the 482 , 421 CpG sites assayed , we filtered out probes with 1000 genomes SNPs or indels at European minor allele frequency >5% and then any other probes that had any of the final set of SNPs used in our study . This yielded a total of 416 , 118 CpG sites to work with . We quantile normalized the data in the following way . The Illumina 450k array is composed of two probe types . Probe type 1 reads the methylated signal and unmethylated signal in the same channel ( red or green ) and probe type 2 reads the methylated signal in the green channel and the unmethylated signal in the red channel . The distribution of β-values before any normalization is quite different for each probe type , thus the two probe types were normalized separately . For probe type 1 , the intensity for the methylated and unmethylated signals were quantile normalized across individuals for each color channel separately . As the density of signal intensities showed that there was a color bias ( at least for the unmethylated probe ) , the color channels were quantile normalized within individuals for the methylated signal and unmethylated signal separately . For probe type 2 , the methylated signals and unmethylated signals were quantile normalized across individual . As for this probe the methylated signal is always in green and the unmethylated signal is always in red . The color bias needs to be corrected before computing β-values . In order to correct the color bias , the methylated ( unmethylated ) signal probe type 2 normalized across individuals is quantile normalized with the methylated ( unmethylated ) signal of probe type 1 that was previously quantile normalized across individuals and within for color correction . In a final step , the β-values of the two probe types were quantile normalized together leading to the final β-values ( Figure 1—figure supplements 5–7 ) . The β-value is defined as the proportion of fluorescent signal from the methylated allele over the total fluorescent signal ( methylated and unmethylated alleles ) ; this represents the percentage of methylation per site ( Bibikova et al . , 2006 ) . We performed Spearman rank correlations ( SRC ) between ( x and y variables ) : SNP genotypes and exon expression levels ( eQTLs ) in 183–185 samples using a 1-MB window to either side of the TSS , SNP genotypes and CpG site methylation levels ( mQTLs ) in 66–111 samples using a 5-kb window to either side of the CpG site , CpG site methylation levels and exon expression levels ( eQTMs ) in 66–118 samples using a 50-kb window to either side of the TSS . Different window sizes were tested for mQTL and eQTM analyses using a subset of 56 individuals . The final window size chosen is the one that yielded the highest enrichment of significant associations . As previously mentioned , association analyses involving genetic variation were ran with SNPs with MAF ≥ 5% and excluding genetic outlier individuals ( Figure 1—figure supplement 1 ) . Similar to previous analyses ( Stranger et al . , 2007; Montgomery et al . , 2010 ) , for each type of association analysis and each cell-type the phenotype belonging to the y variable was permuted 1000 times and the median p value distribution was summarized among the genes or CpG sites . For expression levels this was done on all exons belonging to 1000 randomly selected genes and for each gene the minimum p value distribution out of all its exons was chosen . For methylation levels , 50 , 000 random methylation probes were selected . Based on the null p value distributions yielded by the permutations , we selected for each analysis the associations that passed the permutation threshold that yielded 10% or less False Discovery Rate ( FDR ) in a per gene/CpG basis . We have used the coefficient of variation ( CV ) as a measure of DNA methylation differentiation level , as it is a variability measure that controls for the mean methylation . In detail , for each site we have calculated the standard deviation of the median methylation in each cell-type , divided by the mean of the three medians . By using the median methylation level per site for each cell-type , our differentiation measure is estimating the between cell-type variation only , not taking into account the within cell-type variation . Enhancers and CTCF peaks , were downloaded from the UCSC genome browser tables ( Browning and Browning , 2009 ) and come from ChIP-seq experiments of the ENCODE project ( Birney et al . , 2007; Rosenbloom et al . , 2010; Bernstein et al . , 2012 ) and specific groups ( Boyle et al . , 2008; Ernst and Kellis , 2010; Ernst et al . , 2011; Thurman et al . , 2012 ) . For LCLs we used data belonging to the cell line GM12878 . For fibroblasts we used data belonging to the cell line NHLF . Unfortunately , there was limited data available for T-cells so we used the GM12878 data , given its close lineage relatedness . CpG islands ( CGIs ) were downloaded from the UCSC genome browser ( Karolchik et al . , 2004 ) . The promoter region is defined as going from -1kb of the TSS to +2kb of the TSS , based on methylation pattern observed in this region ( Figure 1—figure supplement 7A ) . CGI promoters are defined as promoter regions overlapping any CGI . The genomic feature enrichment analyses of eQTLs and mQTLs were done by comparing the number of observed overlaps with those found in a null set . The null set selects SNP-exon or SNP-CpG pairs with MAF , distance and expression or methylation values matching the distribution of eQTLs and mQTLs , respectively , from genes or CpG sites lacking any significant associations . The differential genomic feature overlap analyses of eQTMs were done by comparing the number of observed overlaps for positive and negative eQTMs , with the number of expected overlaps in non-eQTM sites . Fisher’s exact tests were used for assessing the significance of each analysis . Intersections of genomic features were done with BEDTools v2 . 7 . 1 ( Quinlan and Hall , 2010 ) . The allelic imbalance ( i . e . , absolute distance from the expected 0 . 5 ratio ) was measured for assayable heterozygote sites . First , we excluded sites that are susceptible to allelic mapping bias: 1 ) sites with 50-bp mapability ( Karolchik et al . , 2004 ) <1 implying that the 50-bp flanking region of the site is non-unique in the genome , and 2 ) simulated RNA-Seq reads overlapping the site show >5% difference in the mapping of reads that carry the reference or non-reference allele . In all the analyses , we used mapping and base quality threshold ≥10 . Next , we calculated the expected reference allele ratio for each individual by summing up reads across all sites separately for each SNP allele combination after down-sampling reads of sites in the top 25th coverage percentile in order to avoid the highest covered sites having a disproportionally large effect on the ratios . These expected REF/TOTAL ratios ( typically 0 . 494–0 . 517 as measured by 10th and 90th quantiles across samples ) correct for remaining slight genome-wide mapping bias as well as GC bias in each individual . Finally , for all the sites covered by ≥16 reads in each individual , we calculated the absolute distance from the expected 0 . 5 ratio described above . Because our genotypes included imputed data , we verified the genotype from the RNA sequencing data by requiring the observation of both of the SNP alleles in at least one cell type per individual . The number of genes in the homozygote and heterozygote eQTL analysis in Figure 3A are 1473 and 1453 in fibroblasts , 2197 and 2177 in LCLs , and 1288 and 1284 in T-cells , respectively . The number of genes in the homo-methylated and semi-methylated eQTM analysis in Figure 3A are 279 and 106 in fibroblasts , 1738 and 794 in LCLs , 1544 and 969 in T-cells , respectively . The number of genes in the imprinted genes analysis in Figure 3A are 29 , 17 , 15 in fibroblasts , LCLs and T-cells , respectively . For measuring allelic imbalance in imprinted genes , all the filtering and requirements mentioned above were applied except we did not require that both SNP alleles are seen . Imprinted genes were taken from http://www . geneimprint . com . The expression values of the exon-SNP-methyl triplets tested were permuted 1000 times to assess the expected p value distributions for the interaction between genetic variation and DNA methylation on gene expression . For each round of permutations a π1 was calculated , creating a null distribution of π1 from which an empirical p value can be inferred ( as reported in the main text ) . Additionally , the 95% confidence intervals were calculated from the 1000 permutations of all tests in each cell-type . False discovery rates of the reported significant results were calculated using the QVALUE R package ( Storey and Tibshirani , 2003 ) . Bayesian networks ( BN ) are directed acyclic graphs where nodes represent random variables and edges represent the conditional dependences among nodes . The direction of the edges between two nodes can be interpreted as a causal relationship . BN have been used before to infer causality in genetic problems ( Schadt et al . , 2005; Zhu et al . , 2008 ) . Likelihood methods are commonly used to estimate the most likely network—that is , the set of causal relationships among the different variables that better agrees with the data . In a BN every node has an associated probability function and , together with the conditional dependencies represented by the edges , they conform the joint probability density of the network . BN satisfy the local Markov property—that is , each variable is conditionally independent of its non-descendants given its parent variables . The Markov property defines the decomposition of the joint probability density of the network into a set of local distributions . Thanks to the Markov property it is easy to calculate the likelihood of a given BN . Since different networks have different complexities , it is common to use a score that takes into account the network complexity instead of the raw likelihood to compare the networks . We used the R package bnlearn ( Scutari , 2010 ) to calculate the maximum likelihood of the networks and the Akaike Information Criterion ( AIC ) score . AIC = 2k − 2ln ( L ) , where k is the number of parameters and L is the maximum likelihood . To calculate how much better one network is compared to another , we used the relative likelihood of one network with respect to the other . If we have two networks , N1 and N2 , with fixed parameters and AIC ( N1 ) ≤ AIC ( N2 ) , then the relative likelihood of N2 with respect to N1 is defined as: exp ( ( AIC ( N1 ) −AIC ( N2 ) ) /2 ) . The networks we have compared ( mentioned in main text ) have five parameters each . To test the networks , we selected three datasets that were then merged and normalized for size in order to calculate the relative frequencies of the inferred best network . In the first dataset we selected the best eQTM and for that methylation site the best mQTL , not requiring statistical significance for the eQTL ( n = 219 , 1074 and 1128 in fibroblasts [F] , LCLs [L] , and T-cells [T] , respectively ) , which would enrich for the SME model since the eQTL is the most distant association . In the second dataset , for each exon we selected its best eQTL and its best eQTM; not requiring significance for the mQTL ( n = 287 , 1316 and 871 in F , L , T ) , which would enrich for the SEM model since the mQTL is the most distant association . And in the third , for each exon we selected its best eQTL and kept the cases where the SNP was the best mQTL for a methylation site; not requiring significance for the eQTM , which would enrich for the INDEP model since the eQTM is the most distant association ( n = 413 , 726 , 775 , in F , L , T ) . In order to create a list of high confidence calls , we have made use of a semi-parametric method , the Causal Inference Test ( Millstein et al . , 2009 ) , that requires a level of significance for calling two out of the three models . The CIT makes SME and SEM calls if either has a p<0 . 05 , makes no call if p<0 . 05 for both models ( highly infrequent ) and makes a call for the INDEP if neither SME or SEM has a significant p-value . We have decided to call a high confidence set of SME and SEM models when they are called as such by both the BN and the CIT methods . Since the INDEP calls in the CIT do not require significance , we called a high confidence INDEP if 1 ) it is called INDEP by both methods , and 2 ) its BN relative likelihood ( RL ) to the second best model is equal or higher to the median RL of the SME and SEM high confidence calls to their second best prediction . As explained in the text , the transcription factor ( TF ) expression levels were Spearman rank correlated to methylation levels of eQTM sites falling in their reported TF binding sites ( Bernstein et al . , 2012; Gerstein et al . , 2012 ) , excluding any sites within 1 Mb of the TF gene transcription start site . The number of expressed TFs in each cell-type is 94 , 106 and 105 , comprising 666 , 851 and 776 exons in fibroblasts , LCLs and T-cells , respectively . The number of targeted eQTM methylation sites in each cell-type is 459 , 3832 and 3497 in fibroblasts , LCLs and T-cells , respectively . The number of tests performed is 17 , 747 , 209136 and 152263 in fibroblasts , LCLs and T-cells , respectively . To show the expected uniform distribution of p values for these correlations and compare it with the ( highly enriched ) observed , we have permuted once the TF expression values for each TF-meth correlation in each cell-type and plotted the p-value distributions ( Figure 4—figure supplement 2 ) . To call significant cases for the target sites of each TF , the false discovery rate was calculated for each using the QVALUE R package ( Storey and Tibshirani , 2003 ) . We tested the interactions between SNPs and TF abundance on their effect on DNA methylation by taking the top mQTL SNPs that fell in TF peaks and whose associated methylation site correlated significantly with the TF expression level of that peak ( 10% FDR for both types of associations ) . We excluded cases in which the SNP and the TF abundance were correlated ( p<0 . 05 ) and in which the number minor allele homozygotes was less than four . This yielded a total set of 114 SNP-TF-meth triplets for which the interaction between SNP and TF abundance was tested by linear regression ( Figure 4—figure supplement 1 ) . Genotyping , RNA-seq and DNA methylation data have been submitted to the EMBL-EBI European Genome-Phenome Archive ( https://www . ebi . ac . uk/ega/ ) under accession number EGAS00001000446 . Zipped text file tables with the identified eQTLs , mQTLs and eQTMs for each cell-type are available as Table 1—source data 1 . Zipped text file tables with the high confidence SME , SEM and INDEP calls for each cell-type are available as Figure 4—source data 1 .	Variations occur throughout our genome . These variations can cause genes to be expressed ( switched on ) in slightly different ways among individuals . Moreover , the same gene can also be expressed in different ways in different cells within an individual . A third level of variation is supplied by epigenetic markers: these are molecules that bind to the DNA at specific points and can have profound effects on the expression of nearby genes . One such epigenetic marker is the addition of a methyl group to a cytosine base , a process that is known as DNA methylation . DNA methylation usually happens when a cytosine base is next to a guanine base , forming a CpG site . In mammals , most CpG sites have methyl groups attached , although regions with a lot of CpG sites ( called CpG islands ) are mostly unmethylated . Initial studies suggested that methylation prevented particular genes from being expressed , but more recent work has indicated that methylation can be associated with both reduced and increased expression of genes . Moreover , it is not clear if this association is active ( i . e . , changes in methylation drive changes in gene expression ) or passive ( DNA methylation is the result of gene regulation ) . Now , Gutierrez-Arcelus et al . have carried out a large-scale study to clarify the relationships between three different types of gene-related variations among individuals . They extracted fibroblasts , T-cells and lymphoblastoid cells from the umbilical cords of 204 babies , and analysed them for variations in DNA sequence , gene expression and DNA methylation . Their results show that the associations between the three are more complex than was previously thought . Gutierrez-Arcelus et al . show that the mechanisms that control the association between the variations in DNA methylation and gene expression in individuals are likely to be different to those that are responsible for the establishment of methylation patterns during the process of cell differentiation . They also find that the association between DNA methylation and gene expression can be either active or passive , and can depend on the context in which they occur in our genome . Finally , where the two copies or alleles of a gene are not equally expressed in a given cell , the difference in expression is primarily regulated by DNA sequence variation , with DNA methylation having little or no role on its own . Equally complex interactions and effects are expected in further studies of genetic and epigenetic variation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2013	Passive and active DNA methylation and the interplay with genetic variation in gene regulation
The sensory and supporting cells ( SCs ) of the organ of Corti are derived from a limited number of progenitors . The mechanisms that regulate the number of sensory progenitors are not known . Here , we show that Fibroblast Growth Factors ( FGF ) 9 and 20 , which are expressed in the non-sensory ( Fgf9 ) and sensory ( Fgf20 ) epithelium during otic development , regulate the number of cochlear progenitors . We further demonstrate that Fgf receptor ( Fgfr ) 1 signaling within the developing sensory epithelium is required for the differentiation of outer hair cells and SCs , while mesenchymal FGFRs regulate the size of the sensory progenitor population and the overall cochlear length . In addition , ectopic FGFR activation in mesenchyme was sufficient to increase sensory progenitor proliferation and cochlear length . These data define a feedback mechanism , originating from epithelial FGF ligands and mediated through periotic mesenchyme that controls the number of sensory progenitors and the length of the cochlea . The Organ of Corti contains mechanosensory hair cells ( HC ) and specialized supporting cells ( SC ) that are required for the transduction of sound ( Wu and Kelley , 2012 ) . The frequency spectrum of sound stimuli is tonotopically represented along the length of the mammalian cochlea ( Fay and Popper , 2000 ) . In mouse , the cochlea begins to grow from the ventral otic vesicle at embryonic day 11 . 5 ( E11 . 5 ) and continues to grow and coil , forming approximately one and a half turns by birth . During its development , the length of the cochlea is limited by the number of progenitors that give rise to sensory HCs and SCs , and is further regulated through a process of convergent extension ( Chen and Segil , 1999; Montcouquiol et al . , 2003; Wang et al . , 2005; Wu and Kelley , 2012 ) . In mouse , sensory progenitors exit the cell cycle by E14 . 5 and begin to differentiate into HCs and SCs . Thus , the size of the progenitor population at this stage of development is the ultimate determinant of the size of the adult cochlea . Progenitor number is determined by proliferation , the timing of differentiation , and in some cases by aberrant cell death . Previous studies indicate that sensory progenitor growth requires mesenchymal signals ( Phippard et al . , 1999; Montcouquiol and Kelley , 2003; Braunstein et al . , 2008 , 2009 ) , however , the identity and source of the factors that control this activity are not known . Fibroblast Growth Factors ( FGFs ) have several stage-specific functions during inner ear development . FGF3 and FGF10 signal from hindbrain and head mesenchyme , respectively , to the overlying ectoderm to induce formation of the otic placode and vesicle ( Urness et al . , 2010 ) . Later in development , FGF20 regulates differentiation of outer hair cells ( OHC ) and SCs , termed the lateral compartment of the cochlea ( Huh et al . , 2012 ) . Phenotypic similarities with mice lacking Fgfr1 in the entire otic epithelium suggest that FGF20 signals directly to FGFR1 , serving as a permissive factor for differentiation ( Pirvola et al . , 2002; Hayashi et al . , 2008; Huh et al . , 2012 ) . FGF9 signaling regulates structural components of the vestibular system , but alone has no effect on cochlear development ( Pirvola et al . , 2004 ) . During postmitotic stages , FGF8 signaling from the inner hair cell ( IHC ) to FGFR3 in SCs regulates pillar cell differentiation ( Colvin et al . , 1996; Mueller et al . , 2002; Jacques et al . , 2007 ) . Here , we identify another critical stage in inner ear development that requires FGF signaling . We show that Fgf9 , expressed in the non-sensory epithelium , and Fgf20 , expressed in the sensory epithelium , regulate the number of cochlear progenitors and the ultimate length of the cochlea through signaling to mesenchymal FGFRs . We find that in vivo FGF9/20 signaling to mesenchymal FGFR1 and FGFR2 is required for sensory progenitor proliferation and that mesenchymal FGFR signaling is sufficient to promote sensory progenitor proliferation and extend the length of the cochlear duct . In addition , we show that prosensory epithelial FGFR1 and FGF20 independently is required for differentiation of outer HCs and SCs . In a prior study , we showed that Fgf20 is required between E13 . 5–14 . 5 for differentiation of cochlear OHCs and SCs in the organ of Corti ( Huh et al . , 2012 ) . However , Fgf20 is expressed in a portion of the otic vesicle sensory epithelium much earlier in development , beginning at E10 . 5 ( Huh et al . , 2012 ) , but analysis of mice lacking Fgf20 did not reveal any function for Fgf20 at this stage of development . Since there are many examples of FGFs functioning redundantly during development , we hypothesized that redundancy could account for the lack of a phenotype in Fgf20 null inner ears between E10 . 5 and 12 . 5 . Fgf9 is closely related to Fgf20 ( Zhang et al . , 2006; Itoh and Ornitz , 2008 ) , and is also expressed in the otic epithelium at E10 . 5–12 . 5 ( Pirvola et al . , 2004 ) ; however , Fgf9−/− mice have normal cochlear development and normal patterning of the organ of Corti ( Pirvola et al . , 2004 ) . We first examined the expression domain of Fgf9 relative to Sox2-expressing sensory progenitors ( and Fgf20 ) using a new Fgf9-βGal reporter allele ( Fgf9lacZ ) in which a splice acceptor-lacZ gene was inserted into the first intron of Fgf9 ( Skarnes et al . , 2011 ) . At E10 . 5 , βGal activity was detected in the otic vesicle epithelium ( Figure 1A ) . Co-staining of βGal and Sox2 at E11 . 5 showed no overlap , indicating that Fgf9 is expressed in the non-sensory epithelium of the otic vesicle ( Figure 1B ) . Taken together with previous Fgf20 expression analysis at this stage ( Huh et al . , 2012 , Figure 1C ) , Fgf9 and Fgf20 are both expressed in the otic vesicle , but in non-overlapping domains in the otic epithelium ( Figure 1D ) . 10 . 7554/eLife . 05921 . 003Figure 1 . Fgf9 and Fgf20 are expressed in distinct regions of the otic vesicle . ( A ) βGal activity in an Fgf9lacZ/+ embryo at E10 . 5 visualized with xGal staining . ( B , C ) βGal ( red ) and Sox2 ( green ) co-immunostaining showing that Fgf9 ( B ) is expressed in Sox2- non-sensory epithelium and Fgf20 ( C ) is expressed in Sox2+ sensory epithelium at E11 . 5 . ( D ) Schematic diagram of FGF9 , FGF20 , and Sox2 immunostaining showing that FGF9 and FGF20 are expressed in distinct domains in the otic vesicle . ov , otic vesicle , scale bars , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 003 To determine whether Fgf9 and Fgf20 could have a redundant role in cochlear development , Fgf9;Fgf20 double knockout cochleae were analyzed at E18 . 5 by staining with phalloidin and with an antibody to p75 to identify sensory HCs and pillar cells , respectively ( Figure 2A–C ) . Control embryos ( Fgf9−/+;Fgf20lacZ/+ ) showed a normal pattern of three rows of OHCs and one row of IHCs throughout the cochlear duct ( Figure 2A–C ) . Fgf9−/− and Fgf9−/−;Fgf20lacZ/+ cochleae showed the same wild type HC pattern as the double heterozygous controls ( Figure 2A–C ) . Fgf20lacZ/lacZ and Fgf9−/+;Fgf20lacZ/lacZ cochleae showed patches of sensory HCs and gaps ( Figure 2A–C ) ( Huh et al . , 2012 ) . Fgf9−/−;Fgf20lacZ/lacZ cochleae also showed a similar patterning phenotype to Fgf20lacZ/lacZ mice ( Figure 2A–D ) . The density ( number of cells per 100 μm ) of OHCs in Fgf9−/− and Fgf9−/−;Fgf20lacZ/+ cochleae was similar to double heterozygous controls ( Figure 2E ) . However , the densities of OHCs in Fgf20lacZ/lacZ , Fgf9−/+;Fgf20lacZ/lacZ and Fgf9−/−;Fgf20lacZ/lacZ cochleae were similar to each other ( ANOVA , p > 0 . 1 ) , and significantly ( ANOVA , p < 0 . 0001 ) decreased compared to double heterozygous controls ( Figure 2E ) . Densities of IHCs were comparable in all genotypes ( Figure 2E ) . To analyze SCs , cochleae were immunostained for Prox1 and Sox2 ( Figure 2D ) . In double heterozygous control cochleae , 5 rows of Prox1+ SCs overlapped with Sox2 staining ( Figure 2D ) . Fgf9−/− and Fgf9−/−;Fgf20lacZ/+ cochleae showed a similar pattern ( Figure 2D ) . In contrast , Fgf20lacZ/lacZ , Fgf9−/+;Fgf20lacZ/lacZ and Fgf9−/−;Fgf20lacZ/lacZ cochleae showed patches of SCs separated by gaps of Sox2+ , Prox1− cells ( Figure 2D ) . The density of SCs in Fgf9−/− and Fgf9−/−;Fgf20lacZ/+ was comparable ( ANOVA , p > 0 . 5 ) to double heterozygous control ( Figure 2F ) . The density of SCs in Fgf20lacZ/lacZ , Fgf9−/+;Fgf20lacZ/lacZ and Fgf9−/−;Fgf20lacZ/lacZ were similar to each other ( ANOVA , p > 0 . 3 ) and significantly ( ANOVA , p < 0 . 0001 ) decreased compared to double heterozygous controls ( Figure 2F ) . 10 . 7554/eLife . 05921 . 004Figure 2 . Fgf9 and Fgf20 regulate cochlear length . ( A , B ) Phalloidin ( A ) and p75 immunostaining ( B ) of E18 . 5 whole cochlea showing hair cells ( HCs ) ( phalloidin ) and pillar cells ( p75 ) in the cochlear duct of Fgf9−/+;Fgf20lacZ/+ , Fgf9−/− , Fgf20lacZ/lacZ , Fgf9−/−;Fgf20lacZ/+ , Fgf9−/+;Fgf20lacZ/lacZ and , Fgf9−/−;Fgf20lacZ/lacZ embryos . ( C ) Phalloidin ( green ) and p75 immunostaining ( red ) showing the orientation of HCs , pillar cells , and gaps in the sensory epithelium . ( D ) Prox1 ( green ) and Sox2 ( red ) co-immunostaining showing supporting cells ( SCs ) ( yellow , Prox1 and Sox2 ) and undifferentiated sensory progenitors ( red , Sox2 ) . ( E–G ) Measurement of HC number ( E ) , SC number ( F ) , and length of cochleae ( G ) of E18 . 5 embryos . Scale bars , A , 500 μm; C , 100 μm . For statistical analysis , all samples were compared with Fgf9−/+;Fgf20lacZ/+ double heterozygous controls . p < 0 . 001 . Sample numbers ( n ) are indicated in data bars . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 004 One of the striking differences among Fgf9;Fgf20 compound mutants was cochlear length . The length of Fgf9−/− and Fgf9−/−;Fgf20lacZ/+ cochleae was comparable ( ANOVA , p > 0 . 1 ) to that of double heterozygous controls ( Figure 2G ) , whereas the length of Fgf20lacZ/lacZ and Fgf9−/+;Fgf20lacZ/lacZ cochleae was 16% and 18% shorter than double heterozygous controls ( p < 0 . 05 and p < 0 . 001 ) , respectively , and the length of Fgf9−/−;Fgf20lacZ/lacZ cochleae was reduced by 58% compared to controls ( p < 0 . 001 ) ( Figure 2G ) . In addition , Fgf9−/−;Fgf20lacZ/lacZ double knockout cochleae were 49% and 51% of the length of Fgf20lacZ/lacZ and Fgf9−/+;Fgf20lacZ/lacZ cochleae ( p < 0 . 001 ) , respectively ( Figure 2G ) . These data identify a redundant role for Fgf9 and Fgf20 to attain the proper cochlear length , while Fgf20 , alone , primarily regulates cochlear patterning and differentiation . We hypothesized that the overall length of the cochlear duct would correlate with the size of the postmitotic prosensory domain . In mouse , cochlear sensory progenitors exit the cell cycle beginning at E12 . 5 in the apex and progressing towards the base by E14 . 5 ( Lee et al . , 2006 ) . E14 . 5 cochleae were dissected and immunostained for Sox2 ( Figure 3A ) , which marks the lineage of cells that will become HCs and SCs ( Kiernan et al . , 2005 ) . The Sox2+ prosensory domain of double heterozygous control and Fgf9−/+;Fgf20lacZ/lacZ inner ears were similar ( Figure 3A ) . However , the Sox2+ prosensory domain of Fgf9−/−;Fgf20lacZ/lacZ inner ears were clearly smaller than that of double heterozygous control or Fgf9−/+;Fgf20lacZ/lacZ cochleae ( Figure 3A ) . Immunostaining of histological sections of E14 . 5 inner ears showed that the Sox2+ prosensory domain was less compact in Fgf9−/−;Fgf20lacZ/lacZ inner ears compared to double heterozygous control and inner ears with one wild type allele of Fgf9 ( Figure 3B ) . Immunostaining for p27kip1 ( Cdkn1b ) showed a very similar pattern to that of Sox2 , with more diffuse cells in the prosensory domain of E14 . 5 inner ears of Fgf9−/−;Fgf20lacZ/lacZ mice ( Figure 3C ) . Jag1 , which marks the medial prosensory cells that will give rise to IHCs , inner SCs , and Kölliker's organ ( Ohyama et al . , 2010; Basch et al . , 2011 ) , showed a similar expression pattern across all three genotypes , indicating that the medial compartment of the cochlea was correctly specified ( Figure 3D ) . 10 . 7554/eLife . 05921 . 005Figure 3 . Fgf9 and Fgf20 are required for sensory progenitor proliferation . ( A ) Sox2 immunostaining of whole E14 . 5 cochlea to identify the progenitor domain ( arrows ) . ( B–D ) Sox2 ( B ) , p27 ( C ) , and Jag1 ( D ) immunostaining of E14 . 5 Fgf9−/+;Fgf20lacZ/+ , Fgf9−/+;Fgf20lacZ/lacZ , and Fgf9−/−;Fgf20lacZ/lacZ embryo sections . Boxed regions of the cochlear duct are magnified below each image and were chosen in regions where the sections perpendicularly transect the cochlear duct . ( E , F ) Sox2 and phospho-Histone H3 ( pHH3 ) co-immunostaining of E11 . 5 ( E ) and E12 . 5 ( F ) Fgf9−/+;Fgf20lacZ/+ , Fgf9−/+;Fgf20lacZ/lacZ and , Fgf9−/−;Fgf20lacZ/lacZ embryo sections . ( G , H ) Measurement of Sox2+ sensory progenitor proliferation at E11 . 5 ( G ) and E12 . 5 ( H ) . All samples were compared with Fgf9−/+;Fgf20lacZ/+ double heterozygous controls . p < 0 . 01 , *p < 0 . 001 . Sample numbers ( n ) are indicated in data bars . See also Figure 3—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 00510 . 7554/eLife . 05921 . 006Figure 3—figure supplement 1 . Proliferation of sensory progenitors . ( A ) Sox2 and EdU staining of E11 . 5 Fgf9−/+;Fgf20lacZ/+ , Fgf9−/+;Fgf20lacZ/lacZ , and Fgf9−/−;Fgf20lacZ/lacZ embryo sections . ( B ) Measurement of Sox2+ sensory progenitor proliferation at E11 . 5 . All samples were compared with Fgf9−/+;Fgf20lacZ/+ double heterozygous controls . p < 0 . 01 . ns , not significant . Sample numbers ( n ) are indicated in data bars . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 00610 . 7554/eLife . 05921 . 007Figure 3—figure supplement 2 . Fgf9 and Fgf20 loss do not cause premature cell cycle exit . ( A–C ) p27 ( green ) and Sox2 ( red ) co-immunostaining of E12 . 5 Fgf9−/+;Fgf20lacZ/+ ( A ) , Fgf9−/+;Fgf20lacZ/lacZ ( B ) , and Fgf9−/−;Fgf20lacZ/lacZ ( C ) embryo sections . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 007 To determine whether the decreased size of the Fgf9−/−;Fgf20lacZ/lacZ prosensory domain resulted from changes in cell proliferation and/or cell death , histological sections of E11 . 5 and E12 . 5 otic vesicles were immunostained for Sox2 and phospho-Histone H3 ( pHH3 ) ( Figure 3E , F ) , or activated Caspase-3 ( aCasp3 ) ( data not shown ) . Quantification of the number of pHH3+ , Sox2+ sensory progenitors showed similar numbers ( p > 0 . 09 at E11 . 5 and p > 0 . 2 at E12 . 5 ) in double heterozygous control and Fgf9−/+;Fgf20lacZ/lacZ cochleae ( Figure 3G , H ) . However , proliferation of Fgf9−/−;Fgf20lacZ/lacZ cochlear epithelial cells was significantly decreased at E11 . 5 ( p < 0 . 001 ) and E12 . 5 ( p < 0 . 01 ) compared to double heterozygous controls ( Figure 3G , H ) . In addition quantitation of cell proliferation using EdU labeling of E11 . 5 embryos showed similar results ( Figure 3—figure supplement 1 ) . No cell death ( aCasp3+ ) was detected in any of the genotypes at E11 . 5 and E12 . 5 ( data not shown ) . Decreased sensory progenitor number could also result from premature cell cycle exit . p27kip1 is one of the cell cycle inhibitors that is expressed in sensory progenitors as they become postmitotic . Expression of p27kip1 begins at E12 . 5 in the apex of the cochlea and progresses towards the base ( Lee et al . , 2006 ) . By E14 . 5 , the entire cochlear progenitor domain becomes p27kip1 positive . Expression of p27kip1 at E12 . 5 in the proximal cochlear duct was not detected in either control or Fgf9−/−;Fgf20lacZ/lacZ embryos suggesting that there is no premature cell cycle exit in mice lacking Fgf9 and Fgf20 ( Figure 3—figure supplement 2 ) . Next , we questioned which cell types are required for sensory progenitor proliferation and/or lateral compartment differentiation . Expression of both Fgfr1 and Fgfr2 have been reported in the otic epithelium and periotic mesenchyme between E10 . 5 and E12 . 5 ( Pirvola et al . , 2000 , 2002 , 2004; Ono et al . , 2014 ) . Epithelial Fgfr1 has been conditionally inactivated in otic epithelium using Foxg1Cre , Six1enh21Cre , and Emx2Cre ( Pirvola et al . , 2002; Ono et al . , 2014 ) . This results in a cochlear epithelium with reduced numbers of HCs , with OHC numbers being more severely affected than IHC numbers . In addition to the loss of differentiated HCs , a 40–50% decrease in cochlear length was reported when Fgfr1 was inactivated with Six1enh21Cre , or Emx2Cre ( Ono et al . , 2014 ) . To directly compare cochlear phenotypes resulting from inactivation of Fgfr1 in otic epithelium with embryos lacking Fgf9 and Fgf20 , we re-created and re-evaluated Fgfr1−/f::Foxg1Cre/+ mutant mice maintained on a 129X1/SvJ;C57BL/6J mixed genetic background . Quantification of the density of OHCs in Fgfr1−/f::Foxg1Cre/+ embryos demonstrated a significant ( p < 0 . 0001 ) decrease compared to controls ( Fgfr1+/f::Foxg1Cre/+ ) ( Figure 4—figure supplement 1A , B , E ) , while the density of IHCs was not changed ( p > 0 . 09 ) . Furthermore , the length of Fgfr1−/f::Foxg1Cre/+ cochleae was only 9% shorter than control ( Figure 4—figure supplement 1A , B , F ) , similar to what was observed in Fgf20lacZ/lacZ embryos ( Figures 1B , 2G and ref . Huh et al . , 2012 ) . Whole mount Sox2 staining of E14 . 5 Fgfr1−/f::Foxg1Cre/+ cochleae was also comparable to control , indicating that the number of Sox2+ progenitors was not changed ( Figure 4—figure supplement 1C ) . In addition , proliferation of the Fgfr1−/f::Foxg1Cre/+ prosensory epithelium was comparable ( p > 0 . 6 ) to that of controls at E12 . 5 ( Figure 4—figure supplement 1D , G ) . Because Fgfr2 often exhibits redundancy with Fgfr1 , it is important to consider potential Fgfr2 function in the inner ear prosensory epithelium . However , Fgfr2 is required for formation of the otic vesicle and Foxg1Cre , which is active before and during the otic vesicle stage ( Hébert and McConnell , 2000 ) , could not be used to investigate the role of Fgfr2 at later stages of otic vesicle development . In addition , due to overall activity of Foxg1Cre in the otic vesicle , cell type specificity of Fgfr1 was still unknown . To study whether Fgfr1 and/or Fgfr2 function cell autonomously or non-cell autonomously in the Fgf20+ domain of the prosensory epithelium , we generated an Fgf20Cre allele ( Figure 4—figure supplement 2A ) to allow conditional gene targeting of the Fgf20 lineage . To assess Cre activity , Fgf20Cre/+;ROSAmTmG/+ mice were generated . Cre activity was detected at E10 . 5 in a subset of the Sox2+ prosensory domain , in a pattern identical to that of Fgf20lacZ embryos ( Figure 4—figure supplement 2B ) . At P0 , all of the components of the organ of Corti were positive for the Fgf20Cre/+;ROSAmTmG/+ lineage tracer , indicating that Fgf20Cre is active in prosensory progenitors or their lineage ( Figure 4—figure supplement 2B ) . To identify potential roles for Fgfr1 and Fgfr2 in the prosensory epithelial lineage , we generated Fgfr1 and Fgfr2 single and double conditional mutant mice using the Fgf20Cre allele . E18 . 5 embryos were harvested and stained with phalloidin and p75 , to visualize cochlear morphology . The phenotype of Fgfr1−/f::Fgf20Cre/+ ( Fgfr1−/f;Fgfr2+/f::Fgf20Cre/+ ) cochleae was similar to that of Fgf20lacZ/lacZ , Fgf9−/−;Fgf20lacZ/lacZ , and Fgfr1−/f::Foxg1Cre/+ cochleae ( Figures 2A–C , 4A–C , Figure 4—figure supplement 1A , B ) . In contrast , the pattern and morphology of Fgfr2−/f::Fgf20Cre/+ ( Fgfr1+/f;Fgfr2−/f::Fgf20Cre/+ ) cochleae was similar to control ( Figure 4A–C ) and Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ cochleae was comparable to Fgfr1−/f::Fgf20Cre/+ ( Figure 4A–C ) . The density of OHCs in Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ was comparable to Fgfr1−/f::Fgf20Cre/+ ( p > 0 . 94 ) and significantly ( p < 0 . 001 ) decreased compared to control ( Figure 4F ) . The density of OHCs of Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ embryos was comparable to control ( p > 0 . 6 ) ( Figure 4F ) and the density of IHCs in Fgfr1−/f::Fgf20Cre/+ , Fgfr2−/f::Fgf20Cre/+ , and Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ cochleae were indistinguishable ( ANOVA , p > 0 . 8 ) from that of controls ( Figure 4F ) . 10 . 7554/eLife . 05921 . 008Figure 4 . Cell-autonomous regulation of sensory progenitor differentiation requires epithelial Fgfr1 but not Fgfr2 . ( A , B ) Phalloidin ( A ) and p75 immunostaining ( B ) of E18 . 5 whole cochlea showing HCs ( phalloidin ) and pillar cells ( p75 ) in the cochlear duct of control , Fgfr1−/f::Fgf20Cre/+ ( Fgfr1−/f;Fgfr2+/f::Fgf20Cre/+ ) , Fgfr2−/f::Fgf20Cre/+ ( Fgfr1+/f;Fgfr2−/f::Fgf20Cre/+ ) and , Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ embryos . ( C ) Phalloidin ( green ) and p75 immunostaining ( red ) showing the patterning of HCs and pillar cells in the cochlear duct . ( D ) Sox2 immunostaining of E14 . 5 whole cochlea to identify progenitor domains ( arrows ) . ( E ) Sox2 and pHH3 co-immunostaining of E12 . 5 embryo sections . ( F , G ) Measurement of HC number ( F ) and length of cochleae ( G ) of E18 . 5 control embryos . ( H ) Measurement of Sox2+ sensory progenitor proliferation in E12 . 5 embryos . All samples were compared with controls . p < 0 . 001; ns , not significant . Sample numbers ( n ) are indicated in data bars . See also Figure 4—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 00810 . 7554/eLife . 05921 . 009Figure 4—figure supplement 1 . Epithelial Fgfr1 is required for lateral compartment differentiation and HC and SC patterning . ( A , B ) Phalloidin ( A ) and p75 immunostaining ( B ) of E18 . 5 whole cochlea showing HCs ( phalloidin ) and pillar cells ( p75 ) in the cochlear duct of control and Fgfr1−/f::Foxg1Cre/+ embryos . ( C ) Sox2 immunostaining of E14 . 5 whole cochlea to identify progenitor domains ( arrows ) . ( D ) Sox2 and pHH3 co-immunostaining of E12 . 5 embryo sections . ( E , F ) Measurement of HC number ( E ) and length of cochleae ( F ) of E18 . 5 control and Fgfr1−/f::Foxg1Cre/+ embryos . ( G ) Measurement of Sox2+ sensory progenitor proliferation in E12 . 5 cochleae . p < 0 . 001 in E and p < 0 . 01 in F . Sample numbers ( n ) are indicated in data bars . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 00910 . 7554/eLife . 05921 . 010Figure 4—figure supplement 2 . Generation of an Fgf20Cre knockin mouse line . ( A ) Schematic diagram showing targeting of the Fgf20 genomic locus . Homologous recombination in mouse ES cells was used to insert a GFP:Cre ( Cre ) gene and neo selection cassette ( flanked by Flip recombination target sequences , grey triangles ) into exon 1 of Fgf20 . F1 mice were bred to mice that express Flip recombinase in the germline to excise the neo selection cassette . Arrows indicate PCR primers used for genotyping . B , BamH1; X , Xho1 . ( B ) Fgf20Cre/+; ROSA26mT/mG/+ ( ROSAmT/mG/+ ) double transgenic mice showing the cumulative lineage of Fgf20Cre expressing cells ( green ) in the Sox2+ prosensory domain ( blue ) at E10 . 5 and in HCs , SCs , and the greater epithelial ridge ( GER ) at P1 . Recombination at the ROSA26mT/mG locus silences membrane localized Tomato ( mT ) and activates membrane localized GFP ( mG ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 010 The length of the cochleae from E18 . 5 Fgfr1−/f::Fgf20Cre/+ and Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ embryos was decreased by 19% and 25% , respectively , compared to controls ( p < 0 . 0001 , Figure 4G ) . However , the length of the cochleae from Fgfr2−/f::Fgf20Cre/+ was comparable ( p > 0 . 5 ) to controls . Together , these data , and those presented above , showed that epithelial Fgfr1 , but not Fgfr2 , is required for lateral compartment differentiation , and has a modest effect on cochlear duct length of a similar magnitude to the 10% reduction in cochlear length seen in Fgf20lacZ/lacZ mice ( Huh et al . , 2012 ) . This reduction in cochlear length could be due to reduced numbers of progenitors or to other effects of FGFR1 signaling on cochlear duct elongation at later stages of development . Whole mount Sox2 staining at E14 . 5 of Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ cochleae showed a similarly sized sensory progenitor domain as compared to controls indicating that the Sox2+ progenitor population was not affected by inactivation of epithelial Fgfr1 and Fgfr2 ( Figure 4D ) . In addition , proliferation of Fgfr1−/f;Fgfr2−/f::Fgf20Cre/+ cochleae was comparable ( p > 0 . 5 ) to controls at E12 . 5 ( Figure 4E , H ) . We next asked whether mesenchymal FGFRs regulate cochlear length . Twist2 ( Dermo1 ) Cre is widely expressed in mesenchymal cells ( Li et al . , 1995; Šošić et al . , 2003 ) . To determine whether Twist2Cre is active in periotic mesenchyme during otic vesicle development , Twist2Cre/+;ROSA26lacZ/+ embryos were stained for lacZ activity at E9 . 5 and E10 . 5 . lacZ activity was observed in all of the mesenchyme surrounding the unstained otic epithelium at both developmental time points ( Figure 5—figure supplement 1 ) . Twist2Cre was then used to inactivate Fgfr1 and Fgfr2 from mesenchymal cells . Cochleae were dissected from E18 . 5 embryos and stained with phalloidin and p75 . All genotypes showed normal sensory HC and SC patterning , with one row of IHCs and three rows of OHCs ( Figure 5A–C ) . The linear density of IHCs and OHCs was comparable ( ANOVA , p > 0 . 2 , p > 0 . 8 , respectively ) among all genotypes ( Figure 5F ) . However , the length of the cochleae of Fgfr1−/f::Twist2Cre/+ ( Fgfr1−/f;Fgfr2+/f::Twist2Cre/+ ) , Fgfr2−/f::Twist2Cre/+ ( Fgfr1+/f;Fgfr2−/f::Twist2Cre/+ ) , and Fgfr1−/f;Fgfr2−/f::Twist2Cre/+ embryos were decreased by 7% , 20% , and 55% , respectively , compared to control ( Figure 5A , B , G ) . Thus , the total number of HCs is decreased in proportion to the decreased length of the cochlea . 10 . 7554/eLife . 05921 . 011Figure 5 . Mesenchymal Fgfr1 and Fgfr2 regulate the length of the cochlear duct and sensory progenitor proliferation . ( A , B ) Phalloidin ( A ) and p75 immunostaining ( B ) of E18 . 5 whole cochlea showing HCs ( phalloidin ) and pillar cells ( p75 ) in the cochlear duct of control , Fgfr1−/f::Twist2Cre/+ ( Fgfr1−/f;Fgfr2+/f::Twist2Cre/+ ) , Fgfr2−/f::Twist2Cre/+ ( Fgfr1+/f;Fgfr2−/f::Twist2Cre/+ ) , and Fgfr1−/f;Fgfr2−/f::Twist2Cre/+ embryos . ( C ) Phalloidin ( green ) staining showing normal HC patterning in the cochlear sensory epithelium . ( D ) Sox2 immunostaining of E14 . 5 whole cochlea to identify progenitor domains ( arrows ) . ( E ) Sox2 and pHH3 co-immunostaining of E12 . 5 cochlea sections . ( F , G ) Measurement of HC number ( F ) and length of cochleae ( G ) of E18 . 5 control , Fgfr1−/f::Twist2Cre/+ ( Fgfr1−/f;Fgfr2+/f::Twist2Cre/+ ) , Fgfr2−/f::Twist2Cre/+ ( Fgfr1+/f;Fgfr2−/f::Twist2Cre/+ ) , and Fgfr1−/f;Fgfr2−/f::Twist2Cre/+ embryos . ( H ) Measurement of Sox2+ sensory progenitor ( green ) proliferation ( red , pHH3 ) in E12 . 5 embryo sections . All samples were compared with controls . p < 0 . 001 . Sample numbers ( n ) are indicated in data bars . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 01110 . 7554/eLife . 05921 . 012Figure 5—figure supplement 1 . Twist2Cre targeting of periotic mesenchyme . Twist2Cre/+;ROSA26lacZ/+ ( ROSAlacZ/+ ) double transgenic mice showing βGal activity ( xGal , blue ) in periotic mesenchymal cells at E9 . 5 and E10 . 5 . OV , otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 012 To determine whether the effect of loss of mesenchymal FGFRs on cochlear length originates early in development , we examined the size of the Sox2+ progenitor domain at the time that HCs commit to differentiate , and cell proliferation within the Sox2+ domain before the onset of differentiation . The size of the Sox2+ progenitor domain , visualized by whole mount Sox2 staining of E14 . 5 cochleae was decreased in Fgfr1−/f;Fgfr2−/f::Twist2Cre/+ embryos compared to control embryos ( Figure 5D ) . In addition , proliferation of Sox2+ progenitors from Fgfr1−/f;Fgfr2−/f::Twist2Cre/+ cochleae was significantly ( p < 0 . 01 ) decreased compared to control cochleae at E12 . 5 ( Figure 5E , H ) . Together , these data show that mesenchymal FGFR signaling is a necessary determinant of cochlear length and sensory progenitor proliferation , but not for cochlear pattern formation or differentiation . To determine whether the FGF signaling pathway is affected in periotic mesenchyme , whole mount RNA in situ hybridization was used to localize expression of Etv4 and Etv5 , two transcription factors that are commonly regulated by FGF signaling ( Raible and Brand , 2001; Firnberg and Neubüser , 2002; Brent and Tabin , 2004; Mao et al . , 2009; Zhang et al . , 2009 ) . Compared to double heterozygous control and Fgf9−/+;Fgf20lacZ/lacZ inner ears , Fgf9−/−;Fgf20lacZ/lacZ inner ears showed decreased expression of Etv4 and Etv5 in mesenchyme surrounding the cochlear duct ( Figure 6—figure supplement 1A , B ) . The only known mesenchymal signaling pathway to regulate sensory progenitor proliferation is a Tbx1/Pou3f4 dependent retinoic acid ( RA ) signaling cascade ( Braunstein et al . , 2008 , 2009 ) . However , Tbx1 and Pou3f4 expression , using RNA in situ hybridization in the embryos lacking Fgf9 and Fgf20 ( Fgf9−/−;Fgf20lacZ/lacZ ) , did not reveal a change in expression of these transcription factors compared to doble heterozygous control and Fgf9−/+;Fgf20lacZ/lacZ embryos ( Figure 6—figure supplement 1C , D ) , suggesting that FGF signaling may function independent of RA signaling . Next we asked whether increased mesenchymal FGFR signaling is sufficient to activate sensory progenitor proliferation . We ectopically expressed a constitutive FGFR1 tyrosine kinase domain in mesenchymal cells by combining the Twist2Cre , ROSArtTA , and the doxycycline-responsive TRE-caFgfr1-myc alleles ( TRE-caFgfr1;ROSArtTA/+::Twist2Cre/+ ) ( Cilvik et al . , 2013 ) . Doxycycline was fed to pregnant female mice beginning at E10 . 5 . Embryos were analyzed at E12 . 5 for prosensory progenitor proliferation using pHH3 and Sox2 co-immunostaining ( Figure 6A ) . The proliferation index in control prosensory cells was 2 . 4 ± 0 . 5/10 , 000 μm2 ( Figure 6B ) . However , embryos in which the caFgfr1-myc allele was induced in mesenchyme showed a significantly increased ( 4 . 1 ± 0 . 6/10 , 000 μm2 , p < 0 . 02 ) proliferation index ( Figure 6B ) . To test the hypothesis that increased proliferation in sensory progenitors could lead to an increase in cochlear length , TRE-caFgfr1;ROSArtTA/+::Twist2Cre/+ embryos were induced from E10 . 5 to E14 . 5 and cochleae were analyzed at E18 . 5 ( Figure 6C ) . Linear densities of IHCs , OHCs , and SCs in TRE-caFgfr1;ROSArtTA/+::Twist2Cre/+ embryos were comparable ( p > 0 . 4 ) to control ( Figure 6E , F ) . However , the cochlear length in TRE-caFgfr1;ROSArtTA/+::Twist2Cre/+ embryos was significantly ( p < 0 . 001 ) increased by 14% compared to control ( Figure 6D ) . 10 . 7554/eLife . 05921 . 013Figure 6 . Ectopic activation of FGFR signaling in mesenchyme increases sensory progenitor proliferation and cochlear length . ( A ) Sox2 ( green ) and pHH3 ( red ) co-immunostaining of E12 . 5 control and TRE-caFgfr1;ROSArtTA/+::Twist2Cre/+ embryo sections . ( B ) Measurement of Sox2+ sensory progenitor proliferation at E12 . 5 . ( C ) Phalloidin ( green ) and p75 immunostaining ( red ) showing the patterning of HCs and pillar cells in the cochlear duct of control and TRE-caFgfr1;ROSArtTA/+::Twist2Cre/+ embryos . Measurement of the length of the cochleae ( D ) , HC number ( E ) , and SC number ( F ) of E18 . 5 control and TRE-caFgfr1;ROSArtTA/+::Twist2Cre/+ embryos . ( G ) Schematic diagram indicating the requirement for epithelial FGF9/20 signaling to mesenchymal FGFR1/2 to induce sensory progenitor proliferation . p < 0 . 01 in B and p < 0 . 001 in D . Sample numbers ( n ) are indicated in data bars . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 01310 . 7554/eLife . 05921 . 014Figure 6—figure supplement 1 . Fgf9 and Fgf20 regulate the expression of Etv4 and Etv5 , but not Pou3f4 or Tbx1 . ( A–D ) Wholemout mRNA in situ staining of Etv4 ( A ) , Etv5 ( B ) , Pou3f4 ( C ) , and Tbx1 ( D ) genes on otic vesicles from E11 . 5 embryos . The genotypes of embryos analyzed are: Fgf9−/+;Fgf20lacZ/+ , Fgf9−/+;Fgf20lacZ/lacZ and , Fgf9−/−;Fgf20lacZ/lacZ . Images are representative of at least three embryos for each probe and genotype . Scale bar , 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05921 . 014 Sensory progenitor proliferation and differentiation are temporally distinct events in cochlear development . In mice , sensory progenitors exit from the cell cycle beginning at the apical end of the cochlea at ∼E12 . 5 and ending at the base at ∼E14 . 5 . In contrast , differentiation begins in the mid-base at ∼E14 . 5 and then extends to the base and apex ( Wu and Kelley , 2012 ) . Under physiological conditions , once progenitors exit the cell cycle , they do not reenter the cell cycle throughout the life of the organism . Previous studies suggested that during development both epithelial and mesenchymal signals are required to regulate cochlear progenitor proliferation and differentiation ( Montcouquiol and Kelley , 2003; Doetzlhofer et al . , 2004 ) . However , the mechanisms that control cochlear sensory progenitor proliferation are not known . In this study , we found that epithelial FGF9 and FGF20 signaling to mesenchymal FGFR1 and FGFR2 is required for normal levels of cochlear sensory progenitor proliferation and that inactivation of either the ligands or the mesenchymal receptors results in a shortened cochlea . We also demonstrated that activation of mesenchymal FGFR signaling is sufficient to increase sensory progenitor proliferation and extend cochlear length . Fgf9 is expressed in non-sensory epithelia of the cochlea and loss of Fgf9 results in defects in periotic mesenchymal cell proliferation , causing a hypoplastic otic capsule ( Pirvola et al . , 2004 ) . Based on known expression patterns in mesenchyme , Fgfr1 and Fgfr2 were considered the most likely targets of FGF9 signaling ( Pirvola et al . , 2004 ) . The critical time window for FGF9 signaling was determined to occur before E14 . 5 . By contrast , Fgf20 is expressed in the sensory epithelium and loss of Fgf20 results in failure of the lateral compartment of the organ of Corti to fully differentiate ( Huh et al . , 2012 ) . Based on expression patterns and phenotypic similarities with epithelial Fgfr1 conditional gene inactivation , FGFR1 was identified as the epithelial target receptor ( Pirvola et al . , 2002; Huh et al . , 2012 ) . The effects of epithelial FGFR1 signaling on the length of the cochlear duct exhibit variability among studies . Ono et al . ( 2014 ) report a 40–50% decrease in cochlear length in Fgfr1f/f::Six1enh21Cre , and Fgfr1f/f::Emx2Cre conditional knockout mice , by contrast , Fgf20lacZ/lacZ , Fgf9−/+;Fgf20lacZ/lacZ , and Fgfr1−/f::Foxg1Cre/+ mice that we studied ( Huh et al . , 2012 , and this study ) showed only a 10–25% decrease in cochlear length . It is clear that in both studies defects in epithelial differentiation is likely to result in some decrease in cochlear length . It is also possible that differences in genetic background could contribute to differences in these two studies . FGF9 and FGF20 are members of the same FGF subfamily and share similar biochemical properties ( Zhang et al . , 2006; Ornitz and Itoh , 2015 ) . Redundancy between these FGFs has also been demonstrated in kidney development , where both ligands are required for nephron progenitor maintenance ( Barak et al . , 2012 ) . Interestingly , in both cases , the expression patterns of these two FGFs do not overlap , but nevertheless they appear to signal to a common target tissue , periotic mesenchyme in the developing inner ear and CAP mesenchyme in the developing kidney . For the evolution of the kidney and inner ear , it is possible that additive expression of these FGFs from distinct sources was required to take advantage of their unique receptor specificities or unique interactions with the extracellular matrix . Tbx1 is a transcription factor that is expressed in both sensory epithelium and mesenchyme ( Vitelli et al . , 2003; Raft et al . , 2004 ) . Deletion of Tbx1 in mesenchymal cells resulted in defects in cochlear epithelial proliferation indicating a non-cell autonomous requirement for Tbx1 for cochlear epithelial development ( Xu et al . , 2007 ) . In addition , the Pou domain containing transcription factor , Pou3f4 , also known as Brn4 , is expressed in mesenchymal cells in the developing inner ear ( Phippard et al . , 1999 ) . Deletion of Pou3f4 resulted in reduction of cochlear length and defects in derivatives of the otic mesenchyme including the spiral limbus , scala tympani , and strial fibrocytes ( Phippard et al . , 1999 ) . Furthermore , decreasing gene dosages of Tbx1 and Pou3f4 resulted in a significant decrease in sensory epithelial proliferation and cochlear length indicating that Tbx1 and Pou3f4 genetically interact . The RA catabolizing genes Cyp26a1 and Cyp26c1 , both targets of Tbx1 and Pou3f4 , were decreased in these mice , suggesting that increased RA signaling could directly or indirectly suppress sensory progenitor proliferation ( Braunstein et al . , 2008 , 2009 ) . Analysis of Fgf9 and Fgf20 double mutant mice showed no change in the expression of Tbx1 and Pou3f4 in mesenchyme surrounding the otic vesicle , suggesting that mesenchymal FGF signaling does not directly affect transcription factors that regulate RA signaling ( Figure 6—figure supplement 1C , D ) . On the other hand , Etv4 and Etv5 function as downstream targets of FGF signaling in other systems including the limb ( Mao et al . , 2009; Zhang et al . , 2009 ) , and Etv4 and Etv5 expression were decreased in Fgf9/Fgf20 mutant ears . Future studies will be needed to determine whether FGF signaling including ETV4 and ETV5 regulates RA signaling downstream of Tbx1/Pou3f4 or act in parallel to the Tbx1/Pou3f4/RA signaling pathway to regulate sensory progenitor proliferation . Whether the cellular target of RA signaling is in the periotic mesenchyme or the sensory progenitor epithelium also remains to be determined . It is also possible that the number of nearby mesenchymal cells may influence sensory progenitor proliferation . However , considering that loss of Fgf9 resulted in decreased mesenchymal cell proliferation ( Pirvola et al . , 2004 ) but did not affect HC formation or cochlear length ( Figure 2 ) , alternative mechanisms may need to be considered . The reactivation of developmental signaling pathways may be important for regeneration . Recent publications showed that inhibition of Notch signaling could induce transdifferentiation of SCs to HCs in a damaged cochlea ( Korrapati et al . , 2013; Mizutari et al . , 2013 ) . In addition , Wnt/β-catenin signaling can induce SC proliferation in neonatal mice ( Chai et al . , 2012; Shi et al . , 2012 ) . One intermediate goal of regenerative biology for the inner ear would be to generate large numbers of sensory progenitor cells that could be differentiated into functional HCs and SCs and then be reintroduced into the damaged inner ear . The studies presented here suggest that in efforts to grow inner ear sensory progenitor cells in vitro , that FGF-induced mesenchyme may be necessary . The identification of mesenchymal factors that are regulated by FGF or RA could also be used to support the growth of sensory progenitor cells . This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . The protocol was approved by the Washington University Division of Comparative Medicine Animal Studies Committee ( Protocol Number 20130201 ) . All efforts were made to minimize animal suffering . Fgf20Cre knock-in mice were generated using a similar method to that reported previously ( Huh et al . , 2012 ) . Briefly , exon1 of Fgf20 was replaced with a Cre-EGFP–FRT-neomycin-FRT cassette to generate Fgf20Cre ( neo ) /+ mice . The neomycin gene was eliminated by mating with CAG-FLPe ( Kanki et al . , 2006 ) mice to generate Fgf20Cre/+ mice . Genotyping was performed using PCR1: CTGCATTC GCCTCGCCACCCTTGCTACACT; PCR2: GGATCTGCAGGTGGAAGCCGGTGCGGCAGT; PCR3: TTCAGGGTCAGCTTGCCGTAGGTGGCATCG primers , which amplify wild type ( 335 bp ) and mutant ( 241 bp ) PCR fragments . Mice were maintained on a 129X1/SvJ;C57BL/6J mixed background . Fgf9lacZ mice were derived from International Knockout Mouse Consortium targeted ES cells ( project number 24486 ) ( Skarnes et al . , 2011 ) . Chimeric mice derived from injected blastocysts were bred to Sox2Cre mice ( Hayashi et al . , 2003 ) to remove the nbactP-neo selection cassette and the second exon of the Fgf9 gene . Genotyping was performed using Wt1: GAAGTCGTGCGTGAGGTGCTCCAGGTCGG; Wt2: CCGCGAATGCTGACCAGGCCCACTGCTAT primers for wild type ( 172 bp ) and mut1: GTT GCA GTGCACGGCAGATACACTTGCTGA; mut2: GCCACTGGTGTGGGCCATAATTCAATTCGC primers for mutant ( 389 bp ) PCR fragments . Mice were maintained on a 129X1/SvJ;C57BL/6J mixed background . Fgfr1f/f , Fgfr2f/f , Twist2 ( Dermo1 ) Cre/+ , Foxg1Cre/+ , R26R , ROSAmTmG/+ , TRE-caFgfr1-myc , ROSArtTA/+ , Fgf20lacZ/+ , and Fgf9−/+ mice lines were reported previously ( Soriano , 1999; Hébert and McConnell , 2000; Colvin et al . , 2001; Pirvola et al . , 2002; Šošić et al . , 2003; Yu et al . , 2003; Belteki et al . , 2005; Muzumdar et al . , 2007; Huh et al . , 2012; Cilvik et al . , 2013 ) . Fgfr1−/+ and Fgfr2−/+ mice were generated by crossing Fgfr1f/f and Fgfr2f/f to Sox2Cre/+ mice , respectively . Embryos were fixed overnight in Mirsky's Fixative ( National Diagnostics , Atlanta , GA ) , washed three times in PBT ( PBS , 0 . 1% Tween-20 ) and incubated in βGal staining solution ( 2 mM MgCl2 , 35 mM potassium ferrocyanide , 35 mM potassium ferricyanide , 1 mg/mg X-Gal in PBT ) at 37°C until color reaction was apparent . Samples were washed in PBS , fixed in 10% formalin and imaged under a dissecting microscope . For frozen sections , embryos were fixed with 4% paraformaldehyde overnight and washed with PBS . Samples were soaked in 30% sucrose and embedded in OCT compound ( Tissue-Tek , Torrance , CA ) . Samples were sectioned ( 12 µm ) and stored at −80°C for immunohistochemistry . Either phalloidin or Prox1 immunostaining were used to identify HCs and SCs , respectively . To measure the density of HCs and SCs , at least 300 µm regions of the base ( 10% ) , middle ( 40% ) , and apex ( 70% ) of the cochleae were counted and normalized to 100 µm along the length of the cochlear duct . Inner and OHCs were identified by location and morphology of phalloidin staining . Cell counting was performed using Image J software . To analyze progenitor proliferation and cell death , frozen sections were prepared from the entire ventral inner ear of E11 . 5 or E12 . 5 embryos . Alternate sections were subjected to staining for pHH3 and Sox2 ( for proliferation ) or activated-Caspase 3 and Sox2 ( for cell death , data not shown ) . For EdU labeling , pregnant females were injected with 50 µg/g ( body weight ) of EdU according to the manufacture's recommendation . Embryos were collected 2 hr after EdU injection . EdU was detected with the Click-iT EdU Alexa Fluor 488 Imaging Kit ( Invitrogen , Carlsbad , CA ) according to manufacture's instructions . The total area of Sox+ cells was measured using Image J software and pHH3+ or activated-Caspase 3+ cells within the Sox2+ domain were counted . Counting was normalized to 10 , 000 μm2 of Sox2+ prosensory epithelium . For whole mount immunofluorescence , cochleae were isolated and fixed in 4% PFA overnight at 4°C . Samples were washed with PBS and blocked with PBS containing 0 . 1% triton X-100 and 0 . 5% donkey serum . Primary antibody was incubated overnight at 4°C . Samples were washed with PBS and incubated with a secondary antibody for 1 hr at room temperature . Samples were washed , placed on a glass microscope slide , coverslipped , and photographed using a Zeiss LSM 700 confocal microscope . For immunofluorescence on histological sections , frozen sections ( 12 µm ) were washed with PBS , blocked with 0 . 1% triton X-100 and 0 . 5% donkey serum , and incubated with primary antibodies in a humidified chamber overnight at 4°C . Sections were then washed and incubated with secondary antibody for 1 hr at room temperature . Samples were washed , coverslipped with Vectashield Mounting Media ( Vector Labs , Burlingame , CA ) , and photographed using a Zeiss LSM 700 confocal microscope . Primary antibodies used: Phallodin ( R&D Systems , Minneapolis , MN , 1:40 ) , Prox1 ( Covance , Princeton , NJ , 1:250 ) , p27 ( Neomarkers , Fremont , CA , 1:100 ) , p75 ( Chemicon , Billerica , MA , 1:500 ) , β-galactosidase ( Abcam , United Kingdom , 1:500 ) , Sox2 ( Millipore , Billerica , MA , 1:250 , Santa Cruz , Dallas , TX , 1:250 ) , Jag1 ( Santa Cruz , Dallas , TX , 1:200 ) , phospho-histone 3 ( Sigma–Aldrich , St . Louis , MO , 1:500 ) , and activated Caspase 3 ( BD Sciences , San Jose , CA , 1:200 ) . Numbers of samples are indicated for each experiment . All data are presented as mean ± standard deviation ( sd ) . The p value for difference between two samples was calculated using a two-tailed Student's t-test or one-way ANOVA where appropriate . p < 0 . 05 was considered as significant .	Mammalian ears contain several structures that are involved in hearing . Within the inner ear is a spiral-shaped structure called the cochlea . This contains an array of cells called sensory hair cells that convert sound vibrations into electrical signals , which are then conveyed to the brain . Sounds of differing pitch are detected at different points along the cochlea , so its overall length helps to determine the range of sounds that an individual can hear . In the embryo , sensory hair cells and their associated supporting cells develop from ‘cochlear progenitor’ cells . The final length of the cochlea is determined by the numbers of progenitor cells that commit to becoming either sensory hair cells or supporting cells . Two proteins called FGF9 and FGF20 are involved in the formation of the cochlea . FGF20 promotes the formation of the hair cells and supporting cells , but the precise roles of both proteins are not clear . Here , Huh et al . studied FGF9 and FGF20 in the inner ear of mice at an early stage of development . The experiments show that these proteins work together to control the number of progenitor cells and the length of the cochlea . FGF20 is produced by the same tissue structure ( called an ‘epithelium’ ) that gives rise to the hair cells and supporting cells . In contrast , FGF9 is produced in another epithelium tissue that produces the cells that line the fluid-filled tubes of the inner ear . The experiments also show that both FGF9 and FGF20 act as signals to cells in an adjacent tissue called the mesenchyme , where they activate other proteins known as FGF receptors . These receptors , in turn , regulate an unknown molecule in the mesenchyme that influences the growth of progenitor cells and the length of the cochlea . Sensory hair cells can be injured or lost by excessive sound exposure , some medications and as part of normal aging . These cells are not replaced , and so their loss is a major cause of permanent hearing loss . Understanding the signals that produce the progenitor cells will take us one step closer to being able to grow these cells in the laboratory for use in therapies to replace or repair damaged sensory hair cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2015	Cochlear progenitor number is controlled through mesenchymal FGF receptor signaling
The role of the hippocampus in spatial cognition is incontrovertible yet controversial . Place cells , initially thought to be location-specifiers , turn out to respond promiscuously to a wide range of stimuli . Here we test the idea , which we have recently demonstrated in a computational model , that the hippocampal place cells may ultimately be interested in a space's topological qualities ( its connectivity ) more than its geometry ( distances and angles ) ; such higher-order functioning would be more consistent with other known hippocampal functions . We recorded place cell activity in rats exploring morphing linear tracks that allowed us to dissociate the geometry of the track from its topology . The resulting place fields preserved the relative sequence of places visited along the track but did not vary with the metrical features of the track or the direction of the rat's movement . These results suggest a reinterpretation of previous studies and new directions for future experiments . When O'Keefe and Dostrovsky discovered that the firing of hippocampal place cells correlates strongly with the position of the rat with respect to discrete locations within the environment ( the cell's place field ) ( O'Keefe and Dostrovsky , 1971 ) , they launched decades of intensive research into how place cells contribute to an internal map of the environment ( Best et al . , 2001 ) . That there is a considerable gulf between place cell firing and a cognitive spatial map , however , has become only clearer with time . Place cells appear to respond to a perplexing array of stimuli , from visual cues to head direction , goal planning , color changes , shape changes , and olfactory , vestibular and kinesthetic inputs , and discrepancies between the expected and actual location of a target ( Harnad , 1994; Frank et al . , 2000; Wood et al . , 2000 ) ; recent work has shown that both spatial and nonspatial cell types in the entorhinal cortex provide myriad inputs to the hippocampus ( Ideker et al . , 2011 ) . The fascination with place cell promiscuity , however , has tended to distract from the fact that cognition is the work not of individual neurons but of large ensembles of cells ( Ludvig , 1999; Fenton et al . , 2008 ) . The nature of the information embedded in the place cell ensemble code and transmitted to downstream neurons has been largely ignored , along with the consequences for the type of spatial properties that might form the basis for the cognitive map . Whatever the identity of these downstream neurons , they clearly have no direct access to the physical environment or to the place fields mapped by experimentalists . The only information they receive is contained in the temporal pattern of the spike trains of the thousands of place cells that are active in a given environment . What sort of information about a space might be constructed from such signals ? The dominant assumption within the field of neuroscience has been that the hippocampal spatial map—if there is one ( Eichenbaum et al . , 1999 ) —is an allothetic , Euclidian representation of the local environment that integrates detailed information about distances and angles arising through self-motion , head direction , and speed in addition to visual and other cues ( Wills et al . , 2010; Erdem and Hasselmo , 2012; Chen et al . , 2013; Rubin et al . , 2014 ) . The compilation of such geometric data would seem to require considerable computational power for the spatial map to be formed quickly enough to be useful under normal conditions of animal navigation ( Wilson and McNaughton , 1993; Frank et al . , 2004 ) . Geometry is not the only aspect of space that can form the basis of a map , however . From a practical standpoint , the topological qualities of a space—relations between locations such as continuity , enclosure , sequence ( Poincaré , 1895; Aleksandrov , 1965 ) —are arguably even more important ( Poucet , 1993 ) . Yet scant attention has been paid to the possible role of topological features in the cognitive map , despite the fact that any movement through space is describable in topological terms ( locations A and B must be somehow connected if movement between them is possible ) . This neglect is even more curious in light of Piaget's early work indicating that children first represent space topologically and only later learn to incorporate metrical details into their mental representations of space ( Poucet , 1993; O'Keefe and Nadel , 1978 ) . A few studies have shown that place cell firing is responsive to topological changes in the environment such as the placement or removal of a barrier across a previously learned route ( Poucet and Herrmann , 2001; Alvernhe et al . , 2011 ) , but as noted above , place cells seem to be responsive to an enormous variety of inputs . Our interest ultimately is not whether place cells can respond to topological features of a space , but whether the information conveyed downstream by place cell spiking might encode topological information , which would offer considerable advantages over geometry in terms of computational speed and flexibility . There are two compelling reasons to consider such a radical departure from the dominant assumptions about the spatial component of the cognitive map . First is the nature of place cell firing . If , in fact , place cell co-firing implies spatial overlap of place fields ( McNaughton et al . , 1983; Brown et al . , 1998; Zhang et al . , 1998; Jensen and Lisman , 2000; Barbieri et al . , 2001 , 2005; Guger et al . , 2011 ) , then a map derived from place cell spiking will necessarily emphasize contiguities between locations as well as the temporal sequence in which they are experienced . ( The reason we include temporal sequence along with more conventional spatial relationships is because movement through space takes place over time; sequence thus embodies connectivity . ) This would seem to be more compatible with the broader role of the hippocampus in representing and storing experience . Second , how geometric changes would be transmitted downstream as metrical changes per se is unclear; precisely because place cells respond to so many inputs , a change in firing rate within a period of overlap might just as well reflect a change in the animal's speed or some other input as it would a change in distance . Muller and Kubie ( 1987 ) found that enlarging their open field set-up caused place fields to enlarge while maintaining the same shape and relative order ( i . e . , the connectivity of the environment remained invariant ) . Similar observations were made by Gothard et al . , 1996a on a linear track , by O'Keefe and Burgess ( 1996 ) in rectangular environments , and by other studies in morphing environments ( Lever et al . , 2002; Leutgeb et al . , 2005; Touretzky et al . , 2005; Wills et al . , 2005; Colgin et al . , 2010 ) . O'Keefe et al . ( O'Keefe and Burgess , 1996 ) reasoned that the responses of the place fields could be described in terms of the geometric characteristics of the deforming environment , such as the altered distances to the walls of the enclosure . If we consider the information available to the downstream neurons that ‘read’ spiking coactivity , however , this explanation becomes a bit less plausible . As adjacent place fields enlarge in response to an expanded environment , the place cells will still be coactive , and the readout to the downstream neurons will be much the same . Thus , even if the place field layout stretches significantly , the corresponding geometrical information will not necessarily be captured by the downstream networks unless the change is quite large . Note that this phenomenon of place field stretching has a topological quality to it , since the fields stretch with the space but maintain their relative positions and order . In this context it is interesting to note that a more detailed analysis of place cell activity in a stretching 1D environment ( Diba and Buzsaki , 2008 ) showed that the temporal gaps between the spikes of the respective place cells remain fixed throughout the stretch , despite the change in the overall firing rate ( rate remapping , Dupret et al . , 2010 ) . Mathematically speaking , this suggests that spatial information contained in the hippocampal map may be invariant with respect to a certain range of geometric transformations , so that the flexible arrangement of place fields does not represent specific locations on a Cartesian grid , but rather a coarser framework of spatiotemporal relationships . Another advantage to the topological model is that it implies that questions of spatial learning might be amenable to recently developed approaches in the field of computational topology . The way place fields cover a space actually calls to mind a fundamental theorem in algebraic topology that holds that a given space , if covered by a sufficient number of discrete sub-regions , can be reconstructed by the pattern of overlaps between those sub-regions ( a simplicial complex ) ( Hatcher , 2002 ) . This theorem is key to the power of topological approaches to resolve otherwise overwhelmingly complex problems in high-dimensional data analysis ( Lum et al . , 2013 ) , and it lends itself to the problem at hand . We recently developed a computational algorithm based on algebraic topology to simulate the behavior of a rat exploring several different environments ( Dabaghian et al . , 2012; Arai et al . , 2014 ) . We created a wide range of testable values for three parameters—place cell firing rate , number of place cells , and place field size—and showed that simulated groups of neurons accurately encode spatial information from topologically distinct environments ( they ‘learn’ the space ) when the parameter values happen to closely parallel biological values derived from animal experiments . Since we made no a priori assumptions about what values would work , this was a promising validation of our model . We subsequently included theta precession in the model and showed that theta oscillation affects the behavior of the neuronal ensemble in ways that augment its performance to enhance learning ( Arai et al . , 2014 ) . Still , there has been no experiment designed explicitly to examine how place cells respond to significant geometric changes when the topology is held invariant . Here we report such an investigation . Since the location-specific firing of place cells is thought to be produced by a path integration mechanism ( accurate idiothetic counting of the elementary displacements and changes of orientation , speed and acceleration produced during movements [McNaughton et al . , 1996; Etienne and Jeffery , 2004; McNaughton et al . , 2006] ) , we used morphing linear tracks whose geometry limits the pool of possible confounding motions . If place cells are primarily concerned with encoding geometric information , then their firing activity should reflect the dynamic planar geometry of the track . If , on the other hand , place cells are primarily concerned with topological information , then variations of the track's configuration should have little effect on where the place cells fire . We devised two elevated U-shaped tracks with a flexible construction that allowed the geometry of one arm to be significantly altered while the rat was on the other part of the track . On the first track , Track A , each arm of the U could bend at four joints to transition from a straight runway in the expanded configuration to a compressed zigzag ( Figure 1; ‘Materials and methods’ ) . The dynamic configurations of the track were mapped against a stable set of spatial bins , which allowed us to dissociate two frames of reference: the effects produced on the CA1 place cells by changes in the track configuration can be described either in the reference frame associated with the room ( the 2D planar reference frame ) or with the track itself ( the 1D linear reference frame ) . If path-integrating the 2D displacements , velocities and acceleration produces place fields that from an allocentric , Cartesian coordinate map of the space , the place fields should be relatively stable in the 2D plane , remaining in the same spatial bins despite changing track segments , and variable in the 1D frame of reference ( See Figure 1 , Figure 1—figure Supplement 1 ) . In contrast , if the place cells attend to the sequence and connectivity of places experienced on the track irrespective of the track's geometry and the directions and angles of the rat's movements , then the place fields will be stable in the linear reference frame , but variable in the planar ( 2D ) reference frame . 10 . 7554/eLife . 03476 . 003Figure 1 . Compressible U/zigzag track ( Track A ) . ( A ) Top-down view of the four-meter long track with both arms in a semi-contracted configuration . ( B ) A schematic representation of the track illustrating how the segments are numbered in the planar ( B ) and the linear ( C ) frames of reference . ( Note that the track depiction in C is drawn at half the scale as in B . The track was designed so that the angle between the segments 1 and 2 and between segments 3 and 4 on the top arm remain the same whether the track arm is extended or compressed , that is , ∠ ( 1 , 2 ) = ∠ ( 3 , 4 ) = α . Similarly , the angles on the bottom arm remain equal , ∠ ( 7 , 8 ) = ∠ ( 9 , 10 ) = β . The magnitude of angle α is defined by the stretch of the top arm ( by the position of the food well F1 ) , and angle β is defined by the position of the food well F2 . Thus , the geometric configuration of the track can be described either by the angles ( α , β ) or the positions of the food wells F1 and F2 . ( D ) To more readily denote track arm configurations , we divided the space that can be occupied by the fully extended top and bottom arms into 12 vertical sections each . Each position of a track arm can be described by the numbered vertical section into which F1 or F2 ( for the top and bottom arms , respectively ) falls . In the depicted track configuration , F1 falls into section 2 and F2 falls into section 9 . However , because we are interested in being able to denote changes in track segment positions in order to compute correlation coefficients , the top arm can be said to move from position i to position j ( numbers in red ) , and the bottom arm from position k to position l ( numbers in blue; see ‘Materials and methods’ ) . So the track configuration depicted here can be fully described by F1 = 2 and F2 = 9 . Given movement of an arm from section i to j ( or k to l ) , the larger the difference between the numbers , d=\|i − j \| or \|k − l\| , the more geometrically dissimilar positions i and j or k and l are . DOI: http://dx . doi . org/10 . 7554/eLife . 03476 . 00310 . 7554/eLife . 03476 . 004Figure 1—figure supplement . 1 . Calculating correlation coefficients for place cells firing in different track configurations . The grid covering the 2D space occupied by the track is binned into squares that are 3 cm × 3 cm . The red line illustrates how we sequenced the bins in the planar and linear reference frames , with m = 1 at the top right square , m = 2 the square below it , etc . The activity vector ui or uj is the firing rate of a place cell given a particular track configuration ( i on the left , j on the right , respectively ) in the 2D frame of reference and is constituted by the firing rates of the cell in particular bins ( ui1 , ui2 , etc ) . In the example in panel A , the same spatial bin produces vector components ui , m ( in configuration i ) and uj , m ( in track configuration j ) . We calculate the correlation coefficient , Ci , j , to understand the relationship between two activity vectors for a given place cell firing in two different track configurations ( for the sake of simplicity , we depict only a change of position in the top arm , from i to j; we can do this safely because the place cells we recorded did not have multiple place fields ) . Note that any given configuration can be arrived at from a number of movements: the top arm could move to position 2 from vertical sections 1 or 3 , 7 , or 10 ( see Figure 1D ) , so the correlation coefficient for a cell that has a place field on the top arm , Ci , j would be denoted by C1 , 2 , C3 , 2 , C7 , 2 , or C10 , 2 , respectively . As explained in Figure 1D , the greater the numeric difference between i and j , the greater the geometric difference between track configurations 1 and 2 . The correlation matrix in the middle graphically depicts correlation coefficients by color . In panel B we show activity vectors vi , vj for linearized trajectories of the animal along the same configurations i and j depicted in 2D in panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 03476 . 004 Three animals were exposed to the initially novel Track A . We recorded a total of 557 pyramidal cells in CA1 and analyzed the pattern of their activity in response to the expansions and contractions of the track in both reference frames ( ‘Materials and methods’ ) and compared the spatial firing rates across track configurations . We found that , for the CA1 place cells active on the mobile sections of the track , spiking was stable in the linear but not in the 2D reference frame . Figure 2A , B shows the spiking pattern of one representative neuron in two different track configurations , each of which is plotted in both the linear and planar reference frames . As evident in the overlay of the two plots ( Figure 2A+B ) , the place cell's spiking was widely dispersed in the 2D reference frame but localized in the linear frame . We quantified this tendency by computing the correlation between the firing rates in the planar and linear reference frames for all pairs of configurations . Because any two linearized track configurations are geometrically identical , whereas only a few parts of the 2D track configurations overlap , we computed the correlations using only the spatial bins that were visited and where at least one spike was fired in both 1D and 2D configurations ( ‘Materials and methods’ ) . 10 . 7554/eLife . 03476 . 005Figure 2 . Place cell spiking across different configurations of Track A shows that place fields remain stable despite large geometric changes . Depicted are the firing patterns of two place cells: one cell that was active primarily on one dynamic segment of the track ( A–D ) and another place cell that was active on a static part of the track ( E–H ) . ( A ) Each green dot represents a spike , which is shown in two reference frames: 2D ( upper row ) and linear ( bottom row ) . The gray line represents the animal's path across all trials with these configurations . ( B ) Overlaying the 2D and linear reference frames shows that the spiking is distributed in 2D space but much more localized in the linearized reference frame . ( C ) The matrix of the correlations between the activity of this place cell over the moving sections of the track defined in planar ( left ) and in the linear ( right ) frame . The color of each square ( Cij ) represents the mean correlation coefficient between two track configurations , configuration1 and configuration2 . The correlations were clearly much higher in the linear frame . ( D ) The mean correlation coefficient values , rd = < Ci , i+d > , averaged over all pairs of the track configurations separated by d bins . The lines show whether the place cell firing was stable across configurations in the linear ( 1D , red ) and the planar ( 2D , black ) reference frames as a function of the difference d between the two configurations; the black line shows clear decay of correlation between the configurations , meaning that place cell firing was not stable in 2D ( See Figure 1 for explanation of D ) . ( E–H ) The same plots for a different neuron ( spikes shown in blue ) that was active on the static track segments . The graph to the right shows a high correlation coefficient for both frames of reference , indicating that the place fields were stable . The asterisk above the SEM error bars indicates p < 0 . 001 . A total of 557 neurons were recorded; the place cells depicted here are representative . DOI: http://dx . doi . org/10 . 7554/eLife . 03476 . 005 Since the track is divided into uniformly spaced vertical sections , the geometric dissimilarity between two configurations can be expressed simply as the difference between the section indices ( See explanation in Figure 1D , ‘Materials and methods’ ) . Therefore , all the correlation coefficients with the same difference between the first and the second index , Ci , i+d , will describe the correlations between ‘equally dissimilar’ track configurations separated by d bins and appear on the d-th subdiagonal of the correlation matrix . Their average value , rd = < Ci , i+d > , thus characterizes the mean correlation for a given degree of dissimilarity , and the decay of rd as a function of d characterizes the decay of correlations as a function of the difference between track geometries . The matrices in Figure 2C show that the correlations computed in the 2D reference frame decayed rapidly as more geometrically dissimilar configurations were compared . In contrast , the linear correlations remained high even for the fully expanded configuration . The difference in the mean correlation is highly significant ( p < 0 . 001 , two sample t tests ) for each point ( all values of d ) on both curves ( Figure 2D ) . The high values of the correlation coefficients for individual neurons in the linear case also show that the place fields were stable , despite the fact that active place cells could produce only a few spikes during the animal's single run through the corresponding place field in each track configuration , the effect of which was to create greater variability in our estimates of the firing rate and reduce the value of correlation coefficients ( Figure 2C ) . To this end , we also examined cells that were active on the static portions of the track ( Figure 2E–F ) and found that they demonstrated uniformly high correlation values for both reference frames ( Figure 2G–H ) . For these cells , the average values of the inter-configuration correlation coefficients in the1D and the 2D case did not diverge: the place fields of these neurons were stable despite the motion of other track elements . This demonstrates that place fields on the moving segments shift in 2D not because of some general instability but because they are , in fact , responding to changes in the configuration of the track . The same place field stability in the linear frame is clearly evident in the ensemble averages for all place cells recorded over all days of recording for each animal ( Figure 3A–C ) and in the combined data across all cells from all animals ( Figure 3D , E ) . As is clear from Figure 3A–C , the correlations were stable in the linear reference frame but decayed rapidly as a function of track configuration changes in the 2D reference frame . At the same time , the correlations for cells active only on the static portion of the track showed no tendency to decay as a function of the distance between vertical sections as the configurations changed ( compare Figure 3D , E ) . The place fields in the linear frame of reference were thus largely invariant with respect to changes in track geometry . 10 . 7554/eLife . 03476 . 006Figure 3 . Place cell firing is stable across different track configurations . ( A–C ) The correlograms for the ensemble averages for each of three animals . The place cell spiking data are taken from only the segments of the track that change position between configurations . The two lines on the graphs in the third column represent the correlation coefficients of place cell ensembles in the 1D ( red ) and 2D ( black ) frames of reference . The correlation coefficients in the 1D configuration ( red lines ) remain stable , but the correlation coefficients for the 2D configuration ( black lines ) decay as the geometric difference between track configurations increases . ( D ) Combined data from all three rats for the correlation coefficients for the combined ensemble averages over all mobile track segments . Only data from the moving segments of the track and the corresponding correlations vs track configuration separation plot for the two reference frames are included . Firing across the population was much more stable in the linear reference frame . ( E ) The same plots as in D for spikes fired over the static sections of the track . Error bars represent SEM and * indicates p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 03476 . 006 The observed stability of place cell activity patterns cannot be attributed to the animals' lack of awareness of the 2D shape of the environment or to their habituation to the environment ( ‘Materials and methods’ ) . The animals were avidly curious about the space around the elevated track , and their deft performance on the constantly changing ( if faintly glowing ) track in a dark room suggests that the proprioceptive and idiothetic information required for such behavior was available to the animals because of the anatomical connections to the hippocampus ( e . g . , with head direction cells [Calton et al . , 2003; Navratilova et al . , 2012] and other inputs [Sharp et al . , 1995; Muir and Bilkey , 2003; Lu and Bilkey , 2010] ) . Thus , it is likely that the recorded activity in area CA1 was stable despite the availability and the variability of these inputs . It might be argued that the above results could be explained by a 1D path integration mechanism that somehow excludes vestibular and some proprioceptive input so that the place cells respond simply to the distance that the animal traveled from/to the food wells ( Gothard et al . , 1996a; Gothard et al . , 1996b ) . Such a reduced path integration mechanism is sometimes expanded ad hoc to include place cell binding to local features of the track , such as joints or segments ( Chen et al . , 2013; O'Keefe and Burgess , 1996 ) . To evaluate these possibilities , we carried out an additional experiment in two animals on a U-shaped track that consisted of 11 segments , three of which could form a protrusion on either of the two sides ( Track B , see Figure 4A , ‘Materials and methods’ ) . The geometric transformation in this case consisted of keeping the protrusion on one side of the track during the first session and then flipping it to the other side during the second session . The starting side for the protrusion alternated each day . On days 1 , 3 , and 5 during the first 20-min running session , the protrusion was formed by segments 3 , 4 and 5; during the following 20 min running session , segments 7-8-9 formed the protrusion ( Figure 4B ) . On days 2 , 4 and 6 the protrusion first appeared in position 7-8-9 and then flipped to the opposite side for the second session ( Figure 4C , D ) . In order to enter the protrusion area , the animal turned to the right on its inbound journeys and to the left on its outbound journeys during both runs , before and after the flip of the protrusion . The protrusion switch thus did not affect the sequence of turns but did produce a local deformation of space , which altered the relative arrangement of the track's segments and changed the distance relationships between locations on it . As before , the deformation did not violate the track's topological structure . Unlike the Track A experiment , however , the geometry of track B changed discontinuously , as the animal was physically taken off the track between the two running sessions . 10 . 7554/eLife . 03476 . 007Figure 4 . Protrusion-flip track ( Track B ) . ( A ) Schematic representation of Track B , with segments numbered , showing the position of the track's protrusion during two successive run sessions . ( B–D ) Each show a representative place cell firing in the two configurations . ( E ) Correlations for each of the four comparisons of regions across the two configurations . The rescaled correlation is highest , indicating that the data are most compatible with a scheme in which the place cell firing does not simply respond to changes in geometry . Error bars represent SEM with * and ** indicating p < 10−8 and < 10−9 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03476 . 007 The geometric transformation provided by Track B allowed us to conduct analyses that complemented our observations on Track A . First , we asked whether place cell activity was bound to the local features on the track ( individual segments and joints ) . If so , place fields would tend to move from one side of the track to the other with the protrusion . Second , we could verify whether place cell activity was determined by the animal's linear distance along the track . If this were the case , a population of place cells would remain active at the same distances from the ends of the track before and after the protrusion switch , that is , they would shift along the track as a result of the protrusion's flip . In contrast , our topological model would predict that the place field in the region of the protrusion should not flip over or shift; instead , it should simply stretch and shrink as the protrusion is inserted and removed . So if place cell spiking activity represents connectivity , the place fields on the section of track replaced by the protrusion should extend continuously into the space provided by the protrusion . We tested these hypotheses by recording from each animal for 7 days: a total of 114 putative excitatory pyramidal cells were recorded while animals traversed Track B . As shown in the examples in Figure 4B–D , individual place cells were generally not bound to the protrusion but instead tended to stretch flexibly along the track when the protrusion was moved . To quantify the changes induced by the movement of the protrusion on the population level , we computed the ensemble averages for several types of correlations between the firing rates on the two protruding regions before and after the protrusion flip . All of these correlations were computed using the linearized data to ensure fair comparisons and included all cells that had a peak occupancy normalized firing rate above a 2 Hz threshold . This low threshold was chosen to be as inclusive as possible of different place cell firing patterns . In our analysis , we saw no tendency of CA1 place fields to remain at fixed distances from the ends of the second track ( Track B ) . Our findings cannot be explained by place cells responding purely to local features ( such as junctions ) or measuring distances from those features , since the place fields showed no tendency to be bound to specific track sections on the protrusion . Instead , place fields stretched or shrank to reflect the insertion or removal of the protrusion . The place fields were best described as remaining in their relative locations on the track . The bar graph in Figure 4E shows the mean of the resulting values of the correlation coefficients for four cases . First , we considered the possibility that the place fields remained bound to specific features of each segment of the track . The first bar , marked ‘Flip’ , shows the average correlations of the firing rate over the protrusion across the two configurations . Since this value is negative and close to zero , it is clear that the place cells on average showed no tendency to be bound to individual segments of the protrusion . The second bar , ‘Distance’ , shows the value of the firing rate correlations between the sections occupied by protrusion in the first configuration ( sections 3–5 or 7–9 , depending on which way the protrusion was facing ) and the sections at the same distance from the wells from the second configuration . Similarly to the previous case , the correlations for the distance-based comparison were close to zero; moreover , the difference in the mean correlation values between cases 1 and 2 was not statistically significant . The third bar , ‘Segment’ , shows the average firing rate correlations over the protruding section in the first configuration ( i . e . , sections 4 or 8 ) and the section onto which the center segment of protrusion projects directly after the flip ( sections 3 and 9 , respectively ) . Clearly this correlation is significantly higher than in the previous two cases . The low protrusion-to-protrusion ‘Flip’ activity correlation values compared to the protruding ‘Segment’ correlations demonstrate that most or all place cells that were active on the protrusion in Configuration I did not follow the protrusion when it was moved to the other side but instead remained in the same relative positions on the track . Lastly , the bar ‘Rescaled’ shows the firing rate correlations following rescaling of the protrusion length ( and the coordinates of the corresponding spikes ) to compensate for the spatial deformation produced by protrusion move ( e . g . , scaling the total length of the sections 3 , 4 and 5 down to the length of the section 3 that replaces the protrusion after the flip ) . Remarkably , the rescaled correlates were not only significantly larger than those of the local features ( case 1 ) and distance ( case 2 ) scenarios , but also were larger than the protruding segment correlations , indicating that local stretching could describe a greater amount of the variability in place fields than path integration ( non-parametric u-tests , which give the p values 10−5 for the differences and one way ANOVA , Tukey–Kramer post-hoc tests , p’s < 10−4 ) . The rescaled case illustrates that rate remapping compensated for the spatial expansion in such a way as to not alter the structural framework of the map . The average correlations on Track B were relatively low compared to those on Track A . This could reflect firing variability because the ‘flip’ of the protrusion Track B happened discontinuously , in one step , leading to some degree of disruption of the continuity of the environment ( similar to [Diba and Buzsaki , 2008] , where the transformation also induced partial remapping ) . We hypothesize that , were we able to carry out this experiment using a continuous transformation from the straight section of the track to the protrusion , as we were able to do with Track A , the hippocampal network would have retained a greater degree of similarity across the configurations . Nevertheless , the difference in the correlations between these four cases was statistically significant . The fact that both the mean and individual place cell activity across the two sessions showed the highest correlation after the rescaling transformation was applied , while the ‘Distance’ and ‘Flip’ correlations were low , suggests that place cell firing , despite its variability , served primarily to represent the connectivity of the environment through the temporal sequence of place cell activity . The CA1 spatial representation thus remained invariant with respect to a wide range of geometric transformations . By explicitly counterposing geometry and topology in this series of experiments , we have demonstrated that hippocampal place cells show a remarkable degree of stability with respect to even dramatic geometrical changes in the environment: they retained their relative order and connectivity despite routes involving diametrically opposite protrusions and orientations . As long as the sequence of spaces experienced remains the same , in other words , and the place fields still overlap , it would appear not to matter to the hippocampus if the place fields are stretched by two centimeters or twenty . We therefore propose that place cells do not represent locations in space but rather provide a spatial context for experience ( Eichenbaum et al . , 1999; Moser et al . , 2008 ) . This would seem to be more compatible with the higher-order functions of the hippocampus in representing and storing experience , and it is consistent with results from experiments that have been conducted in morphing environments , which have an implicit , if overlooked , topological aspect . Muller and Kubie ( 1987 ) found that enlarging their open field set-up caused place fields to enlarge while maintaining the same shape and relative order ( i . e . , the connectivity of the environment remained invariant ) . Similar observations were made by Gothard et al . , 1996a on a linear track , by O'Keefe and Burgess ( 1996 ) in rectangular environments , and by other studies in morphing environments ( Lever et al . , 2002; Leutgeb et al . , 2005; Touretzky et al . , 2005; Wills et al . , 2005; Colgin et al . , 2010 ) . This phenomenon is topological in character , since the fields stretch with the space but maintain their basic position and sequence . A more detailed analysis of place cell activity in a stretching 1D environment reported in Diba and Buzsaki ( 2008 ) shows that the temporal relationships between the firing of the relevant place cells remains fixed throughout the stretch , despite the change in the overall firing rate ( rate remapping , Dupret et al . , 2010 ) . In addition to these previous studies showing that place fields stretch ( maintain their sequence and relative positions ) in morphing environments , we can add several other compelling reasons to doubt that the hippocampal map is exclusively or fundamentally Cartesian , like a street map . There is our recent modeling work showing the feasibility of spatial learning based on topological information ( Dabaghian et al . , 2012; Arai et al . , 2014 ) ; early studies by Piaget and Inhelder showing that children first conceive of spatial relations in topological terms and only later think in terms of geometrical positions ( Dawson and Doddington , 1973 ) ; and previous studies on rodent navigation showing that place cells respond to topological changes in the environment such as the placement or removal of a barrier across a previously learned route ( Poucet and Herrmann , 2001; Alvernhe et al . , 2011 , 2012 ) . We propose that a topological framework , more like a subway map , would provide the hippocampus a more powerful , flexible , and readily formed spatial substrate for creating experiential memories or evaluating the spatial context of behaviors ( Lavenex and Amaral , 2000; Banquet et al . , 2005 ) . Indeed , one study found that cells in the rat medial prefrontal cortex , recipients of place cell efferents , did not fire in response to the rat's position or head direction but rather in response to specific goal-oriented behaviors ( Poucet , 1997 ) . All this is not to say that other brain regions do not contribute metrical information to the internal spatial map—there is abundant evidence that many regions are involved in providing sensory , idiothetic and proprioceptive information for path integration ( Knierim et al . , 1996 , 1998; Fenton et al . , 2000a , 2000b; Quiroga et al . , 2005; Yoganarasimha et al . , 2006; Moser et al . , 2008 ) —only to propose that the hippocampus serves a more general role in spatial cognition by providing a qualitative , flexible representation of the environment . We suggest that place cells emphasize the connections between portions of a given environment and the sequence in which they are experienced . Such an encoded sequence is evident in spontaneous replays of place cell spiking during quiescent ( Muller and Kubie , 1989; Davidson et al . , 2009; Karlsson and Frank , 2009; Carr et al . , 2011; Dragoi and Tonegawa , 2011; Zeithamova et al . , 2012 ) and sleep states ( Skaggs and McNaughton , 1996; Louie and Wilson , 2001; Foster and Wilson , 2006 ) , which recapitulate the order in which the place cells fired during navigation . This implies that when the hippocampal network is driven internally , the structure of sequences remains the same as during navigation , although the activity of place cells under these circumstances is clearly not coupled directly with spatial locations . The encoded map thus appears to serve as a template for generating replay sequences and facilitating imaginative navigation . More recently , studies of hippocampal preplay ( as opposed to replay ) indicate that the hippocampus has a repertoire of preconfigured temporal place cell firing sequences that can be called upon to rapidly encode multiple novel spatial experiences ( Dragoi and Tonegawa , 2013 ) . It is striking that studies from such diverse fields , from electrophysiology to algebraic topology , seem to converge on the notion that the hippocampus operates by providing a flexible , rough-and-ready framework into which spatially grounded experiences can be organized and remembered . If our topological model is borne out , it will have several implications . It is well known that if the environment changes significantly , then place cell activity can undergo global remapping , and if the changes are small , then the place cells preserve the relative order of spiking and respond only by a regular change of their spiking rates ( rate remapping ) ( Colgin et al . , 2008; Allen et al . , 2012 ) . Our topological hypothesis predicts that , within limits ( which are as yet unknown ) , geometric transformations should cause only rate remapping while the sequence of firing remains invariant . Our topological model would also greatly narrow the realm of meaningful correlations between changes in visual cues and responses in place cell spiking . As we have tried to emphasize in this work , a place cell response does not necessarily translate into information communicated downstream . In the absence of global remapping , it may thus not be safe to interpret correlations between changes in stimuli and place cell activity as representing information that is transmitted to other brain regions . A change in response would primarily serve to establish the scope of geometric invariance , which could depend on a variety of biological conditions and which is a question that must be left for future investigations . Moreover , since the sequence of spaces traversed is encoded in ensembles of place cells rather than individual neurons , it is place cell ensembles , not individual neurons , whose responses to changes in the environment need to be studied . This of course poses technical challenges , but such challenges are more likely to be met if we see the necessity of doing so . The notion that the hippocampus encodes topological relationships could help clarify the spatial vs non-spatial debate that has gone on for many years ( Eichenbaum et al . , 1999; O'Keefe and Nadel , 1978 ) . Indeed , it is well known that the rat hippocampus is important for learning and remembering various types of non-spatial information , including visual , odor and action sequences ( Agster et al . , 2002; Fortin et al . , 2002 , 2004; Sauvage et al . , 2008 ) . These functions , and the importance of the human hippocampus for episodic memory ( Squire , 1992 ) , have led to the suggestion that the hippocampus performs a more general associative memory function ( Eichenbaum et al . , 1999; Wood et al . , 1999 ) . It is not entirely clear how the same mechanism would be responsible for representing geometry ( distances and angles ) in the spatial domain and encoding sequences and events in the mnemonic domain . The beauty of the topological hypothesis is that hippocampal spatial maps would have the same structure as memory maps or ‘memory spaces’ ( Eichenbaum et al . , 1999; Eichenbaum , 2000 ) , which would not only explain the stability of relative firing order in flexible environments , but could also provide a common framework for understanding spatial representation , spatial memory and episodic memory mechanisms . We built the tracks as follows: track A ( Figure 1A ) consisted of 10 straight segments 35–40 cm long each ( total length ∼4 m ) , shown in color in Figure 1B . The width of the segments ( ∼8 cm ) and the height of the rim ( 3 cm ) were fixed over the whole length of the track with the exception of the circular joint areas ( 120 mm in diameter ) between the track segments that had no outer rims . While the two leftmost segments , 5 and 6 , remained fixed , the four segments 1–4 of the top arm and the segments 7–10 of the bottom arm could move , so that segments 1 and 10 could be displaced over the full span of ∼140 cm each . During the recordings , each arm was moved over a selected distance on four remotely controlled platforms positioned on a system of rails , visible in Figure 2A , driven independently by two stepper-motors . The distance moved varied pseudo-randomly across trials . The motions of the two arms were unrelated to each other: a contraction phase of one arm could be accompanied either by contraction or by expansion of the other arm , and the distance the arm moved varied pseudo-randomly across trials . Our goal of studying the geometric invariance of the place cell maps imposed a number of constraints on the experimental design . For example , embedding a flexible environment into a stable background would unavoidably produce geometrically complex conflicts between the local and the distal cues ( Knierim et al . , 1996; Shapiro et al . , 1997; Knierim , 2002; Knierim and Rao , 2003 ) . To avoid such issues and to control for information about the track configuration that the rat could infer from changes in the relative positions of the objects in the room , we excluded distal cues by conducting the experiment with no external lighting . To facilitate the rat's navigation , we covered the track surfaces with a scentless , washable , nontoxic glow-in-the-dark paint ( Mister Art ASTM D-4236 ) that provided a soft green glow but not enough light for the distal walls to be visible . The entire visible space was thus concentrated around the track . Furthermore , the view of one glowing arm of the track from the other was blocked by a large screen so that the rat could see only the arm of the track he was traversing . The stepper motors were relatively quiet , and the animals did not display any apparent reactions to the sounds or vibrations generated during the track position shifts . During recordings on Track A , we moved the remotely controlled arms while the animal received the reward ( Hershey's chocolate milk ) at the food well on the tip of opposite arm , so as to avoid cuing the rat about the track configuration change . We adjusted the speed of the platform displacement so that the track arms could be fully expanded or contracted during the period in which the rat was consuming the reward while also minimizing track vibrations and motor noise . There were a total of 7 days of recording for each animal on track A that included 2 run sessions and 1 sleep session . During each run session , both arms reached the maximally contracted and the maximally stretched positions a few times . The exact number of track moves on each day depended on the animal's foraging activity and ranged between 10 and 80 moves , with an average of 40 moves . Track B ( Figure 4A ) , used for control recordings in CA1 ( see below ) , contained 11 segments made with the same material as the Track A , with a total length of 4 . 5 m . During each running session three segments formed a protrusion on one or the other side of the track . Two animals were exposed to Track B . For one of these animals the exposures were interleaved with exposures to Track A ( with 10 hr’ separation between exposures to track A and track B ) , while the other animal experienced only Track B . Each animal had 7 days of experience on Track B . Each recording session consisted of two 20 min runs and two 15 min sleep sessions in a run-sleep-run-sleep sequence . On days 1 , 3 and 5 the protrusion was formed by segments 3 , 4 and 5 during the first run session; during the first sleep session the same physical sections were ‘flipped’ to the opposite side , to positions 7-8-9 ( Figure 4B ) . On days 2 , 4 and 6 the protrusion was first formed by sections 7-8-9 , then flipped to the 5-4-3 position ( Figure 4C , D ) . All the experimental procedures were approved by the Institutional Animal Care and Use Committee at UCSF . A total of four male Long-Evans rats were used in two separate groups of experiments . During the 2 weeks prior to surgery , the animals were pre-trained on a static U-shaped track to alternate between two food wells located at the ends of the track for chocolate milk rewards . During pre-training and after recovery from surgery , the rats were kept at 85% of the free-feeding baseline weight . Outside of the experimental recording periods , the rats had unlimited access to water in their home cages and were kept on a 12:12 hr light–dark cycle . The experimental testing took place during the dark phase of the cycle . At the end of the training period the rats were implanted with a 30-tetrode drive ( see Karlsson and Frank , 2008 for details ) . About half of the tetrodes were lowered directly into the CA1 area of the hippocampus and the other half remained in the parietal cortex , located above the CA1 area ( cannulas at ( −4 , ±2 . 1 ) and ( −4 . 8 , ±4 . 4 ) ) . During the recordings , the animals had no difficulty adapting to the changing geometry of the track and showed no signs of being distracted by the slight noise of the motors or track vibrations . Since the pre-training track was fully stretched and static , the animals showed exploratory behavior on the flexible track during the initial recording sessions , by balancing over the edges and exploring the space surrounding the track segments , especially in the vicinity of the joints . Nonetheless , animals performed the task very well , completing the full trajectory from one food well to the other on 98% of runs . The average time required to complete the track ( ∼34 s ) did not differ significantly from the level of the animals' performance on the static track during pre-training . Given this reliable alternation behavior we were able to quantify the place cell responses on the track . We separated putative excitatory cells from putative inhibitory cells using the standard criteria based on firing rate , spike width and the mean ISI . All analyses excluded times when animals were moving on the track at a speed of less than 3 cm/s . Data analysis was performed using MATLAB . Since we were interested in comparing place cell firing across different track configurations , we needed both a system to describe the track configurations ( and allow comparison between different configurations ) and a way of linearizing the track to allow us to compare firing in the 2-dimensional planar and 1-dimensional linear frames of reference . We used standard methods for binning the space to define precisely where the place cells fired . We describe our methods for each of these needs below . The total number of track moves per day was large ( up to 80 moves ) and the rat spent only a short period of time in each configuration ( about 34 s including the slow exploratory movements of the animal around certain segment junctions ) . In order to make the statistical analysis more robust , we combined similar track configurations in groups . Since the arms contract uniformly ( i . e . , the angles between segments 1 and 2 , 3 and 4 , etc . , remain similar ) , the configuration of each arm can be specified by the position of the endpoints of the track ( the food wells , F1 and F2 ) . Therefore , it is possible to split the full area swept by the moving sections of Track A into a number of vertical sections and then to specify geometrically similar track configurations by the index of vertical sections occupied by the track's ends ( Figure 1D ) . We used Ntot =12 vertical sections on each arm of the track , 9 cm wide , see Figure 1C ( note that the bottom sections are slightly displaced; the bottom arm of the track did not stretch fully because of a mechanical difficulty ) . The number NV of track configurations that fall into a particular section V varied daily depending on the level of the animal's foraging activity . However , each vertical section V was populated by at least three track configurations , NV≥3 , during each day of recordings , that is , only configurations that were experienced at least 3 times were included in our analyses . This method of defining the track configurations , illustrated in Figure 1D , provides the basis for comparison of different track configurations that result in the matrices depicted in Figures 2 and 3 . In order to directly compare the spiking activity in the planar and in the linear frames of reference , we used the planar coordinates ( the usual x-y coordinates of the 2D plane ) and then linearized the coordinates for 1D such that xL provides the distance from one of the food wells , and yL provides the displacement from the midline of the track ( recall that the track is 9 cm wide ) . As a result , the same area of physical space can be described in both the 2D and the 1D representations , which permits a fair comparison of correlations across the two reference frames . We then binned the space into 3 × 3 cm squares in both the linear and in the planar frames of reference and computed the occupancy-normalized spatial rates for each place cell for all the track configurations that fall into a particular spatial bin ( Figure 1—figure supplement 1 ) . Since the global placement of the grid of spatial bins is arbitrary , we averaged the rates over the values obtained by shifting the grid randomly by up to a half-bin width in both the x and y directions . As a result , the firing rates for all cells in the planar and the linear reference frames are roughly the same and can thus be used as characteristics of a given place cell's activity both in the planar and in the linear perspective . No other smoothing was applied to the data . To ensure meaningful comparisons of the correlations between spatial firing patterns in linear and planar reference frames , we considered only those bins ( 1 ) that were actually visited by the rat in both configurations and ( 2 ) in which at least one spike was fired . Since the animals spent only a relatively short time on each segment of the track during each run , an activated place cell typically produced only a few spikes . Given that place cells are highly variable ( Brown et al . , 1998; Fenton and Muller , 1998 ) , we combined multiple passes through a given place field in similar track geometries in order to obtain meaningful results ( ‘Materials and methods’ and Figure 3D , E ) . This reduced the variability of our estimates of the rate while preserving distinct estimates for different shapes . Nonetheless , our estimates were based on relatively few ( 3 to 7 ) passes through each location , which lowers the correlations of place cell activity rates on the moving segments of the track , as compared to the stable segments ( 5 and 6 ) . For this reason , we analyzed the spiking occurring on the mobile sections of the track ( sections 1–4 and 7–10 ) separately from the spiking at fixed regions ( sections 5–6 ) .	The hippocampus is one of the most easily recognizable structures in the brain owing to its characteristic seahorse-like shape . Brain imaging studies in the 1990s famously showed the hippocampus to be larger in London taxi drivers than in other people , suggesting that it plays a role in spatial navigation . This was consistent with previous findings in rodents , which had shown that the hippocampus is active when animals find their way through mazes . Electrode recordings have revealed that whenever an animal is in a specific location of a particular environment ( for example , in the back left-hand corner of a small room with white walls ) one or a small number of cells within the hippocampus will fire to encode that location . When the animal moves to a new location within the same environment , other cells will fire to encode the new location . In this way , the population of cells—which are known as place cells—can together construct a virtual ‘map’ of the environment . It is generally assumed that this hippocampal map represents space in terms of the absolute distances and angles between locations , rather like a street map . However , this type of geometric map appears inconsistent with the results of certain experiments . Dabaghian et al . proposed instead that the hippocampal map is based on topology , or the relative order of locations and the connections between them , rather like a subway map . Subsequently , computer models demonstrated that virtual simulations of place cells could effectively ‘learn’ the topological features of different environments . Now , Dabaghian et al . provide their own empirical data to support the existence of a hippocampal ‘subway-style’ map by recording the electrical impulses from place cells in the rat hippocampus as the animals ran through a U-shaped maze . The maze was constructed so that its arms could either be straight or folded into zigzags . Changing the maze in this way does not alter its topology because the relative order of its various components—such as the positions of food wells in the arms—is unchanged , but it does alter the maze's geometry . Notably , as rats ran through different conformations of the maze , the activity of the place cells in their hippocampi remained largely unchanged , consistent with a map based on topology rather than geometry . By providing evidence that hippocampal maps have more in common with subway maps than street maps , the work of Dabaghian et al . offers an explanation for previously challenging results and provides a framework for further experiments into hippocampal memory function .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Reconceiving the hippocampal map as a topological template
Cardiovascular regulation is integral to life . Animal studies have identified both neural and endocrine pathways , by which the central nervous system adjusts cardiac output and peripheral vascular resistance to changing physiological demands . The outflow of these pathways is coordinated by various central nervous regions based on afferent information from baroreceptors , chemoreceptors , nociceptors , and circulating hormones , and is modulated by physiologic and behavioural state . In humans , however , knowledge on central cardiovascular regulation below the cortical level is scarce . Here , we show using functional MRI ( fMRI ) that at least three hypothalamic subsystems are involved in cardiovascular regulation in humans . The rhythmic behaviour of these systems corresponds to high and low frequency oscillations typically seen in blood pressure and heart rate variability . Given the importance of cardiovascular regulation for our daily survival , it is not surprising that the body has several redundant and mutually interacting systems for this task . Important representatives include the classical baroreceptor ( arterial and cardiopulmonary ) and chemoreceptor reflexes , neuroendocrine systems , like the vasopressin , sympatho-adrenal , renin-angiotensin-aldosterone , and the more recently discovered leptin-melanocortin system ( Sawchenko and Swanson , 1981; Dampney , 1994; Hilzendeger et al . , 2012; Salman , 2016 ) . Much of our present understanding of these systems stems from animals experiments , while mechanistic studies in humans remain scarce , mostly for lack of non-invasive methods to assess subcortical brain activity . Furthermore , most studies have evaluated only one system at a time . We must assume , however , that all neural and neuroendocrine cardiovascular control systems are carefully orchestrated by the central nervous system to achieve optimal regulation . The control centres responsible for such orchestration are presumed to be located in the brainstem and hypothalamus . They are well characterised for the ‘textbook’ baroreflex , but much less for the other systems . To study cardiovascular regulation in human subjects , we devised an MR-compatible lower body negative pressure ( LBNP ) chamber that simulates orthostatic stress via footward blood volume displacement ( Figure 1c; Goswami et al . , 2008 ) . The pressure of −30 mmHg used in our experiment has been shown to recruit both neural and endocrine mechanisms of cardiovascular regulation via activation of cardiopulmonary and arterial baroreceptors ( Loewy , 1981; Mark and Mancia , 1983; Kimmerly et al . , 2005; Salman , 2016 ) . While assessment of endocrine pathways requires invasive methods , neural cardiovascular regulation by the autonomic nervous system can be measured non-invasively as it leads to characteristic rhythms of heart rate and blood pressure variability ( HRV , BPV ) . To assess these rhythmic changes , we recorded blood pressure and heart rate traces during the LBNP-fMRI measurements and subjected them to spectral analysis isolating two frequency bands: low frequency blood pressure variability ( LFBPV , ~0 . 1 Hz ) , reflecting the so-called Mayer waves of sympathetic origin ( Julien , 2006 ) ; and high frequency heart rate variability ( HFHRV , ~0 . 28 Hz ) reflecting the primarily vagally mediated respiratory sinus arrhythmia ( Billman , 2013 ) . Note that both the LF and HF bands lie at the top or above the canonical frequency range of the BOLD signal ( <0 . 1 Hz , Figure 1i ) ; although several studies have recently provided evidence for BOLD oscillations beyond this frequency range ( Chen and Glover , 2015; Lewis et al . , 2016 ) . Subcortical control regions of cardiovascular function are mainly located in the hypothalamus , comprising the paraventricular nucleus ( PVN ) , lateral hypothalamic area ( LH ) , arcuate nucleus ( Arc ) , dorsomedial nucleus ( DMH ) , and median preoptic nucleus ( Supplementary file 1 ) , and in the lower brainstem , comprising the nucleus of the solitary tract ( NTS ) , rostral and caudal ventrolateral medulla ( RVLM/CVLM ) , nucleus ambiguus ( Amb ) , and the caudal raphe nuclei , that is nucleus raphe obscurus ( ROb ) and nucleus raphe pallidus ( RPa ) ( Supplementary file 2; Coote , 2004; Dampney , 1994; Loewy , 1981; Saper et al . , 2015; Benarroch , 1993 ) . Since these areas are notoriously hard to investigate in humans due to their small size , depth within the skull , and physiological noise from surrounding vessels and cerebrospinal fluid ( Brooks et al . , 2013; Beissner , 2015 ) , we devised a high-resolution functional MRI approach and preprocessing pipeline specifically optimised for subcortical imaging . In particular , we balanced spatial and temporal resolution ( 2×2×2 mm³ , 1 . 23 s ) to distinguish neighbouring nuclei , while critically sampling respiratory frequencies . Preprocessing involved maximising anatomical specificity by applying advanced distortion correction , symmetric diffeomorphic image registration to a study template , and omitting any spatial smoothing . In contrast to common practice , we did not regress physiological fluctuations from our data to avoid removing meaningful signal from the cardiovascular regions we were interested in Iacovella and Hasson , 2011 . Instead , we opted for a spatial noise correction approach ( Beissner et al . , 2014 ) as part of our group-level statistical analysis . LBNP led to a significant reduction of systolic blood pressure ( Rest: 132 . 6 ± 20 . 2 mmHg , LBNP: 113 . 4 ± 20 . 2 mmHg , t ( 17 ) =-11 . 3 , p<0 . 001 , Cohen's d = −0 . 95 ) and a compensatory decrease of the interbeat interval ( Rest: 0 . 95 ± 0 . 12 s , LBNP: 0 . 85 ± 0 . 11 s , t ( 20 ) =-4 . 9 , p<0 . 001 , d = −0 . 96 ) , while measures of respiration ( respiratory interval , respiratory volume , and respiratory volume per time ) showed no significant changes . Furthermore , head movement was not significantly different between LBNP and Rest periods . We also observed increased spectral power of LFBPV ( Rest: 0 . 14 ± 0 . 03 , LBNP: 0 . 17 ± 0 . 04 , t ( 17 ) =2 . 9 , p=0 . 011 , d = 0 . 97 ) and reduced spectral power of HFHRV ( Rest: 0 . 18 ± 0 . 08 , LBNP: 0 . 14 ± 0 . 08 , t ( 20 ) =-2 . 4 , p=0 . 0267 , d = −0 . 47 ) indicating sympathetic excitation and vagal inhibition in response to the cardiovascular challenge , respectively ( Figure 1d , e ) . To identify cardiovascular control centres within the hypothalamus , we followed a two-step process; namely segmentation using masked independent component analysis ( mICA ) ( Beissner et al . , 2014 ) of the fMRI time series , followed by testing for cardiovascular involvement using two independent criteria . In the first step , the hypothalamus was segmented into 49 functionally distinct regions ( Figure 1—figure supplement 1 ) , followed by removal of six unspecific components ( Figure 1f ) , leaving 43 components . The initial number 49 was derived by maximising mICA reproducibility , see Figure 1g . This high-dimensional segmentation yielded the characteristic three medio-lateral zones and three rostro-caudal regions expected from post-mortem anatomical studies ( Dudás , 2013; Figure 1j , Figure 1—figure supplement 2 , Supplementary file 3 ) . The lower-dimensional decompositions with 8 and 10 subregions showed only two medio-lateral zones , and were thus not analysed . They were , however , consulted later to elucidate averaged intra-hypothalamic functional connectivity ( see below ) . In the second step , each of the 43 ( 49 - 6 ) hypothalamic regions was tested for two criteria , namely LBNP-related changes in functional connectivity ( ΔfcLBNP ) with any region of the hypothalamus or lower brainstem , and changes in BOLD spectral power in one or both of the two cardiovascular frequency bands ( Figure 1h ) . Our analysis revealed five hypothalamic regions fulfilling both criteria for cardiovascular involvement ( Figure 2 , Supplementary File 4 ) : the right anterior , and bilateral tuberal LH/supraoptic nucleus ( SON , Figure 2a–c ) , the right tuberal PVN/posterior hypothalamic area ( PH ) ( Figure 2d ) , and the arcuate nucleus ( Figure 2e ) . Detailed information on the spatial distribution of BOLD spectral power in the two cardiovascular frequency bands is provided in Figure 2—figure supplement 1 . All identified regions except the bilateral tuberal LH/SON showed positive ΔfcLBNP with the lower brainstem ( Figure 2a , d–e ) . In contrast , ΔfcLBNP of the tuberal LH/SON was restricted to the hypothalamus . Here , both sides showed positive within-nucleus ΔfcLBNP , which can be interpreted as local activity changes , while only the left LH/SON showed additional negative ΔfcLBNP with the arcuate nucleus ( Figure 2b–c ) . Further insights on hypothalamic functional connectivity of the identified regions came from the analysis of the low-dimensional ICA results ( Figure 3 ) . Here , inter-dimensional matching showed that right anterior LH/SON , PVN/PH , and Arc had low-dimensional equivalents ( Figure 3b , d ) , while the bilateral tuberal LH/SON did not . The low-dimensional versions of Arc and right anterior LH both included portions of the PVN/PH and adjacent nuclei . Our supplementary analyses showed that our results are not driven by physiological noise . As the first analysis revealed , physiological noise regression had little effect on the masked ICA results as evidenced by the functional segmentation still showing the same known anatomical subdivisions of the hypothalamus ( Figure 2—figure supplement 2 ) . This was expected as we had previously shown that noise regression has little effect on masked ICA results ( Beissner et al . , 2014 ) . Moreover , the ICs matching our five hypothalamic regions from the main analysis ( spatial correlation r > 0 . 89 ) showed similar , yet smaller functional connectivity changes in response to cardiovascular challenge ( Figure 2 , Figure 2—figure supplement 3 ) . The second supplementary analysis revealed that four of the five hypothalamic regions from the main analysis showed significant functional connectivity with cerebral regions that was clearly dominated by grey matter ( Figure 2—figure supplement 4 ) , while noise components would be expected to mainly show white matter or ventricular connectivity . The one remaining region was the PVN/PH , which showed very little cerebral connectivity at all . We found five hypothalamic regions involved in cardiovascular regulation: the right anterior and bilateral tuberal LH/SON , the right tuberal PVN/PH and the arcuate nucleus . This selection of nuclei agrees with previous results from studies using Fos-like protein expression in response to prolonged hypotension ( Li and Dampney , 1994; Figure 2—figure supplement 5 ) . Based on functional connectivity changes between the original hypothalamic regions and the lower brainstem ( Figure 2 ) , we were able to distinguish three major hypothalamic cardiovascular control systems ( Figure 4c–e ) . The first was characterised by positive ΔfcLBNP of the PVN/PH with a region in the ipsilateral lateral medulla comprising the RVLM , inferior olivary ( IO ) , parvicellular reticular nucleus ( PCRt ) , Amb , and intermediate reticular ( IRt ) nuclei , as well as negative ΔfcLBNP of the PVN/PH with itself ( Figure 2d ) . We suggest that this system represents the ‘textbook’ baroreflex arc with its sympathetic ( RVLM ) and vagal arms ( Amb ) ; both under hypothalamic control by the PVN . Since baroreflex regulation classically involves the NTS and caudal ventrolateral medulla ( CVLM ) , we conducted a supplementary mICA in the lower brainstem . This analysis yielded a functionally distinct region matching the above-mentioned RVLM cluster which was tested for ΔfcLBNP . As expected , a single cluster in the ipsilateral caudal NTS showed positive ΔfcLBNP ( Figure 3—figure supplement 1 ) ; however , we did not observe any functional connectivity changes with the CVLM . There is abundant evidence from animal experiments that the PVN projects heavily to the RVLM and Amb and to a lesser extent to Sp5 and reticular nuclei . These projections involve a wide variety of neurotransmitters including angiotensin-II , vasopressin , glutamate , and corticotropin-releasing hormone for RVLM , as well as oxytocin for the Amb ( Loewy , 1981; Coote , 2004; Geerling et al . , 2010; Sapru , 2013 ) . Damage to the baroreflex arc elicits profound abnormalities in human blood pressure control ( Biaggioni et al . , 1994 ) . Therapeutically , electrical baroreceptor stimulation has been tested for treating arterial hypertension and heart failure , although patients showed disparate treatment outcome ( Heusser et al . , 2016 ) . Thus , a deeper knowledge of central baroreflex control may help understand inter-individual differences and identify patients most likely to respond to such therapies . The second hypothalamic system ( Figure 2a +4d ) involved the right anterior aspect of the LH/SON showing positive ΔfcLBNP with a small region in the ipsilateral dorsal medulla comprising the NTS , dorsal motor nucleus of the vagal nerve ( DMN ) , and PCRt . These findings agree with animal experiments showing afferent connections of the LH/SON from the NTS ( Sawchenko and Swanson , 1982; Card et al . , 2011 ) and efferent projections to the NTS and DMN , the latter of which have been shown to be orexinergic ( Allen and Cechetto , 1992; Coote , 2004 ) . Their exact role in cardiovascular regulation , however , is still unclear . Potentially , this system also represents direct projections from the NTS to the SON that trigger the release of vasoactive hormones from the pituitary gland ( Grindstaff and Cunningham , 2001 ) . This pathway may complement neural cardiovascular regulation and serve as a backup for blood pressure maintenance in humans ( Jordan et al . , 2000 ) . Clinically , hypotension-induced vasopressin release is attenuated in patients with neurodegenerative diseases affecting cardiovascular centres in the brainstem and hypothalamus ( Puritz et al . , 1983 ) . The third hypothalamic system involved the arcuate nucleus ( Arc ) ( Figure 2e +4e ) , and showed positive ΔfcLBNP with three different medullary regions . The first was located in the right dorsal medulla comprising NTS , DMN , PCRt , IRt and dorsal paragigantocellular nucleus ( DPGi ) . The second included the right IRt , right medullary reticular nucleus ( MdRt ) and midline ROb . And finally , a third small region spanned the left NTS/DMN with its rostrocaudal position closely matching that of the contralateral NTS observed for the ‘textbook’ baroreflex control system ( Figure 3—figure supplement 1 ) . Interestingly , the role of the Arc in cardiovascular regulation has received relatively little attention despite several lines of evidence . The Arc has long been known to receive afferent fibres from the NTS ( Ricardo and Koh , 1978 ) . Changes in blood pressure and vascular resistance during electric stimulation , Fos-like protein expression in response to prolonged hypotension and afferent renal nerve stimulation , and retrograde tracing studies link it to key cardiovascular organs ( Li and Dampney , 1994; Sapru , 2013; Rahmouni , 2016 ) . Furthermore , the Arc has been proposed as a potential region in which the leptin-melanocortin and renin-angiotensin-aldosterone systems interact to control sympathetic nerve activity ( Hilzendeger et al . , 2012 ) , relevant to obesity-associated arterial hypertension ( Greenfield et al . , 2009 ) . This issue is of utmost clinical relevance given the pandemic rise in the prevalence of obesity and associated cardiovascular disease . This study combined functional connectivity and spectral analysis of fMRI signals to investigate hypothalamic regions involved during cardiovascular regulation . We found anatomically meaningful connectivity changes in five hypothalamic subregions during the cardiovascular challenge , all of which went along with spectral changes of the fMRI signal in the domains of low and high frequency cardiovascular regulation ( LF , ~0 . 1 Hz; HF , ~0 . 28 Hz ) . In order to interpret these changes as BOLD signals , we need to assume that , at least for the HF domain , BOLD extends beyond the frequency range given by the canonical HRF ( Figure 1i ) . Similar observations have recently been reported by several other groups ( Chen and Glover , 2015; Gohel and Biswal , 2015; Lewis et al . , 2016 ) . However , caution is advisable , when interpreting results derived from BOLD signals . While the majority of our results were robust against the removal of physiological ‘noise’ signals , one should note that completely disentangling neuronal from non-neuronal BOLD signals is close to impossible for several reasons . Firstly , physiological noise is a mixture of several different noise sources , including those related to cardiac activity causing changes in arterial pulsatility , cerebral blood flow , cerebral blood volume and cerebrospinal fluid flow , and those related to respiration causing changes in the magnetic field and arterial carbon dioxide ( Brooks et al . , 2013 ) . To complicate matters further , cardiac and respiratory activity are not independent but intricately linked through phenomena like the respiratory sinus arrhythmia . Secondly , several groups have recently shown that cerebral vascular regulation may be coordinated across long-distance brain regions , thus mimicking the structure of neuronal networks ( Bright et al . , 2020; Chen et al . , 2020 ) . Finally , every temporal regression of physiological ‘noise’ bears the risk of removing meaningful signal from the data ( Iacovella and Hasson , 2011 ) . Thus , the methodological approach used in this study cannot completely rule out the possibility that some of the connectivity we are seeing could be the result of highly structured physiological noise . Nonetheless , we anticipate that the simultaneous in-vivo assessment of the different cardiovascular control systems in humans will deepen our understanding of these systems . Furthermore , access to individual neural signatures may facilitate development of novel pharmaceutical and electroceutical cardiovascular treatments , leading to an era of cardiovascular precision medicine . All MR images were acquired on a Siemens 3T MAGNETOM Skyra using a 64-channel head/neck coil . The scanning protocol consisted of the following sequences ( Supplementary file 5 ) : fMRI data preprocessing was optimised for the brainstem and hypothalamus by avoiding superfluous resampling steps and unnecessary smoothing . On that account , motion correction ( MCFLIRT [Jenkinson et al . , 2002] ) and unwarping ( topup [Andersson et al . , 2003] ) were applied in a single transformation . Afterwards , brain extraction ( BET [Smith , 2002] ) , grand mean scaling and high pass filtering ( 0 . 005 Hz ) were applied . The data were not smoothed . Two study templates were generated using Advanced Normalization Tools ( ANTs [Avants et al . , 2008] ) . The first one using the unwarped EPI reference images ( II ) ; the second one using the T1-images . Functional images were transformed to the T1-template for group analyses in a single transformation . This one-step procedure included a non-linear registration ( symmetric diffeomorphism ) to the EPI-template followed by a six-degrees-of-freedom transformation to the T1-template . Finally , the results were transformed to Montreal Neurological Institute ( MNI ) standard space using a non-linear transformation ( ANTs ) . All transformations used linear interpolation during the resampling . The following physiological measures were acquired with an MR-compatible BIOPAC MP150 system: Pulse data were bandpass filtered ( 0 . 64–2 . 5 Hz ) to remove scanner artefacts before running a semi-automated peak-detection . In the blood pressure recordings we first removed all dropouts and then applied a percentile filter ( 1st-99th percentiles ) . Excluded data points were interpolated linearly . First , a masked independent component analysis ( mICA [Beissner et al . , 2014] ) was performed on the concatenated functional data of all subjects ( Figure 1f ) . We defined the hypothalamic mask for this analysis ( Figure 1b ) using the atlas of Mai et al . , 2016 . The ICA dimensionality of 49 was derived by test-retest reproducibility analysis in the range between 1 and 100 dimensions using 30 random split-half samplings ( mICA toolbox [Moher Alsady et al . , 2016] ) . After matching the components of both half samples by Hungarian sorting of their cross-correlation matrix , mean reproducibility was calculated . After excluding all values with ICA non-convergence , a dimensionality of 49 was found to maximise mean reproducibility ( rmean = 0 . 80 ± 0 . 01 ) ( Figure 1g ) . Independent components ( ICs ) were tested for specificity by running an unmasked dual regression ( Beckmann et al . , 2009 ) to a cuboid volume containing the brainstem and hypothalamus . We calculated the weighted quotient of activation in grey and white matter versus cerebrospinal fluid ( probabilistic masks obtained using FAST [Zhang et al . , 2001] ) . Components were considered unspecific if this quotient was smaller than one standard deviation from the mean . Using this measure , 6 components were excluded from further analysis . Assigning each voxel the component number with the highest z-value at that point yielded a segmentation of the hypothalamus into 43 functional centres ( Figure 1j ) . To discern which of the remaining 43 ICs were involved in cardiovascular regulation , we carried out two separate analyses: a spectral and a functional connectivity analysis . For the spectral analysis we defined two frequency bands , low frequency LF ( 0 . 1±0 . 03 Hz ) and high frequency HF ( 0 . 28 ± 0 . 06 Hz ) by looking at the spectral peaks of the autonomic recordings . The LF band was defined by the position of the Mayer peak in our subjects’ blood pressure variability , whereas the HF band corresponds to the respiration frequency ( mean ± standard deviation ) . We extracted the time series of every IC in every run and calculated their power spectral density for each frequency band normalised to the total power ( similar to an fALFF analysis [Zou et al . , 2008] at higher frequencies ) . Finally , we computed a non-parametric paired t-test to quantify spectral differences in the BOLD signal between LBNP and rest . In the functional connectivity analysis , we performed a dual regression to both , hypothalamus and lower brainstem . Ultimately , we calculated a non-parametric paired t-test thresholded at p<0 . 05 using family-wise error ( FWE ) correction with threshold-free cluster enhancement ( TFCE ) to identify functional connectivity changes between LBNP and rest . Components showing significant changes in both metrics , that is spectral and functional connectivity , were assumed to be involved in cardiovascular regulation . We overlaid the atlas of Mai et al . , 2016 , which is in MNI standard space , to our components and their connectivity changes in order to identify the involved nuclei . To avoid identifying only familiar nuclei , we report every major nucleus from the atlas overlapping with any voxel of a functional connectivity cluster . Furthermore , to ascertain that our results were not driven by physiological noise , we conducted two supplementary analyses . Firstly , we repeated our initial analysis adding an additional physiological noise correction . Specifically , we applied slice-wise regression of cardiac and respiratory influences ( first order , one interaction term ) using FSL PNM ( Brooks et al . , 2008 ) before carrying out the masked ICA . We then matched the independent components to the ones of the original analysis by Hungarian sorting of the spatial cross-correlation matrix , and calculated their functional connectivity changes during LBNP . Since physiological noise regression influences frequency content in a non-linear manner , we did not test for spectral changes in this supplementary analysis . Secondly , using dual regression and a non-parametric one sample t-test thresholded at p<0 . 05 ( FWE corrected ) , we calculated functional connectivity of the identified hypothalamic regions with the whole brain ( smoothed with a Gaussian kernel of 5 mm full width at half maximum ) . This was done to make sure that their functional connectivity was mainly in the grey matter and not in the ventricles or other regions , whose signals are driven by physiological noise .	Stand up too fast and you know what happens next . You will feel faint as the blood rushes away from your head . Gravity pulls the blood into your legs , and your blood pressure drops . To correct this imbalance , the brain sends nerve impulses telling the heart to beat faster and the outer blood vessels to tighten . This is the autonomic nervous system at work . It is how the brain adjusts cardiac output , and quietly controls other internal organs in the body . It involves two key regions of the brain , the hypothalamus and the brainstem , and stimulates smooth muscles and glands around the body . The cardiovascular system also responds to the demands of exercise , with the heart supplying fresh blood laden with oxygen and the blood clearing out waste materials as it flows around the body . Perhaps surprisingly , blood pressure and heart rate fluctuate even at rest . The heart beats faster when breathing in and slower when breathing out . People’s blood pressure , the force that keeps blood moving through arteries , also oscillates in so-called Mayer waves that last about 10 seconds . Much of the current understanding of the inner workings of the cardiovascular system – and how it is regulated by the brain – stems from animal experiments . This is because few attempts have been made to simultaneously measure how a person’s brain and cardiovascular system work with enough detail to see how brain waves and cardiac oscillations might interact . To achieve this , Manuel et al . have now measured the brain activity , pulse and blood pressure of twenty-two healthy people while they were lying down in an MRI machine . This revealed that three distinct parts of the hypothalamus regulate cardiovascular output in humans . These ‘subsystems’ communicate with each other and with the lower brainstem , which sits beneath the hypothalamus . Manuel et al . also observed that the rhythmic activity of these subsystems runs in sync with oscillations typically seen in heart rate and blood pressure . With this work , Manuel et al . have shown that it is feasible to measure different systems of cardiovascular control in humans . In time , with further experiments using this new approach , the understanding of chronic high blood pressure and heart failure may improve .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "neuroscience" ]	2020	Deciphering the neural signature of human cardiovascular regulation
Proteins persist longer in the fossil record than DNA , but the longevity , survival mechanisms and substrates remain contested . Here , we demonstrate the role of mineral binding in preserving the protein sequence in ostrich ( Struthionidae ) eggshell , including from the palaeontological sites of Laetoli ( 3 . 8 Ma ) and Olduvai Gorge ( 1 . 3 Ma ) in Tanzania . By tracking protein diagenesis back in time we find consistent patterns of preservation , demonstrating authenticity of the surviving sequences . Molecular dynamics simulations of struthiocalcin-1 and -2 , the dominant proteins within the eggshell , reveal that distinct domains bind to the mineral surface . It is the domain with the strongest calculated binding energy to the calcite surface that is selectively preserved . Thermal age calculations demonstrate that the Laetoli and Olduvai peptides are 50 times older than any previously authenticated sequence ( equivalent to ~16 Ma at a constant 10°C ) . Ancient protein and DNA sequences are revolutionising our understanding of the past , providing information on phylogeny , migration , evolution , domestication and extinction ( Hagelberg et al . , 2015; Cappellini et al . , 2014 ) . However , the absence of data from warm regions and deep time ( Wade , 2015 ) highlights the fragility of these biomolecules and has so far hampered our ability to answer fundamental evolutionary questions , such as resolving the phylogenetic tree of the genus Homo in Africa . The survival of proteins and DNA in tropical environments and in fossils that go back a few million years ( Ma ) is deemed extremely unlikely and therefore the impact of the 'biomolecular revolution' in palaeontology and palaeoanthropology has so far been relatively limited . Claims for exceptional preservation in the fossil record have been put forward in a number of studies ( Towe and Urbanek , 1972; Bertazzo et al . , 2015; Schweitzer et al . , 2013; Cleland et al . , 2015 ) , but these have not been satisfactorily substantiated . Morphological ( Towe and Urbanek , 1972; Bertazzo et al . , 2015 ) , immunological ( Schweitzer et al . , 2013 ) and spectroscopic ( Bertazzo et al . , 2015 ) evidence of preserved tissues in dinosaurs and other fossils seems to be inconsistent with the observed levels of hydrolysis , dehydration and racemization ( Penkman et al . , 2013 ) in intracrystalline proteins from the fossil mollusc shell ( Sykes et al . , 1995 ) and eggshell ( Brooks et al . , 1990 ) . The mechanisms that might allow preservation over palaeontological and geological time scales are also poorly understood: crosslinking , organo-metallic complexing , including with iron , compression/confinement ( Logan et al . , 1991; Schweitzer et al . , 2014 ) , and mineral stabilization ( Collins et al . , 2000 ) have all been proposed as mechanisms that enhance the survival of ancient biomolecules . A confounding factor when evaluating the authenticity and antiquity of biomolecular sequences is the geographic area of provenance of the fossils and therefore the combined effect of time and temperature on the extent of degradation . Here we have used kinetic estimates of degradation rates of DNA ( Allentoft et al . , 2012 ) , collagen in bones ( Buckley et al . , 2008 ) , and intracrystalline amino acids ( Crisp et al . , 2013 ) to normalize their numerical ( chronological ) ages to thermal age ( Wehmiller , 1977 ) ( Figure 1 , Figure 1—source data 1 , Appendix 1 ) . Thermal age is a measure which enables simple comparison between ancient biomolecular targets by normalising them to an equivalent ( thermal ) age , allowing all samples to be treated as having experienced a constant temperature of 10°C . Thus samples from cooler sites , which experience slower rates of chemical reaction , will have thermal ages younger than their geochronological age , whilst samples from warmer sites will be thermally ‘older’ . Various factors can affect the effective diagenetic temperature experienced by a sample ( and therefore impact on its thermal age ) , from burial depth to seasonal and interglacial / glacial cycles ( Wehmiller , 1977; Huang et al . , 2000; Eischeid et al . , 1995 ) . The greatest absolute ages for recovered DNA ( Orlando et al . , 2013 ) ( 0 . 7 Ma = 0 . 08 Ma@10°C ) and for protein ( Rybczynski , 2013 ) ( 3 . 5 Ma = 0 . 3 Ma@10°C ) sequences are from high latitudes and their survival is consistent with predictions from the kinetic data . Younger samples from more temperate latitudes will have greater thermal ages , yet the oldest of these which has preserved protein ( Weybourne Crag: 1 . 5 Ma = 0 . 2 Ma@10°C ) has a thermal age similar to that of Middle Pleistocene DNA at Sima de los Huesos ( 0 . 4 Ma = 0 . 2 Ma@10°C ) ( Meyer et al . , 2014 ) . 10 . 7554/eLife . 17092 . 003Figure 1 . Eggshell peptide sequences from Africa have thermal ages two orders of magnitude older than those reported for DNA or bone collagen . ( A ) Sites reporting the oldest DNA and collagen sequences are from high latitude sites compared to ostrich eggshell samples from sites in Africa illustrated in ( B ) for which the current mean annual air temperatures are much higher . ( C ) Kinetic estimates of rates of decay for DNA ( Lindahl and Nyberg , 1972 ) , collagen ( Buckley and Collins , 2011 ) and ostrich eggshell proteins ( Crisp et al . , 2013 ) were used to estimate thermal age assuming a constant 10°C ( Figure 1—source data 1; Appendix 1 . For Elands Bay Cave and Pinnacle Point the oldest samples are shown ) . Note log scale on the z-axis: struthiocalcin-1 peptide survival is two orders of magnitude greater than any previously reported sequence , offering scope for the survival of peptide sequences into deep time . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 00310 . 7554/eLife . 17092 . 004Figure 1—source data 1 . Data and calculations for thermal ages reported in Figure 1 and in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 004 Here we explore the impact of strong protein binding in biominerals and its effect on sequence survival , by targeting ancient ostrich eggshell ( Struthio camelus; Struthioniformes ) , which is abundant in archaeological and palaeontological sites throughout Africa ( Materials and methods ) . Our aim was to elucidate a mechanism of preservation and to set out a rigorous methodology for the authentication of ancient protein sequences . We isolated and characterised the intracrystalline proteins ( Figure 2 , Figure 2—figure supplement 1 , Appendix 2 ) and tracked their diagenesis back in time to 3 . 8 Ma ago . Using a systematic approach , we validated the sequences from each of the eggshell samples analysed using amino acid racemization ( AAR ) , organic volatile compounds , ancient DNA and proteomic analyses . All our results are supported by in-depth analysis of patterns of diagenesis in both samples and blanks as well as the evaluation of potential contamination and carry-over . 10 . 7554/eLife . 17092 . 005Figure 2 . Proteome persistence and patterns of degradation . ( A ) Amino acid racemization in ostrich eggshell up to 3 . 8 Ma old from sites in South Africa and Tanzania . ( B ) Linear increase of THAA Val D/L values with the log of thermal age . ( C ) Exponential decrease of the number of identified MS/MS spectra with age ( THAA Val D/L ) . ( D ) The average hydropathicity of the peptides identified remains stable up to Val D/Ls ~1 . Note that Val D/Ls > 1 . 0 are unexpected and may be due to decomposition processes occurring in the closed system . The intracrystalline proteome composition in modern eggshell does not vary across microstructural layers ( Figure 2—figure supplement 1 ) , but patterns of degradation are different between fossil samples and purified proteins degraded at high temperature in the absence of the mineral ( Figure 2—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 00510 . 7554/eLife . 17092 . 006Figure 2—figure supplement 1 . Structure and composition of OES . ( A ) modern ( left ) and fossil ( LOT 13898; right ) OES: crystalline ( 1 ) , prismatic ( 2 ) , cone ( 3 ) and organic ( 4 ) layers . ( B ) comparison of total THAA yields in each layer before and after bleaching . ( C ) comparison between the composition of bleached eggshell powder from the cone , palisade and crystalline layers . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 00610 . 7554/eLife . 17092 . 007Figure 2—figure supplement 2 . Proteome degradation . ( A–B ) fossil OES: ( A ) number of unique proteins; ( B ) mean peptide length ( excluding contaminants ) . ( C–E ) degradation of purified proteins in water: ( C ) number of unique proteins identified; ( D ) number of identified product ion spectra; ( E ) mean peptide length; ( F ) average hydropathicity . No peptides were detected in the 120 hr heated sample . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 007 Twenty-four eggshell samples were sourced from well-dated sites in South Africa and Tanzania ( Figure 1 , Table 1 ) : Elands Bay Cave ( 0 . 3–16 ka BP , Table 3 ) , Pinnacle Point Caves PP 5/6 and PP 30 ( 50–80 ka BP and ~150 ka BP , respectively , Table 4 ) , Wonderwerk Cave ( 1 Ma , Table 5 ) , Olduvai Gorge ( 1 . 34 Ma , Table 6 ) and Laetoli ( 2 . 6–4 . 3 Ma , Table 7 ) . The age and stratigraphy of the oldest fossils , from Laetoli , is well-constrained despite the eggshell fragments being surface finds: their morphology shows no evidence of long-distance transport and the fossil-bearing horizons are well identified within the stratigraphy . The absence of lava flows in stratigraphic proximity also excludes the possibility that these fragments had been exposed to additional heat . The provenance of the fragments from Olduvai Gorge is also secure , as these were found in situ during the excavation of the Bell Korongo site , which overlies directly a volcanic tuff dated by 40Ar/39Ar ( Domínguez-Rodrigo et al . , 2013 ) . 10 . 7554/eLife . 17092 . 008Table 1 . Summary of archaeological and paleontological eggshell samples analysed in this study . See also Tables 3–7 . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 008SiteApproximate age range ( ka ) Approximate thermal age range ( ka@10°C ) Number of specimensElands Bay Cave0 . 3–160 . 5–458Pinnacle Point 5/650–150120–4708Wonderwerk~1000~36001Olduvai~1340~163004Laetoli2600–43008900–20400310 . 7554/eLife . 17092 . 009Table 2 . Binding of proteins to the ( 10 . 4 ) calcite surface . The binding energies calculated as ( a ) mean for the full protein ( by minimization , see Appendix 3 ) ; ( b ) for four individual domains within the proteins ( by molecular dynamics [ovocleidin]; by minimization [struthiocalcin] ) ; ( c ) for the four domains treated as peptides ( by molecular dynamics , see Appendix 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 009OC-17SCA-1SCA-2Charge on the protein+7 ( balanced by Cl- ) −11 ( balanced by Na+ , Ca2+ ) −10 ( balanced byCa2+ ) Binding energy ( mean ) : kJ mol−1−197 ± 22−142 ± 33Binding energy ( domains ) : kJ mol−1−422 ± 43−423 ± 42 ( YHHGEEEEDVWI ) −611 ± 44 ( YSALDDDDYPKG ) −255 ± 72 ( SDSEEEAGEEVW ) −231 ± 68 ( ASIHSEEEHQAIV ) Binding energy ( peptides only ) kJ mol−1−142 ± 19 ( YHHGEEEEDVWI ) −219 ± 24 ( YSALDDDDYPKG ) −131 ± 32 ( SDSEEEAGEEVW ) −122 ± 41 ( ASIHSEEEHQAIV ) Water molecules displaced21 . 320 . 223 . 1Estimated binding free energy: kJ mol−1−188 ± 37−159 ± 24−99 ± 39Residence times for water ( ps ) [average surface bound water molecule = 120 ps]130 ± 3 ( YHHGEEEEDVWI ) 135 ± 3 ( YSALDDDDYPKG ) 124 ± 4 ( SDSEEEAGEEVW ) 123 ± 5 ( ASIHSEEEHQAIV ) 10 . 7554/eLife . 17092 . 010Table 3 . Summary of samples from Elands Bay Cave , South Africa . The stratigraphic layers have been independently dated by radiocarbon . Unpublished uncalibrated dates provided by J . Parkington . Date calibration was performed with OxCal v . 4 . 2 ( Ramsey , 2009 . Calibration curves: IntCal13 for dates obtained on charcoal; Marine13 for dates obtained on shells/crayfish , DeltaR = 93 ± 28 [Dewar et al . , 2012] ) . Age estimates for undated layers based on estimating the median ( mid-point ) of two dates obtained on layers bracketing the layer with OES samples . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 010LOTNEaarLayerAge ( cal BP ) 95 . 4%Material used for 14C dating/notes18686887Kaunda<323 ( estimate ) Layer above dates on layer NKOM18726888George Best322–1008Layer between dates on layers NKOM and EDDI18666889D . Lamour906–2282Layer between dates on layers EDDI and LARM18496891Maroon Robson8773 ± 125Charcoal18506893Nero8748–10096Layer between dates on layers Maroon Robson / Burnt Robeson18236896Crayfish11545 ± 441Crayfish18196899Smoke12589 ± 104Charcoal18406907OBS 115208–15940Layer between dates on layers Smoke and SOSE10 . 7554/eLife . 17092 . 011Table 4 . Sample details for sub-fossil OES analysed for LC-MS/MS; from Pinnacle Point , South Africa . Stratigraphic information and weighted mean OSL age estimates ( ka ) for PP 5–6 ( Karkanas et al . , 2015 ) and PP 30 ( Rector and Reed , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 011SiteLOTNEaarArchaeological sample informationStratigraphic aggregateAge ( ka ) PP5-646137676Plotted Find 102627 , Lot 3151RBSR51 ± 2PP5-646497283Plotted Find 165702 , Lot 8038SGS64 ± 3PP5-646717316Specimen 273467 , Lot 3255SADBS71 ± 3PP5-646057198Specimen 273489 , Lot 3277SADBS71 ± 3PP5-646527286Plotted Find 178331 , Lot 8172ALBS72 ± 3PP5-646757320Specimen 273514 , Lot 7980LBSR81 ± 4PP 3046837328Specimen 66008 , Lot 1795Single horizon~151PP 3046977342Specimen 65168 , Lot 1750Single horizon~15110 . 7554/eLife . 17092 . 012Table 5 . Sample details for sub-fossil OES samples from Wonderwerk Cave , South Africa . Ages based on cosmogenic isotope burial dating and magnetostratigraphy , from Matmon et al . ( 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 012LOTNEaarStratumIndependent age ( Ma ) 1442610581ME46 , SPF#4390 , Exc . 1 , stratum 10 , square Q33 , depth 15–20 cm1 . 07–0 . 9910 . 7554/eLife . 17092 . 013Table 6 . Sample details for fossil OES samples from Olduvai , Tanzania . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 013LOTNEaarLocality/StratumIndependent age ( Ma ) 1557510955Sample BK09-31501 . 338 ± 0 . 0241557810958Sample BK10-53091 . 338 ± 0 . 0241557910959Sample BK09-26271 . 338 ± 0 . 0241558210962Sample BK09-27061 . 338 ± 0 . 02410 . 7554/eLife . 17092 . 014Table 7 . Sample details for fossil OES samples from Laetoli , Tanzania . Ages of the strata and localities ( 40Ar/39Ar ) from Deino ( 2011 ) . LOT 13901 is attributed to Struthio camelus . LOTs 13902 and 13898 are attributed to Struthio kakesiensis ( Harrison and Msuya , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 014LOTNEaarLocality/StratumIndependent age ( Ma ) 1390110574Loc 15 , Upper Ndolanya Beds~2 . 661390210573Loc 10 West , Upper Laetolil Beds~3 . 8–3 . 851389810575Kakesio 1−6 , Lower Laetolil Beds~3 . 85 -> 4 . 3 The chronological ages of the samples were normalised to thermal ages: the mean annual air temperature ( MAT ) for each site was estimated from the NOAA NCDC GCPS monthly weather station ( Eischeid et al . , 1995; Karl et al . , 1990 ) and borehole data ( Huang et al . , 2000; National Climatic Data Center ( NCDC ) , 2012 ) ( Appendix 1—table 1 ) . Samples on the surface or buried at shallow depth will have experienced an effective temperature which is higher than the MAT , as rates of reaction scale exponentially with temperature ( Wehmiller , 1977 ) . The greater the seasonal range at the site , the older the thermal age will be , but the effect of seasonal fluctuations will be mitigated by burial depth , which dampens temperature changes . Holocene sites which today have a MAT of exactly 10°C will have been cooler in the past 500 years due to recent anthropogenic warming . In this study , we used borehole temperature estimates ( Huang et al . , 2000 ) or long-term historic records ( Eischeid et al . , 1995 ) to counter this effect . Pre-Holocene samples from sites which today have an MAT of 10°C will have an even younger thermal age due to the reduction in temperature during glacial periods . This retards the rate of chemical degradation , and therefore slows the advance of thermal age . While we did not correct for seasonal fluctuation , a correction was applied for altitude . The long-term temperature model of Hansen et al . ( 2013 ) , scaled to local or regional estimates of present day values and predicted temperature decline at the last glacial maximum ( LGM ) , was used to project MATs from present day to the time of deposition ( Appendix 1—tables 1 , 2 , 4 ) . Ostrich eggshell protein degradation was compared with the extent of degradation of DNA and bone collagen detected in a variety of Northern Hemisphere sites ( Figure 1 ) . Published kinetic parameters for the degradation of the molecules ( Appendix 1—table 3; [Crisp et al . , 2013; Lindahl and Nyberg , 1972; Holmes et al . , 2005] ) were used to calculate the relative rate difference over a given interval of the long-term temperature record and to quantify the offset from the reference temperature of 10°C , thus estimating the thermal age in years@10°C for each sample ( Figure 1C ) . It is clear that Northern Hemisphere samples are thermally younger than their chronological age ( e . g . Ellesmere Island is ~0 . 02 Ma@10°C ) , while the age of the eggshell samples considered here increases , e . g . the 3 . 8 Ma sample from Laetoli and the 1 . 34 Ma Olduvai samples are estimated to have thermal ages of ~16 Ma@10°C ( Appendix 1—table 4; Figure 1—source data 1 ) . The difference in chronological age between our two oldest sites is therefore minimised by the effect of temperature , which is dampened in Laetoli due to the greater altitude relative to Olduvai . This sample set , spanning the last ~16 Ma@10°C ( Table 1 ) , was chosen in order to explore patterns of diagenesis and protein survival using a well-established experimental approach that can isolate the intracrystalline fraction of proteins enclosed in biominerals , including ostrich eggshell ( bleaching; Crisp et al . , 2013 ) . The intracrystalline fraction of avian eggshell typically contains C-lectins; in ostrich these are struthiocalcin 1 & 2 ( SCA-1 & SCA-2 ) ( Mann and Siedler , 2004 ) . The eggshell proteins were characterised in terms of their amino acid composition across microstructural layers ( Appendix 2 , sections A and B ) and the main proteins sequenced and identified ( Appendix 2 , section C ) in modern eggshell , revealing uniform composition across the eggshell layers . Therefore , samples of the archaeological and paleontological eggshell , usually recovered in a fragmentary state , can be considered to be representative of the overall proteome . The crystallography of SCA-1 ( Ruiz-Arellano et al . , 2015 ) reveals a similar overall structure to ovocleidin-17 ( OC-17 ) , which has previously been proposed to play a catalytic role in the calcification of chicken eggshell via the positively charged cluster of arginine residues interacting with the carbonates on the ( 10 . 4 ) calcite surface ( Freeman et al . , 2010 ) . OC-17 is , however , absent in ostrich; instead , SCA-1 and 2 are negatively charged ( Table 2 ) and thus likely to bind to calcium ions . A molecular dynamics ( MD ) study of the binding of whole SCA molecules at the mineral surface allowed the strongest binding regions of SCA-1 and SCA-2 to be identified , two for each of the two proteins ( Appendix 3 ) . In MD simulations the four peptide sequences that cover the binding regions ( Table 2 ) were moved close to the ( 10 . 4 ) calcite surface from aqueous ( bulk ) solution to determine their respective binding energies ( Appendix 3 ) . All four peptides showed negative binding energies , indicating it was energetically favourable for them to bind to the calcite surface , rather than to remain in solution . SCA-1 bound more strongly than SCA-2 and the binding energies for all four peptides had the same relative order as in the simulations with full proteins . This indicates that the peptides operate as effective proxies for the binding of SCA . The differences between bindings of the different peptides are probably due to the individual amino acids and the primary structure of the peptide enabling favoured binding configurations . When a molecule binds at the surface there will also be changes in entropy - an entropy loss for the molecule as it becomes bound and an entropy gain as water molecules on the surface are displaced . Given only one molecule binds , compared to the displacement of multiple water molecules , this will be an entropically favourable process . We have previously estimated the entropy associated with the water molecules and use this as a correction to the internal energy to estimate the free energy of binding ( Freeman and Harding , 2014 ) . These estimated free energy values ( including the influence of water displacement ) demonstrate the same trends as the configurational energy , since the number of water molecules displaced in all cases is similar . The structure of the water close to the interface is also more ordered than bulk water and has lower energy . Thus , when hydrolysis of the bound protein or peptide occurs , it must react with the stabilized water at the interface , not the water in the bulk . This will raise the barrier to hydrolysis and thus promote the survival of the sequence ( see also the discussion below ) . We would therefore expect that the stronger the peptide binding , the more likely the sequence is to survive in the fossil record , as it is best stabilized by its interactions with the mineral surface and must react with stabilized water . The MD simulations thus predict that the YSALDDDDYPKG sequence , with the lowest binding energy ( Table 2 ) will survive the longest . For fossil eggshell samples the extent of degradation , quantified by chiral amino acid analysis ( AAR ) , shows that both hydrolysis and racemization increase with time , and that racemic equilibrium is reached in samples older than 1 Ma ( ~3 . 6 Ma@10°C; Figure 2; and Appendix 4—table 1 and 2 ) . As degradation proceeds , the complexity of the proteome decreases , until only SCA-1 and SCA-2 are detected by LC-MS/MS in the oldest samples ( Supplementary file 1 ) . These two proteins are extremely well preserved in samples up to 150 , 000 years old , but by 3 . 8 Ma few peptides are recovered ( Figure 3 , Figure 3—figure supplement 1 and 2 ) . A total of 22 peptide sequences was recovered from SCA-1 & SCA-2 in samples from Laetoli ( Appendix 5 , section A; Supplementary file 2 ) , consistent with the idea that dehydration , in addition to mineral binding , may also play a role in retarding degradation of non-binding peptides ( Collins and Riley , 2000 ) . However , 80% of the spectra , consistently identified in ten independent LC-MS/MS analyses of three ostrich eggshell samples from Laetoli were assigned to charged species that contained the four Asp residues found in the peptide YSALDDDDYPKG . The survival of this Asp-rich peptide region is not limited to the samples from Laetoli; the eggshells from Olduvai ( ~16 Ma@10°C ) and Wonderwerk ( ~3 . 6 Ma@10°C ) also show that this region of SCA-1 is preferentially preserved . 10 . 7554/eLife . 17092 . 015Figure 3 . Survival of struthiocalcin 1 and struthiocalcin 2 peptides . Over time ( and increasing THAA Val D/L values ) the spectral count decreases as degradation progresses . Blue bars highlight the peptides investigated computationally ( represented by the filled atoms in the models ) . Highly degraded samples ( Val D/L ~0 . 8–1 . 1 ) preserve the DDDD-containing peptide . The full time series is shown for SCA-1 in Figure 3—figure supplement 1 and for SCA-2 in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 01510 . 7554/eLife . 17092 . 016Figure 3—figure supplement 1 . Frequency of identified spectra of SCA-1 in bleached OES ( fossils ) and purified proteins ( kinetics ) . Spectral count scale: 0–400 for fossil OES; 0–200 for kinetics . Sample 4605 has been recognized as burnt ( Crisp , 2013 ) but excellent sequence preservation is observed . Low spectral counts for sample 4613 are likely due to sample preparation , as AAR did not identify this sample as problematic . Coloured bars show protein structure . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 01610 . 7554/eLife . 17092 . 017Figure 3—figure supplement 2 . Frequency of identified spectra of SCA-2 in bleached OES ( fossils ) and purified proteins ( kinetics ) . Spectral count scale: 0–400 for fossil OES; 0–200 for kinetics . Sample 4605 has been recognized as burnt ( Crisp 2013 ) but excellent sequence preservation is observed . No SCA-2 peptides were found for sample 4613; this is likely due to sample preparation . Wonderwerk and Laetoli samples yielded some peptide sequence , but not consistently . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 017 This peptide does not survive in the absence of the mineral , as shown by the artificial degradation experiments we conducted on purified eggshell proteins heated at 140°C in water ( Figure 2—figure supplement 2 , Appendix 4 , sections B and C ) . Indeed , the same region of the protein was too flexible to be determined when the crystallographic structure of pure SCA-1 was solved ( Ruiz-Arellano et al . , 2015 ) . The patterns of degradation of the same proteins heated in water vs after demineralisation of the eggshell mineral also differ significantly from each other . It is noteworthy that the hydropathicity calculated for the surviving peptides decreases in eggshell mineral and increases in water ( Figure 2 and Figure 2—figure supplement 2 ) , consistent with the hypothesis that mineral binding plays a crucial role in the survival of selected peptide sequences . The authenticity of the peptide sequences recovered in the oldest samples was thoroughly assessed ( Appendix 5 ) . The amino acid concentration was analysed in all bleached eggshell samples and controls ( procedural blanks ) ; concentrations in the blanks were negligible , while the samples retain the original organic fraction within the intracrystalline environment ( Figure 4A ) . In addition , the presence of volatile organic compounds in 2 . 7 Ma ostrich eggshell demonstrates the stability of ostrich eggshell as a closed system ( Appendix 5 , Section E ) . Ratite eggshell has previously proven to be an excellent source of ancient DNA ( Oskam et al . , 2010 ) but , unsurprisingly , NGS sequencing failed to recover avian DNA from the Laetoli eggshell we tested ( Appendix 5 , Section F ) . Water blanks were injected between each LC-MS/MS eggshell sample analysis to assess carry-over ( Figure 4B–F ) . Despite low levels of SCA-1 being occasionally detected in some of the blanks ( Figure 4D ) , the effective carry-over from sample to sample can be estimated to be below 0 . 01% . We also stress that each batch of fossil eggshell was analysed separately in time ( Figure 4C ) and that therefore carry-over between younger and older eggshell samples is impossible . Finally , independent analyses of the results in a second laboratory ( Copenhagen ) also demonstrated the replicability of our results ( Appendix 5 , Section A ) . All peptides and proteins detected in this study presented damage patterns ( i . e . diagenesis-induced modifications , such as deamidation , oxidation ) that are entirely consistent with the age of the samples ( Appendix 5 , Section D; Supplementary file 3 ) . 10 . 7554/eLife . 17092 . 018Figure 4 . Authenticity of the ancient sequences . Amino acid analyses ( A ) : Total concentrations in all eggshell samples ( sum of Asx , Glx , Gly , Ala , Val and Ile ) . Carry-over: ( B ) Total ion chromatogram for one eggshell sample ( EBC_1823 ) and the blank analysed immediately after ( blank_post_EBC1823 ) . ( C ) Spectral abundance of SCA-1 and SCA-2 in LC-MS/MS blanks . ( D ) SCA-1 coverage in the blank analysed after a Pinnacle Point eggshell sample PP_4652 . Note 'DDDD-' and 'EEEED-' peptides and Asn deamidation . ( E ) Extracted ion chromatogram for LDDDDYPK in EBC_1823 , blank_post_EBC1823 and EBC_1819 . ( F ) Absolute and relative total abundance of 'DDDD' peptides in Laetoli samples/blanks . Signal reduction is at least 100-fold ( more often 1000- or 10 , 000-fold ) . Independent replication and manual de novo sequencing of the peptides from Laetoli ( Appendix 5 , section A; Supplementary file 2 ) , consistency of diagenesis-induced modifications ( Appendix 5 , section D; Supplementary file 3 ) and volatile organic compound analyses ( Appendix 5 , section E ) were also used to validate the results obtained . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 018 The breakdown of the proteins and peptides should primarily occur via hydrolysis , involving water and proteins or peptides as reactants . The rate-determining step is the attack of a water molecule ( or molecules , see Pan et al . , 2011 for an extended discussion ) . We schematically map out the pathway in Figure 5 ( red line ) where the reaction coordinate denotes approach of water to the peptide and their subsequent reaction . The process requires energy ( heat ) to be given to the system in order to overcome the energy barrier . In hot environments , such as Tanzania , the high ambient heat means that many interactions have sufficient energy to overcome this barrier , yet our experimental findings demonstrate that some peptides survive . 10 . 7554/eLife . 17092 . 019Figure 5 . Schematic diagram of energy barriers for peptide hydrolysis . A pictorial representation of the energy barriers associated with the lysis of the peptide . The process in bulk water is depicted in red and the process at the surface is depicted in blue . The surface process shows a larger barrier due to the stabilization of the reactants at the surface . DOI: http://dx . doi . org/10 . 7554/eLife . 17092 . 019 We argue that the mechanism allowing the survival of the ancient sequences over ~4 Ma ( ~16 Ma@10°C ) at equatorial sites is the stabilization of optimally configured peptides and associated water molecules by surface binding at this interface . The low , negative free energy of binding ( Table 2 ) of the amino acid residues means that they will readily bind to the calcite surface and remain bound indefinitely and this binding stabilizes the peptides by lowering their configurational energy ( Table 2 ) . Thus , both the position of the ground state and the top of the barrier will be lowered with respect to the situation when the peptide is in bulk water ( Figure 5 ) . The binding of the peptide also forces the hydrolysis reaction to take place with the stabilized water close to the calcite surface . Furthermore , the presence of the calcite surface significantly stabilizes the water molecules surrounding the peptide . Estimates of the residence times ( Table 2 ) and diffusion values of water molecules trapped between the protein and mineral surface indicate that these water molecules have greater residence times and lower diffusion rates than water molecules on the surface with no protein present . This large stabilization of water molecules selectively lowers the ground state energy of the reactants ( protein or peptide plus water ) at the interface with respect to the bulk . Thus the energy barrier will be significantly larger for the bound protein or peptide than for the unbound one . Our surface molecules would therefore need more energy in the system ( i . e . a higher temperature ) to overcome the augmented barrier . The net effect of the binding of the protein or peptide is therefore to retard hydrolysis and prolong peptide sequence survival , albeit of a select ( mineral-binding ) region of the protein . Translating this concept to real samples in geological settings , the burial temperature in Tanzania , which may be high enough to allow rapid hydrolytic breakdown of most proteins , would not be enough to hydrolyse mineral bound peptides over corresponding timescales because of their own stabilization but particularly because they are surrounded by the stabilized water . This is effectively equivalent to a localised 'cooling' effect: the water molecules at the calcite surface would therefore be expected to operate as if they were 'cooler' in terms of reactivity and rates of peptide bond hydrolysis . The survival of 3 . 8 Ma old peptide sequences in equatorial Africa corresponds to an estimated thermal age of ~16 Ma@10°C , two orders of magnitude beyond the oldest recovered DNA ( Figure 1 ) . We explain this exceptional preservation in terms of surface stabilization of both the peptide and water molecules involved in the hydrolytic breakdown of the peptide . Our discovery identifies mineral-binding proteins as the most likely source of ancient biomolecular sequences in the fossil record . In this study , we also set out parameters for the authentication of ancient sequences: the combination of the consistency in patterns of protein degradation and survival of particular peptide regions , independent replication of the results and an in-depth analysis of analytical blanks provide overwhelming evidence for the endogeneity and integrity of the peptides recovered . We suggest that all ancient proteomics studies undertake a similar approach to verify the authenticity of the sequences reported . We anticipate that this study will open up new avenues in palaeontology and palaeoanthropology , for the first time enabling direct comparison between morphological and molecular records of fossils in deep time . Furthermore , the selective preservation of domains associated with biomineralization offers a novel strategy for uncovering functional regions governing mineral formation . Ostrich eggshell samples were ground and bleached for 72 hr ( NaOCl 12% wt/vol ) and rinsed thoroughly before demineralization . Amino acid and mass spectrometry analysis of ancient proteins was conducted using published techniques ( Buckley et al . , 2009; Crisp et al . , 2013 ) . Modifications include the use of trypsin and elastase as digestion enzymes for separate preparations ( Welker et al . , 2015 ) . Liquid chromatography tandem mass spectrometry ( LC-MS/MS ) analyses were performed on Thermo Scientific Orbitrap platforms . Resulting spectra were searched against the Struthioniformes genomes using PEAKS ( version 7 . 5 [Ma et al . , 2003] ) . For PEAKS , FDR rate was set at 0 . 5% , with proteins accepted with −10lgp scores ≥ 40 and ALC ( % ) ≥ 80 . A combination of minimization and conventional MD using the DL_Poly Classic code was used to explore possible protein–calcite binding geometries ( Freeman et al . , 2011 ) . Elands Bay Cave ( EBC ) is located on the present coastline about 200 km north of Cape Town ( South Africa ) . Human occupation occurred repeatedly since the terminal Pleistocene . Ostrich eggshell is present throughout the sequence . The site’s chronology has been firmly established through multiple radiocarbon dates; the samples analysed in this study can each be assigned to an age range on the basis of dates obtained on each layer and/or bracketing the OES ( Parkington , 1980; Stowe and Sealy , 2016 ) . The caves at Pinnacle Point ( PP ) have been in the spotlight of archaeological research for the past few years , and they have yielded extraordinary evidence for early modern human behaviour as well as detailed palaeoclimatic information ( Karkanas et al . , 2015; Bar-Matthews et al . , 2010; Brown et al . , 2009; Marean , 2010; Marean et al . , 2007 ) . The OES fragments analysed in this study come from two sites in the PP complex , PP 5–6 and PP 30 , and were selected from excavations done up to 2010 . PP 5–6 is a well-dated sequence , spanning ca . 50–90 ka . PP 30 is a hyena den ( ~150 ka BP ) and the OES material reflects a single depositional episode . Wonderwerk Cave ( WW ) is located in the arid interior of South Africa , near the southern border of the Kalahari Desert . The site has yielded a unique ca . 2 million years long archaeological sequence ( Horwitz and Chazan , 2015; Berna et al . , 2012 ) . Stratum 10 , from which the OES samples analysed here are derived , bears the earliest evidence of intentional use of fire during the Acheulean , constrained to the Jaramillo subchron ( 1 . 07–0 . 99 Ma ) based on a combination of paleomagnetic and cosmogenic burial age dating ( Horwitz and Chazan , 2015; Berna et al . , 2012 ) . The OES fragments from the cave have been used as an effective proxy for refining palaeoclimatic and environmental reconstructions , especially for the early-mid Pleistocene and Holocene levels ( Ecker et al . , 2015; Lee-Thorp and Ecker , 2015 ) . Olduvai Gorge ( Tanzania ) contains an extensive record of the past two million years of human evolution . The eggshell samples analyzed in the present study were found in situ during the excavation of the BK ( Bell Korongo ) site located in uppermost Bed II . The site is exceptional by the large amount of ostrich eggshell fragments that were found throughout all its stratigraphic sequence . A volcanic tuff just underlying BK was recently dated to ~1 . 34 Ma ( Domínguez-Rodrigo et al . , 2013 ) . The samples analyzed were found in Level 4 , which contains a wealth of fossil bones and associated stone tools . This level has been interpreted as a central-place where the butchery of several animal carcasses ( including megafaunal remains from Sivatherium and Pelorovis ) was carried out ( Domínguez-Rodrigo et al . , 2014 ) . Laetoli ( Tanzania ) is one of the most famous sites for palaeoanthropologists: it has yielded hominin and other animal remains ( Harrison , 2011a , 2011b ) and the first unequivocal evidence for bipedalism thanks to the footprints of Australopithecus afarensis preserved in Pliocene volcanic ash , discovered by Mary Leakey in 1976 ( Leakey and Hay , 1979 ) . The eggshell at Laetoli are surface finds , but visual examinations show no evidence of rolling , transportation or weathering ( having been exposed on the surface for only a very short period of time after having eroded out of the sediment ) . As a consequence , there is no likelihood of long-distance transport . The location and preservation of the fossils , the absence of significant spatial displacement of surface finds , the short stratigraphic sections at each collecting spot , and the identification of the fossil-bearing horizons in each of those sections , allow the fossils to be placed quite precisely in their original stratigraphic context . The age and stratigraphy given for each of the samples can be assigned with a high degree of confidence . There are no lava flows in stratigraphic proximity or direct superposition to the stratigraphic units from which the Lower Laetolil and Upper Ndolanya specimens were recovered . Given that more than 40 m of consolidated sediment , and a time difference of 1 . 5 million years , separate the overlying lava flow ( the Ogol Lavas ) from the stratum from which the Upper Laetolil fossils were obtained , and that the intervening fossil-bearing beds show no geological evidence of having been impacted by heating , we do not believe that the samples have been exposed to additional heating that would have made them thermally older than we predict ( Table 7 ) . On crushing or demineralisation of the subfossil OES , a strong odour was emitted from some samples , so analysis of these volatiles was attempted using gas chromatography mass spectrometry ( GC-MS ) . A sealed container was designed that allowed in-line crushing of a sample under N2 . Volatiles emitted during crushing of the shells were measured using thermal desorption ( Unity , Markes International , Llantrisant , UK ) coupled to gas chromatography with a high-resolution quadrupole time of flight mass spectrometer ( 7200B GC/Q-TOFMS , Agilent Technologies , Wilmington , DE , USA ) . Volatile organic compounds ( VOCs ) were flushed from the shell crusher onto the trap using high purity nitrogen at 100 mL min−1 for 10 min . The trap was held at −30°C during sampling , then ballistically heated to 250°C and held for 5 min to ensure complete desorption . The heating of the trap triggered the start of the GC run . A 5% phenyl-polysilphenylene-siloxane capillary column was used ( 50 m × 0 . 32 mm × 1 mm BPX5 , SGE , Australia ) to separate the VOCs . The oven was held at 40°C for 5 min , followed by a ramp rate of 10°C min−1 to a final temperature of 230°C , which was held for 3 min . High purity helium gas was used as the mobile phase at a flow rate of 4 . 5 ml min−1 . The mass spectrometer was operated in an electron ionisation mode at 70 eV and the ion source was at 250°C . Spectra were collected between m/z 35 and 500 at an acquisition rate of 5 spectra s−1 . The mass spectra obtained were compared to the NIST MS database ( NIST MS Search Program version 2 . 0 ) after background correction . A nitrogen blank , sampled through the shell crusher was used to determine the method background and identify unique VOCs emitted from the egg shells . Analysis was undertaken on a fragment of one of the subfossil OES from Laetoli ( LOT 13901 , ~2 . 7 Ma ) . Two DNA extractions using ~0 . 05 g of eggshell ( sample Laetoli LOT 13902 ) were made following Dabney et al . ( 2013 ) and the extracts combined before the final elution in 25 μL TET . The data discussed in the paper are archived in the following databases: the mass spectrometry proteomics datasets have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD003786; Illumina genetic data have been deposited in the NCBI Short Read Archive ( SRA ) , BioProject ID PRJNA314978; computational modelling data can be found at DOI: 10 . 15131/shef . data . 3491387 ( this contains pdb files giving the initial configurations used for SCA-1 , SCA-2 and the four peptide sequences and input files for DL_POLY that contain a complete specification of the forcefield used and other setting parameters for the simulations ) .	The pattern of chemical reactions that break down the molecules that make our bodies is still fairly mysterious . Archaeologists and geologists hope that dead organisms ( or artefacts made from them ) might not decay entirely , leaving behind clues to their lives . We know that some molecules are more resistant than others; for example , fats are tough and survive for a long time while DNA is degraded very rapidly . Proteins , which are made of chains of smaller molecules called amino acids , are usually sturdier than DNA . Since the amino acid sequence of a protein reflects the DNA sequence that encodes it , proteins in fossils can help researchers to reconstruct how extinct organisms are related in cases where DNA cannot be retrieved . Time , temperature , burial environment and the chemistry of the fossil all influence how quickly a protein decays . However , it is not clear what mechanisms slow down decay so that full protein sequences can be preserved and identified after millions of years . As a result , it is difficult to know where to look for these ancient sequences . In the womb of ostriches , several proteins are responsible for assembling the minerals that make up the ostrich eggshell . These proteins become trapped tightly within the mineral crystals themselves . In this situation , proteins can potentially be preserved over geological time . Demarchi et al . have now studied 3 . 8 million-year-old eggshells found close to the equator and , despite the extent to which the samples have degraded , discovered fully preserved protein sequences . Using a computer simulation method called molecular dynamics , Demarchi et al . calculated that the protein sequences that are able to survive the longest are stabilized by strong binding to the surface of the mineral crystals . The authenticity of these sequences was tested thoroughly using a combination of several approaches that Demarchi et al . recommend using as a standard for ancient protein studies . Overall , it appears that biominerals are an excellent source of ancient protein sequences because mineral binding ensures survival . A systematic survey of fossil biominerals from different environments is now needed to assess whether these biomolecules have the potential to act as barcodes for interpreting the evolution of organisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "biochemistry", "and", "chemical", "biology" ]	2016	Protein sequences bound to mineral surfaces persist into deep time
Among various advantages , their small size makes model organisms preferred subjects of investigation . Yet , even in model systems detailed analysis of numerous developmental processes at cellular level is severely hampered by their scale . For instance , secondary growth of Arabidopsis hypocotyls creates a radial pattern of highly specialized tissues that comprises several thousand cells starting from a few dozen . This dynamic process is difficult to follow because of its scale and because it can only be investigated invasively , precluding comprehensive understanding of the cell proliferation , differentiation , and patterning events involved . To overcome such limitation , we established an automated quantitative histology approach . We acquired hypocotyl cross-sections from tiled high-resolution images and extracted their information content using custom high-throughput image processing and segmentation . Coupled with automated cell type recognition through machine learning , we could establish a cellular resolution atlas that reveals vascular morphodynamics during secondary growth , for example equidistant phloem pole formation . Model organisms have proven essential for dissecting the molecular-genetic control of biological processes in both animals and plants ( Meyerowitz , 2002; Brenner , 2009 ) . Typically , they have been chosen according to a number of criteria , including a small , diploid genome , a short generation time , and easy lab culture . Another frequent feature is their small size , which allows cultivation of numerous individuals to enable large-scale genetic analyses as well as easy observation of developmental processes by microscopy . Fulfilling all these criteria , Arabidopsis thaliana ( Arabidopsis ) , a small , annual dicotyledon of the Brassicaceae family , is the model of choice for developmental biology of higher plants ( Meyerowitz , 1989 ) . Various central processes of the plant life cycle , for example embryogenesis , root meristem organization or flower development can be examined at high spatio-temporal resolution in Arabidopsis . Moreover , in many instances live imaging at ( sub- ) cellular level is possible through microscopy techniques , including confocal microscopy , which is aided by the transparency of whole organs , such as the root , or at least the outermost tissue layers . However , such investigation is limited by organ depth , which can increase dramatically with organ size . For example , while the meristematic and differentiation regions of the root tip comprise a mere 5–6 dozen cells in the radial dimension and can be imaged all across using state-of-the-art microscopes , cell number rapidly increases proximal , towards the mature root ( Dolan et al . , 1993 ) . At the same time , the organization of the root tissue layers rearranges from a partially radial , partially bilateral symmetry towards full radial symmetry , concomitant with the formation of cylindrical secondary meristems and the replacement of the outer cell layers by a new protective outside tissue . Thus , eventually the mature root acquires the same overall organization as the mature aboveground stems , that is a few cell layers of protective tissue produced by an underlying cork cambium that surround the vascular tissues . The latter are produced by another cylindrical secondary meristem , the vascular cambium , which produces xylem tissues towards the inside and phloem tissues towards the outside ( Nieminen et al . , 2004; Groover and Robischon , 2006 ) . The activity of the cambial stem cells drives the radial expansion of roots and stems , a process termed ‘secondary growth’ . Formation of xylem tissues through secondary growth is the main process of durable biomass accumulation in plants and most prominent in tree trunks ( Groover and Robischon , 2006; Spicer and Groover , 2010 ) . In Arabidopsis , substantial secondary growth is not only observed at later stages of root development , but also in the hypocotyl , the embryonic stem ( Chaffey et al . , 2002; Sibout et al . , 2008 ) ( Figure 1A ) . Consistent with the hypocotyl’s role as critical junction between the root and shoot systems that limits the reciprocal transfer of edaphic resources and photosynthetic metabolites , its secondary growth occurs throughout most of the Arabidopsis life cycle and in some ways resembles the radial expansion of tree trunks . The hypocotyl initiates secondary growth shortly after seedling establishment , once its cell elongation growth along the main body axis has seized ( Sibout et al . , 2008; Ragni et al . , 2011 ) . Thus , unlike in post-embryonic stems , secondary growth in the hypocotyl is not obscured by parallel elongation growth , making it an ideal model system for this process . 10 . 7554/eLife . 01567 . 003Figure 1 . Cellular level analysis of Arabidopsis hypocotyl secondary growth . ( A ) Light microscopy of cross sections obtained from Arabidopsis hypocotyls ( organ position illustrated for a 9-day-old seedling , lower left ) at 9 dag ( upper left ) and 35 dag ( right ) . Size bars are 100 μm . Blue GUS staining due to the presence of an APL::GUS reporter gene in this Col-0 background line marks phloem bundles . ( B ) Overview of the developmental series ( time points and distinct samples per genotype ) analyzed in this study . ( C ) Example of a high-resolution hypocotyl section image assembled from 11 × 11 tiles . ( D ) The same image after pre-processing and binarization , and ( E ) subsequent segmentation using a watershed algorithm . ( F ) Number of mis-segmented cells as determined by careful visual inspection in 12 sections , plotted against the total number of cells per section ( log scale ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 003 Previous work has identified two principal phases of hypocotyl secondary growth , an early phase of proportional growth , when the cambium produces phloem and xylem tissues at roughly equal rates , and a later phase of xylem expansion , when the relative production of xylem dominates the radial expansion ( Chaffey et al . , 2002; Sibout et al . , 2008 ) . Early phase xylem consists mainly of the interconnected xylem vessels ( terminally differentiated , dead cells with perforated , thick cell walls that are the actual conducts for water and solutes ) and xylem parenchyma cells . Early phase phloem comprises the sieve elements ( interconnected , enucleated but alive cells that perform the actual transport of the phloem sap ) , companion cells ( which provide basic metabolism for sieve elements and are responsible for loading and un-loading of phloem sap cargo ) and phloem parenchyma cells . In the xylem expansion phase , both xylem and phloem also start to differentiate fibers ( cells with thick secondary cell walls that provide structural support ) , which can be formed from parenchymatic precursor cells . It has been shown that the transition between the early phase and the xylem expansion phase is triggered by the onset of flowering ( Sibout et al . , 2008 ) , through a mobile shoot-derived signal , the plant hormone gibberellin ( Ragni et al . , 2011 ) . In these studies , the transition had been defined as a shift in the relative occupancy of overall xylem vs overall phloem tissue in hypocotyl cross sections . However , as only the overall areas of combined xylem and phloem tissues were considered , it remains unclear what the transition represents at the cellular level . Various scenarios could be envisioned , for instance the relative expansion of xylem might be a consequence of increased post-cambial proliferation during xylem differentiation , or of increased cambial stem cell activity toward the xylem side , or the inverse with respect to phloem . To distinguish between these possibilities proved to be very difficult due to the absence of information about the cellular dynamics during the secondary growth process . Moreover , such investigation is severely hampered by the fact that this process is not amenable to live imaging and can only be monitored invasively , through histological cross sections , thereby killing the individual sample under investigation . Thus , a quantitative understanding of the temporal progression of secondary growth can only be acquired by a high-throughput approach that monitors enough cross-sections from distinct hypocotyls of the same age to provide statistically solid data . In conjunction with the large number and morphological diversity of the cells that constitute this tissue , a quantitative understanding of the cell proliferation , differentiation , and patterning events by conventional means , that is simple visual inspection of cross sections , is out of reach . Therefore , we established an automated histology approach to create a cellular resolution atlas that reveals the vascular morphodynamics during hypocotyl secondary growth . Our data reveal substantially different secondary growth dynamics in two genotypes as well as emerging patterns of cell orientation over time and a constantly equidistant production of phloem poles by the cambium . Based on the secondary growth progression observed during pilot experiments , we chose to analyze five time points in detail , starting at 15 days after germination ( dag ) , when a full cambium is established and the initial outer epidermal and cortex cell layers are already or about to be shed . This was followed by additional sampling at 20 , 25 , 30 and up to 35 dag , when the plants had seized formation of new flowers ( Figure 1B ) . Plants were grown in soil in a 16 hr light–8 hr dark cycle at 22°C with 150 µE light intensity . To minimize variation due to environmental conditions and between experiments , all plants were grown in parallel in a randomized design . In our conditions , all plants of both genotypes flowered at 17 dag ±1 d . For each time point , 50 seedlings were initially planted with the goal to eventually harvest 40 hypocotyls , which were fixed and embedded for sectioning . Embedding was performed using plastic resin , which proved to be the only robust method to acquire 3 µm thin cross-sections while conserving the cellular structure . A first observation by light microscopy after toluidine blue staining confirmed the integrity of the samples and allowed a first rough analysis of secondary growth progression based on overall transverse area ( excluding any remaining epidermal or cortex layers ) and the proportion occupied by the xylem . Whereas the average hypocotyl stele diameter was ca . 0 . 3 mm in Col-0 and ca . 0 . 15 mm in Ler at 15 dag , the radial expansion resulted in an average diameter of 1 . 6 mm in Col-0 and 1 . 1 mm in Ler at 35 dag . Concomitantly , relative xylem area increased from 12% to 29% in Col-0 , and from 31% to 47% in Ler , confirming previous observations ( Ragni et al . , 2011 ) . To obtain accurate quantitative parameters of secondary growth progression , we implemented a segmentation procedure to extract the cellular features from the cross sections . To allow reliable identification of small cells , such as cambial cells , with standard segmentation software , we obtained images of the cross sections with a light microscope at 40 X magnification . Our strategy was to produce ultra-high resolution images of 1024 × 1024 pixels , which would allow a very fine discrimination of individual cell boundaries , the critical requirement for the subsequent image segmentation process . Because the resolution was too high to fit any single cross section into a single image , we used the tiling function of the microscope to fuse 1024 × 1024 pixel subpanels into single images for each cross section . Individual cross section images subjected to segmentation were thus assembled from a minimum of 9 ( 3 × 3 ) up to 144 ( 12 × 12 ) panels ( Figure 1C ) . This procedure permitted information extraction from the whole section without inference or data loss . To this end , we developed a custom , fully automated image processing and segmentation pipeline . This pipeline pre-processes the images ( gamma correction , contrast and brightness adjustment ) and discards noise pixels after binarization ( Figure 1D ) before segmentation using a watershed algorithm ( Figure 1E ) . The pipeline is fully automated and robust and typically performed at more than 99% accuracy ( i . e . , less than 1% of mis-segmented cells ) across the scale of images ( Figure 1F ) . However , because CPU time scaled exponentially with image size , taking ca . 8 min . for a 15 dag sample but ca . 1000 min for a 35 dag sample , computation eventually became limiting for our endeavor . Thus , we restricted our analysis to ca . 20 selected cross sections per genotype and time point ( i . e . , 208 cross sections in total , requiring ca . 800 hr of total CPU time ) ( Figure 1B ) , which gave statistically robust quantitative data . Overall median cell number at 15 dag was 883 for Col-0 and 260 for Ler . At 35 dag , it had increased to 18’124 and 11’026 , respectively , indicating higher overall secondary growth in Col-0 , but higher relative secondary growth in Ler ( i . e . , a ca . 42-fold vs a ca . 21-fold increase in cell number ) . Together with the overall increase in total transverse area ( from ca . 70’000 µm2 at 15 dag to ca . 2 million µm2 at 35 dag and ca . 11’000 µm2 to ca . 1 million µm2 for Col-0 and Ler , respectively ) , this suggests significantly different secondary growth dynamics in the two genotypes . However , these overall averages can be misleading because of the already observed differences in relative tissue abundance . Thus , we advanced towards our goal of a full cellular resolution analysis by computing 16 cellular descriptors that represent the geometric characteristics of cell shape and relative cell position ( Supplementary file 1A ) . The initial set of descriptors was extracted from the segmented images using the EBImage R package ( Pau et al . , 2010 ) . This toolbox computes morphological features by calculating the 2nd-order covariance matrix of the image moments , which is equivalent to fitting an ellipsoid to an object . From these data , we computed additional features , including the position of the cells given by their polar coordinates and the cell incline angle ( see below ) , thereby taking full advantage of the cylindrical morphology of the hypocotyl cross sections . Although the descriptors provided an overview of the cell sizes , shapes and positions within the sections , they did not provide a straightforward indication of the tissue that individual cells belong to . To overcome this limitation , we sought to develop automated cell type recognition that uses the descriptors as an input for cell type classification . To this end , we performed a supervised classification using the support vector machine algorithm ( SVM ) ( Cortes and Vapnik , 1995 ) . Briefly , the SVM classifier principle is to find the optimal decision boundary between classes by maximizing the margin hyperplanes ( the geometrical representation of the decision boundaries in multi-dimension ) between the support vectors . The training set was a subset of our data that comprised a total of 3’144 manually labeled cells , dispatched into two sections per time point and genotype ( Supplementary file 1B ) . This set was split into a learning set comprising two-thirds of the data , and a test set constituted from the remaining cells . The former subset was used to build the classifier whereas the latter was employed for validation . The performance was assessed using the V-fold cross validation method , which consists of five randomly permutated reiterations of training and test sets to maximize the test set prediction error rate . Feature selection is a well-known pivotal issue in machine learning , and indeed the best combination of descriptors was critical in automated cell type classification and varied with the time point and genotype analyzed , mainly because cell type-specific position can vary with the age of the section . Thus , we developed a greedy algorithm for feature selection based on the 16 initial descriptors . This allowed us to select descriptors according to their importance in classifier performance ( Figure 2—figure supplement 1 ) , such that we could build one optimized classifier with respect to a given time point . In general , we selected the combination with the least number of descriptors , the lowest variation and the highest cross-validation performance with respect to the training/test set permutations . Finally , another key criterion in classifier selection was to minimize performance trade-off across different cell types , that is classifiers that scored high in correct recognition of all the different cell types ( the selected classifiers are described in Supplementary file 1C ) . Across all sections and time points , a common set of five distinct cell type categories ( Supplementary file 1D ) could be classified and quantified , that is ( i ) xylem vessels and parenchymatic cells , ( ii ) xylem fibers , ( iii ) cambial cells , ( iv ) phloem bundle cells ( companion cells and sieve elements ) and ( v ) parenchymatic phloem cells ( including any of the rare phloem fibers ) ( Supplementary file 1E ) . Although more categories ( e . g . , xylem vessels and parenchyma cells separately ) could be reliably distinguished at individual time points using other classifiers , we restricted our analyses to these five for the sake of a coherent temporal description of secondary growth progression . For these categories , our purely morphology- and position-based approach identified cells with an average accuracy of 88% and a median accuracy of 95% across the n = 50 cell type category X time point X genotype matrix . Whereas the automated recognition with these classifiers was thus sufficiently accurate for most cell type categories to extract quantitative data about secondary growth progression from the typically thousands of cells per section , the recognition of the xylem vessel and parenchyma cells behaved as an outlier , with lower accuracy especially at later stages . We also noticed that xylem cell types were frequently assigned to cells outside of the xylem area’s average radius . This was particularly prevalent at the later stages of development and could be pinpointed to a frequent confusion between xylem vessels and phloem parenchyma cells , which increasingly resemble each other in their outlines as secondary growth proceeds . However , discarding the problematic sections based on stringent criteria would have meant the exclusion of 33% and 40% of sections for Col-0 and Ler , respectively . To tackle these problems , we developed an automated pipeline for quality control . This procedure was based on manually created mask images that specified both the xylem area and the whole section area of the 208 samples ( Supplementary files 2 and 3 ) . Segmentation of the mask images allowed us to filter out noisy objects outside the sections’ average radius distance , mostly mis-segmented objects that either represented dirt contaminations or shed epidermal or cortex cells . The tool also automatically corrected the mis-assigned xylem cell identities by taking advantage of the mask-defined xylem area of the quality control filter . This correction refined the cell type recognition results and permitted all sections to pass the filtering step . An overview of the entire computational pipeline is shown in Figure 2A . 10 . 7554/eLife . 01567 . 004Figure 2 . The ‘Quantitative Histology’ approach . ( A ) Overview of the computational pipeline from image acquisition to analysis . ( B ) ‘Phenoprints’ for the different genotypes and developmental stages . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 00410 . 7554/eLife . 01567 . 005Figure 2—figure supplement 1 . An example of classifier selection through V-fold cross validation . The green arrow points out the selected feature combination according to the criteria of minimum number of features with the highest performance and the lowest variation ( the radiusV feature was excluded due to its putative variation in tissue location ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 005 For a first overview of secondary growth progression , we used the thus extracted cellular data to define phenotypic profiles ( phenoprints ) for each time point and genotype , comprised of the global ( e . g . , cross-section size or total cell count ) and cell type-specific ( e . g . , relative proportion of a particular cell type category or feature distribution ) statistics ( Figure 2B ) ( the feature description data for all cells of all sections is provided in data files 1 and 2 , the corresponding normalized data used for machine learning and the determined cell type identities are provided in the data files 3 and 4 , all available in the Dryad data repository under doi: 10 . 5061/dryad . b835k ( Sankar et al . , 2014 ) ) . The phenoprints consisted of a set of eight multi-parametric descriptors , which was informative for the normalized values ( Supplementary file 4 ) that were used to perform a principal component analysis ( Figure 3A ) . The computed correlation matrix was projected into a two-dimensional coordinate system , with the first two principal components explaining 76% of the variation . The first component opposed the larger phenoprint stages ( 30–35 dag in both genotypes ) with the smallest ( Ler 15d ) , with proportionally less cambium in the older stages . The second component associated variables of large phloem proportion and inexistent or low fiber content ( Col-0 15 dag , Ler 25 dag , Col-0 20 dag , Col-0 25 dag ) . The analysis also revealed larger angle spans for Ler as compared to Col-0 above all between 15 dag and 25 dag , suggesting substantial morphological changes during the early stages . At later time points , the two genotypes increasingly clustered together , indicating an initially slower development in Ler that however eventually caught up with Col-0 . Overall , the phenoprint clustering suggests a conserved sequence of development from one distinct morphological pattern to another , albeit with a different temporal progression in Col-0 vs Ler . 10 . 7554/eLife . 01567 . 006Figure 3 . Progression of tissue proliferation . ( A ) Principal component analysis ( PCA ) of the phenoprints shown in Figure 2B , performed with normalized values ( Supplementary file 4 ) . The inlay screeplot displays the proportion of total variation explained by each principal component . ( B–E ) Comparative plots of parameter progression in the two genotypes . In ( D ) , xylem represents combined vessel , parenchyma , and fiber cells , phloem represents combined phloem parenchyma and bundle cells . Error bars indicate standard error . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 006 Previous studies ( Ragni et al . , 2011 ) have shown that Ler has a higher ratio of xylem area to phloem area than most other accessions , including Col-0 . Our quantification also confirmed that overall radial expansion of Ler was reduced as compared to Col-0 ( Figure 3B ) . However , xylem area expansion rate was nearly equal in both genotypes , which combined with lower overall radial expansion necessarily resulted in higher xylem occupancy in Ler ( Figure 3C ) . In the temporal trend , two distinct phases of xylem occupancy could be distinguished . Initially , it decreased or remained stable between 15 and 25 dag , followed by an increase between 25 and 35 dag . Whereas these tendencies were similar in both genotypes up to 30 dag , Ler differed in that its xylem area increased steadily , eventually occupying almost 50% of the total transverse area at 35 dag . Quantification of cell proliferation confirmed that the number of xylem cells and the xylem cell proliferation rate were close in both genotypes ( Figure 3D ) , however the total number of cells in Ler was ca . twofold lower than in Col-0 ( Figure 3E ) . Moreover , the phloem proliferation rate was more than twofold lower in Ler , with stagnation in phloem cell number between 30 and 35 dag ( Figure 3D ) explaining the high xylem tissue occupancy at 35 dag . The increase of cambium cell number in Col-0 as compared to Ler at later stages of development ( Figure 3E ) likely contributed to this difference . In summary , our results suggest that a plateau in cambial growth combined with stagnating phloem proliferation is responsible for overall reduced radial growth but relatively increased xylem expansion in Ler . Of the descriptors extracted by our computational approach , the incline angle proved to be most useful in detecting and illustrating the substantial features of vascular organization during secondary growth progression . The incline angle represents the deviation of the major axis of a cell with respect to the radius emanating from the manually defined center point of the cross section ( Figure 4—figure supplement 1 ) . We calculated the incline angle θ ( in radians ) as follows:θ= \|arccos ( x·r‖x‖·‖r‖ ) −π2\|where x and r are vectors , corresponding to the major axis of the cell and the radius running from the cell center , respectively . A value of zero represents perfectly orthoradial ( i . e . , tangential or periclinal ) orientation of the major axis , and a value of π/2 represents perfectly radial ( i . e . , anticlinal ) orientation . Plotting the incline in combination with cell size created informative simplified visualizations of our cross sections ( Figure 4A–B ) . In these , concentric areas of cell orientation are evident , with a central area of mainly large and radially oriented ( high incline ) cells , representing the xylem cell categories . This area is surrounded by the cambium , depicted as a ring of small and orthoradially oriented ( low incline ) cells and , reaching the periphery , a zone comprising a bulk of mainly larger , orthoradially oriented cells representing the phloem area . Following the plots across the time points allowed us to reveal the vascular morphodynamics as a function of incline . 10 . 7554/eLife . 01567 . 007Figure 4 . Bimodal distribution of incline angle according to position . ( A and B ) Spatial distribution of cell incline angle illustrates the vascular organization in Ler ( B ) as compared to Col-0 ( A ) at later stages of development , for example 30 dag . The size of the disc increases with the area of the cell . Blue color indicates radial cell orientation , red orthoradial . ( C and D ) Violin plots of incline angle distribution , illustrating increasingly bimodal distribution coincident with refined vascular organization and different dynamics of the process in the two genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 00710 . 7554/eLife . 01567 . 008Figure 4—figure supplement 1 . An illustration of the incline angle . The incline is the angle between the section radius through the center of an ellipse fit to a cell and the major axis of that ellipse extended towards the x axis . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 008 Interestingly , the spatio-temporal dynamics of the overall incline ( i . e . , covering the whole section cell content at a given time point ) captured the distinct phases of secondary growth progression described above . This could be visualized in violin plots ( Figure 4C–D ) , where the incline angle was uniformly distributed at 15 dag in Col-0 ( Hartigans’ dip test p>10−3 ) , meaning that no distinguishable vascular organization of cell orientation was yet built up . Starting at 20 dag , a first peak towards lower values of incline emerged and persisted until 35 dag . At 30 dag , a second peak towards higher values of incline arose , giving shape to a discernable bimodal distribution ( Hartigans’ dip test p<2 . 2 × 10−6 ) ( Figure 4C ) . In Ler , the pattern was different in that a broad , slightly skewed distribution with a median value towards the lowest values of incline was observed at 15 dag , followed by a broad , slightly bimodal distribution at 20 dag ( Hartigans’ dip test p<10−4 ) ( Figure 4D ) . At the later time points , sharp bimodal-shaped density curves supported the coexistence of two populations of cells , a mostly radially and a mostly orthoradially oriented one ( Hartigans’ dip test p<2 . 2 × 10−6 ) , similar to Col-0 . Plotting the incline of individual cells according to their radial position ( i . e . , distance from a cross section’s center ) and over time points , we could follow the rearrangement in more detail . Normalization allowed us to pool the cells from all sections from a given time point and perform relative comparisons between them . Fitting these cloud distributions with locally weighted linear regression ( i . e . , lowess ) revealed the essential data trends ( Figure 5 ) . In Col-0 , the spatial distribution of the cell incline displayed unexpected temporal dynamics . At 15 dag , a wavy line described the point cloud , meaning that a radial vs orthoradial tissue boundary was not yet distinguishable ( Figure 5A ) . However , around 20 and 25 dag , vascular organization emerged as a plateau of largely radial orientation close to the center that corresponded to xylem cells , followed by a steep decrease to lower incline values in the cambium and phloem tissues ( Figure 5C , E ) . Once this pattern was established , the plateau of xylem enlarged while the span of the orthoradial cell layers narrowed , concomitant with the occurrence of xylem fibers and expansion of the xylem area ( Figure 5G , I ) . We also observed a decrease in the variation spread of incline in cambial cells over time . This reflected the progressive enlargement and organization of the cambium , which appeared to be completed as late as 30 dag , confirming continuous refinement of vascular patterning during secondary growth . A largely similar pattern of events was observed in Ler ( Figure 5B , D , F , H , J ) , however , the final organization appeared more bimodal than in Col-0 , which might reflect the above described decline of relative phloem area size . 10 . 7554/eLife . 01567 . 009Figure 5 . Distinct local organization of incline angle during hypocotyl secondary growth progression . ( A–J ) Density plots of cell incline angle vs radial position for the two genotypes at the indicated developmental stages , representing all cells across all sections for a given time point . The red lines represent the fit of these cloud distributions with locally weighted linear regression ( i . e . , lowess ) , revealing the essential data trends . All sections were normalized from 0 . 0 ( the manually defined center ) to 1 . 0 ( the average radius in a set of sections as determined by the average distance of the outermost cells from the center for individual sections ) . Box plots indicate the quartiles of the radian distribution for each cell-type class and are placed at the average position of the cell type with respect to the y axis . Outliers are shown as circles . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 00910 . 7554/eLife . 01567 . 010Figure 5—figure supplement 1 . Analysis of cell number in defined xylem regions of different size . Cell number in a circle of 200–500 pixels around the section centers for Col-0 . Cell count in a constant area of xylem over time across all averaged across all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 010 The distribution of inclines also had possible implications for the orientation of cell divisions , in the sense that mostly radial orientation could be an indicator for the prevalence of anticlinal division planes , whereas mostly orthoradial orientations could be an indicator for the prevalence of periclinal ones . Visual inspection of cross sections suggested that this is not the case however , also revealing a remarkable rarity of post-cambial cell divisions . In the xylem area , practically no post-cambial divisions were observed ( Figure 5—figure supplement 1 ) and radial cell files were generally continuous with the adjacent cambial files . Following such cell files also suggested that cellular growth led to a switch in xylem cell incline angle orientation . Whereas xylem cells that emerged from the cambium still retained the orthoradial orientation , cellular growth eventually resulted in a switch towards a radial orientation . Such switching was not observed in the phloem , consistent with the prevalence of orthoradial inclines . Similar to the xylem however , phloem cells were typically in continuity with the corresponding cambial cell files , and practically no cell divisions , neither anticlinal nor periclinal , were observed . Importantly however , this was only observed for files of phloem parenchyma cells . The exceptions to this were cell files that ended up in vascular bundles . In these , numerous post-cambial divisions could be observed , both in the anticlinal and periclinal orientations . Finally , as expected the vast majority of cell divisions was observed in the cambium . Mostly , they occurred in a perfect periclinal orientation , but we also observed numerous interspersed anticlinal divisions that are necessary to keep up with overall radial expansion . In summary , the radial expansion of hypocotyls appeared to be mostly driven by cambial activity and very little by post-cambial cell divisions . Since there appeared to be no cessation of cell division in the cell files connecting the cambium and the vascular bundles , the data suggested that the patterning of phloem pole position might already be laid down in the cambium . Although such patterning was not evident from visual inspection of phloem pole distribution , a density map representation of phloem bundle cells suggested a spatial pattern of phloem poles positioning around the central xylem ( Figure 6A ) . These density maps typically had limited resolution power around the cambial area , since newly born poles contain fewer bundle cells but are close in space , leading to a high and broad intensity region . For a more precise mapping of phloem pole positions , we thus analyzed 20 , 25 , and 30 dag sections obtained from transgenic Col-0 plants that expressed a beta-glucuronidase ( GUS ) reporter gene under the control of the phloem bundle-specific ALTERED PHLOEM DEVELOPMENT ( APL ) gene promoter ( Bonke et al . , 2003 ) ( Figure 1A ) . Along a concentric ring-shaped region of interest across the emerging phloem poles , the latter appeared as dark foci of GUS staining with higher pixel intensity . In image analyses , these were detectable as intensity spikes ( after noise reduction through the application of Gaussian blur , mainly to dampen background originating from the opacity of cell walls ) ( Figure 6B ) . Statistical analysis of the position of emerging phloem poles around the cambium revealed their spacing with a constant arc interspace distance . That is , the distance between emerging phloem poles remains constant over time as the cambial circumference enlarges . This was revealed by determination of the corresponding probability density function for the distance between the spikes by an automated Bayesian model ( Granqvist et al . , 2012 ) , which indicates a constant arc interspace distance ( Figure 6C ) with a span of ca . 140 μm , suggesting that vascular bundle formation is a patterned rather than a stochastic process . 10 . 7554/eLife . 01567 . 011Figure 6 . Mapping of phloem pole patterning . ( A ) Example of Gaussian kernel density estimate of the location of predicted phloem bundles cells in a 30 dag Col-0 section . High density represents phloem poles . ( B ) Example of an analysis of emerging phloem pole position in a 30 dag Col-0 section . The plot represents a pixel intensity map after noise reduction along a circular region of interest across the emerging phloem poles . Intensity peaks are due to GUS staining conferred to phloem bundles by an APL::GUS reporter construct . ( C ) Probability density function of the data shown in ( B ) obtained from an automated Bayesian model . The dominant single peak indicates a constant arc distance of ca . 62 pixel between the phloem poles . DOI: http://dx . doi . org/10 . 7554/eLife . 01567 . 011 The principal problems that we faced were the large range of cell sizes as well as the large number of objects within the hypocotyl radius . This required ultra high-resolution imaging of our cross sections as well as an automated segmentation procedure that would not require any seeding . The solution was the assembly of cross sections from tiled , partial high-resolution images and their segmentation through an automated pipeline that relied on a watershed algorithm . This pipeline achieved very good accuracy in object detection , but was still CPU intensive . In part , this could be off set by binarization of the images using an adaptive Gaussian filter , which greatly accelerated the segmentation procedure . We could compensate an associated decrease in segmentation quality ( because watershed segmentation is more accurate on gray scale images ) by effectuating morphological operations on the binarized images , thereby keeping segmentation accuracy high while automating the task . Extending our approach beyond simple cell counting to cell type recognition intrinsically hinged on supervised classification . To this end , we used the Support Vector Machine ( SVM ) method , because it had already proven its efficiency in a broad range of life science applications ( Noble , 2006 ) . Average prediction accuracy based on this method was generally high , however for some cell type categories it was more variable at times . This was due to the nature of the classifiers , which were chosen to optimize for overall accuracy including all cell type categories . Implementation of our quality control tool alleviated this effect , however it is noteworthy that even more accurate classifiers can be identified for analyses that focus on a given cell type or a given time point , extending the range of potential applications of our pipeline . The use of shape characteristics for cell classification was pioneered by Olson et al . , who classified mammalian culture cells into three groups using hierarchical cluster analysis and nearest neighbor analysis ( Olson et al . , 1980 ) . Recent improvements in this area largely benefit from SVM algorithm development , which can take multiple features into account . For instance , a recent study identified factors involved in the transition between cell shapes using automated phenotyping of human cell cultures that took advantage of fluorescent staining for DNA , tubulin and actin on top of cell morphology ( Fuchs et al . , 2010 ) . Conceptually similar , another study exploited cell shape in combination with fluorescent characteristics upon nuclear and cytoskeleton staining in Drosophila ( Yin et al . , 2013 ) . However , classification based solely on cell morphology has also been applied to human cells ( Theriault et al . , 2012 ) . Whereas all of these studies investigated isolated cells in culture , we had to apply morphology-based classification to cells that were embedded in their tissue and in a developmental context . While this complicated the analysis , it also offered the opportunity to assign spatial coordinates to the cells , which could be integrated on top of characteristics of cell geometry to build our classifiers . Average true prediction accuracy in the cited studies was in the range of 83–90% , as compared to 88% in our study . Notably however , our cell type assignment precision was greatly increased by our post-machine learning quality control pipeline , which enabled us to fix the principal classes with lower accuracy , due to frequent SVM confusion between xylem vessels and phloem parenchyma cells . Thereby , we successfully classified up to five cell type categories in a time course experiment where the number of cells ranged from a few hundred to several thousand . The factors that limited our approach were to some degree related to the properties of plant cells , notably that they are encapsulated by rigid cell walls that resist the internal turgor pressure . Their cellular geometry is therefore not only shaped by the material properties of the walls , but also by the permanent force of turgor pressure , manifesting in the reduced variation of cell shape in plants as compared to animals ( Theriault et al . , 2012 ) . To some degree , this uniformity in cell shape hampered the identification of certain cell states by our machine learning approach , for instance the direct identification of dividing cells . Similarly , certain cell types were ticklish to distinguish by their morphology only . For instance , we were not able to separate phloem companion cells from sieve elements or xylem parenchyma cells from xylem vessels across all time points , which therefore had to be grouped into combined categories . Adding tissue-related features , such as cell connectivity ( i . e . , the number of neighboring cells ) , and improving the segmentation algorithm such that cell wall thickness could be incorporated into the analyses might overcome these obstacles and greatly increase performance . Future efforts should go into this direction and could also boost the universal application of our approach . The latter should be possible for any tissue or organ from which cell outlines can be segmented after imaging and for which a reference point can be defined , for example ( partial ) sections from tree trunks or confocal images of root meristems . For the subsequent cellular level analysis , the incline angle descriptor of a cell proved to be particularly valuable . Whereas no temporal changes were discernible for the cell area and the cell eccentricity features , the cell incline distribution varied over time , in a seemingly non-random fashion . Indeed , combination with spatial components ( i . e . , radial cell position in cross sections ) revealed spatio-temporal rearrangement of inclines across a sequence of intertwined morphodynamic events . Our data indicate a gradual increase and arrangement of cambial cells , which together with orthoradial cellular organization of the surrounding tissues appeared to be a prerequisite for proper xylem development and relative xylem expansion around 20–25 dag . One possible explanation for this phenomenon could be tissue mechanics . The growing xylem area might exert a compression force on surrounding cambial and parenchymal cells , forcing them into tangential anisotropic cell elongation . How such mechanical stress is perceived and conveyed into cellular behavior is largely enigmatic and an emerging hot topic in plant biology , where first studies on shoot apical meristem formation have implicated katanins in the dynamic reorientation of microtubules perpendicular to stress direction ( Uyttewaal et al . , 2012 ) . Beyond possible mechanical constraints , molecular genetic patterning is clearly pivotal in vascular morphodynamics . For instance , polarity of the cambium to produce xylem to the inside and phloem to the outside is an inherent feature of secondary growth . A receptor-like kinase—peptide ligand pair is involved in this process and interacts with hormone signaling pathways ( Hirakawa et al . , 2008 , 2010; Etchells et al . , 2012 ) . Notably , the phenotypic penetrance of the respective mutants is background-dependent , with stronger effects in Ler than in Col-0 ( Etchells et al . , 2013 ) . It would be interesting to investigate whether this could result from an interaction with the earlier cessation of phloem production we observed in Ler . The early cessation of phloem production in Ler as compared to Col-0 does , however , not reflect an earlier termination of overall growth in Ler . Rather it appears that phloem production in Ler ceases before xylem production and contributes to the divergent growth dynamics in the two genotypes . The severely reduced overall cell production in Ler as compared to Col-0 can be mainly attributed to reduced phloem and cambium cell number , and is responsible for the higher relative proportion of xylem area that had been reported earlier ( Ragni et al . , 2011 ) . Interestingly , the nearly 50% reduction in overall cell number does not mean that growth is uniformly slower in Ler . Rather , initial secondary growth appears to be particularly slow in Ler as indicated by more than threefold difference in cell number at 15 dag . This is followed by an acceleration of cell production that surpasses Col-0 in relative terms between 15 dag and 25 dag , before dropping to Col-0 levels between 25 dag and 35 dag . This pattern is also evident from the principal component analysis , in which both Col-0 and Ler reach overall similar end points . Thus , our analysis along a series of time points has revealed highly divergent secondary growth dynamics in the genotypes that would not have been evident from a comparison of end points . Beyond the cellular dimensions , our quantitative histology approach also allowed us to conduct follow up analyses to reveal developmental patterns that were not evident from simple visual inspection . For instance , we found a constant arc interspace distance for phloem pole formation along the developmental time series with a concomitant decrease in the interspace angle due to the overall secondary growth . A reaction-diffusion model with a growing boundary ( i . e . , representing the expanding xylem area ) would be consistent with these results . Local production of the above-mentioned mobile ligand and activation of its receptor at a distance are potential candidates for such a mechanism . Alternatively , patterning cues from apical sources might direct phloem pole formation , for instance to coordinate it with phyllotaxy . Application of our quantitative histology approach to complementary stem sections could present one way to explore these possibilities . The Arabidopsis thaliana Col-0 , Ler or APL::GUS ( Bonke et al . , 2003 ) lines were grown in soil , in a 16 hr light–8 hr dark cycle mimicking long day conditions under white light of ca . 150 μE intensity . After harvest , hypocotyls were immediately fixed and embedded in Technovit plastic resin before toluidine blue staining as described ( Sibout et al . , 2008; Ragni et al . , 2011 ) . Sections of 3-μm thickness were then obtained using a Leica RM2255 microtome and were subsequently imaged on a Zeiss LSM 710 confocal microscope in transmitted light mode at 40x magnification using the automated tiling function . Hypocotyls from APL::GUS plants were subjected to GUS staining before fixation , embedding and sectioning as described ( Sibout et al . , 2008; Ragni et al . , 2011 ) and imaged using a Leica DM 5500 microscope . To extract information content of the sections at cellular resolution , we developed an automated image analysis pipeline . The pipeline was written in python with calls to R scripts and ImageJ macros . In brief , images were first pre-processed automatically ( i . e . , gamma correction , contrast , and brightness adjustment ) before their binarization . A series of morphological operations ( two times an erosion operation followed by a dilatation operation ) were applied with the aim to discard noisy pixels and regularize the cell boundaries . These steps were achieved using to the EBImage R package ( Pau et al . , 2010 ) . A variant watershed algorithm with automatic seeding ( http://bigwww . epfl . ch/sage/soft/watershed ) was used to identify the cell boundaries . Each cell was then characterized by a vector composed of 16 components that comprised 10 geometrical and 6 positional features ( Supplementary file 1A ) and was classified into one of the 5 cell-type classes ( Supplementary file 1B ) . One classifier was built per genotype and per time point ( Supplementary file 1C ) using the C-classification with a radial basis function ( RBF ) kernel of the support vector machine ( Cortes and Vapnik , 1995 ) provided by the e1071 package , the R interface for the libsvm library ( Chang and Lin , 2001 ) . The training set for the machine learning comprised 3144 manually labeled cells across 20 sections that covered all time points and genotypes . The optimal parameters , the selected features and the classifier accuracies are given in Supplementary file 1D . To compare secondary growth progression in the two genotypes , we described each developmental stage in a ‘phenoprint’ that represents a vector combined of 8 numerical values ( Figure 2B ) . For principal component analysis ( PCA ) , each observable was scaled with respect to the maximum value to obtain a unit range across variables ( Supplementary file 4 ) . We performed a PCA analysis by computing the eigenvalues and eigenvectors for the correlation matrix . The resulting two first principal components were displayed with a bi-plot representation . The rotation angle between the vector variables represents the correlation between two phenoprints ( >90° meaning no correlation ) . This method allowed direct quantitative comparison of the phenotypic variability of our samples . To automatically map phloem pole positions in sections obtained from APL::GUS plants , a circular region of interest ( ROI ) across the newly generated phloem bundles that was concentric with the section center was defined and GUS staining intensity was measured along the ROI using ImageJ software . For each image , the period between phloem poles was detected using an automated Bayesian model ( Granqvist et al . , 2012 ) , corresponding to the most likely occurring arc interspace distance between two phloem poles .	Our understanding of the living world has been advanced greatly by studies of ‘model organisms’ , such as mice , zebrafish , and fruit flies . Studying these creatures has been crucial to uncovering the genes that control how our bodies develop and grow , and also to discover the genetic basis of diseases such as cancer . Thale cress—or Arabidopsis thaliana to give its formal name—is the model organism of choice for many plant biologists . This tiny weed has been widely studied because it can complete its lifecycle , from seed to seed , in about 6 weeks , and because its relatively small genome simplifies the search for genes that control specific traits . However , as with other much-studied model systems , understanding the changes that underpin the development of some of the more complex tissues in Arabidopsis has been severely hampered by the shear number of cells involved . After it has emerged from the seed , the plant’s first stem will develop from a few dozen cells in width to several thousand cells with highly specialized tissues arranged in a complex pattern of concentric circles . Although this stem thickening process represents a major developmental change in many plants—from Arabidopsis to oak trees—it has been under-researched . This is partly because it involves so many different cells , and also because it can only be observed in thin sections cut out of the plant’s stem . Now Sankar , Nieminen , Ragni et al . have developed a novel approach , termed ‘automated quantitative histology’ , to overcome these problems . This strategy involves ‘teaching’ a computer to automatically recognize different plant cells and to measure their important features in high-resolution images of tissue sections . The resulting ‘map’ of the developing stem—which required over 800 hr of computing time to complete—reveals the changes to cells and tissues as they develop that allow the transport of water , sugars and nutrients between the above- and below-ground organs . Sankar , Nieminen , Ragni et al . suggest that their novel approach could , in the future , also be applied to study the development of other tissues and organisms , including animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2014	Automated quantitative histology reveals vascular morphodynamics during Arabidopsis hypocotyl secondary growth
Over half of breast-cancer-related deaths are due to recurrence 5 or more years after initial diagnosis and treatment . This latency suggests that a population of residual tumor cells can survive treatment and persist in a dormant state for many years . The role of the microenvironment in regulating the survival and proliferation of residual cells following therapy remains unexplored . Using a conditional mouse model for Her2-driven breast cancer , we identify interactions between residual tumor cells and their microenvironment as critical for promoting tumor recurrence . Her2 downregulation leads to an inflammatory program driven by TNFα/NFκB signaling , which promotes immune cell infiltration in regressing and residual tumors . The cytokine CCL5 is elevated following Her2 downregulation and remains high in residual tumors . CCL5 promotes tumor recurrence by recruiting CCR5-expressing macrophages , which may contribute to collagen deposition in residual tumors . Blocking this TNFα-CCL5-macrophage axis may be efficacious in preventing breast cancer recurrence . In 2018 , it is estimated that approximately 270 , 000 women will be diagnosed with breast cancer , and 41 , 000 women will succumb to the disease ( Siegel et al . , 2018 ) . Historically , over half of these deaths are due to recurrence 5 or more years after initial diagnosis and treatment ( Sosa et al . , 2014 ) . This suggests that in a subset of patients , there is a population of clinically undetectable residual tumor cells that survive therapy , and may serve as a reservoir for eventual relapse . The long latency of recurrence has led to speculation that residual tumor cells are slowly growing or even dormant ( Hölzel et al . , 2010; Klein , 2009 ) . Understanding how residual cells survive therapy , persist in a non-proliferative state , and eventually resume proliferation to form recurrent tumors is critical for preventing recurrences . Much of the work examining mechanisms of tumor cell survival and recurrence following therapy has focused on tumor cell-intrinsic pathways ( Sosa et al . , 2011 ) . Genetic mutations that render cells resistant to therapy represent an important mechanism of survival ( Holohan et al . , 2013 ) , but there is emerging evidence that non-genetic pathways can also promote survival in response to therapy . For instance , a population of cells called drug-tolerant persisters has been shown to survive therapy through epigenetic adaptations ( Sharma et al . , 2010 ) . Additionally , epithelial-to-mesenchymal transition has been shown to promote cell survival in response to EGFR inhibitors ( Sequist et al . , 2011 ) . Finally , alterations in apoptotic pathways within tumor cells can promote cell survival in response to both chemotherapy and targeted therapy ( Alvarez et al . , 2013; Damrauer et al . , 2018; Hata et al . , 2016; Holohan et al . , 2013; Mabe et al . , 2018 ) . In spite of this extensive literature on cell-intrinsic mechanisms of therapeutic resistance , much less is known about tumor cell-extrinsic contributions to cell survival following therapy . Specifically , while there has been some recent focus on how the tumor microenvironment can promote tumor cell survival in response to therapy ( Meads et al . , 2009 ) , little is known about whether the microenvironment regulates tumor cell survival , dormancy , and eventual recurrence . We used a conditional mouse model of Her2-driven breast cancer to examine interactions between tumor cells and their microenvironment during tumor dormancy and recurrence . In this model , administration of doxycycline ( dox ) to bitransgenic MMTV-rtTA;TetO-Her2/neu ( MTB;TAN ) mice leads to mammary gland-specific expression of epidermal growth factor receptor 2 ( Her2 ) and the development of Her2-driven tumors . Removal of dox induces Her2 downregulation and tumor regression . However , a small population of residual tumor cells can survive and persist in a non-proliferative state ( Alvarez et al . , 2013; Moody et al . , 2002 ) . These cells eventually re-initiate proliferation to form recurrent tumors that are independent of Her2 . Using this model , we sought to understand how the interplay between tumor cells and their microenvironment regulates residual cell survival and recurrence . To understand how interactions between tumor cells and their environment change in response to therapy , we first examined gene expression changes following Her2 downregulation in Her2-driven tumor cells . Two independent cell lines derived from primary Her2-driven tumors ( Alvarez et al . , 2013; Moody et al . , 2002 ) were cultured in the presence of dox to maintain Her2 expression , or removed from dox for 2 days to turn off Her2 expression . Changes in Her2 expression following dox withdrawal were confirmed by qPCR analysis ( Figure 1—figure supplement 1A ) . Changes in gene expression were measured by RNA sequencing . Her2 downregulation led to widespread changes in gene expression in both cell lines ( Figure 1A ) . Gene set enrichment analysis showed that an E2F signature was the most highly enriched gene set in cells with Her2 signaling on ( +dox; Figure 1—figure supplement 1B ) , consistent with previous literature and the observation that Her2 is required for the proliferation of these cells ( Lee et al . , 2000 ) . Interestingly , the gene sets most significantly enriched in cells following Her2 downregulation ( -dox ) were an inflammatory gene signature and a TNFα/NFκB gene signature ( Figure 1B ) . These gene sets comprised genes encoding chemokines in the CCL family ( CCL2 , CCL5 , and CCL20 ) and CXCL family ( CXCL1 , CXCL2 , CXCL3 , CXCL5 , and CXCL10 ) , proteins that mediate cell-cell interactions ( TLR2 , ICAM1 , and CSF1 ) as well as signaling components of the NFκB pathway ( NFΚBIA and NFΚBIE ) . All these genes were upregulated following Her2 downregulation ( Figure 1C ) . At high concentrations ( >40 μg/ml ) doxycycline itself can inhibit the NFκB pathway ( Alexander-Savino et al . , 2016; Santa-Cecília et al . , 2016 ) . Although the concentrations of dox ( 2 μg/ml ) we use to culture primary tumor cells are well below these levels , we wanted to confirmed that the NFκB pathway activation observed following dox withdrawal was due to loss of Her2 signaling . To do this , we treated primary tumor cells with Neratinib , a small-molecule inhibitor of Her2 , to inhibit Her2 signaling without removal of dox . Neratinib treatment led to an increase in phospho-p65 ( Figure 1—figure supplement 1C ) , increased expression of TNFα ( Figure 1—figure supplement 1D ) , and increased expression of the NFκB targets CXCL5 and CCL5 ( Figure 1—figure supplement 1E and F ) . To further confirm that the low concentrations of dox used to culture primary tumor cells do not directly inhibit the NFκB pathway we treated NIH3T3 cells with TNFα in the presence or absence of 2 μg/ml dox and measured NFκB target genes . Dox treatment had no effect on the induction of NFκB target genes following TNFα treatment ( Figure 1—figure supplement 1G ) . Taken together , these results demonstrate that Her2 inhibition leads to activation of the NFκB pathway . Given the coordinated upregulation of these NFκB target genes , we reasoned that their expression may be induced by a common upstream secreted factor acting in an autocrine manner . To test this , we collected conditioned media from primary tumor cells grown in the absence of dox for 2 days . This conditioned media was supplemented with dox to maintain Her2 expression and added to naive primary tumor cells . Treatment with conditioned media led to a time-dependent upregulation of the pro-inflammatory chemokine CCL5 ( Figure 1D ) . One common upstream mediator of this cytokine response is tumor necrosis factor alpha ( TNFα ) , and we found that TNFα expression is increased between 10-fold and 100-fold following Her2 downregulation ( Figure 1E ) . To test whether this is sufficient to activate downstream signaling pathways , we examined activation of the NFκB pathway following treatment with conditioned media from cells following Her2 downregulation . Indeed , we found that treatment of naive cells with Her2-off ( –dox ) conditioned media led to rapid , robust , and prolonged activation of the NFκB pathway as assessed by phosphorylation of p65 ( Figure 1F ) . Importantly , Her2 levels remained high in these target cells ( Figure 1—figure supplement 1H ) , indicating that Her2-off ( –dox ) conditioned media can activate the NFκB pathway even in the presence of Her2 signaling . In contrast , conditioned media from Her2-on ( +dox ) cells had no effect on p65 phosphorylation ( Figure 1—figure supplement 1I ) . Finally , we tested whether the induction of chemokine genes following Her2 downregulation was dependent upon the NFκB pathway by treating cells with the IKK inhibitor , IKK16 . We found that blocking IKK activity blunted the induction of all chemokine genes following dox withdrawal ( Figure 1G ) . Taken together , these results suggest that Her2 downregulation leads to the induction of a pro-inflammatory gene expression program , likely driven by autocrine-acting TNFα and mediated through the IKK-NFκB pathway . Her2 downregulation in Her2-driven tumors in vivo induces apoptosis and growth arrest , ultimately leading to tumor regression ( Moody et al . , 2002 ) . However , a small population of tumor cells can survive Her2 downregulation and persist for up to 6 months before resuming growth to form recurrent tumors . These residual tumors can be identified histologically ( Figure 2A ) . Many of the cytokines and chemokines induced shortly after Her2 downregulation function as chemoattractants for various immune cells ( Binnewies et al . , 2018; López et al . , 2017 ) . This led us to speculate that Her2 downregulation in vivo may promote infiltration of immune cells into the tumor . We therefore asked whether the immune cell composition of tumors changed during tumor regression and in residual tumors . CD45 staining showed that leukocyte infiltration increased dramatically following Her2 downregulation as compared to primary tumors ( Figure 2B–C , Figure 2—figure supplement 1A ) . Surprisingly , leukocytes remained high in residual tumors ( Figure 2D , Figure 2—figure supplement 1A ) . Masson’s trichrome staining revealed prominent collagen deposition in residual tumors ( Figure 2D ) , consistent with a desmoplastic response in residual tumors . Staining for the macrophage marker F4/80 showed a dramatic increase in macrophage abundance during tumor regression ( Figure 2C , Figure 2—figure supplement 1A ) , and macrophage levels remained elevated in residual tumors ( Figure 2D , Figure 2—figure supplement 1A ) . CD3 staining showed increased T cell infiltration in regressing and residual tumors ( Figure 2—figure supplement 1A , B ) . Taken together , these results indicate that Her2 downregulation leads to the infiltration of CD45+ leukocytes , and specifically F4/80+ macrophages . Residual tumors contain high numbers of macrophages and abundant collagen deposition , consistent with a desmoplastic response . Immune cells can influence tumor cell survival and function ( Flores-Borja et al . , 2016; Pollard , 2004 ) . The large number of immune cells present in residual tumors suggests that these cells may function to regulate the behavior of residual tumor cells . To begin to address this , we sought to identify secreted factors that are expressed in residual tumors . Residual tumor cells in the autochthonous MTB;TAN model are unlabeled and are diffusely scattered throughout the mammary gland , precluding their isolation . Therefore , we used an orthotopic model in which residual tumors can be easily isolated . In this model , primary Her2-driven tumors are digested , cultured , and infected with GFP . Cells are then injected into the mammary fat pad of recipient mice on dox to generate an orthotopic primary tumor . Following dox withdrawal , the fluorescently labeled residual tumors can be easily microdissected ( Figure 2—figure supplement 1C ) . We first confirmed that the orthotopic model exhibited similar patterns of immune cell infiltration as the autochthonous model . Indeed , we found that macrophage staining increased dramatically during tumor regression and in residual tumors ( Figure 2—figure supplement 1D–F ) , suggesting the orthotopic model is appropriate for identifying secreted proteins present in these residual tumors . We generated a cohort of orthotopic primary tumors ( n = 4 ) and residual tumors at 28 days ( n = 6 ) and 56 days ( n = 6 ) following dox withdrawal . Residual tumors were microdissected using a fluorescent dissecting microscope . We then made protein lysates from all samples and measured the expression of cytokines and chemokines using antibody-based protein arrays . Four primary tumors and four 28 day residual tumors were profiled using a commercially available cytokine array , which measures the expression of 20 secreted factors . We then used a second commercially available cytokine array , which measures 40 cytokines and chemokines , to measure cytokine expression in the whole cohort of tumors . This analysis identified eight cytokines that were upregulated in residual tumors as compared to primary tumors ( Figure 3A; fold change >2 , p < 0 . 1 , Figure 3—source data 1 ) , including CCL5 , osteoprotegerin ( OPG ) , and Vascular cell adhesion protein 1 ( VCAM-1 ) ( Figure 3B ) . Interestingly , VCAM-1 has been shown to regulate breast cancer dormancy ( Lu et al . , 2011 ) , while OPG can regulate the survival of breast cancer cells ( Neville-Webbe et al . , 2004 ) . We next asked whether any cytokines were both induced acutely following Her2 downregulation and remained elevated in residual tumors . We found that only two cytokines , CCL5 and OPG , fulfilled these criteria . Given that OPG has previously been associated with dormancy , we focused our attention on CCL5 . We then wanted to determine if CCL5 expression was elevated in human residual breast tumors following treatment . We analyzed a gene expression dataset of residual breast tumors that remain following neoadjuvant targeted therapy . A number of secreted factors were upregulated in residual tumors as compared to primary tumors , and CCL5 was one of the most significantly upregulated cytokines in this group ( Figure 3C–D and Figure 3—figure supplement 1A–M ) . To confirm these results , we examined an independent gene expression data set from breast cancer patients treated with neoadjuvant chemotherapy . We found that CCL5 expression was also increased in residual tumors in this dataset ( Figure 3—figure supplement 1N ) . These results suggest that CCL5 upregulation is a common feature of residual tumors cells that survive both conventional and targeted therapy in mice and humans , suggesting it may be functionally important in mediating the survival of these cells . We next wanted to directly assess whether CCL5 plays a functional role in regulating residual cell survival or recurrence . We first used an ELISA to measure CCL5 levels in orthotopic primary tumors , residual tumors , and recurrent tumors . CCL5 expression was elevated in residual tumors , confirming results from the cytokine array , and increased further in recurrent tumors ( Figure 4A ) . We next engineered primary tumor cells to overexpress CCL5 or GFP as a control ( Figure 4B ) and used these cells in an orthotopic recurrence assay to test the effect of CCL5 expression on tumor recurrence . Control or CCL5-expressing cells were injected orthotopically into recipient mice on doxycycline to maintain Her2 expression . Primary tumors formed with similar kinetics following injection of control and CCL5-expressing cells , indicating that CCL5 expression had no effect on the growth of primary tumors ( data not shown ) . Following primary tumor formation , mice were removed from dox to induce Her2 downregulation and tumor regression . Mice with residual tumors were palpated biweekly to monitor the formation of recurrent tumors . Tumors expressing CCL5 recurred significantly earlier than control tumors , indicating that CCL5 expression is sufficient to accelerate tumor recurrence ( Figure 4C; p = 0 . 023; HR = 2 . 14 ) . We next asked if tumor-derived CCL5 is necessary for recurrence . To this end , we used CRISPR-Cas9 to knock out CCL5 in primary tumor cells ( Figure 4D ) , and tested the effect of CCL5 knockout on recurrence using the orthotopic recurrence assay described above . The growth of CCL5 knockout tumors was not different from control tumors expressing a non-targeting sgRNA ( data not shown ) . Mice were removed from dox , and the latency of recurrence between control and CCL5 knockout tumors was compared . We found that CCL5 knockout had no effect on the latency of recurrence ( Figure 4E ) . Taken together , these results suggest that CCL5 expression is sufficient to accelerate recurrence , but tumor-derived CCL5 is not necessary for recurrence following Her2 downregulation . CCL5 is a chemoattractant for various cell types , including T cells , B cells , eosinophils , basophils , neutrophils , macrophages , and fibroblasts ( Dembic , 2015; Lacy , 2017; Lee et al . , 2017 ) . We observe an increase in CCL5 levels during tumor regression and in residual tumors that is concomitant with immune cell infiltration . We therefore reasoned that the effect of CCL5 overexpression on recurrence may be mediated through its ability to recruit one or more of these cell types to residual lesions and recurrent tumors . CCL5 can signal through multiple receptors , including CCR1 , CCR3 , and CCR5 , but it predominately acts through CCR5 ( Soria and Ben-Baruch , 2008 ) . We therefore examined CCR5 expression on various immune and stromal cells in primary tumors ( +dox ) , regressing tumors ( 5 days –dox ) , residual tumors ( 69 days –dox ) , and recurrent tumors by flow cytometry . As expected , Her2 was downregulated following dox withdrawal in all tumors ( Figure 5—figure supplement 1A ) . For each cell type , we measured the median fluorescence intensity ( MFI ) of CCR5 staining in CCR5+ cells . Interestingly , the level of CCR5 expressed on macrophages increased in residual tumors ( Figure 5A and Figure 5—figure supplement 2 ) . In contrast , CCR5 expression on CD4+ T cells CD8+ T cells increased in regressing tumors , but returned to baseline in residual tumors ( Figure 5B and C , Figure 5—figure supplement 2 ) . Similar to macrophages , the expression of CCR5 on fibroblasts was elevated in residual tumors ( Figure 5D , Figure 5—figure supplement 2 ) . We were also interested in examining CCR5 expression on CD45– tumor cells . We observed a slight increase in CCR5 expression in residual tumor cells , but otherwise there was no change in CCR5 expression on these cells ( Figure 5E ) . To directly compare the expression of CCR5 in macrophages and tumor cells , we sorted these two populations from primary , regressing , residual , and recurrent tumors from MTB;TAN mice and performed qPCR analysis . CCR5 was expressed at higher levels on macrophages than tumor cells at each stage , and its expression was especially high on residual tumor macrophages ( Figure 5—figure supplement 1B ) . Overall , these results identify several cell types – notably macrophages and fibroblasts – that express high levels of CCR5 and so are poised to respond to CCL5 in residual tumors . To determine whether these cell types are recruited by CCL5 in residual tumors , we generated primary and residual tumors overexpressing CCL5 and analyzed the abundance of macrophages and fibroblasts by flow cytometry . Fibroblast levels were not significantly different between control and CCL5-expressing tumors ( Figure 5F , Figure 5—figure supplement 1C ) . In contrast , CCL5-expressing tumors exhibited a modest but consistent increase in macrophage infiltration ( Figure 5G , Figure 5—figure supplement 1D ) . Taken together , these results suggest that CCL5 expression in residual tumors can recruit CCR5-positive macrophages , and suggest that CCL5 may subsequently signal through CCR5 on these cells to modulate macrophage function . We next considered the possibility that CCL5 recruitment of macrophages to residual tumors may promote recurrence through macrophage-tumor cell crosstalk . To address this , we sorted CD45+/CD11b+/F4/80+ macrophages from primary , residual and recurrent tumors from the autochthonous MTB;TAN model by fluorescence activated cell sorting ( FACS ) , and then isolated RNA from the sorted cell populations for RNAseq . Residual tumor-associated macrophages did not yield sufficient RNA for RNAseq , but we were able to sequence RNA from primary , regressing , and recurrent tumor-associated macrophages ( TAMs ) . Examination of differentially expressed genes between primary and recurrent TAMs suggested that FACS-sorted TAMs may have been partially contaminated with tumor cells . For instance , we detected Her2 expression at high levels in primary TAMs and low levels in recurrent TAMs . Therefore , we used a gene expression dataset of primary and recurrent tumor cells cultured in vitro to filter the TAM expression list ( Figure 6—source data 1 ) . After filtering , we were left with approximately 200 genes that were differentially expressed between primary and recurrent tumor macrophages ( Figure 6A , Figure 6—source data 2 ) . Interestingly , genes encoding fibrillar collagen and collagen deposition proteins were more highly expressed in the recurrent TAMs than the primary TAMs or regressing tumor TAMs ( Figure 6B ) . These genes include Collagen alpha-1 ( V ) chain ( COL5A1 ) , Collagen type XXIV alpha 1 ( COL24A1 ) , Procollagen C-endopeptidase enhancer 1 ( PCOLCE ) , and Asporin ( ASPN ) . COL5A1 and COL24A1 encode fibrillar collagens , PCOLCE encodes a glycoprotein that binds and drives the cleavage of type one fibrillar procollagen , and ASPN encodes a protein that binds to fibrillar collagens to regulate mineralization . We next sought to validate these findings by performing qPCR analysis on primary , regressing , residual , and recurrent TAMs . This analysis showed that the expression of these genes progressively increased during tumor regression , residual disease , and recurrence ( Figure 6C ) . Additionally , qPCR on RNA isolated from bulk tumors showed higher expression of COL5A1 and COL24A1 in recurrent tumors , while a subset of recurrent tumors had high expression of ASPN and PCOLCE ( Figure 6D ) . Consistent with this , Masson’s trichrome staining showed increased collagen deposition in residual and recurrent tumors ( Figure 6E , middle and bottom ) . In order to see if similar gene expression patterns are observed in residual disease in breast cancer patients , we examined gene expression data from residual tumors after neoadjuvant targeted therapy . Indeed , expression of these four collagen genes increased in residual tumors following therapy ( Figure 6—figure supplement 1A ) . Finally , we asked whether CCL5 regulates collagen deposition by comparing collagen levels in control and CCL5-expressing recurrent tumors . While control recurrent tumors had uniform levels of collagen deposition ( Figure 6F and Figure 6—figure supplement 1B–C ) , a subset of CCL5-expressing tumors had very high levels of collagen deposition ( Figure 6F and Figure 6—figure supplement 1B–C ) . Taken together , these results suggest that CCL5 promotes macrophage infiltration and collagen deposition . Given the importance of collagen for regulating tumor cell function , this may be one mechanism by which CCL5 expression accelerates recurrence . This is reminiscent of findings in colorectal cancer , where collagen deposition can be mediated in part through CCR2+ macrophages , and depletion of these macrophages inhibits tumor growth ( Afik et al . , 2016 ) . The long-term survival of residual tumor cells following therapy is a major obstacle to obtaining cures in breast cancer . Understanding the pathways that promote residual cell survival – and that induce the reactivation of these cells to generate recurrent tumors – is critical for designing therapies to prevent breast cancer relapse . There has been extensive focus on tumor cell-intrinsic pathways that allow cells to survive therapy ( Holohan et al . , 2013 ) . However , the role of tumor cell-extrinsic factors , including the tumor microenvironment , in regulating the survival and recurrence of residual cells has not been extensively explored . Here , we used a conditional mouse model to investigate how interactions between tumor cells and the tumor microenvironment change during tumor regression , residual disease , and recurrence , and in turn how the microenvironment regulates tumor recurrence . We found that Her2 downregulation led to induction of a pro-inflammatory gene expression program comprising a number of chemokines and cytokines , including CCL5 . This program was mediated by autocrine TNFα and dependent upon IKK/NFκB signaling . Notably , a recent study identified a similar gene expression program in EGFR-mutant lung cancer following treatment with EGFR inhibitors ( Gong et al . , 2018 ) . Consistent with this pro-inflammatory gene expression program , we observed differences in immune and stromal cell infiltration during tumor regression . Both adaptive ( CD4+ and CD8+ T cells ) and innate ( macrophages ) immune cells were recruited to regressing tumors . The residual tumor microenvironment is markedly different from that of primary tumors , with high numbers of macrophages and fibroblasts , abundant collagen deposition , and differential expression of a suite of cytokines , including CCL5 . Functionally , CCL5 overexpression promotes macrophage recruitment , collagen deposition , and promotes tumor recurrence . These results identify CCL5 as a critical regulator of crosstalk between residual tumor cells and the residual tumor microenvironment that promotes tumor recurrence . A number of studies have found that Her2 signaling directly activates the NFκB pathway , and that this is functionally important for tumor growth ( Liu et al . , 2009 ) . Consistent with this , we observed basal levels of p65 phosphorylation in primary tumor cells . Surprisingly , we found that Her2 inhibition further activates the NFκB pathway , and that this occurs through an autocrine pathway that is likely mediated by increased TNFα expression . Hyperactivation of the NFκB pathway in turn leads to the production of a number of cytokines and chemokines which may contribute to the recruitment of immune cells . These findings are consistent with prior work showing that the NFκB pathway is required for macrophage recruitment in a similar Her2-driven mouse model ( Liu et al . , 2010 ) . Our findings add to these previous studies by showing that Her2 inhibition leads to hyperactivation of the NFκB pathway and increased macrophage recruitment . CCL5 has been shown to play an important role in many facets of tumor progression , such as invasion , metastasis , neoangiogenesis , and immune cell infiltration ( Aldinucci and Colombatti , 2014 ) . In glioblastoma , CCL5 upregulation has been correlated with recurrence in post-treatment tumors ( Hudson et al . , 2018 ) . In triple-negative breast cancer , CCL5 expression has also been correlated with residual tumor size and tumor infiltrating lymphocytes after neoadjuvant chemotherapy ( Araujo et al . , 2018 ) . However , CCL5 has not previously been implicated in residual cell survival or recurrence in Her2+ or hormone receptor positive breast cancer . By analyzing gene expression datasets from breast cancer patients treated with neoadjuvant targeted or chemotherapy ( Creighton et al . , 2009; Stickeler et al . , 2011 ) , we show here that CCL5 expression is elevated in residual tumor cells that survive therapy . A notable observation in our study is that while CCL5 expression promoted recurrence ( Figure 4C ) , knockout of CCL5 in tumor cells did not delay recurrence ( Figure 4E ) . This suggests that CCL5 may be at least partially redundant with other chemokines , such as CCL2 and CXCL1 and 2 , in recruiting macrophages to promote recurrence . Mechanistically , we show that CCL5 acts to recruit CCR5+ macrophages to residual tumors , consistent with its known role as a chemoattractant factor for macrophages ( Mantovani et al . , 2017 ) . RNAseq analysis of primary and recurrent TAMs suggested that recurrent TAMs have high expression of genes encoding fibrillar collagen and proteins required for collagen deposition . qPCR analysis indicated that residual TAMs shared this gene expression program . Consistent with this , collagen deposition is high in residual and recurrent tumors , and CCL5 expression promotes collagen deposition . Collagen deposition is traditionally thought to be driven by fibroblasts in the microenvironment ( Thannickal , 2012 ) . However , a recent report showed that macrophages are responsible for collagen deposition in a mouse model of colorectal cancer ( Afik et al . , 2016 ) . Collagen deposition is important for tumor progression and invasiveness ( Provenzano et al . , 2008 ) . Collagen bundles can potentiate cell migration and increase tissue stiffness , and enzymes which crosslink collagens are often upregulated in breast cancer and are correlated with a poor prognosis ( Lu et al . , 2012 ) . It is possible that collagen deposition may promote the survival or proliferation of residual tumor cells , and that this mediates the effect of CCL5 on tumor recurrence . The findings reported here suggest that efforts to block CCL5-driven macrophage infiltration and subsequent collagen deposition may have therapeutic benefit . Possible therapies include the use of Maraviroc , a CCR5 antagonist ( Velasco-Velázquez et al . , 2012 ) , and agents that block macrophage infiltration or function , such as the CSF-1R inhibitor PLX3397 ( DeNardo et al . , 2011; Strachan et al . , 2013; Zhu et al . , 2014 ) . It is also possible that , because CCL5 is sufficient but not necessary for tumor recurrence , it would be preferable to block the induction of the pro-inflammatory program that is induced following Her2 downregulation using agents targeting TNFα or the NFκB pathway . It is important to note that while our studies focus on the function of CCL5 in recruiting CCR5+ macrophages , breast cancer cells themselves can also express CCR5 . Indeed , previous studies have found that CCR5 acts in tumor cells to promote stem cell expansion and metastasis in breast cancer ( Jiao et al . , 2018; Velasco-Velázquez et al . , 2012 ) . Although in the current study we find that in residual tumors CCR5 is expressed at higher levels in macrophages than on tumor cells , it is possible that tumor cell-expressed CCR5 may mediate at least some of the effects of CCL5 on tumor recurrence . Future work with mice lacking CCR5 on specific cell types will clarify the relative important of CCR5 on macrophages and tumor cells . The survival and recurrence of residual tumor cells is a critical clinical problem in breast cancer . The results identified here show that interactions between residual tumor cells and their microenvironment are critical for recurrent tumor formation . Targeting tumor cell-microenvironment interactions may hold promise for preventing recurrent breast cancer . Orthotopic tumor recurrence assays were performed as described ( Alvarez et al . , 2013 ) . Briefly , cohorts of 6-week-old recipient mice ( nu/nu or TAN ) on doxycycline were injected bilaterally in the #4 inguinal mammary fat pad with 1 × 106 primary tumor cells ( expressing either a control sgRNA , a sgRNA targeting CCL5 , CCL5 cDNA , or GFP cDNA ) . Once tumors reached 5 mm ( 2–3 weeks ) , doxycycline was removed to initiate oncogene down-regulation and tumor regression . Mice were palpated biweekly to monitor tumor recurrence , and sacrificed when recurrent tumors reached 10 mm . Differences in recurrence-free survival between control and experimental cohorts were compared using Kaplan-Meier survival curves ( Kaplan and Meier , 1958 ) and evaluated by the p-value from a log-rank test and the hazard ratio from the Cox proportional hazard regression , as described previously ( Alvarez et al . , 2013 ) . Power calculations were used to determine cohort size for each in vivo experiment . Briefly , in order to detect a 2 . 5-fold difference in recurrence-free survival between control and experimental groups , given a median recurrence-free survival of 60 days for the control group and a 300 day follow-up , we estimated we would need to enroll 22 tumors per group ( 80% power , p<0 . 05 ) . We enrolled extra mice in each cohort to account for tumor take rates and unexpected mortality . Final cohort sizes were: GFP tumors , 17 mice ( 34 tumors ) ; CCL5 tumors , 18 mice ( 36 tumors ) ; sgControl tumors , 20 mice ( 40 tumors ) ; sgCCL5 tumors , 20 mice ( 40 tumors ) . Cell lines derived from primary MTB;TAN tumors were grown as previously described in media containing 2 μg/ml dox ( Alvarez et al . , 2013 ) . For conditioned media experiments , primary tumor cell lines were plated on 10 cm plates . 24 hr later , media was changed to media without dox , and conditioned media was collected 1 or 2 days later . Media was centrifuged to remove cells , supplemented with 2 μg/ml dox , and applied to naive primary tumor cells . Cells treated with conditioned media were harvested 1 or 2 days later for qPCR or Western blot analysis . For dox withdrawal experiments , primary tumor cell lines were plated 10 cm plates . 24 hr later , media was changed to media without dox and cells were collected 1 or 2 days later for qPCR or western blot analysis . IKK16 ( Selleckchem , Houston , TX ) was used at 100 nM , TNFα ( BioLegend , San Diego , CA ) was used at 10 ng/ml . Primary cells derived from MTB;TAN tumors ( 54074 and 99142 cells ) were generated by our lab , are used at early passages , and as a result have not been authenticated . NIH3T3 cells were tested by the Duke Cell Culture Facility for mycoplasma contamination and tested negative . The facility was not able to perform STR authentication on these mouse cells . Tumors were harvested and digested as previously described ( Mabe et al . , 2018 ) . Cells were aliquoted at 1 × 106 cells per 5 mL falcon tube . CD16/CD32 Fc Block antibody was added for 10 min at 4°C ( 2 μL/1 × 106 cells ) . Tumors were then stained with antibody cocktails listed below for 30 min at 4°C , and then washed three times with FACs buffer ( BD Biosciences , Billerica , MA ) . Cells were analyzed using a FACSCanto analyzer ( BD Biosciences ) and data were analyzed using FlowJo software ( TreeStar , Ashland , OR ) . Gating of the CCR5-high population was determined by using a fluorescence minus one ( FMO; cells stained with antibodies for cell type markers , lacking the CCR5 antibody ) histogram in the fluorescence channel for the CCR5 antibody as a negative control . The FMO negative control histogram was plotted with a positive control of the single stain ( cells stained only with CCR5 antibody ) from the same tumor . Percent of CCR5+ cells were gated according to the positive control . RNA was isolated from tumors and cells using RNeasy columns ( Qiagen , Hilden , Germany ) . 1 μg of RNA was reversed transcribed using cDNA synthesis reagents ( Promega , Madison , WI ) . qPCR was performed using 6-carboxyfluorescein labeled TaqMan probes ( Thermo , Waltham , MA ) : CCL5 ( Mm01302427_m1 ) , CXCL1 ( Mm04207460_m1 ) , CXCL2 ( Mm00436450_m1 ) , CXCL5 ( Mm00436451_g1 ) , CCL2 ( Mm00441242_m1 ) , Actin ( Mm02619580_g1 ) , ASPN ( Mm00445945_m1 ) , PCOLCE ( Mm00476608_m1 ) , COL5A1 ( Mm00489299_m1 ) , COL24A1 ( Mm01323744_m1 ) , and read on a Bio-Rad ( Hercules , CA ) CFX qPCR machine . Western blotting was performed as described ( Alvarez et al . , 2013 ) using the following antibodies: NFκB p65 ( D14E12 , Cell Signaling , Danvers , MA ) , p-NFκB p65 ( 93H1 , Cell Signaling ) , and tubulin ( TU-02 , Santa Cruz , Dallas , TX ) , all at a 1:1000 dilution . Secondary antibodies conjugated to Alexa Flour 680 ( Life Technologies , Carlsbad , CA ) or 800 ( LI-COR Biosciences , Lincoln , NE ) were detected with the Odyssey detection system ( LI-COR Biosciences ) . For p-p65 detection , secondary antibodies conjugated to HRP were used and blots were developed using Classico or Crescendo reagent ( Millipore , Burlington , MA ) and exposed to film ( VWR , Radnor , PA ) . Secondary antibodies were used at a 1:5000 dilution . For cytokine array analysis , tumor lysates were made in 2X lysis buffer ( RayBiotech , Norcross , GA ) and diluted to 50 μg per 100 μL in diluent provided . Tumor lysates and standards were run on both Quantibody Mouse Cytokine Array Q1 and Q4 ( RayBiotech ) . Slides were scanned and quantified by RayBiotech . pLenti CMV GFP Puro was purchased from Addgene ( Watertown , MA ) . A CCL5 cDNA encoding the full-length mouse protein was amplified by RT-PCR from recurrent MTB;TAN tumor cells and cloned into the retroviral expression vector pK1 using the following primers: Forward: TAACCTCGAGATGAAGATCTCTGCAGCTG , Reverse: TAACGCGGCCGCCAGGGTCAGAATCAAGAAACC . A CCL5 cDNA encoding the full-length mouse protein was amplified by RT-PCR from recurrent MTB;TAN tumor cells and cloned into the lentiviral expression vector pLenti CMV using the following primers: Forward: TAACTCTAGAATGAAGATCTCTGCAGCTG , Reverse: TAACGTCGACCAGGGTCAGAATCAAGAAACC . CCL5 CRISPR sgRNAs: CCL5_1 ( TGTAGAAATACTCCTTGACG ) , CCL5_2 ( TACTCCTTGACGTGGGCACG ) , CCL5_3 ( TGCAGAGGGCGGCTGCAGTG ) . A small guide against AAVS was used as control . sgRNAs were cloned into Lentiguide puro ( Sanjana et al . , 2014 ) . Cas9 infection was with lentiguide Cas9 blast ( Sanjana et al . , 2014 ) . Retrovirus was produced by transfecting the packaging lines 293T Ampho and 293T Eco with the retroviral construct pK1 empty or CCL5 using Lipofectamine 2000 . Retroviral supernatant was collected 48 hr post-transfection , filtered , and used to transduce cells in the presence of 6 μg/mL polybrene ( Sigma , St . Louis , MO ) . Lentivirus was produced by transfecting 293 T cells with the packaging plasmids psPAX2 and pMD2 . G and lentiviral construct pLenti CMV GFP or CCL5 using Lipofectamine 2000 . Lentiviral supernatant was collected 48 hr post-transfection , filtered , and used to transduce cells in the presence of 6 μg/mL polybrene ( Sigma ) . RNA was isolated from tumors or tumor cells using RNeasy columns ( Qiagen ) . For TAM sequencing , macrophages were isolated by FACS using the antibody panel described above , and RNA was isolated using RNeasy columns ( Qiagen ) . RNA was sequenced using the Illumina HiSeq 4000 libraries and sequencing platform with 50 base pair single end reads by the Duke GCB Sequencing and Genomic Technologies Shared Resource ( Durham , NC ) . Sequencing data have been deposited in SRA as PRJNA506006 for cell line data and PRJNA505845 for macrophage data . Publicly available microarray data from human primary and residual breast cancer datasets GSE10281 and GSE21974 and their corresponding clinical annotation were downloaded , converted to log2 scale , and median centered . Heatmaps were created using R ( R Development Core Team , 2013 ) . Tumor sections were fixed in 10% normal formalin for 16 hr , then washed twice with PBS and transferred to 70% ethanol for storage . Stored tumor sections were paraffin imbedded and cut on the microtome in 5 μm sections . Sections were stained using a regressive H and E protocol , immunohistochemistry , or Masson’s Trichrome . The regressive H and E protocol is as follows: dewax and rehydrate slides . Incubate slides in Harris Modified Hematoxylin with Acetic Acid ( Fisher , Hampton , NH ) for 5 min . Incubate in Eosin ( Sigma ) for 1:30 min . Then dehydrate slides and mount slides with permount and coverslip . Let dry overnight . For cytokeratin eight staining ( Troma 1 , Brulet , P . , Kemler , R . Institut Pasteur , Paris , France ) immunohistochemistry slides were dewaxed and rehydrated as above . Slides were boiled in antigen retrieval buffer ( 1X in ddH2O ) for 5 min and allowed to cool . Slides were washed in PBS and then incubated in 0 . 3% H2O2 . Slides were washed , blocked and stained according to the protocol from the rabbit secondary Vectastain ABC kit ( Vector Labs , Burlingame , CA ) . Primary antibody was used at a dilution of 1:50 . CD45 ( 30-F11 , BD Biosciences , 1:200 ) , CD3 ( SP7 , Thermo , 1:100 ) , and F4/80 ( Cl:A3-1 , Bio-Rad , 1:1000 ) staining were performed by the Duke Pathology core ( Durham , NC ) . Trichrome stain was performed using a staining kit from Abcam ( Cambridge , UK ) ( ab150686 ) . To quantify the amount of positive staining for CD3 , CD45 , and F4/80 and for Masson’s Trichrome , we used Fiji ( Schindelin et al . , 2012 ) . The ‘Color Deconvolution’ function was used to separate the colors into positive staining and hematoxylin for normalization . We then converted each image to 8-bit and applied a threshold of positive staining to each image and used this same threshold across all images . We then measured the pixel area of the positive staining and normalized this to the hematoxylin staining for each image . For the primary tumors and 5 day -dox tumors , the whole image was used for quantification . For residual tumors , we manually selected regions-of-interest to exclude adipose tissue from the quantification . For GSEA , the normalized enrichment score ( NES ) is reported . The normalized enrichment score accounts for differences in gene set size and in correlations between gene sets . The NES is based on all dataset permutations , to correct for multiple hypothesis testing . The nominal p value is also reported and is the statistical significance of the enrichment score , without adjustment for gene set size or multiple hypothesis testing . A reported p value of zero ( 0 . 0 ) indicates an actual p-value of less than 1/number-of-permutations . ( Subramanian et al . , 2005 ) . Two-tailed unpaired t-tests were used to analyze significance between primary tumor samples and all other time points for qPCR , cytokine array , and flow cytometry analysis . For the cytokine array , appropriate same size was calculated using JMP Pro ( SAS Institute Inc , Cary , NC ) . A standard deviation of 20% was assumed , with a power of 0 . 8 , fold change of 2 , and p-value ( alpha ) of 0 . 05 . This power calculation indicated that a sample size of 8 ( 4 tumors per cohort ) was required . The same parameters were used for sample size calculation for flow cytometry analysis of control and CCL5-expressing tumors . For recurrence free survival ( RFS ) , statistical analysis methods are listed in orthotopic recurrence assays . Outliers were never excluded except for in flow cytometry experiments . Tumors that were >90% CD45+ were excluded from analysis to avoid analyzing tumors with potential contamination from the inguinal lymph node . For all other experiments where no power analysis was used , sample size was chosen based upon previous experience ( Alvarez et al . , 2013 ) . Animal care and all animal experiments were performed with the approval of and in accordance with Duke University IACUC guidelines . Mice were housed under barrier conditions .	Breast cancer is the second-leading cause of cancer-related deaths in women . Recurrence of breast-cancer five or more years after initial diagnosis and treatment causes more than half of these deaths . This suggests that some tumor cells survived treatment and persisted undetected . These residual tumor cells may not grow for years and are often surrounded by other cells , including immune system cells . What role these surrounding immune cells play in triggering future growth of these residual tumor cells is not clear . Many breast cancer patients receive chemotherapy , which kills all quickly dividing cells . Targeted therapies , which block signals necessary for cancer cell growth , are also used often . More recently , scientists have developed treatments that use a patients own immune system to fight off cancer . Scientists are currently studying whether combining these immunotherapies with chemotherapy or targeted therapies increases the likelihood of eliminating cancer . Learning more about the role surrounding immune cells play in allowing residual tumor cells to persist and regrow is important to understanding how to treat cancer more successfully and prevent recurrence . Now , Walens et al . show that immune cells called macrophages supply residual breast cancer cells in mice with a protein called collagen that they need to grow . In the experiments , mice with an aggressive form of breast cancer called Her2 received targeted cancer therapy . After the treatment , tumor cells in the mice released small molecules called cytokines that attract immune system cells . Levels of one cytokine called CCL5 rose after treatment and remained high in residual tumors in the mice . The experiments also revealed that CCL5 levels were high in residual breast cancer tumors collected from women . This shows that high levels of CCL5 appear to shorten the amount of time between tumor treatment and recurrence because CCL5 attracts macrophages that deposit collagen in the residual tumors . Scientists believe collagen promotes tumor growth because recurrent tumors have high levels of collagen and breast cancer patients with high levels of collagen in their tumors often have worse outcomes . Treatments that prevent or block the release of CCL5 or that stop macrophages from supplying the residual tumor cells with collagen may help prevent recurrence .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation", "cancer", "biology" ]	2019	CCL5 promotes breast cancer recurrence through macrophage recruitment in residual tumors
During outbreaks of high-consequence pathogens , airport screening programs have been deployed to curtail geographic spread of infection . The effectiveness of screening depends on several factors , including pathogen natural history and epidemiology , human behavior , and characteristics of the source epidemic . We developed a mathematical model to understand how these factors combine to influence screening outcomes . We analyzed screening programs for six emerging pathogens in the early and late stages of an epidemic . We show that the effectiveness of different screening tools depends strongly on pathogen natural history and epidemiological features , as well as human factors in implementation and compliance . For pathogens with longer incubation periods , exposure risk detection dominates in growing epidemics , while fever becomes a better target in stable or declining epidemics . For pathogens with short incubation , fever screening drives detection in any epidemic stage . However , even in the most optimistic scenario arrival screening will miss the majority of cases . International air travel drove the spread of SARS in 2003 and influenza A/H1N1p in 2009 ( Brockmann and Helbing , 2013 ) , and has since led to imported cases of influenza A/H7N9 ( William et al . , 2015 ) , MERS-CoV ( Cauchemez et al . , 2014 ) and Ebola virus infection ( McCarthy , 2014 ) . Traveller screening policies , including fever screening and/or questionnaires at point of departure and/or arrival , have been proposed to limit the geographic spread of infection ( Malone et al . , 2009; World Health Organization , 2009; Cowling et al . , 2010; Khan et al . , 2013; Bogoch et al . , 2015; Centers for Disease Control and Prevention , 2014a ) . Fever screening at point of arrival has been criticized , however , because long incubation periods and imperfect efficacy of fever screening devices reduce the probability of detecting symptoms in infected arriving passengers ( Pitman et al . , 2005; Bitar et al . , 2009; Mabey et al . , 2014 ) . As the effectiveness of integrated screening programs will depend both on the pathogen-specific natural history of infection and epidemiological knowledge of exposure risk , as well as travel time and efficacy of screening methods , it is important to understand how these different factors contribute to screening effectiveness at departure and arrival . During screening initiatives for influenza A/H1N1p , MERS-CoV , SARS-CoV and Ebola virus , large numbers of travellers were detained for in-depth assessment , but few or no cases were ultimately detected ( Table 1 ) . Although fever is the symptom most commonly measured during screening , it might not be detected in all infected individuals for several reasons . First , those with recent exposure may not yet have progressed to a symptomatic stage ( Pitman et al . , 2005; Mabey et al . , 2014 ) . Second , travellers might be symptomatic but not febrile; the probability a symptomatic patient will have a fever varies by pathogen ( Donnelly et al . , 2004; Cao et al . , 2009; Louie et al . , 2009; Assiri et al . , 2013; Cowling et al . , 2013; Gao et al . , 2013; Gong et al . , 2014; Sun et al . , 2014; WHO Ebola Response Team , 2014 ) . Third , the sensitivity of non-contact infrared thermometers ( the devices most often used for airport fever screening ) is limited , so passengers with fever may pass through symptom screening undetected ( Hausfater et al . , 2008; Bitar et al . , 2009; Nishiura and Kamiya , 2011 ) . Fourth , passengers may conceal fever and other symptoms during screening using antipyretic drugs ( Nishiura and Kamiya , 2011 ) . At the same time , fever is notoriously non-specific as a symptom , leading to high opportunity costs from detaining travellers with non-target illnesses ( Anderson et al . , 2004; Gunaratnam et al . , 2014; Mabey et al . , 2014 ) . 10 . 7554/eLife . 05564 . 003Table 1 . Airport screening measures during past disease outbreaksDOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 003PathogenDateLocationDirectionScreenedDetainedPositiveSourceInfluenza A/H1N1p27 April–22 June 2009Auckland , New ZealandInbound456 , 5184064 ( Hale et al . , 2012 ) 28 April–18 June 2009Sydney , AustraliaInbound625 , 14758453 ( Gunaratnam et al . , 2014 ) 28 April–18 June 2009Tokyo , JapanInbound471 , 73380515 ( Nishiura and Kamiya , 2011 ) SARS Co-V5 April–16 June 2003AustraliaInbound1 , 840 , 0007940 ( Samaan et al . , 2004 ) 31 March–31 May 2003SingaporeInbound442 , 9731760 ( Wilder-Smith et al . , 2003 ) 14 May–5 July 2003Toronto , CanadaInbound349 , 75412640 ( St John et al . , 2005 ) 14 May–5 July 2003Toronto , CanadaOutbound495 , 4924110 ( St John et al . , 2005 ) MERS Co-V24 September 2012–15 October 2013EnglandInboundNR772 ( Thomas et al . , 2014 ) Ebola virusAugust–September 2014Guinea , Liberia , Sierra LeoneOutbound36 , 000770 ( Centers for Disease Control and Prevention , 2014a ) 11 October–22 October 2014United StatesInbound76230 ( Apuzzo and Fernandez , 2014; CBS , 2014 ) Self-reporting of symptoms or potential recent exposure to infection via mandatory questionnaires is also a common component of traveller screening programs ( St John et al . , 2005; Nishiura and Kamiya , 2011; Hale et al . , 2012; Centers for Disease Control and Prevention , 2014a; Cho and Yoon , 2014; Gunaratnam et al . , 2014 ) . Because information about risk factors does not depend on the presence of detectable symptoms at the time of screening , there is potential to identify a broader set of exposed travellers . However , epidemiological knowledge on factors linked to risk of infection is limited for some pathogens—particularly for novel emerging pathogens that are often the focus of screening programs . Even for pathogens with well-characterized routes of transmission , not all cases will necessarily have a known source of exposure ( Lau et al . , 2004; Cao et al . , 2009; Tuite et al . , 2010; Cowling et al . , 2013; Gao et al . , 2013; Gong et al . , 2014; Sun et al . , 2014; WHO Ebola Response Team , 2014 ) . Thus , the contribution of questionnaires to the overall effectiveness of traveller screening programs is unclear . Screening initiatives have also been implemented both at points of departure and arrival . It has been suggested that departure screening is more efficient than entry screening because it needs to be implemented in only a few airports rather than globally ( Khan et al . , 2013; Bogoch et al . , 2015 ) , but there is often local political pressure for arrival screening as well . To understand how departure and arrival screening combine with pathogen natural history , epidemiological knowledge , efficacy of screening methodology , and human behavioral factors to determine overall screening outcomes , we developed a general modelling framework ( Figure 1 ) for the screening process . We used this framework to assess outcomes for six pathogens of current or recent concern: influenza A/H7N9 , influenza A/H1N1p , SARS-CoV , MERS-CoV , Ebola virus , and Marburg virus . By separating the contribution of different factors to the probability of detecting infectious travellers , we evaluated pathogen-specific strengths and weaknesses of different screening strategies . We considered scenarios in which the source epidemic is growing or stable , as epidemic phase influences the distribution of times since exposure in potential travellers . We also identified factors that could improve the effectiveness of screening programs for future emerging pathogens . 10 . 7554/eLife . 05564 . 013Figure 1 . Model of traveller screening process . ( A ) Upon airport arrival , passengers passed through screening for fever , followed by screening for risk factors . We assumed a one-strike policy: passengers identified as potentially infected by any single screening test were detained . ( B ) Passengers who did not present with fever would always pass through symptom screening , but could still be identified during questionnaire screening . ( C ) Passengers who were not aware of exposure risk would always pass through questionnaire screening . ( D ) Passengers with neither fever nor knowledge of exposure would go undetected . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 01310 . 7554/eLife . 05564 . 014Figure 1—figure supplement 1 . Detailed model formulation with parameters . Each case represents a different detectability class . Travellers are assigned to detectability classes with probabilities f ( presence of fever ) and g ( awareness of exposure risk ) . Values for f and g are given in Table 2 . θ ( d ) describes the infection age distribution ( times since exposure ) in individuals attempting travel . δ ( d ) is the incubation period cumulative distribution function , which describes the probability that travellers have progressed to symptom onset at the time of attempted travel . ε describes the efficacy of each respective screening module . S is the probability that travellers develop symptoms in flight , given that they did not yet have symptoms at departure . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 014 For each pathogen , the natural history of infection and state of epidemiological knowledge determined the potential for successful screening at various points in the process . For all six emerging pathogens considered here , the majority of identified cases exhibited a fever ( Figure 2A ) . However , the proportion of confirmed cases who were aware of their exposure risk varied greatly . Influenza A/H1N1p , which can have generic symptoms and can be transmitted via the airborne route , had the lowest reported proportion; Ebola virus , which requires close contact with infected individuals who have conspicuous symptoms , had the highest . Influenza A/H7N9 , Marburg virus and SARS-CoV had similar proportions of cases that present with fever , and that had knowledge of exposure risk . We excluded MERS-CoV from the natural history space in Figure 2A because there are no established risk factors for exposure . Moreover , there was limited information available for the fever parameter for MERS-CoV: in a hospital outbreak of MERS-CoV , 20 out of 23 cases presented with fever at onset ( Assiri et al . , 2013 ) ; the small size of this sample means there is greater uncertainty surrounding the estimate for proportion of cases that exhibit fever . 10 . 7554/eLife . 05564 . 004Figure 2 . Parameters characterizing natural history of infection and epidemiological knowledge . ( A ) Proportion of infected individuals who report known exposure risk and show fever at onset . Point shows median estimate , using data in Tables 2 , 3; circle shows joint 95% binomial confidence interval . Red , influenza A/H7N9; purple influenza A/H1N1p; blue , MERS; green , SARS; orange , Ebola; black , Marburg . ( B ) Incubation period and fever at onset . Point shows median estimate , circle shows joint 95% CI , generated using a binomial distribution for fever symptoms and fitted parametric distributions given by references in Table 3 for incubation period . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 004 The mean and variance of the incubation period have been recognized as key drivers of the effectiveness of fever screening at arrival , since shorter incubation periods mean a greater likelihood that travellers will progress to symptoms during travel ( Pitman et al . , 2005; Al-Tawfiq et al . , 2014; Mabey et al . , 2014 ) . There was considerable variability in incubation period among different pathogens ( Figure 2B ) . Influenza A/H7N9 and A/H1N1p have the shortest incubation periods , while Ebola virus and Marburg virus have the longest . For some pathogens , the estimated variance in incubation period could increase with the addition of more data , which would improve characterization of the tails of the distribution . For instance , the incubation period distribution for Marburg virus was estimated from just five cases with a single exposure opportunity ( Martini , 1973 ) ; observing more cases might give rise to a right-skewed distribution as seen for Ebola virus . Similarly , the incubation period distribution for MERS-CoV is determined using data from only 23 confirmed cases , and its variance might also expand with the addition of more data . The possibility of a lengthy incubation period presents challenges for symptom screening . Our focus on the natural history and epidemiology of infection revealed the crucial influence of the time between exposure and the departing flight . When we included the natural history parameters in the model , ( see ‘Materials and methods’ ) we found that the contributions of each component of a screening program depend strongly on the time between exposure to infection and intended departure from the airport ( Figure 3 ) . Individuals with more recent exposure were less likely to display symptoms at the time of screening , and hence less likely to be identified by fever screening at departure . For pathogens with long incubation periods , the marginal value of fever screening at arrival was also lower . Note , however , that the 70% efficacy of non-contact infrared thermoscanners means that arrival screening can contribute by catching symptomatic cases missed at departure . Thus , the bulk of the contribution of arrival fever screening is mediated by equipment efficacy rather than natural history . 10 . 7554/eLife . 05564 . 005Figure 3 . Impact of infection age on effectiveness of screening measures . Expected fraction of passengers detected by fever and risk factor screening , at arrival and departure , as a function of the time between an individual’s exposure and the departure leg of their journey . We assume a 70% probability that fever screening will identify febrile patients , and a 25% probability that a traveller with a known history of risky exposure will report it on a questionnaire . We assume 24 hr travel time . The white lines denote the point at which travellers board their flight; the black dashed line shows the median time from exposure to hospitalization for each pathogen . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 00510 . 7554/eLife . 05564 . 006Figure 3—figure supplement 1 . Expected proportions detected by screening when efficacy of fever screening is 50% and proportion of cases with known exposure history who report correctly is 0 . 25 . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 00610 . 7554/eLife . 05564 . 007Figure 3—figure supplement 2 . Expected proportions detected by screening when efficacy of fever screening is 70% and proportion of cases with known exposure history who report correctly is 0 . 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 00710 . 7554/eLife . 05564 . 008Figure 3—figure supplement 3 . Expected proportions detected by screening when efficacy of fever screening is 50% and proportion of cases with known exposure history who report correctly is 0 . 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 008 Shortly after exposure , we found that detection was typically possible only by risk factor questionnaire screening , as most cases had not yet progressed to symptoms and were undetectable by fever screening ( Figure 3 ) . ( Again , questionnaire screening at arrival contributes by catching some individuals who did not disclose their exposure risk at departure . ) The duration of this phase depended on the incubation period , which is shortest for influenza A/H7N9 and A/H1N1p , and longest for Ebola and Marburg viruses; for MERS-CoV , despite a mid-length incubation period , questionnaire screening contributes nothing due to our ignorance of risk factors . As time since exposure elapsed , fever screening made a greater contribution to case detection , with pathogen natural history factors ( i . e . , incubation period , and fraction presenting with fever ) becoming the primary determinants of screening effectiveness . We found similar qualitative patterns when we assumed reduced efficacy for fever screening devices ( Figure 3—figure supplement 1 ) , questionnaire reporting ( Figure 3—figure supplement 2 ) , or both tests ( Figure 3—figure supplement 3 ) . The striking patterns in Figure 3 highlight the important role of the ‘infection age structure’ ( i . e . , the distribution of times from exposure to departure ) of the traveller population . As a basic consideration , cases are more likely to have progressed to severe disease or death as time since exposure increases , so the population of infected individuals able to attempt air travel will be skewed toward more recent exposures ( i . e . , younger infections ) . The distribution of time since exposure will be influenced by the epidemic phase in the source population ( Figure 4 , Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 05564 . 009Figure 4 . Proportion of infected travellers that would be missed by each of four screening scenarios . ( A ) Proportion of 50 infected travellers that would be missed by both departure and arrival screening in a growing epidemic . Figure shows three possible screening methods: fever screen , exposure risk questionnaire , or both . Lines show 95% bootstrapped CI . ( B ) Proportion of infected travellers missed by both departure and arrival screening in a stable epidemic . ( C ) Proportion of infected individuals who fly that are missed by arrival screening in a stable epidemic . ( D ) Proportion of infected arrivals missed by point of entry screening in a stable epidemic . We assume 25% probability traveller will report if they know exposure and 70% probability screening with identify visibly febrile patients . We assume R0 = 2 and a 24 hr travel time . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 00910 . 7554/eLife . 05564 . 010Figure 4—figure supplement 1 . Different time from exposure to departure functions used in model . Red , influenza A/H7N9; purple influenza A/H1N1p; blue , MERS; green , SARS; orange , Ebola; black , Marburg . ( A ) Growing epidemic with R0 = 1 . 5 . ( B ) Stable situation . DOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 010 Overall screening effectiveness was greater in stable than growing epidemics ( Figure 4A–B ) . These gains were driven by increased potential for fever detection in stable epidemics , where cases are less likely to be recently exposed and asymptomatic . In contrast , exposure risk detection does not vary with epidemic phase because exposure risk awareness does not depend on the infection age distribution . Regardless of epidemic phase , the full screening program fails to detect at least 25% of infected travellers , despite our optimistic assumptions . Focusing on the contribution made by screening at point of arrival , our projections suggest that arrival screening will still miss half to three-quarters of infected travellers that manage to complete their flights ( Figure 4C–D ) . For pathogens with short incubation periods ( i . e . , influenza virus ) fever detection was responsible for the majority of case identification in all epidemic phases . However , for pathogens with longer incubation periods ( i . e . , Ebola , Marburg , and SARS-CoV ) , exposure risk screening was responsible for half or more of case detection in growing epidemics . For these pathogens , fever detection was dominant only in stable epidemics ( Figure 4 ) . We assessed the influence of pathogen natural history , knowledge of exposure risk , efficacy of screening techniques , and epidemic phase on the ability to detect infected passengers using integrated traveller screening programs . By incorporating pathogen natural history and epidemiology into a mathematical model , we compared screening effectiveness for different pathogens , and showed that detection is driven by screening for exposure risk in travellers with recent exposure and by screening for fever in travellers with older infections . We found that natural history , epidemiological knowledge and epidemic phase combined to determine overall screening outcomes , as well as the relative contribution of each screening method to case detection . Exposure risk screening made a greater contribution to case detection during growing epidemics and for pathogens with longer incubation periods , and when exposure risk factors were well-characterized epidemiologically . Our results highlight distinct taxonomic patterns in the effectiveness of screening measures . For influenza viruses , which have the shortest incubation periods , our findings suggest that fever screening would be responsible for the majority of case detection in any epidemic context . The picture is more subtle for filoviruses , which have longer incubation periods and distinctive symptoms . When the source epidemic is still growing and many travellers are recently exposed , our model suggests half or more of case detection for filoviruses will be driven by exposure risk screening; when the epidemic has stabilized , the shift in the age structure of infections means fever is likely to become the dominant mechanism of case detection ( Figure 4 ) . Our analysis of coronaviruses illustrates two important features of the screening process . First , the variance of incubation period distributions can strongly modulate screening outcomes: the median incubation period for SARS-CoV is similar to influenza viruses , but because its distribution has a long right tail , the expected incubation time is longer . Thus , despite a short , influenza-like median incubation time , screening outcomes for SARS-CoV are filovirus-like and less favorable . Second , epidemiological knowledge is required to implement risk factor screening . Our results show that for pathogens with long incubation periods , early characterization of exposure risk factors is a powerful screening tool at the beginning of an outbreak; for these pathogens , robust characterization of and screening for specific exposure risk factors can contribute more to case detection than rapid implementation of fever screening ( Figures 3–4 ) . However , risk factor screening cannot improve screening effectiveness if factors that increase risk of exposure are not well characterized . For MERS-CoV , recent findings have strengthened the evidence that dromedary camels play a role in primary exposure of some cases , but it remains unclear whether the majority of transmission is driven by contact with infected humans or repeated zoonotic spillovers ( Azhar et al . , 2014; Cauchemez et al . , 2014 ) . When novel pathogens emerge , it is important to prioritize case-control and other epidemiological studies to establish which factors contribute to exposure risk . As well as enhancing potential case detection using risk factor screening , such knowledge could have the additional benefit of promoting general awareness of exposure risk factors , and hence contribute to reductions in risky behavior in affected regions . Further , when risk factors are known there is potential to conduct extended follow-up with travellers who have exposure risks , but no symptoms at the time of flight . For example , patients with known risks for Ebola exposure should be monitored by local health authorities near their travel destination until the maximum plausible incubation period has elapsed ( Brown et al . , 2014 ) . Our results also illustrate that well-characterized exposure risks are not always sufficient: other factors of pathogen natural history can limit the potential effectiveness of risk detection . For example , exposure risks for influenza A/H7N9 are well defined and typically identifiable ( exposure to live poultry or infected human contacts [Cowling et al . , 2013; Gao et al . , 2013] ) . However , because influenza A/H7N9 infection has a short incubation period and a high probability of generating febrile symptoms , fever screening becomes a potent tool within just a few days of exposure and eclipses the potential contribution of risk factor-driven detection . Thus , even during a growing epidemic , where the proportion of asymptomatic travellers is highest , fever detection is likely to remain the dominant mechanism of detection for pathogens with short incubation periods , such as the influenza viruses ( Figure 4A ) . We found that once the source epidemic stabilizes or begins to decline , screening has greater overall potential effectiveness: the infection age structure in a stable epidemic allows more fever-driven detection , which has much higher efficacy than risk screening ( Figure 4B ) . This result reinforces other studies that emphasize the need to control outbreaks at the source during the growth phase ( Khan et al . , 2013; Bogoch et al . , 2015; Mabey et al . , 2014 ) . Moreover , it suggests that stabilizing the growth of an epidemic could have the added benefit of making passenger screening more effective . Even in this stable phase , however , our results suggest that screening at point of arrival would still miss more than half of incoming infected passengers . As well as being influenced by the natural history and epidemiological factors described above , the overall effectiveness of traveller screening depends on the efficacy of particular screening techniques . This efficacy in turn depends on a combination of instrument and human factors . We assumed that fever and risk factor screening are implemented with 70% and 25% efficacy , respectively , which we consider upper bounds ( see ‘Materials and methods’ ) . Although our results are qualitatively insensitive to these assumptions ( Figure 3—figure supplements 1–3 ) , the quantitative results of our main analysis likely represent best-case scenarios . The estimated efficacy of fever screening reflects the sensitivity of non-contact infrared thermal scanner equipment , but human factors may further reduce the efficacy of screening techniques . For example , data from influenza A/H1N1p screening in Tokyo in 2009 suggested antipyretic drug use could have been widespread among febrile travellers ( Nishiura and Kamiya , 2011 ) . Additionally , outbreak-affected countries with heavily burdened public health systems may have limited resources to invest in departure screening , while limited preparation , awareness or focus may lower efficacy during arrival screening in countries outside the epidemic region . Empirical evidence suggests that the majority of travellers with known exposure would not self-report ( Hale et al . , 2012; Gunaratnam et al . , 2014 ) , so the absolute effectiveness of risk factor screening in our model was lower than the effectiveness of symptom screening once cases progressed to onset , even if many exposed travellers were not yet symptomatic and symptom screening was only 70% effective ( Bitar et al . , 2009 ) . For questionnaire-based screening , an essential unknown is the probability that travellers will divulge their exposure history if it puts them at risk of detainment or delay . We arrived at a rough , upper-bound estimate of 25% probability of honest reporting for the 2009 influenza A/H1N1p pandemic ( see ‘Materials and methods’ ) , but we emphasize that this is a topic in need of further study . Even more valuable would be effective ways to motivate travellers to honestly report their risky exposures . Increasing honest exposure reporting not only has the potential to enhance detection of infected travellers , but is essential for implementation of follow-up monitoring of travellers who may have been exposed but have not yet developed symptoms . Data from past and ongoing screening initiatives support our suggestion that outcomes predicted in this study should be interpreted as plausible best-case scenarios . For example , during the 2009 influenza A/H1N1 pandemic , arrival screening in Sydney , Australia detected 3 of an estimated 48 infected travellers , giving an empirical sensitivity of 7% ( 95% CI , 1–18% ) . This initiative used a combination of risk factor and fever detection in a growing epidemic , and yielded less favorable results than the effectiveness of 32% ( 95% CI 20–46% ) predicted by our model . Also , questionnaire-based arrival screening for influenza A/H1N1 in Auckland , New Zealand detected 4 of 69 infected individuals , for a sensitivity of 6% ( 95% CI 2–14% ) . In this case , our model's predicted sensitivity ( 6% , 95% CI 2–16% ) matches the observed pattern well . Between August and January 2015 , screening for Ebola in United States detected neither of two case importations and screening in the United Kingdom did not detect the single known case importation ( Department of Health , 2015 ) . Therefore observed sensitivity for Ebola screening has been 0% , which is lower than our model-predicted value of around 50% . While the comparison is not statistically significant with only three data points , these outcomes underscore that screening is inherently imperfect and can be expected to reduce—but not to prevent—disease importations . These comparisons illustrate that actual screening efficacies and honest reporting fractions may vary considerably , and in some cases appear to be quite poor . Even under best-case scenario assumptions , our model suggests arrival screening will miss half or more of infected travellers . Thus , for screening to be implemented with reasonable effectiveness there is a need to identify behavioral incentives that encourage much better self-reporting and efficacy than the current data indicate . There are some additional limitations to our framework . Because our model is structured so that fever screening precedes exposure risk screening , case detection through fever screening increasingly overlaps with potential detection through questionnaires as time since exposure increases . The model results appear to show that the effectiveness of exposure risk screening decreases with time since exposure ( Figure 3 ) , but in fact this shows that risk factor screening becomes increasingly redundant when passengers are subject to fever screening first . We have treated fever screening as the primary means of detection , as it is much easier to conclusively diagnose infection when symptoms are present ( Towner et al . , 2004; Centers for Disease Control and Prevention , 2013; Centers for Disease Control and Prevention , 2014b ) . The overall effectiveness of traveller screening , and the total proportion of cases detected before and after a flight , are independent of screening order . We did not consider the potential for case detection using symptoms other than fever . While other symptoms may aid in case detection , many ( coughing , sneezing , etc ) are also non-specific and more easily concealed than fever in the early stages of infection ( Donnelly et al . , 2004; Cao et al . , 2009; Louie et al . , 2009; Assiri et al . , 2013; Gao et al . , 2013; Gong et al . , 2014; Sun et al . , 2014; WHO Ebola Response Team , 2014 ) . Episodic symptoms such as vomiting , and internal symptoms such as gastrointestinal distress , would be difficult to detect via point-screening but could be incorporated into questionnaires . We also concentrated on the sensitivity rather than specificity of screening measures . This is another respect in which our results should be considered a best-case scenario projection of detection outcomes , assuming that the financial and opportunity costs of imposing additional clinical assessment on a large number of uninfected individuals can be neglected . Past screening programs have been implemented at large financial cost and have delayed large numbers of travellers , while detecting only a few cases ( Table 1 ) , reflecting the fact that fever and many risk factors ( e . g . , contact with live poultry ) have low positive predictive value for infection with rare pathogens . Though the high cost and low effectiveness of screening have been noted ( St John et al . , 2005; World Health Organization , 2009; Cowling et al . , 2010; Gunaratnam et al . , 2014; Mabey et al . , 2014 ) , to the best of our knowledge no formal cost analysis of traveller screening policies has ever been conducted . Such information would greatly aid future policy decisions about screening measures . Screening at departure rather than arrival has been suggested as a more cost-effective and logistically feasible policy , as departure screening need be implemented only in affected regions , rather than globally ( Khan et al . , 2013; Bogoch et al . , 2015 ) . Our analysis suggests that arrival screening has the potential to make a non-negligible contribution to overall case detection , not only by detecting travellers who develop symptoms in flight , but also by detecting travellers who were missed by imperfect screening at departure . Hence , the additional benefit of arrival screening is greatest when efficacy of departure screening is relatively low , for example if potentially infected travellers primarily depart regions with limited public health re-sources and arrive in regions where public health resources are more abundant . Yet even costly policies that combine exit and arrival screening lack the potential to prevent all case importations . Our analysis suggests that in any context screening would miss a substantial proportion of infected travellers; this result is consistent with other analyses that highlight the limited effectiveness of screening ( Pitman et al . , 2005; Bitar et al . , 2009; Mabey et al . , 2014 ) and with previous or ongoing screening outcomes ( Table 1 ) . Policy makers should carefully consider whether resources are better spent on arrival screening , which will reduce but not eliminate the importation of cases , or instead on tracing and containing cases that inevitably do arrive . Screening policies have been implemented during several recent epidemics ( Samaan et al . , 2004; Pitman et al . , 2005; St John et al . , 2005; Nishiura and Kamiya , 2011; Hale et al . , 2012; Khan et al . , 2013; Centers for Disease Control and Prevention , 2014a; Gunaratnam et al . , 2014 ) , and will likely continue to be discussed in response to future disease outbreaks . Certain aspects of screening , particularly fever screening at arrival , have been criticized as having little scientific justification ( Pitman et al . , 2005; Bitar et al . , 2009; Mabey et al . , 2014 ) , but political leaders and health policy makers are likely to consider implementing screening programs when public pressure becomes intense . Thus there is a need to characterize the potential contributions of screening programs when implemented at different times , in different combinations , and for different pathogens; ultimately a quantitative understanding will be needed , to factor into cost-benefit calculations . In this study we begin to address these issues by demonstrating that screening outcomes depend strongly on pathogen natural history and epidemiological features , as well as human factors in implementation and compliance . Our results emphasize the need to characterize basic properties of emerging pathogens , as this knowledge can enhance disease control measures . Using previously published studies of influenza A/H7N9 , influenza A/H1N1p , SARS-CoV , MERS-CoV , Ebola virus , and Marburg virus , we assembled a set of four parameters describing the natural history and epidemiology of each pathogen . To describe epidemiology we established the proportion of cases that had a known source of exposure , and for natural history we established the proportion of symptomatic cases that exhibited fever ( Table 2 ) . For pathogen natural history we also gathered estimates for incubation period ( i . e . , time from exposure to onset of symptoms ) , and the time from onset to hospitalization ( Table 3 ) , which we use to approximate the period after which most exposed individuals have progressed to severe illness and will not attempt travel ( details in ‘Appendix’ ) . To estimate the proportion of individuals with known source of exposure for a particular pathogen , we identified the fraction of confirmed cases in descriptive epidemiological studies who reported contact with a known source of infection ( e . g . , poultry for influenza A/H7N9; close contact with an infected human for SARS-CoV , influenza A/H1N1p , Marburg and Ebola ) . As we could not find published estimates for the proportion of MERS-CoV cases with known source of exposure outside a hospital setting ( Assiri et al . , 2013 ) , we assumed that exposure risk would not typically be known for MERS-CoV cases; this is consistent with recent publications highlighting the crucial knowledge gap in risk factors for MERS-CoV infection ( Al-Tawfiq et al . , 2014; Zumla et al . , 2014 ) . 10 . 7554/eLife . 05564 . 011Table 2 . Natural history parameters: f is the proportion of cases with fever , g is the proportion of cases aware of exposure riskDOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 011PathogenParameterMeanSample sizeReferenceA/H7N9f0 . 7985 ( Cowling et al . , 2013 ) f1 . 0046 ( Gong et al . , 2014; Sun et al . , 2014 ) f1 . 00111 ( Gao et al . , 2013 ) g0 . 75123 ( Cowling et al . , 2013 ) g0 . 56111 ( Gao et al . , 2013 ) g0 . 7846 ( Gong et al . , 2014; Sun et al . , 2014 ) A/H1N1f0 . 67426 ( Cao et al . , 2009 ) f0 . 891088 ( Louie et al . , 2009 ) g0 . 29426 ( Cao et al . , 2009 ) SARSf0 . 941452 ( Donnelly et al . , 2004 ) g0 . 291192 ( Lau et al . , 2004 ) MERSf0 . 8723 ( Assiri et al . , 2013 ) g010 , 000 ( Cauchemez et al . , 2014 ) Ebolaf0 . 871151 ( WHO Ebola Response Team , 2014 ) g0 . 86142 ( Pattyn , 1978 ) Marburgf0 . 93129 ( Bausch et al . , 2006 ) f0 . 4715 ( Bausch et al . , 2003 ) g0 . 6739 ( Roddy et al . , 2010 ) 10 . 7554/eLife . 05564 . 012Table 3 . Time from exposure to onset ( i . e . , incubation period ) and onset to hospitalization for different pathogensDOI: http://dx . doi . org/10 . 7554/eLife . 05564 . 012PathogenTime fromMean ( days ) ReferenceInfluenza A/H7N9Exposure-to-onset4 . 3 ( Cowling et al . , 2013 ) Onset-to-hospitalization5 ( Gao et al . , 2013; Sun et al . , 2014 ) Influenza A/H1N1Exposure-to-onset4 . 3 ( Tuite et al . , 2010 ) Exposure-to-onset2 . 05 ( Ghani et al . , 2009 ) Onset-to-recovery7 ( Tuite et al . , 2010 ) SARS-CoVExposure-to-onset6 . 4 ( Donnelly et al . , 2003 ) Onset-to-hospitalization4 . 85 ( Donnelly et al . , 2003 ) MERS-CoVExposure-to-onset5 . 2 ( Assiri et al . , 2013 ) Exposure-to-onset5 . 5 ( Cauchemez et al . , 2014 ) Onset-to-hospitalization5 ( Assiri et al . , 2013 ) EbolaExposure-to-onset9 . 1 ( WHO Ebola Response Team , 2014 ) Onset-to-hospitalization5 ( WHO Ebola Response Team , 2014 ) MarburgExposure-to-onset6 . 8 ( Martini , 1973 ) Onset-to-hospitalization5**As there was limited data for onset-to-hospitalization for Marburg , we assumed the same value as for Ebola . A review of studies of non-contact infrared thermometer efficacy , when applied to forehead ( as is typical for airport screening ) , suggested that the scanners had an average efficacy of 70% ( Bitar et al . , 2009 ) . In our main analysis , we therefore assumed that the probability that febrile travellers would be detected by fever screening was 70% . This is an optimistic estimate , ignoring possible challenges in implementation in outbreak-affected regions and oversights made by device operators in arrival sites where risk may seem remote . Another important parameter is the fraction of travellers who will report honestly about known exposure to risk factors in a screening questionnaire . This quantity is intrinsically difficult to measure , and to our knowledge it has not been estimated before . We estimate an upper bound on this quantity by drawing on information from studies of influenza A/H1N1p . As summarized in Table 1 , one study from the early phase of the pandemic in China showed that 29% ( 95% binomial CI: 25–33% ) of cases were aware of their exposure to earlier cases ( Cao et al . , 2009 ) . Studies of influenza A/H1N1p screening in New Zealand ( Hale et al . , 2012 ) and Australia ( Gunaratnam et al . , 2014 ) estimated that self-reported exposure screening identified 3/45 and 4/69 infected passengers respectively . Assuming the lower limit of the CI , that is , 25% of the infected passengers knew about their exposure history , the New Zealand and Australia studies suggest that a proportion 3/ ( 0 . 25 × 45 ) = 0 . 27 and 4/ ( 0 . 25 × 69 ) = 0 . 23 of infected travellers who knew their exposure history reported so on the questionnaire . Based on this , in our main analysis we assumed a 25% probability of honest self-reporting of exposure risk in each questionnaire . The assumption of independent decisions on each questionnaire is optimistic: it allows travellers who did not report honestly at departure to report honestly at arrival . The estimated probability is also optimistic , since we used a low-end estimate of the fraction of travellers who knew their exposure history: if more travellers were aware of their exposure , our estimate for the proportion who reported correctly would have been lower . We used a probabilistic model to assess the influence of pathogen natural history and epidemiological factors on screening outcomes . Upon airport arrival , we assumed that passengers pass through screening for fever , followed by screening for risk factors ( Figure 1A ) . We assumed a one-strike policy: infected passengers who were identified by any single screening test were successfully caught by the screening program . We used the incubation period distribution to estimate the proportion of passengers who progressed to symptom onset in flight . The number of opportunities to detect each infected traveller varied depending on whether they displayed the symptom of fever and whether they knew their exposure history . We assumed passengers who did not present with fever would always pass through symptom screening , but could still be identified during questionnaire screening ( Figure 1B ) . Passengers who are not aware of exposure risk will always pass through questionnaire screening ( Figure 1C ) , and passengers with neither fever nor knowledge of exposure will go undetected ( Figure 1D ) . The model is described in full in Figure 1—figure supplement 1 . Source code for model analyses can be found in Source Code 1 . This should be a citation for the source code file .	International air travel has contributed to the spread of several recent disease epidemics . For example , travelers infected with severe acute respiratory syndrome ( or SARS ) in 2003 carried the disease around globe . One infected air traveler can carry a disease to a new continent: in 2014 , a man infected with Ebola in West Africa flew to the United States and infected two healthcare workers in Dallas during treatment . Efforts to prevent the spread of SARS , Ebola and other disease outbreaks have included screening air passengers for infection prior to boarding , or immediately after arrival . In these situations , infrared thermometers are often used to check for symptoms of fever and passengers may be asked to fill out questionnaires to assess their risk of exposure to the disease . However , the effectiveness of these airport screenings is questionable . 1000s of air travelers have been screened during several recent disease outbreaks , but few disease cases were detected . There are many reasons why an infected individual may be missed in airport screens . Passengers who have recently been infected may not yet display any symptoms and some passengers may be able to hide a fever or other symptoms by taking medication . Even if an individual has a fever , infrared thermometers will only detect it about 70% of the time . Also , screening questionnaires may miss passengers who are infected if they lie about any possible exposure to the disease . Gostic et al . created a mathematical model to help assess how useful airport screening is for detecting cases of disease caused by the SARS coronavirus , Ebola , influenza H1N1 and several other viruses . The model reveals that the effectiveness of airport screening depends on several factors including: how long it takes for symptoms to develop after infection ( the incubation period ) , how much is known about the virus and how it spreads , and whether the epidemic is still growing in size or is starting to slow down . For influenza H1N1 and other viruses with short incubation periods , fever screening is the most successful method to detect cases throughout the epidemic . However , for viruses with long incubation periods—such as Ebola—questionnaires are more useful in the early stages of an epidemic when the number of cases is rapidly rising . Fever screening becomes more useful later in the epidemic when new cases start to fall because the people who are infected are more likely to be displaying symptoms . Even so , Gostic et al . point out that in all of these scenarios airport screening will still miss many infected passengers . Thus , a challenge for future outbreaks will be to identify situations in which screening is worthwhile , and obtain better measurements of the factors that influence detection rates .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "epidemiology", "and", "global", "health" ]	2015	Effectiveness of traveller screening for emerging pathogens is shaped by epidemiology and natural history of infection
Agricultural soil harbors a diverse microbiome that can form beneficial relationships with plants , including the inhibition of plant pathogens . Pseudomonas spp . are one of the most abundant bacterial genera in the soil and rhizosphere and play important roles in promoting plant health . However , the genetic determinants of this beneficial activity are only partially understood . Here , we genetically and phenotypically characterize the Pseudomonas fluorescens population in a commercial potato field , where we identify strong correlations between specialized metabolite biosynthesis and antagonism of the potato pathogens Streptomyces scabies and Phytophthora infestans . Genetic and chemical analyses identified hydrogen cyanide and cyclic lipopeptides as key specialized metabolites associated with S . scabies inhibition , which was supported by in planta biocontrol experiments . We show that a single potato field contains a hugely diverse and dynamic population of Pseudomonas bacteria , whose capacity to produce specialized metabolites is shaped both by plant colonization and defined environmental inputs . Plant pathogenic microorganisms are responsible for major crop losses worldwide and represent a substantial threat to food security . Potato scab is one of the main diseases affecting potato quality ( Larkin et al . , 2011 ) and presents a significant economic burden to farmers around the world . The Gram-positive bacterium Streptomyces scabies , which is the causal organism of potato scab , is ubiquitous and presents a threat in almost all soils ( Bignell et al . , 2010; Lerat et al . , 2009 ) . Properly managed irrigation is a reasonably effective control measure for potato scab . However , scab outbreaks still regularly occur in irrigated soil , and with increasing pressures on water use it is clear that alternative approaches to the control of scab are needed . An attractive potential alternative involves the exploitation of soil microorganisms that suppress or kill plant pathogens , known as biocontrol agents ( Köhl et al . , 2019; Weller , 2007 ) . Many soil-dwelling Pseudomonas species form beneficial relationships with plants , positively affecting nutrition and health ( Cheng et al . , 2017; Loper et al . , 2012; Zamioudis et al . , 2013 ) and exhibiting potent antagonistic behavior towards pathogenic microorganisms ( Biessy et al . , 2019; Haas and Défago , 2005; Weller , 2007 ) . Pseudomonas influence the plant environment using a diverse range of secondary metabolites ( Arseneault et al . , 2013; Gross and Loper , 2009; Nguyen et al . , 2016; Stringlis et al . , 2018 ) and secreted proteins ( Ghequire and De Mot , 2014; Rangel et al . , 2016 ) . As such , Pseudomonas sp . have been identified as key biocontrol organisms in numerous plant-microbe systems ( Mauchline et al . , 2015; Wei et al . , 2019 ) , and these bacteria have potential applications as agricultural biocontrol agents and biofertilizers ( Kwak and Weller , 2013; Weller , 2007 ) . Many soil pseudomonads belong to the Pseudomonas fluorescens group , which consists of over 50 subspecies and exhibits huge phenotypic and genetic diversity ( Biessy et al . , 2019; Gomila et al . , 2015; Loper et al . , 2012; Melnyk et al . , 2019; Silby et al . , 2009 ) , with a core genome of about 1300 genes and a pan-genome of over 30 , 000 genes ( Garrido-Sanz et al . , 2016 ) . These bacteria use a variety of mechanisms to colonize the plant rhizosphere ( Little et al . , 2019 ) , communicate with plants ( Zamioudis et al . , 2013 ) , and suppress a range of plant pathogens ( Haas and Défago , 2005 ) , including bacteria ( Arseneault et al . , 2015 ) , fungi ( Michelsen et al . , 2015 ) , and insects ( Flury et al . , 2017 ) , although a single strain is unlikely to have all of these attributes . Specialized metabolites are critical to many of these ecological functions , and the Pseudomonas specialized metabolome is one of the richest and best characterized of any bacterial genus ( Gross and Loper , 2009; Nguyen et al . , 2016; Stringlis et al . , 2018 ) . Various studies have associated pseudomonads with potato scab suppression . A significant increase in the abundance of Pseudomonas taxa has been observed for irrigated fields , correlating with reduced levels of potato scab ( Elphinstone et al . , 2009 ) . Naturally scab-suppressive soils have also been shown to contain a greater proportion of Pseudomonas when compared to scab-conducive soils ( Meng et al . , 2012; Rosenzweig et al . , 2012 ) , and phenazine production by P . fluorescens can contribute to scab control ( Arseneault et al . , 2015; Arseneault et al . , 2013; Arseneault et al . , 2016 ) . Differences between soil microbial populations that enable effective pathogen suppression are routinely assessed using amplicon sequencing ( Fierer , 2017; Rosenzweig et al . , 2012 ) . However , the heterogeneity of the P . fluorescens group limits the usefulness of these methods for observing changes at the species or even the genus level . To effectively determine the relationship between the soil Pseudomonas population and disease suppression , it is important to accurately survey genotypic and phenotypic variability at the level of individual isolates , and to determine how this variation is linked to agriculturally relevant environmental changes ( Mauchline and Malone , 2017 ) . To investigate the genetic bases for S . scabies inhibition by P . fluorescens and to assess whether the scab-suppressive effects of irrigation derive from increased populations of biocontrol genotypes in the soil or on the plant , we focused on the Pseudomonas population from a potato field susceptible to potato scab . We first employed a phenotype-genotype correlation analysis across P . fluorescens strains isolated from a single potato field . We hypothesized that an unbiased correlation analysis would identify genetic loci and biosynthetic gene clusters ( BGCs ) that may be overlooked by screening for bioactive small molecules or by focusing on the biosynthetic repertoire of a limited number of strains . Here , we correlated phylogeny , phenotypes , specialized metabolism , and accessory genome loci , then investigated the importance of strong correlations by genetic manipulation of selected wild isolates . In total , 432 Pseudomonas strains were phenotyped ( with 69 whole genomes sequenced ) . This approach also enabled us to answer a number of ancillary questions: how diverse is the P . fluorescens population from a single field location ? Do the phenotypes associated with a P . fluorescens strain correlate with its biosynthetic capacity ? What does irrigation do to both the population structure of the P . fluorescens group and to the wider bacterial community ? Using this approach , we identify the P . fluorescens genes , gene clusters , and natural products that are required for potato pathogen suppression in vitro . We use this data to inform the discovery that the cyclic lipopeptide ( CLP ) tensin is a key determinant of in planta pathogen suppression by a Pseudomonas species . We show that irrigation induces profound and repeatable changes in the microbiome , both on a global level and within the P . fluorescens species group . Finally , we propose a model for the relationship between irrigation , pathogen suppression , and population-level shifts within the plant-associated P . fluorescens population . The ability of irrigation to protect root vegetables against S . scabies infection is agriculturally important and widespread , but poorly understood . It is likely that the irrigated soil microbiota plays a role in mediating scab suppression , but how this occurs is unclear . We therefore assessed the impact of irrigation on the total bacterial population of a commercial potato field in the United Kingdom . Multiple soil samples were taken from two sites ( A1 and B1 ) within this field , immediately prior to potato planting in January . Following potato planting , one site was irrigated as normal ( site A ) , while the second was protected from irrigation ( site B ) . Tuber-associated soil was sampled from both sites in May ( A2 and B2 ) when tubers were just forming , and the plants were most susceptible to S . scabies infection . Total genomic DNA was then extracted from replicate samples of each site after each sampling event , and 16S rRNA amplicon sequencing was used to examine the bacterial population in each of these sites ( Figure 1 ) . We used voom ( Law et al . , 2014 ) with a false discovery rate ( FDR ) of 0 . 05 to assess population changes across the four sampling sites . In total , changes were observed for 26 bacterial orders ( Figure 1A ) , with the most significant changes observed between January and May regardless of irrigation . This partially reflects an increase in bacterial orders that have previously been associated with the potato root microbiome ( Pfeiffer et al . , 2017; Weinert et al . , 2011 ) , including Rhizobiales , Sphingomonadales , and Flavobacteriales . Significant population changes ( FDR < 0 . 05 ) were also observed for eight bacterial orders between the irrigated ( A2 ) and nonirrigated ( B2 ) sites ( Figure 1B and C ) , including a larger proportion of Pseudomonadales bacteria in the irrigated site . In contrast , despite the potential for microbial heterogeneity across the fertilized field prior to planting , no significant changes were observed between pre-planting sites A1 and B1 . Taxonomic identifications using 16S rRNA amplicon analysis showed order-level changes to the field microbiome between sites ( Figure 1—figure supplement 1 ) but were unable to accurately capture diversity within genera or species groups . Therefore , biologically relevant variation within the populations of genetically diverse species groups such as P . fluorescens is potentially overlooked . To investigate the diversity of the fluorescent pseudomonad population , we isolated 240 individual Pseudomonas strains from our pre- and post-irrigation field sites ( Supplementary file 1 ) . These strains were screened for multiple phenotypes including motility , protease production , fluorescence ( siderophore production ) , and on-plate suppression of S . scabies using a cross-streak assay ( Figure 2—figure supplements 1 and 2 ) . Each phenotype was scored on an ordinal scale between 0 ( no phenotype observed ) and 3 ( strong phenotype ) . The cross-streak assay provided a rapid read-out of bacterial antagonism for both contact-dependent and diffusible mechanisms of growth inhibition . On-plate suppression of S . scabies was a surprisingly rare trait , with 79% of Pseudomonas isolates outcompeted by S . scabies in this assay . To determine whether this suppressive activity correlated with specific genetic loci , 69 isolates were selected for whole-genome sequencing , where almost half ( 32 strains ) exhibited on-plate suppression of S . scabies and the remaining strains represented a diverse selection ( based on phenotypic variation and 16S rRNA sequencing ) of nonsuppressive strains . We hypothesized that a comparative analysis of a similar number of genomes from suppressive and nonsuppressive strains would identify those BGCs that play important roles in suppressive activity . The phylogeny of the 69 sequenced strains was analyzed alongside various model pseudomonads , including representatives of the eight phylogenomic P . fluorescens groups defined by Garrido-Sanz et al . , 2016 . Our sequenced strain collection spans much of this characterized global phylogenetic diversity and contains representatives of at least five of the eight P . fluorescens phylogenomic groups ( Garrido-Sanz et al . , 2016 ) , as well as strains belonging to the Pseudomonas putida and Pseudomonas syringae groups ( Figure 2—figure supplement 3 ) . This genetic heterogeneity was also reflected in the diverse specialized metabolome of these strains , as predicted by a detailed analysis of the BGCs encoded in their genomes . Each genome was subjected to antiSMASH 5 . 0 analysis ( Blin et al . , 2019 ) , which was further refined by extensive manual annotation to improve the accuracy of predicted pathway products . This second annotation step was particularly important for BGCs that are atypically distributed across two distinct genomic loci ( e . g . , viscosin and pyoverdine ) . Our analysis was further expanded to include BGCs not identified by antiSMASH 5 . 0 , including BGCs for hydrogen cyanide ( HCN ) ( Pessi and Haas , 2000 ) , microcin B17-like pathways ( Metelev et al . , 2013 ) , and the auxin indole-3-acetic acid ( IAA ) ( McClerklin et al . , 2018; Palm et al . , 1989 ) . This was achieved by searching the genomes with a curated set of known Pseudomonas BGCs using MultiGeneBlast ( Medema et al . , 2013 ) ( see Appendix 1 for further details ) . This manual annotation provided a level of resolution superior to that provided by automated cluster-searching algorithms alone and provided confidence that the majority of natural product biosynthetic potential had been identified . Within a given pathway type ( e . g . , nonribosomal peptide synthetases [NRPSs] ) , likely pathway products were assigned where possible ( e . g . , CLPs ) or assigned a code when a conserved uncharacterized BGC was identified ( e . g . , NRPS 1 ) . All BGCs were mapped to strain phylogeny ( Figure 2 ) . Multiple BGCs were commonly found across the sequenced strains ( Figure 2 , Supplementary file 1 ) , including BGCs predicted to make CLPs ( Raaijmakers et al . , 2010 ) , arylpolyenes ( Cimermancic et al . , 2014 ) , and HCN ( Pessi and Haas , 2000 ) . In addition to pyoverdine BGCs ( Cézard et al . , 2015 ) in almost all strains , numerous other siderophore BGCs were identified , including pathways predicted to make achromobactin ( Berti and Thomas , 2009 ) , ornicorrugatin ( Matthijs et al . , 2008 ) , pyochelin-like molecules ( Patel and Walsh , 2001; Appendix 1—figure 1 ) , and a pseudomonine-like molecule ( Mercado-Blanco et al . , 2001 ) . A variety of polyketide synthase ( PKS ) , terpene , and NRPS BGCs with no characterized homologues were also identified ( Figure 2 ) . Furthermore , BGCs were identified that were predicted to make compounds related to microcin B17 ( Metelev et al . , 2013 ) , fosfomycin-like antibiotics ( Kim et al . , 2012 ) , lanthipeptides ( Repka et al . , 2017 ) , safracin ( Velasco et al . , 2005 ) , a carbapenem ( Coulthurst et al . , 2005 ) , and an aminoglycoside ( Kudo and Eguchi , 2009; Appendix 1—figure 3 ) . Each of these natural product classes is predicted to have potent biological activity and some are rarely found in pseudomonads . In addition to these potentially antibacterial and cytotoxic compounds , all genomes contain BGCs predicted to produce the plant auxin IAA , while 23 genomes contained genes for IAA catabolism ( Leveau and Gerards , 2008 ) . All 69 strains had at least one BGC for the production of the electron-transport cofactor pyrroloquinoline quinone ( PQQ ) ( Puehringer et al . , 2008; Appendix 1—figure 4 ) , reported to function as a plant growth promoter ( Choi et al . , 2008 ) . Surprisingly , BGCs for numerous well-characterized Pseudomonas specialized metabolites were not found , including phenazine , pyrrolnitrin , or 2 , 4-diacylphloroglucinol BGCs ( Gross and Loper , 2009 ) . In total , 787 gene clusters were identified that could be subdivided into 61 gene cluster families ( Figure 2 ) . The P . fluorescens species group possesses a highly diverse array of nonessential accessory genes and gene clusters . These are often critical to the lifestyle of a given strain and can include motility determinants , proteases , secretion systems , polysaccharides , toxins , and metabolite catabolism pathways ( Mauchline et al . , 2015 ) . These accessory genome loci were identified using MultiGeneBlast ( details in Appendix 1 ) , which revealed a high degree of genomic diversity across strains . Specialized metabolism BGCs and accessory genome loci were mapped to strain phylogeny ( Figure 2 ) , which indicated that for some loci ( e . g . , the psl operon , auxin catabolism , HCN biosynthesis ) there is a close , but not absolute , relationship between phylogeny and the presence of a gene cluster . We hypothesized that genes associated with suppression of S . scabies could be identified by a correlation analysis between S . scabies cross-streak inhibition and the presence of BGC families or accessory genes . We therefore calculated Pearson correlation coefficients for each BGC with S . scabies inhibition ( Figure 3—figure supplement 1 ) . The top 10 positively correlating genotypes and phenotypes ( Figure 3 ) comprised four BGC families ( Pep1 , CLP , Pep2 , and HCN ) ( Appendix 1—figure 2 ) , four accessory genome loci ( chitinase ChiC , protease AprA , chitinase class 1 , and the Pga operon ) , and two phenotypes ( motility and secreted protease production ) . The production of HCN and/or CLPs by Pseudomonas strains has been previously associated with the suppression of various plant pathogens including fungi ( Michelsen et al . , 2015; Zachow et al . , 2015 ) and oomycetes ( Hultberg et al . , 2010; Hunziker et al . , 2015 ) , and can also contribute to insect killing ( Flury et al . , 2017; Jang et al . , 2013 ) , but have not been linked to the suppression of bacteria . A variety of genotypes associated with plant-microbe interactions were moderately negatively correlated with suppression ( ρ < –0 . 3 ) , including BGCs for PQQ biosynthesis and catabolism of the plant auxins IAA and phenylacetic acid ( PAA ) ( Figure 3A ) . Interestingly , while certain BGC loci ( e . g . , CLP ) positively correlated with both suppression and motility , this relationship was not seen for every locus ( e . g . , HCN correlates with suppression but is less strongly correlated with motility ) . Correlation does not equate to causation , especially considering the significant evolutionary association seen for some BGCs ( Figure 2 ) . The importance of correlating BGCs to S . scabies suppression was therefore investigated experimentally using a genetically tractable subset of suppressive isolates . The strong positive correlation between putative CLP gene clusters and S . scabies suppression prompted us to investigate whether CLPs play a role in suppressive activity . Pseudomonas CLPs have previously been associated with a wide array of functions , including fungal growth inhibition , plant colonization , and promotion of swarming motility ( Alsohim et al . , 2014; Raaijmakers et al . , 2010 ) , although there are no reports of Pseudomonas CLPs functioning as inhibitors of streptomycete growth . However , prior work has shown that surfactin , a CLP from Bacillus subtilis , inhibits Streptomyces coelicolor aerial hyphae development ( Straight et al . , 2006 ) , while iturin A , a CLP from Bacillus sp . sunhua , inhibits S . scabies development ( Han et al . , 2005 ) . To determine the identity of each CLP , we combined bioinformatic predictions of the NRPS products ( Blin et al . , 2019 ) with experimental identification using liquid chromatography–tandem mass spectrometry ( LC-MS/MS ) . In every strain that contained a CLP BGC , a molecule with an expected mass and MS/MS fragmentation pattern was identified ( Figure 4 , Figure 4—figure supplements 1–7 ) . These data showed that P . fluorescens strains from a single field have the collective capacity to make viscosin ( m/z 1126 . 69 , identical retention time to a viscosin standard ) ( de Bruijn et al . , 2007 ) , a viscosin isomer ( m/z 1126 . 69 , different retention time to viscosin standard ) ( Figure 4—figure supplements 1 and 2 ) , as well as compounds with BGCs , exact masses , and MS/MS fragmentation consistent with tensin ( m/z 1409 . 85 , Figure 4—figure supplement 3; Nielsen et al . , 2000 ) , anikasin ( m/z 1354 . 81 , Figure 4—figure supplement 4; Götze et al . , 2017 ) , and putisolvin II ( m/z 1394 . 85 , Figure 4—figure supplement 5; Kuiper et al . , 2004 ) . In addition , an array of related metabolites were observed that differed by 14 or 28 Da , which is characteristic of different lipid chain lengths . This analysis also proved that the linear lipopeptides syringafactin A ( m/z 1082 . 74 ) and cichofactin ( m/z 555 . 38 , [M + 2 H]2+ ) were made by strains harboring BGCs predicted to make these phytotoxins ( Götze et al . , 2019; Pauwelyn et al . , 2013; Figure 4—figure supplements 6 and 7 ) . The metabolic capacity of all strains was mapped using mass spectral networking ( Aron et al . , 2020; Wang et al . , 2016 ) , which showed that CLPs were strongly associated with strains that inhibit S . scabies ( Figure 4 ) . To assess the potential role of CLPs in mediating the interaction between P . fluorescens and S . scabies , an NRPS gene predicted to be involved in the biosynthesis of a viscosin-like molecule in Ps682 was deleted by allelic replacement ( Figure 5A ) . The resulting Ps682 ∆visc strain was unable to make the viscosin-like molecule ( m/z 1126 . 69 , Figure 5B ) or to undergo swarming motility ( Figure 5—figure supplement 1 ) . This is in agreement with earlier work on the role of viscosin in the motility of P . fluorescens SBW25 ( Alsohim et al . , 2014 ) and the observation that possession of a CLP BGC was the genotype that most strongly correlated with motility ( ρ = 0 . 65 , Figure 3—figure supplement 1 ) . A cross-streak assay with S . scabies revealed an active role for this CLP in on-plate S . scabies inhibition ( Figure 5D ) . Wild-type ( WT ) Ps682 appeared to specifically colonize the S . scabies streak , whereas Ps682 ∆visc was unable to restrict S . scabies growth . Alternatively , it was possible that this instead could reflect diffusible inhibition of Streptomyces development by WT Ps682 , leading to a ‘bald’ S . scabies phenotype ( Flärdh and Buttner , 2009 ) . To distinguish between these possible inhibition modes , a constitutively expressed lux operon was integrated into the chromosomal att::Tn7 site ( K . -H . Choi et al . , 2005 ) of Ps682 to visualize this interaction by bioluminescence . This clearly showed viscosin-dependent Pseudomonas colonization of the Streptomyces streaks ( Figure 5D ) . To quantitatively assess the antagonistic effect of the Ps682 CLP , it was purified and structurally characterized using MS/MS ( Figure 5—figure supplement 2 ) and nuclear magnetic resonance ( NMR ) spectroscopy ( 1H , 13C , COSY HSQC , TOCSY , HMBC , Figure 5—figure supplements 3–13 , Supplementary file 2D ) . NMR analysis revealed that the molecule has an identical amino acid composition to viscosin ( 3-hydroxydecanoic acid-Leu1-Glu2-Thr3-Val4-Leu5-Ser6-Leu7-Ser8-Ile9 , Figure 5B ) , which was fully supported by detailed high-resolution MS ( calculated viscosin [M + H]+ = 1126 . 6970 , observed [M + H]+ = 1126 . 6964 ) and MS/MS fragmentation data ( Figure 5—figure supplement 2 ) . The LC retention time of this CLP is different to viscosin , but is almost identical to WLIP ( Figure 5—figure supplement 2C ) , which is a viscosin isomer that has a D-Leu5 residue instead of L-Leu5 ( Rokni-Zadeh et al . , 2012 ) . However , comparison of NMR data in DMF-d7 revealed some minor shift differences between published WLIP spectra ( Rokni-Zadeh et al . , 2012 ) and the Ps682 CLP , such as the γ-CH2 group of Glu2 ( WLIP = δH 2 . 54 ppm , δC 30 . 3 ppm; Ps682 CLP = δH 2 . 24 ppm , δC 34 . 8 ppm ) . Therefore , we could not conclusively confirm the absolute configuration of the Ps682 CLP and thus named it viscosin I ( for viscosin Isomer ) . A disk diffusion assay of purified viscosin I with S . scabies ( Figure 5C ) demonstrated that it directly inhibited S . scabies growth with a minimum inhibitory concentration of approximately 20 µg/mL . Long-term growth of S . scabies in the presence of viscosin I ( Figure 5—figure supplement 14 ) indicated that the inhibition of S . scabies is temporary and growth partially resumes after several days . These data show that in addition to its role as a surfactant viscosin I functions by inhibiting the growth rate of S . scabies , consistent with the on-plate data for Ps682 ∆visc . Pan-genome analysis showed that HCN production was predicted for a significant number of suppressive strains ( ρ = 0 . 47 , Figure 3—figure supplement 1 ) , where 17 of the 19 strains containing HCN gene clusters were inhibitory towards S . scabies ( Figure 2 ) . The HCN pathway is encoded by the hcnABC gene cluster ( Appendix 1—figure 2 ) and has previously been associated with insect and fungal pathogen inhibition in other Pseudomonas strains ( Flury et al . , 2017; Hunziker et al . , 2015; Siddiqui et al . , 2006 ) . HCN is toxic to a wide variety of organisms , but not to Pseudomonas owing to their branched aerobic respiratory chain that has at least five terminal oxidases , including a cyanide-insensitive oxidase ( Comolli and Donohue , 2002; Ugidos et al . , 2008 ) . We confirmed that nearly every strain with the hcnABC gene cluster produced HCN ( 18 out of 19 ) using the Feigl–Anger colorimetric detection reagent ( Feigl and Anger , 1966; Supplementary file 1 ) and used this assay to identify HCN producers across the original collection of 240 Pseudomonas strains . This wider analysis showed that HCN production strongly correlated with S . scabies inhibition ( ρ = 0 . 52 , Figure 6—figure supplement 1 ) , in accordance with our analysis of the sequenced strains . To examine the role of HCN in S . scabies suppression and whether it exhibited a synergistic effect with CLP production , Ps619 was investigated as this strain produces both HCN and a tensin-like CLP ( Figures 2 and 6A ) . A tensin BGC has not previously been reported , but the predicted amino acid specificity , mass ( Figure 6A ) , and MS/MS fragmentation ( Figure 4—figure supplement 3 ) indicated that seven isolates produce tensin-like CLPs ( Figure 4 ) . The hcn and ten gene clusters were inactivated by in-frame deletions to generate single and double mutants of Ps619 , and the resulting Δhcn , Δten , and ΔhcnΔten mutants were subjected to cross-streak assays ( Figure 6B ) . A comparison of WT , single , and double mutants showed that HCN inhibits S . scabies growth and development across the entire plate , while tensin is important for Pseudomonas motility and helps the Pseudomonas to grow competitively at the cross-streak interface . Furthermore , this suppressive effect is additive: the Ps619 Δhcn and Δten single mutants both retained some inhibitory activity towards S . scabies , whereas the Ps619 ΔhcnΔten double mutant could not inhibit S . scabies . In drier growth conditions expected to favor streptomycete growth and limit motility , the role of tensin-mediated motility was abrogated , yet tensin and HCN still possessed an additive inhibitory effect at the microbial interface ( Figure 6—figure supplement 2A ) . Notably , Ps619 Δhcn was able to induce a developmental defect in S . scabies at the microbial interface that was not present in Ps619 Δten or Ps619 ΔhcnΔten , showing that the tensin-like CLP induces a developmental defect in S . scabies that is independent of Pseudomonas motility , comparable to the inhibitory effect of isolated viscosin I . This analysis also clearly showed that at areas distant from the bacterial interaction S . scabies grew more vigorously when cultured with Δhcn strains , consistent with the volatility of HCN enabling a long-range inhibitory effect . A similar volatile effect was seen when Ps619 strains were separated from S . scabies by a barrier , where only those strains producing HCN inhibited growth and development ( Figure 6—figure supplement 2B ) . To further probe how tensin and HCN affected the interaction between Ps619 strains and S . scabies , the interfacial regions of cross-streaks were imaged using cryo-scanning electron microscopy ( cryo-SEM ) . WT Ps619 was able to colonize the S . scabies streak , meaning that the interfacial region imaged was further from the cross-streak intersection than all other co-cultures ( Figure 6C ) . Here , Ps619 inhibited S . scabies development , which appears as a mixture of deformed aerial hyphae and vegetative growth reminiscent of a ‘bald’ phenotype ( Tschowri et al . , 2014 ) . Cryo-SEM indicated that both the Ps619 Δhcn and Δten mutants induced a similar partially bald phenotype in S . scabies , but the ΔhcnΔten double mutant was unable to trigger the same developmental defect as S . scabies could develop aerial mycelia close to the microbial interface ( Figure 6C , Figure 6—figure supplement 3 ) . This appears as a clear boundary between Ps619 ΔhcnΔten ( single cells in background , bottom right panel of Figure 6C ) and S . scabies ( hyphae in the foreground ) . The volatile HCN can inhibit growth and development at a distance , whereas CLP inhibition of development only occurs close to the microbial interface . Both inhibitory mechanisms enable Ps619 to obtain a competitive advantage at the microbe-microbe interface ( Figure 6C , Figure 6—figure supplement 3 ) , while the CLP also functions as a surfactant enabling Ps619 motility , promoting Pseudomonas invasion of the Streptomyces cross-streak . To examine the in planta biocontrol properties of Ps619 and Ps682 , and to determine the contribution of HCN and CLPs to activity , potato scab suppression assays were carried out in glasshouse trials . Maris Piper potatoes were infected with S . scabies 87-22 and scored for disease severity after 16 weeks using the method of Andrade et al . , 2019 . A subset of plants was also treated with Pseudomonas spp . and associated BGC mutants . Ps619 conferred significant protection against potato scab , where disease severity was reduced to levels similar to uninfected control plants ( Figure 7 ) . This suppressive ability was lost for Ps619 Δten and Ps619 ΔhcnΔten , resulting in disease severity similar to scab-infected tubers . In contrast , Ps619 Δhcn was just as effective as WT Ps619 at suppressing potato scab , which differed from the on-plate results for HCN . The significance of these results was supported by an independent in planta biocontrol experiment , where equivalent results were obtained for each strain ( Figure 7—figure supplement 1 ) . This result indicates that tensin plays an important role in the biocontrol of potato scab . In contrast to its on-plate suppressive activity , potato scab assays showed no significant antagonistic activity for Ps682 against S . scabies infection . Unfortunately , this meant that the role of viscosin I could not be determined in planta . To determine whether the strains and metabolites we identified have suppressive activity towards a range of plant pathogens , we investigated the ability of the potato field strain collection to suppress the growth of Phytophthora infestans , the oomycete that causes potato blight ( Nowicki et al . , 2012 ) , and Gaeumannomyces graminis var . tritici , the fungus that causes take-all disease of cereal crops ( Mauchline et al . , 2015 ) . These assays revealed strong congruence between the genotypes that correlated with suppression of each pathogen ( Appendix 2—figure 1A ) . HCN and CLPs have both previously been identified as inhibitors of oomycete and fungal growth ( Hunziker et al . , 2015; Michelsen et al . , 2015 ) . To assess whether these natural products are critical for inhibition of P . infestans and G . graminis by Ps619 and Ps682 , the HCN/CLP mutants were tested for inhibitory activity ( Appendix 2—figure 1 ) . Surprisingly , neither HCN or tensin were required for Ps619 inhibition of either G . graminis or P . infestans ( Appendix 2—figure 1 ) , indicating the production of at least one other secreted inhibitory factor . In contrast , inactivation of the viscosin I pathway in Ps682 abolished activity towards both pathogens ( Appendix 2—figure 1D ) . These data indicate that a subset of pseudomonads can function as generalist pathogen suppressors , possessing multiple growth inhibition mechanisms ( e . g . , Ps619 ) and/or by producing molecules with broad bioactivity ( e . g . , Ps682 ) . Multiple other genome loci are strongly correlated with pathogen suppression ( Figure 3A , Appendix 2—figure 1A ) , including chitinases ( Folders et al . , 2001 ) and the extracellular metalloprotease AprA ( Laarman et al . , 2012 ) . Phenotypically , extracellular protease activity also positively correlates with suppression . The BGC that correlated most strongly with S . scabies suppression was Pep1 ( ‘Peptide 1’ ) , while the related Pep2 also correlated strongly ( Figure 3A ) . These were identified by antiSMASH as putative ‘bacteriocin’ BGCs and encode short DUF2282 peptides alongside DUF692 and DUF2063 proteins ( Appendix 1—figure 2 ) . The DUF692 protein family includes dioxygenases involved in methanobactin ( Kenney et al . , 2018 ) and 3-thiaglutamate ( Ting et al . , 2019 ) biosynthesis . Other studies indicate that DUF692 and DUF2063 proteins may be involved in heavy metal and/or oxidative stress responses ( Clark et al . , 2014; Price et al . , 2018; Sarkisova et al . , 2014 ) . Further work is required to determine the significance of both the Pep BGCs and the accessory genome loci for pathogen inhibition . Irrigation is currently the only effective way to control potato scab , so we hypothesized that this may lead to an increase in the number of inhibitory bacteria associated with the soil and/or tuber , especially as the Pseudomonadales population moderately increased in irrigated soil ( Figure 1C ) . However , a greater number of strongly suppressive strains ( inhibition score ≥2 ) were isolated from nonirrigated sites ( 7/60 strains ) than from irrigated sites ( 1/60 strains ) . A similar pattern was observed for strongly motile ( score ≥2 ) strains ( six nonirrigated versus two irrigated ) . Analysis of the BGCs in our sequenced strains revealed a similar result , where 5/18 unirrigated strains contained CLP BGCs versus 0/16 irrigated strains . This counterintuitive observation led us to hypothesize that irrigation enables nonmotile , nonsuppressive bacteria to survive and colonize plant roots , whereas highly motile bacteria that produce multiple biological weapons can more effectively colonize plants in drier , more ‘hostile’ conditions . To test these hypotheses , we sampled irrigated and unirrigated sites in a neighboring field 2 years after the first sampling event . 48 strains were isolated from bulk soil and the rhizospheres of tuber-forming potato plants , with and without irrigation , providing a total of 192 P . fluorescens strains ( Supplementary file 1 ) . These strains were scored for motility , HCN production , and S . scabies suppression ( Figure 8A ) . Our results were in strong agreement with the first sample set , including strong positive correlations between S . scabies inhibition , motility , and HCN production ( Figure 8B ) . A negative correlation was observed between irrigation and S . scabies suppression on the plant roots , but not in the surrounding soil . This appeared to be driven primarily by differences in the unirrigated samples , where a substantially greater proportion of suppressive isolates were associated with roots than with the surrounding soil . We observed a strong positive correlation between motility and root association for unirrigated samples , while the reverse was true for irrigated plants ( Figure 8A ) . This effect of irrigation on the distribution of motile bacteria was striking – in dry plants , the motile population was almost entirely associated with roots , while in irrigated plants a comparable proportion of motile bacteria were found in the soil and roots ( Figure 8A ) . This analysis therefore supports the root colonization hypothesis , where a lack of irrigation leads to a more specialized pseudomonad population colonizing the root . Upon irrigation , the difference between the bulk soil and root pseudomonad populations is much less significant . The mechanism for this population change is not yet defined , and these changes are counterintuitive in relation to the suppression of potato scab upon irrigation , given there is a drop in suppressive strains colonizing the potato root following irrigation . Irrigation did lead to moderately more motile pseudomonads in bulk soil versus unirrigated conditions , but this was not associated with more suppressive strains or HCN producers ( Figure 8A and B ) . The mechanism and significance of this irrigation effect require further investigation . It is possible that a protective microbiome in irrigated conditions actually contains a mixture of S . scabies-suppressive ‘biocontrol’ Pseudomonas strains alongside other nonmotile pseudomonads that interact with the plant in important ways due to traits usually absent from the ‘biocontrol’ strains , such as their ability to produce PQQ and catabolize auxins ( Choi et al . , 2008; Leveau and Gerards , 2008 ) . Profound irrigation-associated changes in antibiotic-producing Pseudomonas populations have previously been observed for the wheat rhizosphere ( Mavrodi et al . , 2018; Mavrodi et al . , 2012 ) . Prior studies on the suppression of potato scab have indicated a potential biocontrol role for Pseudomonas bacteria ( Arseneault et al . , 2015; Arseneault et al . , 2013; Elphinstone et al . , 2009; Rosenzweig et al . , 2012 ) . Fluorescent pseudomonads form multiple beneficial relationships with plants , including growth promotion and biocontrol ( Haas and Défago , 2005; Zamioudis et al . , 2013 ) . However , there is limited understanding of the genetic factors that are critical for such activity , and little is known about the diversity of the P . fluorescens species group within a given agricultural field or how this population is shaped by environmental changes . In this study , we integrated genomics , metabolomics , phenotypic analysis , molecular biology , and in planta assays to identify the genetic determinants of Pseudomonas antagonism towards S . scabies . This population-level approach shows that the P . fluorescens population in a single field is highly complex , heterogeneous , and dynamic ( Figures 2 , 3 and 8 ) , where the overall genotypic diversity is similar to the global diversity of P . fluorescens ( Garrido-Sanz et al . , 2016 ) . Pan-genome analysis and metagenomics represent increasingly powerful routes to understanding the genetic determinants of biological activity in plant-associated microbes ( Beskrovnaya et al . , 2020; Biessy et al . , 2019; Carrión et al . , 2019; Melnyk et al . , 2019; Mullins et al . , 2019; Tracanna et al . , 2021 ) . Multiple BGCs and accessory genome loci were identified that correlated with on-plate inhibition of S . scabies growth and development ( Figure 3 ) , including BGCs for CLPs and HCN . These loci also correlated with inhibition of P . infestans and G . graminis , and their contribution to suppression was validated genetically . This confirmed a role for both molecules in S . scabies inhibition ( Figures 5 and 6 ) , representing a new function for these Pseudomonas specialized metabolites . Co-culture assays and cryo-SEM imaging ( Figure 6 ) showed that HCN and a tensin-like CLP produced by Ps619 arrest the formation of streptomycete aerial hyphae and subsequent sporulation , providing the pseudomonad with a competitive advantage at the microbial interface . In planta experiments confirmed that Ps619 could suppress potato scab and that CLP production was a key determinant of this inhibitory effect ( Figure 7 ) . In contrast , HCN production was not a requirement for potato scab suppression by Ps619 . It is possible that HCN is not produced in sufficient amounts during root colonization for S . scabies inhibition , or that it instead has an alternative natural role , such as metal chelation ( Rijavec and Lapanje , 2016 ) . The roles of the CLPs are reminiscent of the interaction between B . subtilis and S . coelicolor , where the CLP surfactin functions as a surfactant required for the formation of aerial structures in B . subtilis and arrests aerial development in S . coelicolor ( Straight et al . , 2006 ) . Collectively , these results are surprising given that streptomycetes themselves use surfactants to assist in the erection of aerial mycelia ( Kodani et al . , 2004; Willey et al . , 2006 ) and points to secondary antagonistic roles for these molecules beyond the reduction of surface tension . This is strongly supported by the inhibitory effect of purified viscosin I towards S . scabies ( Figure 5C ) . HCN and CLPs have also been associated with insect ( Flury et al . , 2017 ) and nematode ( Siddiqui et al . , 2006 ) killing , as well as the suppression of pathogenic fungi ( Michelsen et al . , 2015; Fukuda et al . , 2021 ) . This indicates that a subset of pseudomonads are generalist suppressors of pathogens ( and presumably also nonpathogenic organisms ) due to the production of these broad range antimicrobials . Genetic analysis indicates that these strains are more likely to produce multiple suppressive metabolites and proteins . Evidence for this is provided by the inhibition of both P . infestans and G . graminis by Ps619 ΔhcnΔten ( Appendix 2—figure 1 ) . A study of the inhibitory properties of bacteria associated with the Arabidopsis leaf microbiome showed that a large proportion of the total inhibitory activity was due to Pseudomonadales bacteria and that a subset of individual strains were active against a wide array of bacteria ( Helfrich et al . , 2018 ) . Unexpectedly , irrigation led to a decrease in the proportion of suppressive pseudomonads on potato roots ( Figure 8A ) even though irrigation is one of the most effective ways to suppress potato scab . One possible reason for this discrepancy is that irrigation enables nonsuppressive Pseudomonas spp . with low motility to be transported to plant roots more effectively . Recruitment of nonsuppressive pseudomonads to the rhizosphere may benefit the plant in other ways , such as immune system priming ( Bakker et al . , 2007; Teixeira et al . , 2021 ) or modulation of auxin biosynthesis . For example , Cheng et al . , 2017 showed that auxin biosynthesis was linked to plant growth promotion and induced systemic resistance by P . fluorescens SS101 . An alternative hypothesis is that changes in the overall relative abundance of soil Pseudomonas over Streptomyces resulting from irrigation may override the observed shift towards less-suppressive Pseudomonas genotypes . In support of this , drought-induced enrichment for commensal Streptomyces and depletion of Proteobacteria in sorghum and rice plants have been shown to be reversed by irrigation ( Santos-Medellín et al . , 2021; Xu et al . , 2018 ) . In this model , irrigation may reduce the relative fitness of S . scabies versus Pseudomonas spp . , while the microbiome of irrigated roots simultaneously becomes less optimal for disease suppression . Our data show that Ps619 is highly effective at inhibiting potato scab , yet Ps619-like strains are naturally less abundant in irrigated conditions . Therefore , possible future efforts to control potato scab could combine irrigation with pretreatment with effective biocontrol strains , like Ps619 , to ensure tubers are colonized by a significant proportion of biocontrol strains . Such a strategy could reduce the quantity of water required for effective scab suppression . While our study was focused on fluorescent pseudomonads , interactions between these bacteria and the wider microbiome ( Figure 1 ) may also have a key role in potato scab suppression . Moving forward , systematic analyses of individual organisms within microbiomes will continue to help answer questions relating to microbial communities and host interactions that are difficult to address using global ‘omics approaches alone . For example , the role of many bacterial specialized metabolites in nature is poorly understood , especially for prolific producers such as the pseudomonads and the streptomycetes ( van der Meij et al . , 2017 ) . Future studies could examine whether the host selects for bacterial populations enriched in specific BGCs and whether environmental stimuli modulate the abundance of these BGCs . Synthetic microbial communities based on well-characterized natural communities could then be used to test hypotheses on the role of specialized metabolites in shaping the community or modulating the health of the host organism . All strains used in this study are listed in Supplementary file 2A . Unlessfigurotherwise stated , chemicals were purchased from Sigma-Aldrich , enzymes from New England Biolabs , and molecular biology kits from GE Healthcare and Promega . All P . fluorescens strains were grown at 28°C in L medium ( Luria base broth , Formedium ) and Escherichia coli at 37°C in lysogeny broth ( LB ) ( Miller , 1972 ) . 1 . 3% agar was added for solid media . Gentamicin was used at 25 μg/mL , carbenicillin at 100 μg/mL , and tetracycline ( Tet ) at 12 . 5 μg/mL . S . scabies spore suspensions were prepared using established procedures ( Kieser et al . , 2000 ) . Soil samples were collected from potato fields at RG Abrey Farms ( East Wretham , Norfolk , UK , 52 . 4644° N , 0 . 8299° E ) . The first sampling was conducted in 2015 from two adjacent plots in a single field . Soil samples were taken on 22 January 2015 , immediately prior to planting . One plot was then covered loosely in polythene to protect it from irrigation . The same field sites were sampled again in May at the point of maximum scab impact , once potato tubers had begun to form . In this case , soil samples were taken from the base of the plants , near the root system . For each sampling event , a total of 12 samples were taken from three parallel potato beds at regularly spaced intervals approximately 1 m apart . For the second experiment , 12 irrigated and 12 nonirrigated potato plants were uprooted from field sites in June 2017 and returned to the laboratory in large pots . Bulk soil samples were taken from these pots alongside an equivalent number of rhizosphere-associated samples , which were defined as isolated root systems gently shaken to remove bulk soil before processing as below . Samples were collected in sterile 50 mL tubes and stored at 4°C . Sample processing was conducted at 4°C throughout . 10 mL of sterile phosphate-buffered saline ( PBS , per liter: 8 g NaCl , 0 . 2 g KCl , 1 . 44 g Na2HPO4 , 0 . 24 g KH2PO4 , pH 7 . 4 ) were added to 50 mL tubes containing 20 g of soil or root material , and vortexed vigorously for 10 min . Samples were then filtered through a sterile muslin filter to remove larger debris . The resulting suspension of soil and organic matter was centrifuged at 1000 rpm for 30 s to pellet remaining soil particles , before serial dilution in PBS and plating on Pseudomonas selective agar . The selection media comprised Pseudomonas agar base ( Oxoid , UK ) supplemented with CFC ( cetrimide/fucidin/cephalosporin ) Pseudomonas selective supplement ( Oxoid , UK ) . Plates were incubated at 28°C until colonies arose , then isolated single colonies were patched on fresh CFC agar and incubated overnight at 28°C before streaking to single colonies on King’s B ( KB ) agar plates ( King et al . , 1954 ) . Six isolates were selected at random per soil sample and subjected to phenotypic/genomic analysis . Genomic DNA was isolated from 3 g of pooled soil samples using the FastDNA SPIN Kit for soil ( MP Biomedicals , UK ) following the manufacturer’s instructions . Genomic DNA concentration and purity was determined by NanoDrop spectrophotometry as above . Microbial 16S rRNA genes were amplified from soil DNA samples with barcoded universal prokaryotic primers ( F515/R806 ) targeting the V4 region ( Caporaso et al . , 2011 ) , and then subjected to Illumina MiSeq sequencing ( 600-cycle , 2 × 300 bp ) at the DNA Sequencing Facility , Department of Biochemistry , University of Cambridge ( UK ) . The data were analyzed using the MiSeq Reporter Metagenomics Workflow ( Illumina , UK ) to acquire read counts for all taxonomic ranks from phylum to genus . MiSeq data were visualized and analyzed using Degust 3 . 1 . 0 ( http://degust . erc . monash . edu/ ) and Pheatmap ( https://CRAN . R-project . org/package=pheatmap ) in R 3 . 5 . 1 . All phenotyping assays were conducted at least twice independently , and where disagreements were recorded in the ordinal data , additional repeats were conducted until a firm consensus was reached . Single colonies of each isolate to be sequenced were picked from L agar plates and grown overnight in L medium . DNA was then extracted from 2 mL of cell culture using a GenElute Bacterial Genomic DNA Kit ( Sigma-Aldrich , USA ) . DNA samples were subjected to an initial quality check using a NanoDrop spectrophotometer ( Thermo Scientific , Wilmington , DE ) before submission for Nextera library preparation and paired-end read sequencing on the Illumina MiSeq platform ( 600-cycle , 2 × 300 bp ) at the DNA Sequencing Facility , Department of Biochemistry , University of Cambridge ( UK ) . Reads from 35 pseudomonads collected in February 2015 were assembled into genomes using MaSuRCA v3 . 2 . 6 ( Zimin et al . , 2013 ) with the following settings: GRAPH_KMER_SIZE = auto; USE_LINKING_MATES = 1; LIMIT_JUMP_COVERAGE = 60; CA_PARAMETERS = ovlMerSize = 30 cgwErrorRate = 0 . 25 ovlMemory = 4 GB; NUM_THREADS = 16; JF_SIZE = 100000000; DO_HOMOPOLYMER_TRIM = 0 . Reads from 32 samples collected in May 2015 were assembled into genomes using SPAdes v3 . 6 . 2 ( Bankevich et al . , 2012 ) with k-mer flag set to -k 2133557799127 . All assembly tasks were conducted using 16 CPUs on a 256 GB compute node within the Norwich Bioscience Institutes ( NBI ) High Performance Computing cluster . An additional strain from May 2015 ( Ps925 ) was sequenced and assembled by MicrobesNG ( http://www . microbesng . uk ) , which is supported by the BBSRC ( grant number BB/L024209/1 ) . The 69 assembled genome sequences were annotated using Prokka ( Seemann , 2014 ) , which implements Prodigal ( Hyatt et al . , 2010 ) as an open-reading frame calling tool . Assembly qualities were assessed using CheckM ( Parks et al . , 2015 ) . Genome assemblies are available at the European Nucleotide Archive ( http://www . ebi . ac . uk/ena/ ) with the project accession PRJEB34261 . The gyrB housekeeping gene sequence was identified in each newly sequenced genome by BLAST comparison with the sequence of gyrB from P . fluorescens SBW25 . The full-length gyrB sequences from these strains and several reference strains were aligned using MUSCLE 3 . 8 . 31 ( Edgar , 2004 ) with default settings , then a maximum likelihood tree was calculated using RAxML 8 . 2 . 12 ( Stamatakis , 2014 ) on the CIPRES portal ( Miller et al . , 2015 ) with the following parameters: raxmlHPC-HYBRID-AVX -T 4f a -N autoMRE -n result -s infile . txt -c 25 m GTRCAT -p 12345k -x 12345 . Genomes were subjected to bioinformatic analysis as described in Appendix 1 . Phylogenetic trees and presence/absence data for accessory genes were visualized using Interactive Tree of Life ( iTOL ) ( Letunic and Bork , 2016 ) , with Pseudomonas aeruginosa PAO1 gyrB as the outgroup . Cloning was carried out in accordance with standard molecular biology techniques . P . fluorescens deletion mutants were constructed by allelic exchange as described previously ( Campilongo et al . , 2017 ) . Up- and downstream flanking regions ( approximately 500 bp ) to the target genes were amplified using primers listed in Supplementary file 2B . PCR products in each case were ligated into pTS1 ( Scott et al . , 2017 ) between XhoI and BamHI . The resulting deletion vectors were transformed into the target strains by electroporation , and single crossovers selected on L + Tet and re-streaked to isolate single colonies . 100 mL cultures in L medium from single crossovers were grown overnight at 28°C , then plated onto L + 10% sucrose plates to counter-select for double crossovers . Individual colonies from these plates were then patched onto L plates ± Tet , with Tet-sensitive colonies tested for gene deletion by colony PCR using primers external to the deleted gene in each case ( Supplementary file 2B ) . Luminescent-tagged strains were produced by introduction of the Aliivibrio fischeri luxCDABE cassette into the neutral att::Tn7 site in Pseudomonas chromosomes using the Tn7-based expression system described in Choi et al . , 2005 . Strains were electroporated with plasmids pUC18-miniTn7T-Gm-lux and the helper pTNS2 , and transformant colonies were grown on solid L medium+ gentamicin for 2–3 days at 28°C . Integration of the lux cassette into Pseudomonas genomes was confirmed by PCR and with a luminometer . Luminescent cells were then tracked using the NightOWL visualization system ( Berthold Technologies , Germany ) . All plasmids used in this study are reported in Supplementary file 2C . Pseudomonas isolates were grown overnight in L medium ( 10 mL ) for 16 hr at 28°C . 100 μL of each culture was used to inoculate 40 mm diameter KB agar plates . Plates were incubated for 24 hr at 28°C , before the agar from each plate was decanted into a sterile 50 mL tube and extracted with 10 mL 50% EtOH with occasional vortexing for 3 hr . 2 mL was taken from each sample and centrifuged in 2 mL tubes for 5 min at 16 , 000 × g . The supernatant was collected and stored at –80°C . Samples were diluted with an equal volume of water , then subjected to LC-MS analysis using a Shimadzu Nexera X2 UHPLC coupled to a Shimadzu ion-trap time-of-flight ( IT-TOF ) mass spectrometer . Samples ( 5 μL ) were injected onto a Phenomenex Kinetex 2 . 6 μm XB-C18 column ( 50 × 2 . 1 mm , 100 Å ) , eluting with a linear gradient of 5–95% acetonitrile in water +0 . 1% formic acid over 6 min with a flow rate of 0 . 6 mL/min at 40°C . To compare the retention times of viscosin I ( Ps682 ) , WLIP ( Pseudomonas sp . LMG 2338 ) , and viscosin ( P . fluorescens SBW25 ) , extracts were prepared from their producing organisms as described above . The same chromatography conditions as above were used , but with a linear gradient of 5–100% acetonitrile in water + 0 . 1% formic acid over 15 min . Positive mode mass spectrometry data were collected between m/z 300 and 2000 with an ion accumulation time of 10 ms featuring an automatic sensitivity control of 70% of the base peak . The curved desolvation line temperature was 300°C , and the heat block temperature was 250°C . MS/MS data were collected in a data-dependent manner using collision-induced dissociation energy of 50% and a precursor ion width of 3 Da . The instrument was calibrated using sodium trifluoroacetate cluster ions prior to every run . A molecular network was created using the online workflow at the Global Natural Product Social Molecular Networking ( GNPS ) site ( https://gnps . ucsd . edu/; Aron et al . , 2020 ) . The data were filtered by removing all MS/MS peaks within ±17 Da of the precursor m/z . The data were then clustered with MS-Cluster with a parent mass tolerance of 1 Da and an MS/MS fragment ion tolerance of 0 . 5 Da to create consensus spectra . Consensus spectra that contained less than two spectra were discarded . A network was then created where edges were filtered to have a cosine score above 0 . 6 and more than four matched peaks . Further edges between two nodes were kept in the network if each of the nodes appeared in each other’s respective top 10 most similar nodes . The spectra in the network were then searched against GNPS spectral libraries . The library spectra were filtered in the same manner as the input data . All matches kept between network spectra and library spectra were required to have a score above 0 . 7 and at least four matched peaks . Networks were visualized using Cytoscape v3 . 8 . 2 ( Shannon et al . , 2003 ) , and the data were manually filtered to remove duplicate nodes ( same m/z and retention time ) . The data are available as MassIVE dataset MSV000084283 at https://massive . ucsd . edu , and the GNPS analysis is available at https://gnps . ucsd . edu/ProteoSAFe/status . jsp ? task=51ac5fe596424cf88cfc17898985cac2 . High-resolution mass spectra were acquired on a Synapt G2-Si mass spectrometer equipped with an Acquity UPLC ( Waters ) . Aliquots of the samples were injected onto an Acquity UPLC BEH C18 column , 1 . 7 μm , 1 × 100 mm ( Waters ) and eluted with a gradient of acetonitrile/0 . 1% formic acid ( B ) in water/0 . 1% formic acid ( A ) with a flow rate of 0 . 08 mL/min at 45°C . The concentration of B was kept at 1% for 1 min followed by a gradient up to 40% B in 9 min , ramping to 99% B in 1 min , kept at 99% B for 2 min and re-equilibrated at 1% B for 4 min . MS data were collected in positive mode with the following parameters: resolution mode , positive ion mode , scan time 0 . 5 s , mass range m/z 50–1200 calibrated with sodium formate , capillary voltage = 2 . 5 kV; cone voltage = 40 V; source temperature = 125°C; desolvation temperature = 300°C . Leu-enkephalin peptide was used to generate a lock-mass calibration with 556 . 2766 , measured every 30 s during the run . For MS/MS fragmentation , a data-directed analysis ( DDA ) method was used with the following parameters: precursor selected from the four most intense ions; MS2 threshold: 5000; scan time 0 . 5 s; no dynamic exclusion . In positive mode , collision energy ( CE ) was ramped between 10–30 at low mass ( m/z 50 ) and 15–60 at high mass ( m/z 1200 ) .	Potato scab and blight are two major diseases which can cause heavy crop losses . They are caused , respectively , by the bacterium Streptomyces scabies and an oomycete ( a fungus-like organism ) known as Phytophthora infestans . Fighting these disease-causing microorganisms can involve crop management techniques – for example , ensuring that a field is well irrigated helps to keep S . scabies at bay . Harnessing biological control agents can also offer ways to control disease while respecting the environment . Biocontrol bacteria , such as Pseudomonas , can produce compounds that keep S . scabies and P . infestans in check . However , the identity of these molecules and how irrigation can influence Pseudomonas population remains unknown . To examine these questions , Pacheco-Moreno et al . sampled and isolated hundreds of Pseudomonas strains from a commercial potato field , closely examining the genomes of 69 of these . Comparing the genetic information of strains based on whether they could control the growth of S . scabies revealed that compounds known as cyclic lipopeptides are key to controlling the growth of S . scabies and P . infestans . Whether the field was irrigated also had a large impact on the strains forming the Pseudomonas population . Working out how Pseudomonas bacteria block disease could speed up the search for biological control agents . The approach developed by Pacheco-Moreno et al . could help to predict which strains might be most effective based on their genetic features . Similar experiments could also work for other combinations of plants and diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "genetics", "and", "genomics" ]	2021	Pan-genome analysis identifies intersecting roles for Pseudomonas specialized metabolites in potato pathogen inhibition
In cell culture , genetically identical cells often exhibit heterogeneous behavior , with only ‘lineage primed’ cells responding to differentiation inducing signals . It has recently been proposed that such heterogeneity exists during normal embryonic development to allow position independent patterning based on ‘salt and pepper’ differentiation and sorting out . However , the molecular basis of lineage priming and how it leads to reproducible cell type proportioning are poorly understood . To address this , we employed a novel forward genetic approach in the model organism Dictyostelium discoideum . These studies reveal that the Ras-GTPase regulator gefE is required for normal lineage priming and salt and pepper differentiation . This is because Ras-GTPase activity sets the intrinsic response threshold to lineage specific differentiation signals . Importantly , we show that although gefE expression is uniform , transcription of its target , rasD , is both heterogeneous and dynamic , thus providing a novel mechanism for heterogeneity generation and position-independent differentiation . Multicellular development requires the stereotypical and robust restriction of pluripotent cells to specific lineages . In many cases , this is dependent on positional information , where the relative position of a cell within the embryo determines the nature or amount of instructive differentiation signals received . However , there are also a growing number of examples of position independent patterning ( Kay and Thompson , 2009 ) . In these , different cell types firstly arise scattered in a ‘salt and pepper’ fashion before sorting out . To understand this mechanism , it will be important to understand why some cells differentiate , whereas neighboring cells within the same environment do not . One possible clue comes from cell culture studies that have revealed that genetically identical populations of cells exhibit heterogeneous behavior ( Chambers et al . , 2007; Chang et al . , 2008; Wu et al . , 2009 ) . When these cells receive identical doses of defined differentiation inducing signals , only a small fraction of ‘lineage primed’ cells actually respond . In this scenario , a higher inducer concentration increases the number of responding cells without affecting the magnitude of the response of individual cells . This suggests that cells exhibit different intrinsic response biases or discrete transcriptional activation thresholds to signals . There is now evidence to support the idea that the mechanisms underlying heterogeneous responses observed in cell culture could in fact regulate differentiation and developmental patterning in multicellular organisms ( Kaern et al . , 2005 ) . For example , in one of the earliest lineage choices made during mouse embryogenesis , cells of the inner cell mass ( ICM ) adopt either primitive endoderm ( PrE ) or epiblast ( EPI ) fates . This occurs in a position independent fashion with ICM cells exhibiting seemingly stochastic expression of PrE and EPI markers ( Dietrich and Hiiragi , 2007; Plusa et al . , 2008 ) . It has been proposed that heterogeneity in responsiveness to differentiation inducing signals , such as the PrE inducer FGF , underlies this salt and pepper differentiation ( Yamanaka et al . , 2010 ) . Crucially , in this model , it is not necessary for cells to receive different levels of FGF , only that they exhibit heterogeneity in their response thresholds to the signal . Finally , following this period of ‘symmetry breaking’ , coherent tissues can emerge due to a process of sorting out . Sorting is likely caused by differential gene expression resulting in differential cell motility , which can be driven by chemotaxis or differential cell adhesion ( with the elimination of misplaced cells also possible ) . Pattern formation based on stochastic salt and pepper differentiation and sorting out is likely to be a fundamental and deeply conserved developmental patterning mechanism ( Kay and Thompson , 2009 ) . However , our knowledge of the underlying molecular mechanism , as to how heterogeneity affects responsiveness to differentiation signals , is still in its infancy . One route to understanding this phenomenon comes from the finding that initial cell fate choice and pattern formation in Dictyostelium discoideum , a genetically and biochemically tractable organism with a small number of easily recognizable cell types , is based on this mechanism ( Thompson et al . , 2004b ) . Upon starvation , Dictyostelium cells enter a developmental cycle that begins with the aggregation of several thousand cells to make a multicellular mound . Within the mound , cells differentiate intermingled into prestalk or prespore cells . After sorting out , the different cell types are organized into discrete tissues in the migratory slug . Prestalk cells occupy the anterior 25% of the slug , with anterior like cells also found scattered within the prespore zone . Upon culmination , prestalk cells undergo extensive rearrangements to populate distinct parts of the fruiting body , including the stalk and basal disc , as well as ancillary spore head supporting structures known as the upper and lower cup . Prestalk cells can further be subdivided into several major subtypes ( pstA , pstO and pstB ) . Each cell type is defined largely by the pattern of expression of the classical prestalk specific transcripts ecmA and ecmB ( Jermyn et al . , 1989; Williams et al . , 1989; Jermyn and Williams , 1991; Early et al . , 1993 ) , and by their position in the migratory slug and final fruiting body ( Figure 1—figure supplement 1 ) . There is now good evidence to suggest that heterogeneities during growth of Dictyostelium , such as nutritional status or cell cycle position , affect developmental fate ( Leach et al . , 1973; Gomer and Firtel , 1987; Thompson and Kay , 2000a; Chattwood and Thompson , 2011 ) . Firstly , cells grown without glucose preferentially become stalk cells when mixed in chimeric development with glucose grown cells ( Leach et al . , 1973; Noce and Takeuchi , 1985; Blaschke et al . , 1986 ) . Importantly , when cells from different growth conditions are compared , a hierarchy of biases results , where cells can be ‘stalky’ in one mixture but ‘sporey’ in another ( Leach et al . , 1973; Thompson and Kay , 2000a ) . Therefore , growth history does not commit a cell to a given fate , but is instead context dependent and relative to the growth history of the entire population . Secondly , biases established during growth have been shown to affect the sensitivity of cells to differentiation signals experienced during development , such as the chlorinated alkyl phenone DIF-1 ( hereafter termed DIF ) , which is primarily required for pstB cell differentiation , with a more minor role in pstO cell differentiation ( Thompson and Kay , 2000b; Thompson et al . , 2004a; Zhukovskaya et al . , 2006; Huang et al . , 2006b; Keller and Thompson , 2008; Saito et al . , 2008 ) . For example , ‘stalky’ biased cells are more sensitive to DIF in monolayer culture assays than ‘sporey’ biased cells ( Thompson and Kay , 2000a; Kubohara et al . , 2007 ) . Finally , immediate-early gene responses to DIF exhibit all-or-none behavior , leading to population level heterogeneity ( Stevense et al . , 2010 ) . This provides a basis for salt and pepper differentiation in vivo , where it is presumed all cells experience similar DIF concentrations . From these studies it is clear that in order to understand patterning by salt and pepper differentiation followed by sorting , it will first be important to understand why some cells respond to differentiation inducing signals whereas others do not . To address this , we have used a novel genetic approach to identify genes involved in lineage priming . Here we describe the characterization of one such gene , gefE , and show that its expression during growth is specifically required to determine the number of cells that respond to DIF , both in cell culture and in vivo . Finally , we show that although expression of gefE itself is not heterogeneous during growth , expression of its target , rasD , is both heterogeneous and dynamic . These studies thus establish how salt and pepper differentiation in response to a uniform diffusible signal can be achieved through the stochastic expression of a Ras gene , the activity of which in turn is dependent on a ubiquitously expressed regulatory Gef . These studies therefore provide key insights into the basis of gene regulatory networks that can underpin patterning during a poorly understood but fundamental mode of pattern formation . The nutritional conditions under which cells are grown have previously been shown to bias lineage choice during Dictyostelium development ( Leach et al . , 1973; Thompson and Kay , 2000a ) . Consistent with these observations , we find that cells grown in the absence of glucose ( G− ) produce significantly fewer spores than cells grown in the presence of glucose ( G+ ) when developed together in chimera ( Figure 1A ) . Importantly , our analysis of the patterning of cell types at the slug and fruiting body stages reveals that this is because G− cells are biased towards the DIF dependent pstO and pstB lineages ( Figure 1B , Figure 1—figure supplement 2 ) . This is consistent with previous studies showing G− cells are more sensitive to DIF than G+ grown cells ( Thompson and Kay , 2000a ) . 10 . 7554/eLife . 01067 . 003Figure 1 . RasGEFE mutant cells are enriched in a genetic screen for modulators of nutritional bias . ( A ) G− cells produce fewer spores than G+ cells in chimeric development . GFP-labelled Ax3 wild type cells were grown in either G+ or G− conditions and mixed 10:90 with wild type G+ cells . GFP spores were quantified by counting . Dotted line indicates the percentage GFP spores expected if there is no fate bias . Error bars represent SEM , p<0 . 0001 . ( B ) G− growth biases cells towards pstO and pstB cell fates . Diagram shows organisation of different cell types along the anterior-posterior axis of the Dictyostelium slug . Patterning of GFP-labelled ( * ) G+ ( i ) and G− ( ii ) cells when mixed at 10:90 ratio with G+ cells . The reciprocal pattern was observed when GFP-labelled G+ ( iii ) and G− ( iv ) were mixed at 10:90 ratio with G− cells . ( C ) Schematic diagram of the genetic selection . REMI mutant cells were grown in G− and mixed 10:90 with wild type G+ GFP cells . Chimeric fruiting bodies were harvested and spores returned to growth medium after each developmental cycle . Wild type cells were removed with Blasticidin . ( D ) Generation of gefE− mutants . REMI plasmid , pBSR1 , inserted into 42 bp exon 1 . RasGEF catalytic domain ( blue ) deleted by homologous recombination . ( E ) gefE− mutant cells produce more spores than Ax3 wild type cells after G− growth . RFP-labelled wild type cells were grown in G+ medium and mixed at a 50:50 ratio with unlabelled wild type or gefE− mutant cells grown in G− . Number of unlabelled spores was quantified by counting . Error bars represent SEM , p<0 . 04 . ( F ) Comparison of the patterning of GFP-labelled ( * ) wild type ( i ) or gefE− mutant ( ii ) cells grown in G− conditions when mixed at 10:90 ratio with unlabelled wild type G+ cells . AP axis in all slug images oriented from right-left with white bars showing regions of GFP enrichment . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 00310 . 7554/eLife . 01067 . 004Figure 1—figure supplement 1 . Organisation of cell types in Dictyostelim slug and culminants . Nomenclatures for parts of Dictyostelium slug ( left ) and fruiting body ( right ) are depicted . Yellow ( prespore ) cells eventually form sorus and red cells ( pstA or AB ) form stalk cells . Blue ( pstO ) and green ( pstB ) cells are the DIF dependent cell types which eventually form upper/lower cup and basal disc . ecmA promoter can be subdivided to pstA and O , and ecmB promoter to pstAB and pstB cell types . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 00410 . 7554/eLife . 01067 . 005Figure 1—figure supplement 2 . Cell type specific effects of nutritional history in Dictyostelium slugs . ( A ) and culminants ( B ) 10% labelled G+ cells ( upper panels ) or G− cells ( lower panels ) mixed with 90% unlabelled G+ or G− cells . Closed arrows indicate enrichment of label , open arrows an absence . In each panel , the top row shows the control mix and is comparable to the image directly beneath . General patterning differences between G+ and G− cells ( A–D ) are described in the main text and Figure 1 . G+ cells labelled with ecmAO:lacZ occupy anterior-most pstA regions when mixed with G− cells ( E and F ) . In the reciprocal mix , G− cells labelled with ecmAO:lacZ were absent from the tip and formed a collar in the pstO regions when mixed with unlabelled G+ cells ( G and H ) . Expression of ecmB:lacZ was decreased in pstB regions of chimeras when labelled G+ cells were mixed with unlabelled G− mutant cells ( I and J ) and increased when labelled G− mutant cells were mixed with unlabelled G+ cells ( K and L ) . Wild type cells labelled with pspA:lacZ occupy the prespore region when mixed with unlabelled G− cells ( M and N ) . In the reciprocal mix , where G− cells labelled with pspA:lacZ were mixed with unlabelled G+ cells , expression was greatly reduced and restricted to the rear of the prespore compartment ( O and P ) . The AP axis of all structures is aligned from right-left . Refer to Figure 1—figure supplement 1 to link marker expression to cell type localization within slug or culminant . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 00510 . 7554/eLife . 01067 . 006Figure 1—figure supplement 3 . Enrichment of REMI mutants during screen correlates with change in cell fate preference during development . ( A ) Patterning phenotype of REMI mutant pools in chimera with wild type cells . 10% wild type GFP cells were mixed with 90% REMI mutant cells from each pool at each round of selection . The dark regions in each slug show the location of the REMI mutant cells . In Rd0 , REMI mutant cells localize in the pstO and pstB regions . By Rd7 , wild GFP cells are almost entirely absent from prespore compartment , especially in pool 2 . In pool 1 ( top row ) , REMI mutant cells are enriched in the collar and back until Rd3 . By Rd6 and 7 , wild type GFP cells are enriched in the anterior prestalk region , suggesting the REMI mutant cells are in the prespore population . ( B ) Proportion of REMI mutants identified by iPCR that mapped to exon 1 of gefE locus in each round of selection . ( C ) Measurements of growth rate of wild type and gefE− mutant cells in G+ or G− medium . ( D ) Measurement of spore hatching rate of wild type and gefE− mutant cells . The percentage of hatched spores was scored at the indicated times after spores were placed in growth medium . The cell density after 24 hr of induction was also measured , revealing that the faster hatching rate of gefE− mutant resulted in approximately twofold higher cell density after 24 hr . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 006 To better understand this lineage bias , we took a genetic approach in which a REMI mutant library estimated to initially contain approximately 1000 individual blasticidin resistant clones was hatched from a frozen stock two times . Each replicate pool was then grown separately in G− medium and developed in chimera with G+ wild type cells ( Figure 1C ) . After each round of development , chimeric spore heads were transferred into growth medium containing blasticidin , thus killing all wild type cells and only allowing mutant spores to grow . This provides a selective pressure to enrich for mutants that adopt the spore cell fate . The selection was repeated on each pool for seven rounds . The success of the selection was assessed qualitatively by examining the distribution of mutant and wild type cells in chimeric slugs . As expected , in initial rounds , G− grown mutant cells were enriched in the pstO and pstB populations . However , in later rounds the distribution of mutant cells changed with a clear enrichment within the prespore zone , suggesting that mutants defective in bias imposed by G− growth were becoming enriched ( Figure 1—figure supplement 3A ) . Inverse PCR revealed the pool of cells in later rounds of each pool selection to be largely composed of clones in which the REMI plasmid had integrated into a small 5′ 42 bp exon within the gefE gene ( Figure 1D , Figure 1—figure supplement 3B ) . In one pool , gefE mutant cells were enriched by round 2 , whereas in the second it was only strongly enriched by round 6 . These differences could be due to the effects of random genetic drift caused by harvesting a finite number of spores after each round . Alternatively , it has recently been shown that when multiple clones are hatched and grown on bacteria , small differences in growth rate at the feeding edge can be amplified , leading to overrepresentation of some clones ( Buttery et al . , 2012 ) . We next performed a reconstruction experiment to verify that enrichment of gefE− mutant cells was due to being overrepresented in the spore population , rather than the mutation simply conferring advantages during other stages of the selection process . For this , because the catalytic domain of GefE is encoded by the larger 3′ exon , we generated a mutant allele in which the catalytic domain was deleted ( Figure 1D ) . For all subsequent studies , the catalytic domain disruption allele was used , but because both mutants exhibit identical phenotypes , both likely represent null alleles . Reconstruction experiments revealed no difference in the rate of growth of gefE− mutant cells in either G+ or G− medium , although gefE− mutant spores did , however , hatch at a faster rate than wild type spores , perhaps helping to explain the unexpectedly strong enrichment for this mutant within the pools ( Figure 1—figure supplements 3C , D ) . Most importantly , when wild type or gefE− mutant G− grown cells were developed in chimera with G+ wild type cells , mutant G− cells were significantly overrepresented in the spore population of chimeric fruiting bodies ( Figure 1E , p<0 . 04 ) , thus providing a simple explanation for the enrichment of this mutant in the genetic selection . This is because mutant G− cells do not show the strong overrepresentation in the collar and back populations of chimeric slugs seen when wild type G− cells are mixed with wild type G+ grown cells ( Figure 1F ) . Growth in the absence of glucose specifically biases cells towards DIF dependent prestalk lineages . Because the gefE− mutant fails to respond to this G− growth bias , we next tested whether gefE gene disruption also resulted in specific defects in DIF dependent lineage bias . To test this , wild type and mutant cells were grown under identical growth conditions in the presence of glucose before mixing for chimeric development . Under these conditions , gefE− cells still formed more spores than wild type cells ( Figure 2A , p<0 . 01 ) . In chimeric slugs , labelled mutant cells were enriched in the prespore region and excluded from the DIF dependent pstB and pstO cell fates , whilst labelled wild type cells were localized in the pstB and pstO regions of the slug ( Figure 2B , Figure 2—figure supplement 1 ) . Similar results were seen when cells were grown in the absence of glucose ( Figure 2—figure supplement 2 ) . Finally , when GefE expression was driven by a strong constitutive promoter in wild type cells , cells expressing high gefE levels were found largely in the pstO and pstB regions of slugs and in the lower cup of fruiting bodies when mixed with wild type cells ( Figure 2C ) . Together , these data suggest that expression of gefE is required for biasing cells to adopt DIF dependent pstO and pstB cell fates . 10 . 7554/eLife . 01067 . 007Figure 2 . gefE− mutant cells avoid pstO and pstB fates . ( A ) RFP-labelled AX3 wild type cells were mixed at a 50:50 ratio with unlabelled wild type or gefE− mutant cells . Both strains were grown in presence of glucose . Number of unlabelled spores quantified by counting . Dotted line indicates the percentage RFP spores expected if there is no fate bias . Error bars represent SEM , p<0 . 01 . ( B ) Chimeras of 10% GFP-labelled ( * ) wild type or gefE− mutant cells mixed with 90% unlabelled wild type or gefE− mutant cells . AP axis in all slug images oriented from right-left with white bars showing regions of GFP enrichment . ( C ) Cells transfected with RFP control vector ( a and b ) , or GefE-RFP fusion vector ( c and d ) , under control of the constitutive actin promoter , mixed with unlabelled wild type cells at 20:80 ratio and observed during slug ( a and c ) and culminant ( b and d ) stages of development . Closed arrows indicate relative enrichment in reporter gene expression . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 00710 . 7554/eLife . 01067 . 008Figure 2—figure supplement 1 . Cell type specific effects in chimeric gefE−/wt slugs . ( A ) and culminants ( B ) 10% labelled Ax3 wild type ( upper panels ) or gefE− mutant cells ( lower panels ) mixed with 90% unlabelled wild type or gefE− mutant cells . Closed arrows indicate enrichment of label , open arrows an absence . In each panel , the top row shows the control mix and is comparable to the image directly beneath . General patterning differences between wild type and gefE− mutant cells ( A–D ) are described in the main text and Figure 2 . Wild type cells labelled with ecmAO:lacZ were absent from the tip and formed a collar in the pstO regions when mixed with gefE− mutant cells ( E and F ) . In the reciprocal mix , gefE− mutant cells labelled with ecmAO:lacZ occupied the anterior-most pstA regions when mixed with unlabelled wild type cells ( G and H ) . Expression of ecmB:lacZ was increased in pstB regions of chimeras when labelled wild type cells were mixed with unlabelled gefE− mutant cells ( I and J ) and reduced when labelled gefE− mutant cells were mixed with unlabelled wild type cells ( K and L ) . At culminant stage , ecmB expression is upregulated in pstA cells . Blue arrows in panel B highlight the tendency of wild type cells to avoid this fate ( I and J ) at the expense of gefE− mutant cells ( K and L ) . Expression of pspA:lacZ was reduced and restricted to the prespore region when labelled wild type cells were mixed with unlabelled gefE− mutant cells ( M and N ) . The reciprocal pattern was seen when labelled gefE− mutant cells were mixed with unlabelled wild type cells ( O and P ) . The AP axis of all structures is aligned from right-left . Refer to Figure 1—figure supplement 1 to link marker expression to cell type localization within slug or culminant . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 00810 . 7554/eLife . 01067 . 009Figure 2—figure supplement 2 . gefE− mutant cells avoid pstO and pstB fates when both partners have a G- growth history . All cells were grown in absence of glucose . 10% GFP-labelled Ax3 wild type or gefE− mutant cells were then mixed with 90% wild type or gefE− mutant cells and developed to slug stage . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 009 Genes that affect cell fate choice can either affect lineage commitment or lineage bias . A lineage bias gene is one that is not required for cell fate commitment per se , but required for the relative probability of commitment to one fate over another . If the gefE gene is involved in lineage priming rather than commitment , an effect on development will not be observed if all cells in the population have the same bias ( e . g . , normal clonal development ) . We therefore examined cell fate choice and patterning of the gefE− mutant during clonal development , with all cells grown under identical conditions . Clonal development was found to be morphologically identical to that of the wild type . Furthermore , the number of prespore cells in gefE− mutant slugs was found to be the same as that observed in wild type slugs , and the total number of spores produced in terminal fruiting bodies was the same for the gefE− mutant as for the wild type ( Figure 3A ) . There was also no defect in prestalk cell specific gene expression in slugs or fruiting bodies ( Figure 3B ) . The only difference between the clonal development of the gefE− cells and the wild type was a slight , although reproducible , delay in the initiation of aggregation ( data not shown ) . These results are , therefore , consistent with the gefE gene being involved in lineage priming rather than lineage commitment . Furthermore , if the gefE gene is involved in lineage priming , rather than commitment , then the bias of the gefE− cells to avoid a DIF dependent fate , that is observed when they are mixed with wild type cells , should be reversed when they are mixed with cells of a stronger bias . Consistent with this idea , the lineage bias of the gefE− mutant cells was found to be dependent on the growth history of the partner genotype . For example , gefE− mutant cells were strongly biased towards DIF dependent fates when grown in the G− medium and then mixed with mutant cells grown in G+ medium ( Figure 3C ) . These experiments provide strong evidence that gefE− cells are totipotent and that GefE regulates context dependent/relative DIF dependent lineage bias rather than cell fate choice . 10 . 7554/eLife . 01067 . 010Figure 3 . Cell type differentiation is unaffected during clonal development in the gefE− mutant . ( A ) Quantification of prespore:prestalk ratio and total number of spores produced by Ax3 wild type and gefE− mutant at slug and fruiting body stages respectively . ( B ) Expression of prestalk markers ecmA and ecmB in clonal Ax3 wild type and gefE− mutant culminants . ( C ) Chimeras of 10% GFP-labelled ( * ) gefE− mutant cells grown in G+ or G− mixed with 90% unlabelled gefE− mutant cells grown in G+ or G− conditions . AP axis in all slug images oriented from right-left with white bars showing regions of GFP enrichment . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 010 G+ grown cells are less likely to adopt DIF dependent lineages than G− cells and G+ cells have been shown previously to be less sensitive to DIF than G− grown cells ( Thompson and Kay , 2000a ) . Therefore , we next tested whether gefE− mutant cells also exhibit differences in DIF responsiveness when compared to wild type cells . We found that gefE− mutant cells make fewer stalk cells than wild type across a 1000-fold range of DIF concentrations in the cAMP removal monolayer assay but exhibit similar sensitivity to changes in DIF concentration ( Figure 4A ) . Importantly , growth in the absence of glucose results in a similar increase in DIF sensitivity in both wild type and mutant cells ( Figure 4A ) , thus suggesting that GefE is only required to set basal levels of DIF sensitivity . Similar results were seen when the effect of gefE disruption was observed on the expression of primary DIF target genes . In the first experiment , cells were transformed with ecmAO:lacZ or ecmB:lacZ reporter genes . DIF induced gene expression of these reporters was reduced in the gefE− mutant compared to wild type at all DIF concentrations , but showed a similar sensitivity to changes in DIF concentration ( Figure 4B ) . A similar result was seen when endogenous ecmA and ecmB transcripts were measured in response to DIF induction ( Figure 4C ) . Finally , we determined the effects of gefE disruption on an immediate-early DIF response . GATAc is a transcription factor that translocates to the nucleus minutes after DIF stimulation and is specifically required for DIF dependent pstB cell differentiation ( Keller and Thompson , 2008 ) . In gefE− mutant cells basal levels of nuclear GATAc were similar to wild type . Furthermore , when gefE− mutant cells were stimulated with DIF , the rate and duration of GATAc-GFP nuclear accumulation were largely unaffected . However , large differences were seen in the number of gefE− mutant cells that exhibit nuclear GATAc accumulation compared to wild type ( Figure 4D ) . Together , these results suggest that gefE− cells are less sensitive to DIF than wild type cells , consistent with the idea that GefE biases cells towards DIF sensitive fates . 10 . 7554/eLife . 01067 . 011Figure 4 . gefE− mutant cells are less sensitive to DIF . ( A ) Quantification of stalk cell formation 22 hr after DIF induction . ( B ) Expression level of lacZ reporter gene fused to prestalk specific promoter of ecmA ( left ) or ecmB ( right ) 22 hr after DIF induction . ( C ) qPCR analysis of endogenous ecmA ( left ) or ecmB ( right ) transcript levels after 3 hr induction with 100 nM DIF . ( D ) Nuclear translocation of DIF-induced transcription factor , GATAc-GFP in response to 100 nM DIF . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 011 The above data demonstrate that gefE− mutant cells are less likely to respond to DIF . It has recently been shown that DIF induced transcription of the prestalk gene , ecmA only occurs once a cell intrinsic threshold is exceeded; increased DIF dose only results in a small change in target gene expression level in each cell , but a much larger increase in the number of cells that respond ( Stevense et al . , 2010 ) . We , therefore , hypothesised that the reduced DIF responsiveness of the gefE− mutant was due to an increase in the threshold required to trigger a transcriptional response . To test this idea further , we developed a novel FACS based approach that permits ecmA expression levels to be measured in individual wild type or gefE− mutant cells in response to DIF . For this we replaced the endogenous copy of the ecmA gene with GFP ( Figure 5A ) , and showed that GFP was expressed from the ecmA promoter in the same localization in slugs and fruiting bodies ( Figure 5B ) as has been previously described for ecmA-LacZ ( Jermyn and Williams , 1991 ) . As expected , in wild type controls , increasing DIF concentrations resulted in greater numbers of GFP expressing cells , with only a small change in GFP expression level per cell ( Figure 5C , D ) . Most importantly , at all concentrations of DIF , the number of responding gefE− mutant cells was lower than wild type ( Figure 5C , D ) . However , there was little difference in the level of ecmA:GFP expression between individual wild type or gefE− mutant responding cells ( Figure 5C , D ) . Taken together , these data provide strong support for the idea that gefE regulates the cell-intrinsic threshold at which cells respond to DIF . 10 . 7554/eLife . 01067 . 012Figure 5 . GefE regulates the DIF response threshold . ( A ) Replacement of the endogenous ecmA gene with GFP ( B ) prestalk specific expression of GFP knock-in strain at slug and culminant stages . ( C ) FACS analysis of Ax3 wild type and gefE− mutant GFP knock-in strains stimulated with 0 . 01–100 nM DIF for 9 hr . Y-axis shows the number responders and X-axis the GFP expression level per cell . ( D ) Median GFP expression level of wild type and gefE− mutant populations stimulated with 0 . 01–100 nM DIF . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 012 We next sought to understand the molecular mechanism by which GefE affects lineage bias . As gefE is predicted to encode a RasGEF , whose primary function is to activate Ras proteins by catalyzing the exchange of GDP for GTP , it therefore seemed likely that GefE would exert its effects on DIF dependent lineage bias through Ras regulation . Although the Dictyostelium genome encodes 11 Ras proteins , only one , RasD , exhibits strong developmental expression ( Reymond et al . , 1984 ) . Furthermore , we found that the level of both RasD expression and GTP bound activated RasD are significantly higher in cells grown in the absence of glucose ( Figure 6A ) . Finally , GefE and RasD mutant cells have previously been shown to exhibit identical slug phototaxis defects ( Wilkins et al . , 2000 , 2005 ) , raising the possibility that RasD could be a target for GefE . To test this idea directly , we firstly compared the relative levels of RasD-GTP in wild type and mutant cells by Western blot , using a pull down assay and a specific antibody ( Wilkins et al . , 2000 ) . In 12 hr developed cells , the level of activated RasD was significantly lower in gefE− mutant cells , although there was no difference in the total level of the RasD protein ( Figure 6B ) . In addition , there was no corresponding decrease in the levels of activated RasG , RasB , RasC or Rap1 ( data not shown ) . These results suggest that GefE is the predominant activator of RasD and is specific for RasD . We therefore next tested whether rasD− and gefE− mutants exhibit similar cell fate choice phenotypes . When labelled wild type cells were mixed with rasD− mutant cells , the labelled cells accumulated in the pstO and pstB populations and were underrepresented in the prespore population ( Figure 6C , Figure 6—figure supplement 1 ) . Moreover , when RasD expression levels were elevated in wild type cells or cells were forced to express constitutively active rasD ( G12T ) , these cells exhibited a strong localization towards the pstB region at the rear of slug and in the lower cup of fruiting bodies when mixed with wild type cells ( Figure 6D ) . Importantly , unlike wild type cells , this effect was only observed in gefE− mutant cells which expressed constitutively active rasD ( G12T ) . These findings thus suggest that RasD is normally converted to its GTP bound form upon activation by GefE and support the idea that GefE and RasD coordinate DIF dependent lineage bias . 10 . 7554/eLife . 01067 . 013Figure 6 . GefE activates RasD . ( A ) Comparison of the levels of activated RasD-GTP and total RasD by Western blot in vegetative or 12 hr starved cells grown in the presence or absence of glucose . ( B ) Comparison of activated RasD-GTP and total RasD levels by Western blot in wild type Ax3 , gefE− or rasD− cells . ( C ) GFP-labelled ( * ) Ax3 wild type cells mixed at 10:90 ratio with unlabelled Ax3 or rasD− mutant cells . Closed arrows show enrichment of wild type cells in pstO and pstB populations . Open arrows show reciprocal enrichment of rasD− cells . ( D ) RasD overexpression results in GefE dependent bias towards the pstO and pstB cell fates . 20% cells constitutively expressing RFP , RFP-RasD or RFP-RasD ( G12T ) were mixed with 80% unlabelled parental cells . When wild type Ax3 cells overexpress RFP-RasD or RFP-RasD ( G12T ) they become enriched in the pstO and pstB populations ( arrows ) . Only gefE− cells that express constitutively activated RasD ( G12T ) are enriched in pstO and pstB populations ( arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 01310 . 7554/eLife . 01067 . 014Figure 6—figure supplement 1 . rasD− mutant cells avoid pstO and pstB fates in chimera with wild type cells . 10% Ax3 labelled wild type were mixed with 90% unlabelled rasD− mutant cells . The top row shows the control mix and is comparable to the image directly beneath . Closed arrows indicate enrichment of label , open arrows an absence . The AP axis of all slugs is aligned from right-left . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 014 Studies of the behavior of gefE− and rasD− mutant cells during chimeric development and in cell culture suggest that the levels of activated RasD are critical to set the DIF response threshold within populations of cells . This suggested that heterogeneous activity of RasD within a growing population of cells could perhaps underpin the salt and pepper differentiation of DIF dependent lineages during normal development . Therefore , we next tested whether gefE or rasD gene expression was heterogeneous during growth . For this , gefE or rasD promoter sequences were used to drive RFP expression . Although gefE was found to be uniformly expressed in all cells during growth , rasD expression was highly heterogeneous ( Figure 7A ) . The rasD promoter RFP reporter construct used here shows strong developmental regulation and drives prestalk cell specific expression at the slug and culminant stages ( Figure 7—figure supplement 1 ) , consistent with previous reports using rasD:lacZ ( Esch and Firtel , 1991; Jermyn and Williams , 1995 ) and observations of the endogenous transcripts ( Maeda et al . , 2003 ) , validating its use as a measure of rasD promoter expression . 10 . 7554/eLife . 01067 . 015Figure 7 . RasD expression is heterogeneous in growth phase populations . ( A ) Dual promoter vectors used to drive constitutive GFP expression and gefE promoter ( left ) or rasD promoter ( right ) driven RFP expression . Cells growing in tissue culture plates were photographed with a fluorescence microscope on red channel ( a and d ) , green channel ( b and e ) and both ( c and f ) . ( B ) Ax3 cells transformed with rasD promoter vector were fractionated into RFP high and RFP low populations by FACS . These populations were mixed in a 5:95 ratio with unlabelled Ax3 cells . Cell fate choice was traced by constitutive expression of GFP at slug ( a and b ) and culminant ( c and d ) stages . White bars show regions of fewer GFP cells . ( C ) FACS sorted low or high rasD:RFP cells were cultured back in HL-5 medium . The ratio of RFP:GFP cells was scored over time . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 01510 . 7554/eLife . 01067 . 016Figure 7—figure supplement 1 . RasD is expressed in prestalk cells during development . Expression of rasD:RFP ( ‘text’ ) in the prestalk populations of a clonal slug ( left ) and culminant ( right ) . AP axis of both structures oriented top-bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 01067 . 016 We next tested whether differences in rasD expression reflect differences in lineage bias . Firstly , we found that growth of cells in G− medium ( conditions that increase the likelihood of DIF dependent lineage choice ) resulted in a significant increase in the number of rasD:RFP expressing cells ( 73 ± 3% to 89 ± 5% ) whereas no change was observed in gefE:RFP expressing cells . Secondly , growing rasD:RFP expressing cells were sorted by FACS based on their fluorescence and developed in chimera with wild type cells . Cells expressing lower levels of RasD were found to be less likely to differentiate as prestalk cells than prespore cells ( Figure 7B ) . Surprisingly , however , cells expressing high rasD transcript levels can form both prespore and prestalk cells , as evidenced by their random distribution at both slug and culminant stages ( Figure 7B ) . So , while there is an increased likelihood that high RasD cells will become prestalk cells compared to low RasD cells , they do not sort to pstO and pstB regions as might be expected . One likely explanation for this discrepancy is that the stability of the RFP protein may be different to endogenous RasD . For example , if RasD protein has a shorter half-life than RFP , then cells expressing RFP may no longer express RasD . These cells would thus not exhibit the expected prestalk bias and could differentiate as prespore cells . Population heterogeneity in gene expression , such as that observed for rasD expression , can be achieved if gene expression fluctuates stochastically or in response to other intrinsic or extrinsic factors ( Raj and van Oudenaarden , 2008; Eldar and Elowitz , 2010 ) . This often leads to the establishment of a dynamic equilibrium that allows populations to return to normal after perturbation or respond appropriately to environmental cues ( Chang et al . , 2008; Eldar and Elowitz , 2010 ) . In order to test whether RasD behaves in this fashion , rasD high or rasD low expressing cells were purified by FACS and then regrown in culture . Each population quickly returned to its pre-sorted equilibrium ( Figure 7C ) . It is interesting to note , however , that the low RasD expressing population exhibits slightly faster dynamics than the high RasD expressing population , thus supporting the idea that chimeric behavior may be affected by differences in RasD transcription/translation and RFP stability . Together these findings suggest that population heterogeneity is generated by dynamic rasD expression in vegetative cells and that extrinsic factors such as nutritional history affect the position of this dynamic equilibrium . Studies on the possible role played by ras genes in Dictyostelium development have not been conclusive ( Chubb and Insall , 2001 ) . Initial studies carried out in cell lines expressing constitutively activated Ras ( G12T ) suggested Ras signaling played a key role in prestalk cell differentiation ( Jaffer , 2000; Jaffer et al . , 2001 ) . When these cells were developed in chimera with wild type cells , they were shown to occupy the collar and back of slugs , consistent with our present findings using RasD ( G12T ) ( Figure 6C ) . However , disruption of the gene encoding RasD , the major developmentally expressed Ras protein , was reported not to affect cell fate specification during clonal development ( Wilkins et al . , 2000 ) . This led to the suggestion that observations with constitutively activated Ras could be an artifact or that lack of RasD could be compensated by upregulation of other Ras genes ( Wilkins et al . , 2000 ) . We found that there was indeed compensatory upregulation of the highly related RasG protein during development in both rasD− and gefE− mutant backgrounds ( data not shown ) but that this did not affect lineage bias in chimeras with wild type cells . This suggests that GefE-mediated activation of RasD is uniquely important to confer lineage bias in Dictyostelium . Such bias can be viewed as giving some cells either a head start or handicap in a ‘race’ to adopt fates . Only a limited number of ‘winning’ cells will differentiate before feedback mechanisms restrict other cells from adopting this fate , a process akin to lateral inhibition . Indeed , an analogous form of cell competition has been observed during Drosophila development when a clone of slow growing mutant cells is surrounded by wild type cells ( Levayer and Moreno , 2013; Vincent et al . , 2013 ) . In this case , slow growing mutant cells are often eliminated by the faster growing wild type cells , which overgrow and compensate for their absence ( Simpson and Morata , 1981 ) . Like cell fate bias , cell competition is thus based on measurements of relative fitness between neighboring cells . Consequently , disruption of competition or bias genes only results in subtle defects during when all developing cells are the same genotype , such as a delay in the timing of cell fate determination . Ras signaling has been the subject of intense study , with upstream and downstream components of Ras signaling pathways now well defined ( Karnoub and Weinberg , 2008 ) . However , we have less knowledge as to the possible importance of Ras regulation by transcriptional or post-translational control . Our studies using a reporter gene in which the rasD promoter drives RFP expression , reveals heterogeneous transcriptional activation of a ras gene within a population of growing Dictyostelium cells . Furthermore , this proportion increases when cells are transferred into medium lacking glucose . This indicates that there is a stochastic component to rasD gene expression that is influenced by external factors . In addition , when high or low rasD expressing cell populations are separated by FACS , they return to the pre-sorting equilibrium after a short period of growth . This suggests that cells freely transition between high and low rasD expressing subpopulations . Similar dynamic transcription has been observed in mammalian cell populations and is thought to represent slowly fluctuating transcriptomes that prime cell fate decisions both in the petri dish and during early embryogenesis ( Chang et al . , 2008; Silva and Smith , 2008; Kalmar et al . , 2009; Canham et al . , 2010; Abranches et al . , 2013 ) . Similarly , growing Dictyostelium cells expressing low levels of rasD transcripts are primed to become prespore cells during development , whilst RasD hyperactivation is sufficient to drive cells towards a prestalk cell fate . Surprisingly , however , we found that cells expressing high rasD:RFP transcript levels only exhibit a relatively weak prestalk preference . One interpretation of this finding is that if RasD protein has a shorter half life than RFP , then not all cells that express RFP would actually express RasD . Consequently , not all RFP expressing cells would exhibit a prestalk bias and may be free to differentiate as prespore cells , thus hindering experimental observation of their behavior . It is interesting to note , however , that similar observations of cell populations behaving more promiscuously in their lineage restriction than might be expected , have also been seen when fluorescent protein reporters are used in studies of the early mouse embryo and in embryonic stem cell cultures ( Canham et al . , 2010; Grabarek et al . , 2012; Morgani et al . , 2013 ) . Most importantly , all these observations highlight the need for faithful real time reporters of gene expression , which will permit future studies of the regulation of RasD promoter activity , the molecular basis of transcriptional heterogeneity and lineage priming in general . Immediate-early DIF responses have been shown to be digital , with the number of responding cells being strongly affected by changes in DIF concentration , whereas the magnitude of the response in individual cells is largely unaffected ( Stevense et al . , 2010 ) . This observation implies that ( 1 ) a response is only triggered once an internal threshold is exceeded and ( 2 ) the position of the threshold varies across a population . Work described in this report provides further support for this idea and gives mechanistic insight into how the DIF response threshold is set and , thus , how cell fate choice is regulated in vivo . Specifically , we find that GefE dependent heterogeneous RasD activation during the growth phase is required to set the normal DIF response threshold . However , several major questions emerge . For example , DIF is a developmental inducer that is present only after cells have been starved; yet the DIF response threshold is set ( by RasD levels ) before DIF exposure , in the growth phase . Indeed , heterogeneity in rasD expression is independent of DIF , as RasD expression is unaffected by DIF treatment ( Jermyn et al . , 1987 and data not shown ) . One clue to how this heterogeneity is generated comes from our observation that RasD protein levels are higher in cells grown in medium lacking glucose . This suggests that heterogeneity in metabolic state between cells may play a role in rasD regulation . However , other growth phase heterogeneities , including differences in cell cycle position ( Gomer and Firtel , 1987; Araki et al . , 1994; Thompson and Kay , 2000a ) , intracellular calcium concentration ( Baskar et al . , 2000; Azhar et al . , 2001 ) and intracellular pH ( Gross et al . , 1983; Kubohara et al . , 2007 ) have all also previously been shown to affect cell fate choice . However , the molecular basis underlying these heterogeneities , or whether they affect RasD expression , is currently unknown . It thus remains to be determined whether they are a cause or consequence of the observed differences in RasD levels . Furthermore , it is currently unknown how RasD activation affects the sensitivity of cells to DIF at the molecular level . Because gefE− and rasD− cells are capable of forming DIF dependent cell types during development and rasD transcription is not induced by DIF ( Jermyn et al . , 1987 and data not shown ) , these observations strongly suggest that RasD is not a central component of the DIF response pathway and that positive feedback through rasD transcriptional activation does not contribute to the binary switch . It thus appears that RasD only ‘tunes’ responses to DIF , rather than being integral to the DIF response per se . One clue to the molecular basis of this regulation , however , comes from the finding that the rapid ( within 10 min ) nuclear translocation of GATAc-GFP in response to DIF is affected in the gefE− mutant . Consequently , it seems likely that the tuning effects of RasD act during the immediate-early phase of the DIF response . However , to fully understand how this tuning of DIF sensitivity occurs will require a greater knowledge of the DIF signaling pathway itself , and crucially how each component is regulated ( as all components represent potential RasD targets ) . For example , to date , only a small number of transcription factors and one protein kinase have been shown to be crucial for DIF signal transduction and responses ( Fukuzawa et al . , 2001; Thompson et al . , 2004a; Zhukovskaya et al . , 2006; Huang et al . , 2006a; Keller and Thompson , 2008; Araki et al . , 2012 ) . However , the regulation of their activity is complex , requiring control by posttranslational modifications , as well as subcellular localization ( Fukuzawa et al . , 2001 , 2003; Thompson et al . , 2004a; Zhukovskaya et al . , 2006; Huang et al . , 2006a; Araki et al . , 2008 , 2012; Keller and Thompson , 2008 ) . It is therefore possible that RasD alters the activity of specific protein kinases or transcription factors or hitherto unknown DIF signaling components , making them more likely to be DIF induced . In this way , because the DIF signaling pathway is intact in rasD− and gefE− mutant cells , they would still respond to DIF ( albeit less efficiently than wild type ) when given prolonged exposure to the signal . Finally , it is unclear to what extent inputs from different signaling pathways are also integrated to modulate the sensitivity of cells to DIF in response to nutritional bias , or other modulators of cell fate bias . Indeed , it is likely that the effects of nutrition are only partially explained by GefE activity . For example , when gefE− mutant cells are grown in the absence of glucose are compared to wild type cells grown in the presence of glucose , they do not produce as many spores in chimeric development . Furthermore , the Dictyostelium retinoblastoma ortholog , RblA , is a cell cycle regulated protein whose expression level during the growth phase is correlated with both DIF sensitivity and cell fate preference during development ( MacWilliams et al . , 2006 ) . Interestingly , rblA− mutant cells display hypersensitivity to DIF and adopt collar and back cell fates when mixed with wild type cells in chimera , perhaps indicating a degree of antagonism between Ras and Retinoblastoma signaling . Similarly , knock out of the Glycogen Synthase Kinase 3 homolog , gskA , results in DIF hypersensitivity and the gskA− strain preferentially forms basal disc cells at the expense of spores when developed clonally ( Harwood et al . , 1995; Schilde et al . , 2004 ) . Such similarities , as well as clear phenotypic differences ( data not shown ) , suggest this level of complexity allows the DIF signal to be integrated into multiple fate choices during development . Development in Dictyostelium is unusual , being based on aggregation of separate cells , rather than division of a fertilized egg . However , remarkable similarities are emerging between this evolutionarily ancient organism and the behavior of embryonic stem cell cultures , and indeed lineage specification in the early mouse embryo . ( 1 ) Salt and pepper differentiation of prestalk and prespore cells: During mouse embryogenesis , the PrE and EPI lineages arise from the ICM . Each lineage can be clearly defined at later stages through specific gene expression profiles , as well as their position within the blastocyst . However , at the earliest stages cells expressing PrE and EPI lineage markers are intermingled , with active sorting out observed at later stages ( Dietrich and Hiiragi , 2007; Plusa et al . , 2008; Yamanaka et al . , 2010 ) . ( 2 ) Dynamic expression of lineage priming/specific gene expression: observations in cell culture reveal expression of ICM cell fate markers such as Hex ( Canham et al . , 2010 ) and Nanog ( Chambers et al . , 2007; Kalmar et al . , 2009b ) is highly heterogeneous . When high or low level expressing cells are separated by FACS , the population structure returns to an equilibrium on time-scale of hours to days . Most importantly , differences in expression levels also correlate well with fate preferences in chimeric development ( Canham et al . , 2010 ) . ( 3 ) Modulation of Ras activation affects lineage choice: although FGF activates multiple pathways , genetic studies indicate that the Grb2 > Sos ( RasGEF ) > Ras > Erk axis is the most important during differentiation of both ES cells and PrE ( Chazaud et al . , 2006; Lanner and Rossant , 2010 ) . ( 4 ) Heterogeneous responses/thresholds to differentiation signals: FGF signaling is required for normal EPI/PrE differentiation ( Yamanaka et al . , 2010 ) . However , it has recently been shown that salt and pepper expression pattern of EPI and PrE markers does not require FGF signaling ( Kang et al . , 2013 ) . Rather , activation of this pathway is associated with exit from pluripotent state and commitment to the PrE fate . Finally when FGF levels are manipulated , the proportion of PrE:EPI cells correlates well with FGF levels , implying existence of cell intrinsic threshold to FGF ( Yamanaka et al . , 2010; Grabarek et al . , 2012 ) . While it is clear that multicellular development requires extreme precision , there is a growing realization that key decisions are biased by heterogeneous behavior between equipotent cells . Population level variation in sensitivity to a global signal is one way to generate diversity . However , our understanding of the mechanistic basis of lineage priming is still in its infancy . One reason for this is that mutants in lineage priming genes may only show subtle defects during clonal development . Importantly , our study highlights that testing chimeric effects can uncover important regulatory roles for these genes . Due to the relative ease with which chimeric development can be assayed in Dictyostelium , it seems likely that studies in this model system will lead to a better understanding of how lineage biases are acquired and potentially how they might be manipulated to increase the efficiency of stem cell based therapeutics . Dictyostelium strains were cultured on lawns of Klebsiella aerogenes or in HL5 medium with ( G+ ) or without ( G− ) 86 mM glucose . Cells were grown in shaken suspension for 2–4 days for G− phenotypes . All cultures were maintained at log phase ( 1–4 × 106 cells/ml ) during this period . Cells transformed by electroporation were selected with blasticidin ( 10 µg/ml ) or G418 ( 20–40 µg/ml ) . For development , amoebae were washed with KK2 ( 16 . 1 mM KH2PO4 , 3 . 7 mM K2HPO4 ) and deposited onto KK2 plates containing 1 . 5% purified agar ( Oxoid ) at a density of 3 . 5 × 106 amoebae/cm2 . Plates were kept for 14–16 hr at 22°C in a dark , moist box then removed and allowed to complete development in the light . REMI mutagenesis ( Kuspa and Loomis , 1992 ) was performed as described in Parkinson et al . , 2011 . 25 individual transformations producing approximately 1000 mutant clones were pooled after 4 days growth in 10 µg/ml blasticidin . This library of clones was used to perform two selections in parallel . Mutant G− cells were mixed with GFP labelled wild type G+ cells at a 10:90 ratio and developed to the mature fruiting body stage . Spores were collected first into spore buffer ( 10 mM EDTA , 0 . 1% NP-40 ) and then into HL5 +86 mM glucose . After hatching , amoebae were incubated with 10 µg/ml blasticidin for 4 days to remove wild type cells and enrich for mutants . This selection was repeated for seven rounds . Mutant loci were identified by iPCR ( Keim et al . , 2004 ) . Stalk cell differentiation was quantified using the cAMP removal described in Thompson et al . , 2004a . Expression of ecmA:lacZ and ecmB:lacZ reporter genes was induced by addition of 0 . 01–100 nM DIF-1 and 50 µM cerulenin to stalk medium ( 10 mM MES , pH6 . 2 , 1 mM CaCl2 , 2 mM NaCl , 10 mM KCl , 200 μg/ml streptomycin sulphate ) containing 5 mM cAMP . After 22 hr incubation at 22°C , levels of lacZ in cell lysates measured as described in Parkinson et al . , 2011 . Endogenous levels of ecmA and ecmB after 9 hr induction with 0 . 01–100 nM DIF were measured by qPCR as described in Huang et al . , 2006 . Nuclear translocation of GATAc-GFP in response to 100 nM DIF was measured as described in Keller and Thompson , 2008 . Individual-cell responses to DIF quantified by knocking in a single copy of GFP at the endogenous ecmA locus . Cells were incubated in stalk medium + 5 mM cAMP for 9 hr . Then , another 200 µl dose of 200 mM cAMP was added with 0 . 01–100 nM DIF . After a further 9 hr incubation , cells were resuspended in 1 ml KK2 and analysed on a Beckman Coulter Cyan ADP FACS machine . Exponentially growing rasD:RFP promoter cells were subjected to FACS analysis . Cells expressing low levels of actin:GFP were removed and high and low RasD expressing cells were collected . Those cells were mixed with unlabelled wild type cells in a 5:95 ratio and developed in chimera . To measure dynamics of RasD expression , FACS sorted low or high rasD:RFP cells were returned to HL-5 medium . As the numbers of recovered cells was quite low , unlabelled wild type cells were added to ensure exponential growth . The number of RFP/GFP expressing cells was scored every 12 hr for 2 days . Developmental structures were observed under a Leica S6D dissection microscope or a Leica MZ16FA stereoscope . Whole-mount lacZ staining performed as described in Dingermann et al . , 1989 . For measurement of prespore:prestalk ratio , dissociated slug stage cells were disaggregated in KK2/20 mM EDTA with 21 G needle . Cells were then fixed and stained with prespore-specific anti-psv antibody as described in Forman and Garrod , 1977 . For measurement of total spore number , 2 × 106 cells in 20 µl were spotted onto KK2 agar and incubated at 22°C in the dark for 2 days . Spores were harvested in KK2 with 0 . 1% NP40/20 mM EDTA and scored with a hemocytometer . The gefE− mutant allele ( Wilkins et al . , 2005 ) was obtained from the Dictybase Stock Center . In this study , two additional gefE− mutant alleles were generated . To disrupt the catalytic domain , a 2 . 1 kb fragment of the gefE gene was amplified by PCR ( 5′-GTAATGGCTCGAAAGTCCTTC and 5′-TTAAGACTTAAAAGAATT ) and cloned into the TOPO pcr2 . 1 vector . A floxed blasticidin resistance cassette derived from pLPBLP by Sma1 digestion was then inserted by blunt ligation at the endogenous Bst171 site in the catalytic domain of gefE . A second construct was made to delete a region of 855 bp from the catalytic domain . Two fragments of the gefE gene incorporating restriction sites were amplified by PCR ( gefE_sal–AGGCGTCGACCACCCTATAGTCCAGATAC , gefE_nco–AGGCATCGGAATCTCTGTTGGATC , gefE_pst–ATGCTGCAGCAGTTTCTGATCGACC , gefE_not–ATAGCGGCCGCCGTTGATTATGAGCAT ) and cloned into pLPBLP , one on each side of the floxed blasticidin . The rasD− mutant allele described in this paper was recreated with the pATW1 construct used to generate the original rasD− knockout ( Wilkins et al . , 2000 ) . For constitutive reporter gene expression , pDM318 or pDM324 ( Veltman et al . , 2009 ) were used . For cell type specific expression studies pDd19 ( pspA ) , pEcmAO-gal and pEcmB-gal vectors ( Jermyn and Williams , 1991 ) were used . Reporter genes were cloned into these vectors as required . The complete 3 . 1 kb gefE coding sequence was amplified from wild type cDNA by designing primers ( gefEF BamHI EEtag–CCCGGATCCAAAATGGAATATATGCCAATGGAAATGGATCATACTGAGTGTAAC , gefER SpeI-AATTACTAGTAGACTTAAAAGAATTTAAAAG ) and cloned into pDM324 with the appropriate restriction sites . The complete rasD gene was amplified from wild type gDNA by designing primers ( rasDF BglII EEtag–CCCAGATCTAAAATGGAATATATGCCAATGGAAACAGAATATAAATTAGTTATTGTAGG , rasDR SpeI–CCCCACTAGTTAAAATTAAACATTGTTTTTTC ) and cloned into pDM318 with the appropriate restriction sites . Using this vector as a template , primers were designed ( G12TF–GTTATTGTAGGTGGTACTGGTGTTGGTAAAAGTGC , and G12TR–GCACTTTTACCAACACCAGTACCACCTACAATAAC ) to generate the rasD ( G12T ) vector by PCR using a QuikChange I Site-Directed Mutagenesis Kit ( Agilent Technologies , UK ) . For gefE and rasD promoter studies , 900 bp or 700 bp of upstream sequence plus the first 7 or 13 codons , respectively , were amplified by designing primers ( gefEpF XhoI–AATTCTCGAGCAAAATTGATTGTAAAGCTGG , gefEpR BglII–AATTAGATCTGTTACACTCAGTATGATCCAT , rasDpF XhoI–ACTGCTCGAGTCTTATAATTTGGTTAAATCGATG , and rasDpR BglII–AAAAGATCTACCACCACCACCTACAATAAC ) and cloned into pDM324 with appropriate restriction sites . To ensure variation in RFP expression was specific to gefE or rasD promoters , actin promoter-driven GFP was cloned from pDM327 ( Veltman et al . , 2009 ) into NgoMIV site of each vector . First , the full length of GFP gene was amplified from pTX-GFP ( Levi et al . , 2000 ) by PCR ( GFPf_SalI–acgcGTCGACGGAACCAGTAAAGGAGAAGAAC and GFPr_stop_HindIII–acgcAAGCTTTTATGCATCTCGAGTGGAACC ) and cloned into SalI/HindIII sites of pLPBLP ( Faix et al . , 2004 ) . Then , two 2000 bp fragments of the upstream sequence plus the first codon and the downstream sequence including the stop codon of the ecmA gene were amplified from wild type genomic DNA by designing primers ( ecmA_up_F_KpnI acgcGGTACCATAGTCGATAATTTCATTACATCATC , ecmA_up_R_SalI acgcGTCGACCATTTTCAACGTTATAATTTTTAAAC , ecmA_down_F_PstI acgcCTGCAGTAAATAACTCTTTTTATTTAATTATATTTT , and ecmA_down_R_BamHI acgcGGATCCTACGTATTGAAATTCATCATCC ) . These PCR products were cloned into the GFP integrated pLPBLP with appropriate restriction sites . This construct was digested with KpnI/BamHI and electroporated into wild type or gefE− mutant cells to generate GFP knock-in recombinants . Vegetative or 12 hr filter developed cells were washed three times in Bonner’s salts ( 10 mM NaCl , 10 mM KCl , 2 mM CaCl2 ) . Cells were lysed with 2 X lysis buffer ( 20 mM sodium phosphate , PH 7 . 2 , 2% Triton X-100 , 20% glycerol , 300 mM NaCl , 20 mM MgCl2 , 2 mM EDTA , 2 mM Na3VO4 , 10 mM NaF , containing two tablets of protease inhibitor [Roche Complete] per 50 ml buffer ) . 400 μg of protein was then incubated with 100 μg of GST-Byr2-RBD on glutathione-sepharose beads at 4°C for 1 hr ( Bolourani et al . , 2010 ) . The glutathione-sepharose beads were harvested by centrifugation and washed three times in 1 X lysis buffer . 50 μl of 1 X SDS gel loading buffer was then added to the pelleted beads and the suspension boiled for 5 min . Samples were subjected to SDS-PAGE and Western blots probed with anti-RasD specific antibody .	How genetically identical cells develop into distinct cell types is one of the fundamental questions in biology . Certain molecules are known to act as signals that tell progenitor cells what type of cell they should become . The position of a cell within an embryo can determine which of these signals it is exposed to and thus influence its fate . However , it is also possible for a group of cells to be exposed to the same signal , but for only a few to respond . This gives rise to ‘salt and pepper’ differentiation—in which the cells differentiate in an apparently random manner to produce a mixture of different cell types—but the molecular basis of this phenomenon is unclear . An organism called Dictyostelium discoideum , commonly known as slime mould , is often used to study these processes . Dictyostelium has an unusual life cycle; existing as individual cells when its bacterial food source is plentiful , with the cells coming together when food is scarce to form a multicellular slug that can move around . Cells within the slug turn into spores or into stalk cells , which lift the spores above the ground so that they can disperse . Under the right conditions , a single cell hatches from each spore; upon finding a new food source , this cell begins dividing thus allowing the life cycle to begin again . The formation of stalk and spore cells occurs in a ‘salt and pepper’ pattern . A chemical messenger called DIF triggers cells to become stalk cells irrespective of their position within the aggregated mass of cells . Now , Chattwood et al . have shown that this process depends on the activity of two proteins; GefE and its substrate RasD . Surprisingly , both proteins are expressed many hours before cells differentiate , when cells are still well fed and dividing . Although GefE is uniformly expressed in these cells , its substrate—a protein called RasD—is expressed in only a subset of cells , and it is these cells that will later respond to DIF and ultimately become stalk cells . The variable expression of RasD explains how ‘salt and pepper’ patterning arises following uniform exposure of apparently identical cells to DIF . It is likely that similar mechanisms have been conserved in higher organisms , so these findings could lead to a better understanding of how progenitor cells develop into specific cell types in multicellular plants and animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2013	Developmental lineage priming in Dictyostelium by heterogeneous Ras activation
Goal-directed attention is usually studied by providing individuals with explicit instructions on what they should attend to . But in daily life , we often use past experiences to guide our attentional states . Given the importance of memory for predicting upcoming events , we hypothesized that memory-guided attention is supported by neural preparation for anticipated attentional states . We examined preparatory coding in the human hippocampus and mPFC , two regions that are important for memory-guided behaviors , in two tasks: one where attention was guided by memory and another in which attention was explicitly instructed . Hippocampus and mPFC exhibited higher activity for memory-guided vs . explicitly instructed attention . Furthermore , representations in both regions contained information about upcoming attentional states . In the hippocampus , this preparation was stronger for memory-guided attention , and occurred alongside stronger coupling with visual cortex during attentional guidance . These results highlight the mechanisms by which memories are used to prepare for upcoming attentional goals . Humans continuously experience rich perceptual input — input that exceeds the brain’s information processing capacity ( Luck and Vogel , 1997; Pylyshyn and Storm , 1988; Raymond et al . , 1992 ) . As a result , only a small portion of the information that is encountered on a moment-by-moment basis is fully processed . Indeed , unless attended , even very salient information can go undetected ( Neisser and Becklen , 1975; Simons and Chabris , 1999 ) . Despite this severe limitation in information processing capacity , we can adaptively and efficiently function in the complex environment around us . How do we figure out what to attend and what to ignore in the face of rich , multidimensional input ? In laboratory studies , goal-directed attention is typically studied by providing explicit instructions to participants ( Posner , 1980; Kastner and Ungerleider , 2000; Wolfe et al . , 1989 ) . For example , in cued attention tasks , participants are given particular target images or object categories that should be attended and detected ( e . g . , ‘find a human in this picture’; Wolfe et al . , 2011 ) . These studies have very compellingly shown that humans can guide attention based on top-down goals and highlighted the neural mechanisms that allow this to happen ( Gazzaley and Nobre , 2012; Hopfinger et al . , 2000 ) . However , in daily life , it is exceedingly rare to receive explicit instructions on how we should direct our attention . Instead , our attentional states are often guided by past experiences in similar situations ( Awh et al . , 2012 ) . Such memory-guided attention is effective in guiding goal-directed behavior ( Aly and Turk-Browne , 2017; Chen and Hutchinson , 2018; Nobre and Stokes , 2019 ) but is relatively under-explored . Here , we examine the mechanisms underlying memory-guided attention with the aim of determining the nature of neural representations that enable past experiences to be used to prepare for upcoming attentional states . We define ‘attentional state’ as the prioritized processing of particular environmental features in order to perform a given task . This entails focusing on task-relevant features , often at the expense of task-irrelevant features . Attentional states can be considered an instance of a task representation or a task set ( Mayr and Kliegl , 2000; Sakai , 2008 ) , with the task defining what should be attended to . What brain regions may establish memory-guided attentional states ? We focus on two candidate regions , the hippocampus and medial prefrontal cortex ( mPFC ) . Interactions between these regions have been linked to a variety of goal-directed behaviors that are guided by long-term memory ( Euston et al . , 2012; Kaplan et al . , 2017; Shin and Jadhav , 2016 ) . Furthermore , both the hippocampus ( Aly and Turk-Browne , 2016a; Aly and Turk-Browne , 2016b; Aly and Turk-Browne , 2018; Córdova et al . , 2019; Fenton et al . , 2010; Mack et al . , 2016; Muzzio et al . , 2009; Ruiz et al . , 2020 ) and mPFC ( Mack et al . , 2016; Small et al . , 2003 ) contribute to attentional processing . These findings suggest that the hippocampus and mPFC may work together to guide attentional behaviors on the basis of memory . Below , we explore their potential roles in more detail . Previous work from our lab has demonstrated that the hippocampus represents online attentional states ( Aly and Turk-Browne , 2016a; Aly and Turk-Browne , 2016b; Córdova et al . , 2019 ) . Moreover , decades of work have highlighted the critical role of the hippocampus in encoding and retrieving long-term memories ( Lepage et al . , 1998; Shapiro and Eichenbaum , 1999 ) . These findings therefore suggest that the hippocampus might play an important role in establishing memory-guided attentional states . In line with this , several studies have found that hippocampal activity levels are higher for memory-guided vs . explicitly instructed attention ( Aly and Turk-Browne , 2017; Stokes et al . , 2012; Summerfield et al . , 2006 ) . This activity enhancement for memory-guided attention is present as soon as information from memory is available , and even prior to attentional guidance . This suggests that the hippocampus may be using memory to direct attentional states in a preparatory fashion: Hippocampal memories might prepare perception for attentional requirements that are anticipated based on previous experiences ( Stokes et al . , 2012 ) . However , enhanced activity levels are ambiguous and do not by themselves establish what a brain region is doing to guide attention on the basis of memory . One possibility is that the hippocampus simply retrieves a memory that is then used by other brain areas to guide attention . An alternative possibility is that the hippocampus is itself engaged in the process of guiding attention based on past experience . For example , when using past experience to anticipate a navigational goal on the right-hand side , it could be that ( 1 ) the hippocampus retrieves a memory that your desired location is on the right , and other brain areas use that information to guide attention; or ( 2 ) the hippocampus itself codes for a rightward attentional bias in preparation for detecting the navigational goal . Because our prior studies have indicated that the hippocampus can represent attentional states that are currently in play ( Aly and Turk-Browne , 2016a; Aly and Turk-Browne , 2016b; Córdova et al . , 2019 ) we hypothesized that it can also represent attentional goals that are retrieved from memory , and use those to prepare for upcoming attentional tasks ( Stokes et al . , 2012; Summerfield et al . , 2006 ) . Beyond the hippocampus , mPFC may play an important role in memory-guided attention . In rodents , increased neural synchrony between the hippocampus and mPFC has been observed at decision points in which memory must be used to guide future behavior ( Benchenane et al . , 2010; Jones and Wilson , 2005 ) . In humans , functional magnetic resonance imaging ( fMRI ) studies have demonstrated that the orbitofrontal cortex ( a region in the ventral medial prefrontal cortex ) represents goal state representations that are not explicitly instructed but rather inferred on the basis of past experience ( Niv , 2019; Schuck et al . , 2015; Schuck et al . , 2016 ) . Moreover , the hippocampus and ventromedial PFC ( vmPFC ) show functional coupling as individuals learn which features of an object are relevant for determining its category , and thus should be attended ( Mack et al . , 2016 ) . Based on these studies , we predicted that vmPFC might also represent memory-guided attentional states . To test if the hippocampus and vmPFC represent attentional states that are guided by memory , we used a novel behavioral task in conjunction with representational similarity analyses ( Kriegeskorte et al . , 2008 ) . We were inspired by past work that demonstrated enhanced hippocampal activity in anticipation of attentional goals that were known based on memory ( Stokes et al . , 2012 ) as well as findings that link enhanced vmPFC activity to behavioral benefits that are attributed to the preparatory allocation of attention ( Small et al . , 2003 ) . Based on this work and the other findings noted above , we predicted that the hippocampus and vmPFC will establish memory-based attentional states prior to when those states must be used . To this end , we first sought to determine whether these regions can differentiate between different online attentional states , and then tested whether neural signatures of these states can be detected prior to the attentional task itself — with the hypothesis that these regions will prepare for upcoming attentional states primarily when they are guided by memory . We therefore compared attention in two tasks: One where attention was explicitly instructed , and one where attention was guided by memory . These tasks were modifications of ones we have previously used to demonstrate hippocampal representations of online attentional states ( Aly and Turk-Browne , 2016a; Aly and Turk-Browne , 2016b ) . One key feature of these tasks is that they require relational representations , which are known to be strong drivers of hippocampal function ( Aly et al . , 2013; Aly and Turk-Browne , 2018; Brown and Aggleton , 2001; Cohen and Eichenbaum , 1993; Davachi , 2006; Hannula and Ranganath , 2008 ) . Participants were shown sequentially presented images of 3D-rendered rooms , each of which had several pieces of furniture , unique configurations of wall angles , and a single painting ( Figure 1 ) . In the explicitly-instructed task , participants received a cue prior to the first image ( the base image ) that told them to pay attention to either the style of the paintings ( ‘ART’ ) or the spatial layout of the rooms ( ‘ROOM’ ) . Following the base image , participants viewed a search set of 4 other images . On ‘art’ trials , they were to attend to the style of the paintings , and indicate whether any of the paintings in the search set could have been painted by the same person who painted the painting in the base image . On ‘room’ trials , they were to attend to the layout of the rooms , and indicate whether any of the rooms in the search set had the same spatial layout as the base image , but viewed from a slightly different perspective . Finally , participants received a probe ( ‘ART ? ’ or ‘ROOM’ ? ) and had to indicate if any of the search images matched the base image in the probed category ( i . e . , painting by the same artist , or room with the same spatial layout ) . The memory-guided task had the same basic structure , except the attentional cue ( ‘ART’ or ‘ROOM’ ) was not overtly instructed at the beginning of each trial . Instead , attentional states were chosen by the participant based on stay and switch cues that were learned in an earlier phase of the experiment . Specifically , participants first learned four stimuli , two that signaled that they should stay in the same attentional state on the following trial ( ‘stay cues’ ) and two that signaled that they should switch to the other attentional state on the following trial ( ‘switch cues’ ) . During the subsequent attention task , a stay or switch cue could be embedded in the search set for any given trial . Thus , memory for the stay/switch cue on trial N , as well as memory for what that cue signaled , had to be used to guide attention on trial N+1 . In sum , we compared attention in two tasks: One where attentional goals were instructed at the beginning of each trial with an explicit cue , and one in which memory for specific images had to be used to select attentional goals . The tasks were identical otherwise — same stimuli , same motor demands — allowing us to rigorously test whether and how the hippocampus and vmPFC support memory-guided attention . Our main prediction was that these regions would prepare for upcoming attentional states , primarily when those states were guided by memory . We first examined behavioral performance with two goals in mind: First , to determine if attention was effectively manipulated , and second , to determine if performance was roughly equivalent across the memory-guided and explicitly instructed tasks . This would ensure that differences in brain activity levels across the tasks are unlikely to be driven by differences in task difficulty ( Barch et al . , 1997; McKiernan et al . , 2003 ) . To determine if attention was effectively engaged , we compared behavioral performance ( A’: 1 = perfect , 0 . 5 = chance , and response times ) on valid vs . invalid trials . On valid trials , the attentional cue at the beginning of the trial — whether it was selected by the participant based on memory , or explicitly instructed — matched the probe at the end ( e . g . , participants were attending to room layouts , and at the end of the trial were probed as to whether there was a room match ) . On invalid trials , the attentional cue at the beginning of the trial did not match the probe at the end ( e . g . , participants were attending to room layouts , and at the end of the trial were probed as to whether there was an art match ) . If attention is effectively engaged by the cue at the beginning of the trial , participants should be more accurate and faster on valid vs . invalid trials . This should be the case whether the attentional cue was selected by the participant based on memory , or explicitly instructed . We tested this with a 2-by-2 repeated measures ANOVA with the factors task ( memory-guided , explicitly instructed ) and cue validity ( valid , invalid ) . Indeed , behavioral sensitivity ( i . e . , A’ for detecting art or room matches ) was higher on valid trials ( M = 0 . 809 , 95% CI [0 . 787 , 0 . 831] ) compared to invalid trials ( M = 0 . 508 , 95% CI [0 . 451 , 0 . 565] ) , as revealed by a main effect of cue validity , F ( 1 , 28 ) =128 . 13 , p<0 . 0001 , ηp2 = 0 . 82 ( Figure 2 ) . In fact , sensitivity was higher than chance only on valid trials ( memory-guided: t ( 28 ) = 20 . 25 , p<0 . 0001 , d = 3 . 76 , 95% CI [0 . 768 , 0 . 828] , explicitly instructed: t ( 28 ) = 26 . 01 , p<0 . 0001 , d = 4 . 83 , 95% CI [0 . 795 , 0 . 846] ) , and not on invalid trials ( memory-guided: t ( 28 ) = 0 . 66 , p=0 . 51 , d = 0 . 12 , 95% CI [0 . 412 , 0 . 545] , explicitly-instructed: t ( 28 ) = 1 . 08 , p=0 . 29 , d = 0 . 20 , 95% CI [0 . 468 , 0 . 606] ) . Moreover , response times were slower on invalid compared to valid trials , F ( 1 , 28 ) =76 . 50 , p<0 . 0001 , ηp2 = 0 . 73 . These results suggest that our manipulation of attentional states was successful: Participants selectively attended to the category ( art; room ) that they chose in the memory-guided task and that they were instructed to attend in the explicitly instructed task . We next examined behavioral performance across the memory-guided and explicitly instructed tasks , and found that the difference between them was not statistically significant ( i . e . , no main effect of task ) , F ( 1 , 28 ) =3 . 20 , p=0 . 084 , ηp2 = 0 . 10 . The task by validity interaction was also not significant , F ( 1 , 28 ) =1 . 11 , p=0 . 30 , ηp2 = 0 . 04 . Because only valid trials were used in some fMRI analyses ( see Methods ) , we also compared task performance on valid trials only . Again , the difference in A’ for the memory-guided vs . explicitly instructed tasks was not statistically significant , t ( 28 ) = 1 . 32 , p=0 . 20 , d = 0 . 25 , 95% CI [−0 . 058 , 0 . 012] . Therefore , the tasks were of comparable difficulty , with similar modulations of attentional behavior by cue validity . To ensure that , in the memory-guided task , individuals were indeed using the stay and switch cues to guide their attentional states , we examined their accuracy in choosing the correct attentional state based on the stay/switch cue in the previous trial ( e . g . , choosing ‘room’ as the attentional goal when the previous trial contained either a room ‘stay’ cue or an art ‘switch’ cue ) . Decision accuracy was high and was not significantly different between ‘stay’ cues ( M = 0 . 949 , 95% CI [0 . 932 , 0 . 955] ) and ‘switch’ cues ( M = 0 . 967 , 95% CI [0 . 954 , 0 . 978] ) , t ( 28 ) = 1 . 68 , p=0 . 10 , d = 0 . 31 , 95% CI [−0 . 004 , 0 . 040] . Thus , participants were successfully able to use stay/switch cues to select memory-guided attentional goals . In daily life , we often use our memories to guide attention . For example , we use memory to decide where to attend when we navigate familiar routes , or which parts of the street to avoid because of dangerous potholes . However , attention in laboratory studies is typically investigated by providing explicit instructions to participants about what or where to attend . To bridge real-world behavior and laboratory studies , we explored the neural mechanisms underlying memory-guided vs . explicitly instructed attention . We designed two tasks that differed only in their requirement to use memory to guide attention . In the explicitly-instructed attention task , participants were given randomly determined attentional goals on each trial . In the memory-guided attention task , participants chose their attentional goals based on cues that had to be stored in memory . Based on previous studies implicating the hippocampus and vmPFC in memory-guided behaviors ( Euston et al . , 2012; Kaplan et al . , 2017; Shin and Jadhav , 2016 ) , we predicted that these regions would support the ability to use memory to prepare for anticipated attentional states . Extending prior work ( Stokes et al . , 2012; Summerfield et al . , 2006 ) , we found that activity levels in both hippocampus and vmPFC were higher for memory-guided vs . explicitly instructed attention . Furthermore , the memory-guided activity enhancements in hippocampus and vmPFC were correlated across individuals , suggesting that these regions may play a common role or work together for memory-guided attention . To further examine their role in memory-guided attention , we used representational similarity analyses ( Kriegeskorte et al . , 2008 ) to identify the information present in these regions in preparation for , and during , attentional guidance . Activity patterns in the hippocampus and vmPFC contained information about current and upcoming attentional states . Importantly , in the hippocampus , preparatory attentional state representations were stronger for memory-guided vs . explicitly instructed attention . Further analyses confirmed that these preparatory attentional states did not reflect retrieval of past attentional goals , but rather the anticipation of upcoming attentional states . Lastly , the hippocampus and early visual cortex ( V1-2 ) showed increased covariation in their attentional state representations in the memory-guided vs . explicitly instructed task . Together , these results elucidate how the hippocampus and vmPFC support memory-guided attention , and show that the hippocampus is preferentially involved in preparing for anticipated attentional goals that are guided by memory . Its role in memory-guided attention may be supported via its interactions with early visual cortex . These interactions may be the means by which mnemonically relevant information in the environment is detected and used to guide attention and perception . Thus , our work demonstrates the adaptive function of memories by highlighting the mechanisms by which past experiences can be used to prepare for future behaviors ( Nobre and Stokes , 2019 ) . Many studies of memory have focused on the importance of the hippocampus and vmPFC for memory-guided behaviors , such as navigational decisions ( Euston et al . , 2012; Kaplan et al . , 2017; Shin and Jadhav , 2016 ) . Because the world is complex and contains many more features than those that are currently relevant for our needs , memory can only guide effective behavior insofar as it can guide attention . Yet , studies of attention almost entirely ignore memory systems of the brain , and instead focus on sensory regions and frontoparietal control networks ( e . g . , Corbetta et al . , 2005; Ester et al . , 2016; Serences et al . , 2005 ) . To determine how memories can flexibly guide behavior , we must understand how memories , and memory systems of the brain , guide attention . We suggest that representations in , and coordination between , the hippocampus , early visual cortex , and vmPFC allow past experiences to trigger anticipation of upcoming attentional targets . In this way , memories of the past can be used to prepare for , and behave adaptively in , predicted environments . Our work therefore complements prior studies on predictive coding in the hippocampus ( Hindy et al . , 2016; Kok et al . , 2012 ) . Many such studies , however , focus on the representation of future navigational trajectories or navigational goals ( Brown et al . , 2016; Johnson et al . , 2007; Pfeiffer and Foster , 2013 ) . Here , we show that non-navigational , abstract attentional states are also represented in the hippocampus in a preparatory manner . To our knowledge , our study is the first to show that the hippocampus and vmPFC can prepare for anticipated attentional states . In this way , the current work takes principles and findings from research on memory and discovers their applicability to goal-directed attention . The current study also broadens the research literature on hippocampal contributions to attention ( Aly and Turk-Browne , 2017 ) . We have previously shown that attention modulates hippocampal representations ( Córdova et al . , 2019 ) and that this modulation predicts both online attentional behavior ( Aly and Turk-Browne , 2016a ) and memory formation ( Aly and Turk-Browne , 2016b ) . Furthermore , hippocampal damage impairs performance on attention tasks that require processing of spatial relations ( Ruiz et al . , 2020 ) . However , these studies are limited because they investigate attentional behaviors that are explicitly instructed , and thus are less ecologically valid than studies of memory-guided attention . Here , we expand on the contributions of the hippocampus to attentional behaviors by investigating scenarios in which attentional goals must be decided on the basis of past experience . Our work was inspired by studies of memory-guided attention ( e . g . , Stokes et al . , 2012; Summerfield et al . , 2006 ) but it differs from them in a number of ways . One key difference is that many of these prior studies involved teaching participants the relationship between particular memory cues ( e . g . , scenes ) and locations to be attended . Thus , participants were able to use memory to guide spatial attention , with knowledge of what visual content will be experienced . In contrast , participants in our study learned that particular memory cues signaled to either stay in the same attentional task or switch to a different one . This is akin to studies in which learned attention cues direct individuals to either hold or shift their current attentional focus ( e . g . , Chiu and Yantis , 2009; Greenberg et al . , 2010; Yantis et al . , 2002 ) . Furthermore , the current study involved some trials in which participants were free to choose what to attend; this is similar to studies investigating the neural correlates of self-directed attentional decisions ( Taylor et al . , 2008 ) . Although our study shares similarities with these latter investigations , it differs from studies of memory-guided attention in that memory did not allow individuals to anticipate specific visual content . Instead , it enabled participants to anticipate the upcoming task and , at a high-level , the types of visual features relevant for that task . Despite these differences , however , prior studies and ours share similarities . First , like other studies of attention , we found that manipulations of attentional cue validity led to robust behavioral consequences ( Posner , 1980; Stokes et al . , 2012; Summerfield et al . , 2006 ) : participants were faster and more accurate on valid vs . invalid trials , and their performance on invalid trials was not different from chance . Thus , although our study manipulates a more abstract form of attention relative to other studies , it replicates a key behavioral marker that is used as evidence for an attentional manipulation . Second , our study converges with other studies of memory-guided attention in suggesting that the hippocampus plays a role in guiding attentional behaviors on the basis of past experience ( see Aly and Turk-Browne , 2017 , for a review ) . For example , during the attentional search task ( i . e . , during the image period ) , hippocampus and vmPFC univariate activity levels were higher for memory-guided vs . explicitly instructed attention ( Figure 3 ) . This finding broadly replicates other studies of memory-guided attention , but enhanced univariate activity is somewhat ambiguous . Here , this difference could be a result of the demand to monitor the search set for remembered stay/switch cues , identify the meaning of those stay/switch cues , or it could reflect another cognitive process arising from the dual-task nature of the memory-guided condition . Thus , many potential cognitive functions can account for the univariate activity enhancement in hippocampus and vmPFC during memory-guided attention in this study . We also found that these regions showed no difference in univariate activity levels between the memory-guided and explicitly instructed conditions during the orienting period . This null univariate effect is in contrast to previous studies of memory-guided attention , which observed higher univariate activity in the hippocampus during preparation for memory-guided attention ( Stokes et al . , 2012 ) . Why might there be this difference between our findings and those of Stokes et al . ( 2012 ) ? One potential reason is the difference in information provided by memory . In Stokes et al . ( 2012 ) , the memory cues carried content-related information about target items: the cues signaled where in space a target will appear . Conversely , the memory ( stay/switch ) cues in the current study ( indirectly ) signaled the task that will be carried out on the upcoming trial , with no indication of specific visual content or targets that would appear . Furthermore , there was a long and variable blank delay between the orienting period and the attentional task in the Stokes et al . ( 2012 ) study; in the current study , the length of the orienting period was variable , but there was no blank delay between it and the attentional task . Thus , differences in the kind of information carried by memory ( specific content vs . abstract task set ) , as well as in the timing of the orienting periods and the attention task , could have led to the observed differences in univariate activity during preparatory attention . That said , another difference could be in the relative timing of memory retrieval in the two tasks . In order to use memory to anticipate upcoming attentional goals , one must first retrieve the relevant memory and then use it to prepare for the upcoming task at hand . The retrieval of an attentional goal and the use of this goal to prepare for upcoming tasks may be inextricably intertwined , but they may also be partly dissociable in time . One possibility , although speculative , is that hippocampal activity enhancements reflect memory retrieval of particular associations ( as in Stokes et al . , 2012 ) , and such memory retrieval occurred earlier in our task vs . that of Stokes et al . ( 2012 ) . Specifically , it is possible that individuals retrieved the meaning of stay/switch cues before the orienting period , e . g . , during the inter-trial interval or during the previous trial . This retrieved information may then be used to prepare for upcoming attentional states during the orienting period . Indeed , the image-period univariate activity enhancement in the hippocampus for memory-guided attention may reflect such memory retrieval ( Figure 3 ) . Future studies using methods with high temporal resolution ( e . g . , MEG/EEG ) will be useful for determining the temporal dynamics by which the hippocampus switches from retrieving a past memory to using that memory to anticipate upcoming attentional states — if indeed , these are separable processes as opposed to inherently linked . One final possibility for the different findings in our study and that of Stokes et al . ( 2012 ) is that univariate activity and multivariate activity patterns in the hippocampus are differentially sensitive to different kinds of information , e . g . , retrieval of specific memories ( Stokes et al . , 2012 ) vs . abstract task sets ( current study ) . Although once again speculative , this could potentially help explain why we observed effects during the orienting period in multivariate activity patterns but not overall univariate activity . Such a dissociation in the information present in univariate activity vs . pattern similarity is consistent with the finding that multivariate attentional state representations are dissociable from changes in overall activity levels ( Aly and Turk-Browne , 2016a ) . When a brain region prepares for , or anticipates , an upcoming task , what is being represented ? We have referred to the orienting period activity patterns in hippocampus and vmPFC as reflecting preparatory attentional states . This is because activity patterns prior to , or in preparation for , an upcoming attentional task resembled those during the task itself . However , a number of different cognitive processes can lead to overlap in brain representations for engaging in a task and anticipating it . We consider these below . One possibility is that preparatory attentional states observed in our study reflect the anticipated difficulty of art and room attentional states . For example , if a participant finds attending to art more challenging than attending to rooms , they may modulate arousal or effort when anticipating an art trial . This modulation of arousal or effort may have an effect on activity patterns in the hippocampus or vmPFC . As a result , activity patterns during the anticipation and execution of an art trial would be similar due to shared effort- or arousal-related components . If this is the case , individuals who found one attentional state much more difficult than the other ( e . g . , art harder than room or vice versa ) should show stronger evidence of neural preparation . However , we did not find any significant correlations between performance differences on art and room trials and the strength of anticipatory attentional state representations ( all ps > 23 ) . Thus , we argue that differences in difficulty between art and room trials are unlikely to be the driving factor for pattern similarity across the orienting period and image period . That said , differences in subjective assessments of difficulty may nevertheless contribute to the extent of neural preparation , even if objective performance differences do not seem to . Previous studies have shown preparatory coding for concrete shapes and locations in the hippocampus and sensory regions ( Battistoni et al . , 2017; Corbetta et al . , 2005; Hindy et al . , 2016; Kok et al . , 2012; Stokes et al . , 2009 ) . Preparatory representations of anticipated shapes or locations , in turn , are thought to facilitate the perception of task-relevant information in the external world ( Battistoni et al . , 2017 ) . Is the preparatory coding observed in our study indicative of the brain’s anticipation of particular objects or locations , or is it more abstract in nature ? Accordingly , another possibility is that participants , upon anticipating an art or room attentional state , start to represent concrete visual features related to those categories . For example , they might bring to mind paintings or rooms that were previously seen in the experiment . However , this approach may not be effective , because the particular paintings or rooms imagined are unlikely to be the specific ones relevant on that trial ( because of the large number of images used in the experiment ) . A mismatch between imagined visual features and those that end up being relevant might hurt performance instead of boosting it . As a result , it may not be adaptive for individuals to bring to mind specific paintings or rooms in preparation for the upcoming attentional state . Instead , it may be beneficial to prioritize the visual system and hippocampus to process spatial/global information in general ( for the room task ) or color/object/local information in general ( for the art task ) . Thus , the preparatory attentional states that we observed may be relatively abstract in nature . This is particularly likely because the presence of these preparatory states was established by examining the similarity between activity patterns related to preparation ( during the orienting period ) and activity patterns related to attentional guidance ( during the image period ) . Given that these image period activity patterns were calculated across trials that used many different visual images , they presumably reflect attentional states that are abstracted away from specific visual features on any given trial . However , what those abstractions are is not clear from the current study . The preparatory signals in hippocampus and vmPFC might reflect an abstract attentional orientation ( attend to local features vs . global features; attend to color vs . geometry ) , maintenance of a task instruction ( find a similar painting vs . find a similar room ) , or even a metacognitive state ( ‘The art task is harder for me , so I should expend more effort’ ) . As long as these cognitive processes occur during both the orienting period and the image period , they may be components of the observed preparatory signals . The representational nature of the preparatory attentional states that are observed in the present study therefore deserves further investigation . One key limitation of the current study is the absence of a long period of no visual stimulation between the orienting period and the image period . A long blank period would have allowed cleaner isolation of preparatory signals from those related to carrying out the task itself . However , several measures were taken to reduce autocorrelation when comparing activity patterns from the orienting period to those from the image period , and we argue that the current results are difficult to explain with autocorrelation ( see Robustness of preparatory attentional states and Methods ) . Nevertheless , it would be ideal for future studies to include a longer delay between the orienting period and image period , for better isolation of anticipatory neural states . This would be particularly useful if fMRI were complemented with EEG , to incorporate the high temporal resolution of the latter method ( e . g . , Stokes et al . , 2012 ) . Attention can be guided by many forms of memory at multiple timescales ( Nobre and Stokes , 2019 ) . Which are at play in the current study ? We believe that long-term memory , intermediate-term memory , and working memory all contribute . We elaborate on these below . Long-term memory plays an essential role in our memory-guided task because the stay/switch cues that were used to select attentional states were well-learned ~30 min prior to the fMRI scan . Participants showed near-perfect performance in using these cues to select the correct attentional state . Moreover , the ability to detect art or room matches did not differ between the memory-guided and explicitly instructed tasks ( Figure 2 ) , suggesting that the additional demand to identify stay/switch cues in the memory-guided task might have been relatively automatized ( Logan , 1988 ) . Therefore , the long-term memories used to identify the stay/switch cues and retrieve their meanings were well-learned , and possibly partly semanticized . Indeed , semantic memories can contribute to the guidance of attention ( Brockmole and Le-Hoa Võ , 2010; Moores et al . , 2003; Olivers , 2011; Torralba et al . , 2006 ) . This is common in daily life , where many cues that are used to direct attention ( e . g . , traffic signs ) are extensively practiced and retained in semantic memory . However , memories for the stay/switch cues in the current study are likely not semantic to the same extent as memories for traffic signs , the latter of which are learned and practiced over a lifetime rather than ~30 min . Thus , although the stay and switch cues were well-learned , they were learned the same day as the fMRI scan and thus unlikely to be truly semanticized . Instead , they might more closely resemble episodic memories . The second timescale of memory that may have contributed to attentional guidance in the current study lies somewhere between long-term and working memory: the relatively intermediate-term memory for what occurred on the previous trial . Specifically , when a new trial starts , participants have to remember their attentional state on the previous trial , and whether there was a stay or switch cue in the previous trial , to select their attentional state . Alternatively , participants may decide their attentional state for the following trial as soon as they see a stay/switch cue , and then store the intention in memory until the following trial starts . This memory — whether it is a memory for the intention or a memory for the stay/switch cue — might be stored as an episodic trace during the inter-trial interval and recalled at the beginning of the next trial . This would be consistent with work demonstrating that episodic memories can bias attention ( Stokes et al . , 2012; Summerfield et al . , 2006 ) . Alternatively , this information may be maintained in working memory throughout the inter-trial interval until the onset of the following trial . Finally , once an individual decides what to attend to — or is told what they should attend to based on an explicit instruction — this attentional state is likely represented in working memory over the course of visual search . Indeed , attentional templates stored in working memory guide attention and bias perception in a way that aligns with attentional goals ( Carlisle et al . , 2011; Chelazzi et al . , 1998; Desimone , 1996; Gunseli et al . , 2014a; Gunseli et al . , 2014b; Olivers et al . , 2011; Gunseli et al . , 2016 ) . This form of working-memory-guided attention should contribute to performance in both the memory-guided and explicitly instructed tasks . In sum , multiple timescales of memory likely contributed to performance in the current task ( Hutchinson and Turk-Browne , 2012; Nobre and Stokes , 2019 ) : long-term , overlearned memories; intermediate-term episodic memories; and working memory . Future studies will be useful for understanding the similarities and differences between attentional guidance by memories at these timescales . For example , one question is whether the hippocampus can be involved in the guidance of attention by semantic memories ( e . g . , when detecting and responding to a traffic sign ) or if it is preferentially involved when episodic memories guide attention ( e . g . , when avoiding a pothole that we noticed yesterday ) . Such a question can also help better isolate the complementary roles of the hippocampus and vmPFC in memory-guided attention . It is possible that more semanticized or consolidated episodic memories might call on vmPFC to guide attention , while the hippocampus is more important for the guidance of attention by relatively recent or rich episodic memories . This would be consistent with the differential role of these regions in semanticized vs . vivid episodic memories ( Bonnici and Maguire , 2018; Sekeres et al . , 2018 ) . The current study confirmed our hypothesis that the hippocampus and vmPFC prepare for upcoming attentional states . However , contrary to our hypotheses , only the hippocampus — and not vmPFC — showed stronger preparation for memory-guided attention . Why might this be ? There are at least two possible explanations . First , vmPFC might weight explicit instructions and memories equally when preparing for upcoming task goals , while the hippocampus may prioritize information that is retrieved from memory . Given the importance of the hippocampus for memory retrieval , it is reasonable that information that arises from within the hippocampus itself might , at least in some situations ( Tarder-Stoll et al . , 2020 ) be prioritized relative to information from the external environment . An alternative possibility is that the hippocampus is capable of preparing for upcoming attentional states equally strongly regardless of how these states are guided ( i . e . , by memories vs . explicit instructions ) — but we were not able to observe this in our task because of limitations of the experimental design . In particular , the upcoming attentional state was known for longer in the memory-guided vs . explicitly instructed task: attentional states for trial N were known as soon as trial N - 1 was over for the memory-guided task , but only known when the attentional cue was displayed on trial N for the explicitly instructed task . Furthermore , the attention task started relatively soon after the attentional cue was shown . Thus , it is possible that vmPFC is able to rapidly prepare for upcoming attentional states regardless of how they are known , but the hippocampus needs more time in order to represent attentional goals that are cued by the environment . Future studies that use methods with higher temporal resolution ( e . g . , EEG/MEG ) , and longer delays between when attentional goals are known and when they must be used , will be needed to explore this question . Such methods can establish the temporal dynamics by which memory-guided vs . explicitly instructed attention influence representations across different brain regions . What is the benefit of preparatory attentional states ? Previous research has shown that representations in early visual cortex are sharpened for anticipated stimuli ( e . g . , Kok et al . , 2012 ) . Furthermore , attentional modulation of early visual cortex can bias the detection of goal-relevant information over distractors ( Peelen and Kastner , 2011; Reynolds et al . , 1999; Stokes et al . , 2009 ) . Such a biasing process has primarily been studied when attention is explicitly instructed . When attention is guided by memory , the hippocampus might be important for preparing visual cortex for task-relevant features ( Stokes et al . , 2012 ) . For example , hippocampal anticipation of upcoming attentional states might enable visual cortex to prioritize the processing of task-relevant information . Indeed , hippocampal pattern completion is associated with predictive coding in early visual cortex ( Hindy et al . , 2016 ) . The potential importance of hippocampal interactions with visual cortex for memory-guided attention was also evident in our study: The attentional states of hippocampus and early visual cortex were more strongly coupled for memory-guided vs . explicitly instructed attention . Such covariation may allow mnemonically relevant information to be detected in the environment , and then subsequently used by the hippocampus to prepare for upcoming attentional states . Future studies that investigate the direction of information flow between hippocampus and early visual cortex can test whether visual cortex first influences the hippocampus to cue the retrieval of relevant information , and whether this direction of influence reverses once hippocampal memories can be used to anticipate attentional states ( Place et al . , 2016 ) . We have largely considered the complementary functions of attention and memory: how memories can be used to guide attentional behavior . Yet , there can also be a tension between attention and memory , particularly when attention to the external world has to be balanced against the processing of internally retrieved memories . How does the hippocampus balance the demand between externally and internally oriented attention ? This is particularly interesting to examine in cases like the current study , where both external attention and memory retrieval are needed for the effective guidance of behavior . One hypothesis is that the hippocampus might rapidly fluctuate between internal and external modes , prioritizing either attention/encoding or memory retrieval at different timepoints ( Hasselmo , 1995; Hasselmo and Fehlau , 2001; Hasselmo and Schnell , 1994; Hasselmo et al . , 1996; Honey et al . , 2018; Meeter et al . , 2004; Patil and Duncan , 2018; Tarder-Stoll et al . , 2020 ) . Although there are ‘background’ fluctuations between external and internal attention in the hippocampus , top-down goals or external factors ( e . g . , surprise ) can also affect these fluctuations ( Sinclair and Barense , 2019 ) . Thus , one possibility is that the appearance of a stay/switch cue briefly switches the hippocampus from an externally oriented state to an internally focused one . Future studies will be needed to explore how the demands of internal and external attention are balanced by the hippocampus in the context of memory-guided attention . Memories frequently guide attention in the real world , but how they do so is relatively under-explored . We have shown that the hippocampus and vmPFC prepare for anticipated attentional states , and the hippocampus does so more strongly for attentional states that are selected on the basis of memory . Furthermore , attentional states in the hippocampus correlate , on a trial-by-trial basis , with those in early visual cortex when attention is guided by memories . This informational connectivity may be essential for enabling perceptual signals to cue memory-guided goals and for memory-guided goals to bias perception . Together , these findings suggest that memories can be flexibly used to guide attentional behavior , and that this process calls on representations in , and coordination between , systems involved in memory and perception . Thirty individuals from the Columbia University community participated for monetary compensation ( $12/hour for behavioral sessions and $20/hour for the fMRI session; $72 in total ) . The study was approved by the Institutional Review Board at Columbia University . Written informed consent was obtained from all participants . One participant did not perform well on the memory-guided attention task , as indicated by poor accuracy in using stay/switch cues to guide attention ( M = 0 . 847 ) . This person’s accuracy was more than three standard deviations below the group average ( M = 0 . 954; SD = 0 . 0317 ) , suggesting that they were not effectively using memory to select attentional goals . We therefore excluded this participant from the analyses , leaving 29 participants ( 17 female; one left-handed; all normal or corrected-to-normal vision; 18–35 years old , M = 26 , SD = 4 . 07; 13–21 years of education , M = 17 . 1 , SD = 2 . 2 ) . MRI data were collected on a 3 T Siemens Magnetom Prisma scanner with a 64-channel head coil . Functional images were obtained with a multiband echo-planar imaging ( EPI ) sequence ( repetition time = 1 . 5 s , echo time = 30 ms , flip angle = 65° , acceleration factor = 3 , voxel size = 2 mm iso ) , with 69 oblique axial slices ( 14° transverse to coronal ) acquired in an interleaved order . There were eight functional runs , four for the explicitly instructed task and four for the memory-guided task . Whole-brain high-resolution ( 1 . 0 mm iso ) T1-weighted structural images were acquired with a magnetization-prepared rapid acquisition gradient-echo sequence ( MPRAGE ) . Field maps were collected to aid registration , consisting of 69 oblique axial slices ( 2 mm isotropic ) .	At any given moment , humans are bombarded with a constant stream of new information . But the brain can take in only a fraction of that information at once . So how does the brain decide what to pay attention to and what to ignore ? Many laboratory studies of attention avoid this issue by simply telling participants what to attend to . But in daily life , people rarely receive instructions like that . Instead people must often rely on past experiences to guide their attention . When cycling close to home , for example , a person knows to watch out for the blind junction at the top of the hill and for the large pothole just around the corner . Günseli and Aly set out to bridge the gap between laboratory studies of attention and real-world experience by asking healthy volunteers to perform two versions of a task while lying inside a brain scanner . The task involved looking at pictures of rooms with different shapes . Each room also contained a different painting . In one version of the task , the volunteers were told to pay attention to either the paintings or to the room shapes . In the other version , the volunteers had to use previously memorized cues to work out for themselves whether they should focus on the paintings or on the shapes . The brain scans showed that two areas of the brain with roles in memory – the hippocampus and the prefrontal cortex – were involved in the task . Notably , both areas increased their activity when the volunteers used memory to guide their attention , compared to when they received instructions telling them what to focus on . Moreover , patterns of activity within the hippocampus and prefrontal cortex contained information about what the participants were about to focus on next – even before volunteers saw the particular picture that they were supposed to pay attention to . In the hippocampus , this was particularly the case when the volunteers based their decisions on memory . These results reveal a key way in which humans leverage memories of past experiences to help optimize future behavior . Understanding this process could shed light on why memory impairments make it harder for people to adjust their behavior to achieve specific goals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Preparation for upcoming attentional states in the hippocampus and medial prefrontal cortex
Changing receptor abundance at synapses is an important mechanism for regulating synaptic strength . Synapses contain two pools of receptors , immobilized and diffusing receptors , both of which are confined to post-synaptic elements . Here we show that immobile and diffusing GABAA receptors are stabilized by distinct synaptic scaffolds at C . elegans neuromuscular junctions . Immobilized GABAA receptors are stabilized by binding to FRM-3/EPB4 . 1 and LIN-2A/CASK . Diffusing GABAA receptors are stabilized by the synaptic adhesion molecules Neurexin and Neuroligin . Inhibitory post-synaptic currents are eliminated in double mutants lacking both scaffolds . Neurexin , Neuroligin , and CASK mutations are all linked to Autism Spectrum Disorders ( ASD ) . Our results suggest that these mutations may directly alter inhibitory transmission , which could contribute to the developmental and cognitive deficits observed in ASD . Fast synaptic inhibition is primarily mediated by the neurotransmitter GABA and GABA-activated chloride channels ( GABAA receptors ) . Several studies suggest that an important mechanism for modulating inhibitory transmission is altered abundance of synaptic GABAA receptors . In mammalian neurons , variation in the amplitude of miniature inhibitory post-synaptic currents ( mIPSCs ) is caused by corresponding differences in the abundance of GABAA receptors at synapses ( Nusser et al . , 1997 ) . Long term potentiation of GABAergic transmission is associated with increased mIPSC amplitudes and increased GABAA abundance at synapses ( Petrini et al . , 2014 ) while the converse effects are associated with long term depression ( Bannai et al . , 2009 ) . GABAA receptors on the cell surface are mobile , undergoing lateral diffusion in the plasma membrane ( Jacob et al . , 2005 ) . Like all synaptic receptors , GABAA diffusion is significantly reduced at synapses , resulting in accumulation of receptors at the synapse ( Jacob et al . , 2005; Thomas et al . , 2005; Bannai et al . , 2009; Petrini et al . , 2014 ) . Local confinement of receptors at synapses is termed diffusional trapping and is mediated by binding to cytoplasmic scaffolds ( Choquet and Triller , 2013 ) . The post-synaptic scaffold that immobilizes GABAA receptors is proposed to consist of a ternary complex of Gephyrin , Neuroligin-2 ( NL2 ) , and collybistin ( Jacob et al . , 2005; Poulopoulos et al . , 2009 ) . Gephyrin binds directly to the large cytoplasmic loop between the third and fourth transmembrane domains ( TM3-4 loop ) of GABRA1 and 2 subunits ( Tretter et al . , 2008 ) , thereby confining these receptors at synapses ( Jacob et al . , 2005; Mukherjee et al . , 2011; Saliba et al . , 2012 ) . Genetic manipulations impairing the Gephyrin/NL2/Collybistin complex invariably decrease but fail to eliminate synaptic GABAA receptors ( Kneussel et al . , 1999; Papadopoulos et al . , 2007; Poulopoulos et al . , 2009 ) . Thus , it is likely that additional proteins are involved in this process . Within a post-synaptic element , receptors exhibit heterogenous behavior ( Choquet and Triller , 2013 ) . At both excitatory and inhibitory synapses , super-resolution imaging suggests that a subset of receptors are localized in immobile nanoclusters ( ∼75 nm in diameter ) ( Nair et al . , 2013; Specht et al . , 2013 ) . These immobile receptors undergo dynamic exchange with diffusing receptors that are confined to synapses . These studies highlight several important questions . Do immobile and diffusing receptors both contribute to IPSCs ? Current models propose that post-synaptic currents are mediated by immobilized receptors and synaptic plasticity is mediated by the dynamic exchange of receptors between the diffusing and immobile pools ( Choquet and Triller , 2013 ) . It has not been possible to genetically test these models because mutations that selectively disrupt the two receptor pools are not available . What are the synaptic scaffolds that stabilize immobilized and diffusing GABAA receptors ? What controls the exchange between the two receptor pools ? Here we utilize the C . elegans neuromuscular junction ( NMJ ) as a model to address these questions . We show that immobilized and diffusing GABAA receptors are stabilized by two distinct post-synaptic scaffolds both of which contain subunits encoded by genes linked to ASD . C . elegans body muscles receive direct inhibitory input from GABAergic motor neurons ( White et al . , 1986 ) . The GABAA receptors found at these NMJs contain two subunits ( UNC-49B and C ) both encoded by the unc-49 gene ( Bamber et al . , 1999 ) . Mutants lacking UNC-49 receptors have defects in GABA-activated muscle currents , as assessed by recording miniature inhibitory post-synaptic currents ( mIPSCs ) and muscle currents evoked by an exogenous GABA agonist ( muscimol ) ( Figure 1A–E ) . An mIPSC corresponds to the current evoked by the fusion of a single synaptic vesicle at a GABAergic NMJ and , consequently , measures the function of synaptic UNC-49 receptors . Muscimol activates all surface UNC-49 receptors ( including non-synaptic receptors in the nerve cord and muscle cell bodies ) . The mIPSC ( Figure 1A–C ) and muscimol-evoked current ( Figure 1D , E ) defects of unc-49 mutants were rescued by a transgene expressing GFP-tagged UNC-49B receptor in body muscles . Muscimol-evoked currents were significantly larger in GFP-UNC-49B transgenic animals ( Figure 1E ) , presumably because this transgene is expressed at higher levels than the endogenous unc-49 gene . These results demonstrate that the GFP-tag ( inserted into the TM3-4 loop ) did not impair UNC-49B receptor function . 10 . 7554/eLife . 09648 . 003Figure 1 . Inhibitory NMJs contain both mobile and immobile UNC-49/GABAA receptors . ( A–E ) mIPSCs and muscimol-evoked currents were abolished in unc-49 mutants , and were restored by a transgene expressing GFP-tagged UNC-49B in body muscles . mIPSCs ( A–C ) and muscimol-evoked currents ( D , E ) were recorded from adult body wall muscles . For mIPSCs , representative traces ( A ) , mean current amplitude ( B ) and mean frequency ( C ) are shown . For muscimol-evoked currents , a representative wild type response ( D ) , and mean current amplitude ( E ) are shown . GFP-tagged UNC-49B is localized to GABAergic NMJs . ( F ) The distribution of muscle expressed GFP-UNC-49B ( Green ) is compared to presynaptic RAB-3::mCherry ( Red ) , expressed in GABAergic motor neurons ( scale bar 5 μm ) . ( G–I ) Synaptic UNC-49B consists of both mobile and immobilized receptors . The mobility of synaptic GFP-UNC-49B and pHluorin-tagged UNC-49B ( pH-UNC-49B ) was analyzed by FRAP . Representative images of GFP-UNC-49B FRAP ( G ) , a representative scatter plot of GFP-UNC-49B fluorescence recovery ( solid line indicates a single exponential fit ) ( H ) , and summary data for fluorescence recovery of GFP- and pH-UNC-49B ( I ) are shown . Examples of scatter plots for pH-UNC-49B recovery are shown in Figure 4—figure supplement 1A . Pre-synaptic RAB-3::mCherry fluorescence was captured as control . Values that differ significantly are indicated ( * , p < 0 . 001; ns , not significant ) . The number of animals analysed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 003 In wild type animals , GFP-UNC-49B fluorescence exhibits a punctate distribution where each punctum is closely apposed to GABAergic motor neuron nerve terminals ( labelled with mCherry-tagged RAB-3 ) ( Figure 1F ) , consistent with prior studies ( Thompson-Peer et al . , 2012 ) . To assess the mobility of synaptic UNC-49B receptors , we measured fluorescence recovery after photobleaching ( FRAP ) of GFP-UNC-49B puncta in the dorsal nerve cord ( Figure 1G–I ) . FRAP assesses the mobility of proteins in vivo whereby increased mobility is indicated by increased FRAP . UNC-49B puncta that were co-localized with mCherry-tagged RAB-3 expressed in GABAergic motor neurons were considered synaptic . In wild type controls , 40% of UNC-49B puncta fluorescence was mobile in FRAP experiments , with recovery occurring several minutes after photobleaching ( Figure 1G , H ) . GFP-UNC-49B puncta fluorescence could comprise receptors on the cell surface and those in intracellular organelles . To more accurately assess the mobility of surface receptors , we analyzed UNC-49B receptors containing a pH-sensitive GFP ( pHluorin ) tag in the ecto-domain ( pH-UNC-49B ) . pHluorin fluorescence is quenched in intracellular acidic compartments ( e . g . , endosomes ) ; consequently , pHluorin fluorescence primarily results from molecules in the plasma membrane ( Miesenbock et al . , 1998 ) . In wild type animals , 46% of pH-UNC-49B synaptic fluorescence recovered following photobleaching ( Figure 1I ) . Because similar mobile fractions were observed with the GFP and pHluorin tagged receptors , these results suggest that the majority of UNC-49B puncta fluorescence results from receptors in the plasma membrane . Collectively , these results suggest that UNC-49B synaptic puncta comprise a mixture of mobile ( ∼40% total ) and immobilized ( ∼60% total ) receptors on the cell surface . Prior studies utilizing both FRAP and single molecule tracking techniques reported similar proportions of mobile and immobilized receptors at both excitatory and inhibitory synapses in cultured mammalian neurons ( Jacob et al . , 2005; Ashby et al . , 2006; Heine et al . , 2008 ) . ERM ( Ezrin/Radixin/Moesin ) domain containing proteins couple cell surface receptors to the actin cytoskeleton ( Tepass , 2009 ) and are implicated in targeting synaptic glutamate receptors and extra-synaptic GABAA receptors in neurons ( Biederer and Sudhof , 2001; Loebrich et al . , 2006 ) . To identify ERM proteins that could play a role in UNC-49B targeting , we screened all C . elegans ERM proteins and found that FRM-3 binds the UNC-49B TM3-4 loop in yeast 2-hybrid assays ( Figure 2A ) . FRM-3 is a band 4 . 1 ( EPB4 . 1 ) paralog . We did three additional experiments to determine if FRM-3 binds UNC-49B in vivo . First , we showed that a frm-3 promoter construct expressed GFP in body muscles ( Figure 2—figure supplement 1A ) , consistent with FRM-3 function in muscles . Second , we showed that GFP-tagged FRM-3 expressed in body muscles formed puncta in the nerve cord that were co-localized with a post-synaptic marker for GABAergic NMJs ( mCherry-tagged NLG-1/Neuroligin ) ( Pearson's correlation R = 0 . 80 ± 0 . 028 , p = 0 . 02 , n = 8 ) ( Figure 2B ) ( Maro et al . , 2015; Tu et al . , 2015 ) . Third , we showed that FLAG-tagged FRM-3 and GFP-UNC-49B co-immunoprecipitated from worm extracts , when both were expressed in body muscles ( Figure 2—figure supplement 1B ) . Collectively , these results suggest that FRM-3 is localized to GABAergic synapses where it may directly bind UNC-49B receptors . 10 . 7554/eLife . 09648 . 004Figure 2 . FRM-3 EPB4 . 1 binds UNC-49B and is required for its synaptic targeting . ( A ) FRM-3′s ERM domain binds the UNC-49B TM3-4 loop in yeast 2-hybrid assays . Growth of Y2HGold cells on selective media ( –Trp/-Leu/-His/-Ade ) is shown . Yeast cells were transformed with vectors expressing the indicated fusion proteins . Positive ( + , pGBKT7-53 and pGADT7-T ) and negative ( − , pGBKT7-Lam and pGADT7-T ) controls are indicated . ( B ) Muscle expressed FRM-3::GFP ( Green ) and NLG-1::mCherry ( Red ) are co-localized in the nerve cord ( scale bar 5 μm ) . ( C , D ) GFP-UNC-49B puncta fluorescence in the nerve cord was decreased in frm-3 mutants . This defect was rescued by transgenes expressing FRM-3 in body muscles ( M ) but not by those expressed in GABAergic neurons ( N ) . Representative images ( C , scale bar 5 μm ) and mean puncta intensity ( D ) are shown . ( E–G ) mIPSC amplitude was decreased in frm-3 mutants and this defect was rescued by restoring FRM-3 expression in body muscles ( M resc . ) . mIPSCs were recorded from adult body wall muscles . Representative traces ( E ) , mean amplitude ( F ) and mean frequency ( G ) are shown . ( H ) The function of total surface UNC-49 receptors was unaltered in frm-3 mutants . Muscimol-activated currents were recorded from adult body muscles . Mean peak currents are shown . Values that differ significantly are indicated ( * , p < 0 . 001; , p < 0 . 01; ns , not significant ) . The number of animals analysed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 00410 . 7554/eLife . 09648 . 005Figure 2—figure supplement 1 . FRM-3 is expressed in body muscles and binds to UNC-49B . ( A ) Transgenes containing the frm-3 promoters express GFP in body muscles . Muscle arms are indicated by the asterisks . Scale bar , 25 μm . ( B ) FLAG-tagged FRM-3 and GFP-UNC-49B co-immunoprecipitate from worm extracts . FLAG-FRM-3 was immunoprecipitated from worm membrane extracts and bound proteins were analyzed by immunoblotting with FLAG ( top ) and GFP ( bottom ) antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 005 To estimate the abundance of synaptic GABAA receptors , we measured the intensity GFP-UNC-49B puncta in the dorsal nerve cord . In frm-3 mutants , UNC-49B puncta fluorescence was significantly reduced and this defect was rescued by transgenes restoring FRM-3 expression in body muscles but not by those expressed in GABAergic motor neurons ( Figure 2C , D ) . To determine if expression of endogenous UNC-49 receptors was also altered , we patch clamped body muscles and recorded mIPSCs and muscimol-evoked currents . mIPSC rate was unaffected in frm-3 mutants , implying that pre-synaptic GABA release was not significantly altered ( Figure 2E , G ) . mIPSC amplitude was significantly reduced in frm-3 mutants ( Figure 2E , F ) , consistent with decreased UNC-49 abundance at synapses . The frm-3 mIPSC amplitude defect was rescued by transgenes restoring FRM-3 expression in body muscles ( Figure 2E , F ) , implying that FRM-3 acts in muscles to promote the function of synaptic UNC-49 receptors . Muscimol-evoked currents were unaltered in frm-3 mutants ( Figure 2H ) ; consequently , the frm-3 mutant mIPSC defect is unlikely to result from decreased bulk expression and surface delivery of UNC-49 receptors . Collectively , these results suggest that muscle FRM-3 promotes the localization and function of synaptic UNC-49 receptors but is not required for the function or trafficking of non-synaptic UNC-49 receptors . CASK is a synaptic scaffolding protein that binds directly to EPB4 . 1 ( Biederer and Sudhof , 2001 ) . C . elegans has two predicted CASK isoforms ( LIN-2A and B ) , both encoded by the lin-2 gene ( Hoskins et al . , 1996 ) . LIN-2A and B share the PDZ , SH3 , and GK domains while only LIN-2A contains the CaMK homology domain ( Hoskins et al . , 1996 ) . LIN-2A is required for targeting epidermal growth factor receptors ( EGFRs ) to the basolateral domain of epithelial cells ( Simske et al . , 1996 ) . The impact of LIN-2/CASK on GABAA receptors has not been determined . Like their mammalian counterparts ( Biederer and Sudhof , 2001 ) , LIN-2A/CASK interacted with FRM-3/EPB4 . 1 in yeast 2-hybrid assays ( Figure 3A ) . We did several experiments to determine if LIN-2A acts in muscles to promote UNC-49 targeting to NMJs . A lin-2 promoter construct expressed GFP in body muscles , indicating that LIN-2A may function in muscles ( Figure 3—figure supplement 1A ) . mCherry-tagged LIN-2A expressed in muscles formed puncta in the nerve cords and these LIN-2A puncta were co-localized with GFP-UNC-49B at NMJs , consistent with LIN-2A binding to FRM-3 at NMJs ( Pearson's correlation R = 0 . 68 ± 0 . 057 , p = 0 . 04 , n = 5 ) ( Figure 3B ) . Like frm-3 mutants , lin-2 mutants had decreased UNC-49B puncta fluorescence ( Figure 3C , D ) and decreased mIPSC amplitudes ( Figure 3E , F ) , both implying that synaptic UNC-49 levels were decreased . mIPSC rates were unaltered in lin-2 mutants ( Figure 3G ) , indicating that pre-synaptic GABA release was unaffected . Muscimol-activated current was unaffected in lin-2 mutants ( Figure 3H ) , indicating that the lin-2 puncta and mIPSC defects were not caused by decreased bulk expression or surface delivery of UNC-49 receptors . The lin-2 puncta and mIPSC defects were rescued by transgenes restoring LIN-2A expression in body muscles but not by those expressed in GABAergic motor neurons ( Figure 3D , F ) . If LIN-2A and FRM-3 function together to localize UNC-49 receptors , lin-2 and frm-3 mutations should not have additive effects in double mutants . Consistent with this idea , UNC-49B puncta fluorescence and mIPSC amplitudes in frm-3 lin-2 double mutants were not significantly different from those in either single mutant ( Figure 3D , F ) . Collectively , these results suggest that LIN-2A/CASK and FRM-3/EPB4 . 1 function together in body muscles to localize UNC-49B at NMJs but are not required for the expression or function of non-synaptic UNC-49 receptors . 10 . 7554/eLife . 09648 . 006Figure 3 . LIN-2A/CASK binds FRM-3 and is required for UNC-49 synaptic targeting . ( A ) FRM-3′s ERM domain binds LIN-2A in yeast 2-hybrid assays . Growth of Y2HGold cells on selective media ( –Trp/-Leu/-His/-Ade ) is shown . Yeast cells were transformed with vectors expressing the indicated fusion proteins . Positive ( + , pGBKT7-53 and pGADT7-T ) and negative ( − , pGBKT7-Lam and pGADT7-T ) controls are indicated . ERM domains derived from FRM-1 , FRM-2 and FRM-3 were tested for interaction with LIN-2A . ( B ) Muscle expressed GFP-UNC-49B ( Green ) and LIN-2::mCherry ( Red ) are co-localized in the nerve cord . A representative image is shown ( scale bar 5 μm ) . ( C , D ) GFP-UNC-49B puncta fluorescence in the nerve cord was decreased in lin-2 mutants . This defect was rescued by transgenes expressing LIN-2A in body muscles ( M ) but not by those expressed in GABergic neurons ( N ) . Representative images ( C , scale bar 5 μm ) and mean puncta intensity ( D ) are shown . ( E–G ) mIPSC amplitude was reduced in lin-2 mutants and this defect was rescued by a transgene expressing LIN-2 in body muscle ( M resc ) . mIPSCs were recorded from adult body muscles . Representative traces ( E ) , mean amplitude ( F ) , and mean frequency ( G ) are shown . ( H ) Muscimol-activated currents in adult body muscles were unaffected in lin-2 mutants , indicating that the function of total surface UNC-49 receptors was unaltered . Mean peak currents are shown . lin-2 and frm-3 mutations did not have additive effects on UNC-49B puncta fluorescence ( D ) or mIPSC amplitudes ( F ) in double mutants . Values that differ significantly are indicated ( * , p < 0 . 001; , p < 0 . 01; ns , not significant ) . The number of animals analysed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 00610 . 7554/eLife . 09648 . 007Figure 3—figure supplement 1 . mIPSC amplitudes were unaltered in lin-7 and lin-10 mutants . ( A ) Transgenes containing the lin-2 promoters express GFP in body muscles . Muscle arms are indicated by asterisks . Scale bar , 25 μm . ( B , C ) mIPSCs were recorded from adult body wall muscles of lin-7 and lin-10 mutants . Representative traces ( B ) and mean mIPSC amplitude ( C ) are shown . No significant differences were observed . The number of animals analysed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 007 LIN-2/CASK associates with another scaffolding complex that contains LIN-7/Velis and LIN-10/Mint subunits ( Butz et al . , 1998; Kaech et al . , 1998 ) . We found that mIPSC amplitudes were unaltered in lin-7 and lin-10 mutants ( Figure 3—figure supplement 1B , C ) , indicating that this complex is not required for UNC-49 synaptic targeting . The decreased UNC-49B synaptic abundance in lin-2 and frm-3 mutants could reflect a loss of either mobile or immobilized receptors . To distinguish between these possibilities , we measured FRAP of UNC-49B puncta in the dorsal nerve cord ( Figure 4 ) . In both frm-3 ( Figure 4A ) and lin-2 ( Figure 4B ) mutants , GFP-UNC-49B FRAP was significantly increased ( Figure 4C ) . The frm-3 FRAP defect was rescued by transgenes restoring FRM-3 expression in body muscles ( Figure 4C ) . A similar increase in FRAP was observed for pH-UNC-49B receptors in frm-3 mutants ( Figure 4—figure supplement 1A , B ) . The GFP-UNC-49B transgene is likely to be expressed at higher levels than endogenous UNC-49 . Over-expression could alter UNC-49B mobility at synapses . To address this possibility , we repeated the FRAP measurements using a single copy RFP-tagged UNC-49B transgene ( krSi2 ) ( Pinan-Lucarre et al . , 2014 ) . Using this single copy transgene , a similar increase in FRAP of RFP-UNC-49B was observed in frm-3 mutants ( Figure 4—figure supplement 1C , D ) . Increased FRAP suggests that synaptic UNC-49B receptors had increased ability to undergo exchange in the nerve cord , most likely due to increased diffusional mobility in the plasma membrane . These results suggest that LIN-2A and FRM-3 function as a scaffold that stabilizes an immobile pool of UNC-49 receptors in the plasma membrane at post-synaptic elements . 10 . 7554/eLife . 09648 . 008Figure 4 . LIN-2A and FRM-3 stabilize immobile UNC-49B receptors at synapses . Mobility of synaptic GFP-UNC-49B was analyzed by FRAP . Representative scatter plots of fluorescence recovery ( solid lines indicate single exponential fits ) ( A , B ) and summary data ( C ) are shown . Fluorescence recovery was increased in frm-3 and lin-2 mutants , indicating increased mobility of synaptic UNC-49B . The frm-3 mutant FRAP defect was rescued by a transgene expressing FRM-3 in body muscles ( Resc ) . Values that differ significantly are indicated ( * , p < 0 . 001; * , p < 0 . 05 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 00810 . 7554/eLife . 09648 . 009Figure 4—figure supplement 1 . FRAP analysis of pH-UNC-49B and single copy RFP-UNC-49B in frm-3 mutants . To assess the mobility of surface UNC-49B receptors at synapses , we analyzed FRAP of pHluorin-tagged UNC-49B ( pH-UNC-49B ) and of an RFP-UNC-49B single copy transgene ( krSi2 ) . FRAP of pH-UNC-49B ( A , B ) and RFP-UNC-49B ( C , D ) were both significantly increased in frm-3 mutants . Representative scatter plots of fluorescence recovery and single exponential fits ( solid lines ) ( A , C ) , and summary data ( B , D ) are shown . The number of animals analysed is indicated for each genotype . Error bars indicate SEM . Values that differ significantly from WT controls are indicated ( *** , p < 0 . 001; * , p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 009 Synaptic UNC-49 receptor levels and mIPSC amplitudes were decreased but not eliminated in frm-3 and lin-2 mutants; consequently , other proteins must also play a role in confining UNC-49 receptors to these synapses . In mammals , Neuroligin 2 binds gephyrin and collybistin and is required to recruit GABAA receptors to synapses ( Jacob et al . , 2005; Poulopoulos et al . , 2009 ) . Prompted by these results , we tested the idea that post-synaptic Neuroligin also plays a role in confining UNC-49 receptors at synapses . Several results support this idea . First , an nlg-1 promoter construct expressed GFP in body muscles ( Hunter et al . , 2010 ) , consistent with NLG-1 function in muscles . Second , mCherry-tagged NLG-1 expressed in body muscles was co-localized with GFP-tagged UNC-49B receptors at NMJs ( Pearson's correlation R = 0 . 81 ± 0 . 02 , p = 0 . 01 , n = 8 ) ( Figure 5A ) . Third , GFP-UNC-49B puncta fluorescence was significantly reduced in nlg-1 mutants ( Figure 5B , C ) , consistent with a decrease in total synaptic receptors . Fourth , FRAP of GFP-UNC-49B ( Figure 5D , E ) , pH-UNC-49B ( Figure 5—figure supplement 1A , B ) , and single copy RFP-UNC-49B ( Figure 5—figure supplement 1C , D ) were all significantly reduced in nlg-1 mutants , indicating that the residual synaptic UNC-49B receptors were largely immobile . These results suggest that NLG-1 stabilizes a mobile pool of surface UNC-49B receptors at synapses . 10 . 7554/eLife . 09648 . 010Figure 5 . NLG-1 stabilizes mobile UNC-49B at synapses . ( A ) Muscle expressed GFP-UNC-49B ( Green ) and NLG-1::mCherry ( Red ) are co-localized in the nerve cord . A representative image is shown ( scale bar 5 μm ) . ( B , C ) GFP-UNC-49B synaptic abundance was decreased in nlg-1 mutants and this defect was rescued by transgenes expressing NLG-1 in body muscles ( M ) but not by those expressed in motor neurons ( N ) . Representative images ( B , scale bar 5 μm ) and mean puncta intensity ( C ) are shown . ( D , E ) Mobility of synaptic GFP-UNC-49B was analyzed by FRAP . Representative scatter plots of fluorescence recovery and single exponential fits ( solid lines ) ( D ) and summary data ( E ) are shown . Fluorescence recovery was decreased in nlg-1 mutants , indicating that synaptic UNC-49B mobility was decreased . The nlg-1 mutant FRAP defect was rescued by a transgene expressing NLG-1 in body muscles ( M Resc ) ( E ) . ( F–H ) mIPSC amplitude was reduced in nlg-1 mutants and this defect was rescued by a transgene expressing NLG-1 in body muscle ( M resc . ) . mIPSCs were recorded from adult body muscles . Representative traces ( F ) , mean frequency ( G ) , and mean amplitude ( H ) are shown . ( I ) Muscimol-evoked currents ( mean peak amplitude ) was unaltered in nlg-1 and in frm-3 nlg-1 double mutants . nlg-1 and frm-3 mutations had additive effects on UNC-49B puncta fluorescence ( B , C ) and mIPSCs ( F ) in double mutants . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . Values that differ significantly are indicated ( * , p < 0 . 001; , p < 0 . 01; * , p < 0 . 05; ns , not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 01010 . 7554/eLife . 09648 . 011Figure 5—figure supplement 1 . FRAP analysis of pH-UNC-49B and single copy RFP-UNC-49B in nlg-1 mutants . To assess the mobility of surface UNC-49B receptors at synapses , we analyzed FRAP of pHluorin-tagged UNC-49B ( pH-UNC-49B ) and of an RFP-UNC-49B single copy transgene ( krSi2 ) . FRAP of pH-UNC-49B ( A , B ) and RFP-UNC-49B ( C , D ) was significantly decreased in nlg-1 mutants . Representative scatter plots of fluorescence recovery and single exponential fits ( solid lines ) ( A , C ) , and summary data ( B , D ) are shown . The number of animals analysed is indicated for each genotype . Error bars indicate SEM . Values that differ significantly from WT controls are indicated ( * , p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 011 To determine if the function of endogenously expressed UNC-49 was altered , we measured GABA activated currents in body muscles . The mIPSC rate was significantly reduced in nlg-1 mutants ( Figure 5F , G ) . The mean mIPSC amplitude ( Figure 5H ) was also significantly reduced , consistent with decreased abundance of synaptic UNC-49 . The nlg-1 mIPSC rate and amplitude defects were both rescued by a transgene restoring NLG-1 expression in body muscles ( Figure 5G , H ) . The decreased mIPSC rate in nlg-1 mutants could be a secondary consequence of the smaller mIPSC amplitudes ( i . e . due to decreased detection of mIPSCs ) . Alternatively , the decreased mIPSC rate could result from decreased pre-synaptic GABA release . The muscimol-evoked current was unaffected in nlg-1 mutants ( Figure 5I ) , indicating that the nlg-1 puncta and mIPSC defects were not caused by decreased bulk expression or surface delivery of UNC-49 receptors . Collectively , these results suggest that NLG-1 stabilizes a mobile pool of UNC-49 receptors at synapses , and that this receptor pool contributes to post-synaptic currents . These results are consistent with two recent studies , which also showed that NLG-1 promotes UNC-49 targeting to synapses ( Maro et al . , 2015; Tu et al . , 2015 ) . C . elegans expresses long ( NRX-1α ) and short ( NRX-1β ) Neurexin isoforms , both encoded by the nrx-1 gene . NLG-1 binds to the sixth LNS repeat of NRX-1 ( Hu et al . , 2012 ) and , consequently , could bind to both NRX-1α and β . To test the impact of NRX-1 on UNC-49B localization , we isolated an nrx-1 null allele ( nu485 ) that inactivates both NRX-1α and β . Mean mIPSC amplitude was significantly increased in nrx-1 null mutants ( Figure 6A , B ) , indicating an increased number of functional UNC-49 receptors at synapses . The mIPSC rate was unaltered in nrx-1 mutants ( Figure 6C ) , suggesting that presynaptic GABA release occurs normally . The amplitude of muscimol-activated current was also unaltered in nrx-1 mutants ( Figure 6D ) ; consequently , the increased mIPSC amplitude was unlikely to be caused by increased bulk expression or surface delivery of UNC-49 receptors . GFP-UNC-49B puncta fluorescence was also unaltered in nrx-1 null mutants ( Figure 6E ) , suggesting that the mIPSC amplitude increase was not caused by increased abundance of synaptic UNC-49B receptors . Although UNC-49B puncta fluorescence was unaltered , FRAP of synaptic GFP-UNC-49B was significantly reduced in nrx-1 mutants ( Figure 6F , G ) . The nrx-1 mutant defects in mIPSC amplitude ( Figure 6B ) and UNC-49B FRAP ( Figure 6G ) were both rescued by transgenes expressing NRX-1α in GABAergic motor neurons but not by those expressing NRX-1β . NRX-1α transgenes expressed in body muscles lacked rescuing activity ( Figure 6B , G ) . Collectively , these results suggest that the total number of synaptic UNC-49B receptors was unaltered in nrx-1 mutants; however , there was a shift in receptor mobility whereby the pool of immobilized synaptic UNC-49B was enlarged ( resulting in increased mIPSC amplitude ) while the mobile pool was diminished . These results support the idea that pre-synaptic NRX-1α inhibits the immobilization of mobile UNC-49B receptors at synapses . 10 . 7554/eLife . 09648 . 012Figure 6 . Pre-synaptic NRX-1α inhibits immobilization of synaptic UNC-49B . ( A , B ) Mutations inactivating nrx-1 increased mIPSC amplitude and this defect was rescued by transgenes expressing NRX-1α in motor neurons ( Nα ) but not those expressing NRX-1β ( Nβ ) . Transgenes expressing NRX-1α in body muscles ( Mα ) lacked rescuing activity . mIPSCs were recorded from adult body muscles . Representative traces ( A ) , mean amplitude ( B ) and mean frequency ( C ) are shown . The effect of nrx-1 mutations on mIPSC amplitudes was eliminated in nrx-1;nlg-1 double mutants but was unaffected in nrx-1; frm-3 double mutants ( B ) . ( D ) Muscimol-evoked currents ( mean peak amplitude ) was unaffected in nrx-1 mutants . ( E ) GFP-UNC-49B synaptic abundance was unaltered in nrx-1 mutants Representative images ( top , scale bar 5 μm ) and mean puncta intensity ( below ) are shown . ( F , G ) FRAP analysis suggests that mobility of synaptic GFP-UNC-49B was decreased in nrx-1 mutants . This FRAP defect was rescued by transgenes expressing NRX-1α in motor neurons ( Nα ) but not those expressing NRX-1β ( Nβ ) . Transgenes expressing NRX-1α in body muscles ( Mα ) lacked rescuing activity . Representative scatter plots of fluorescence recovery and single exponential fits ( solid lines ) ( F ) and summary data ( G ) are shown . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . Values that differ significantly are indicated ( * , p < 0 . 001; , p < 0 . 01; ns , not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 012 Which pool of UNC-49 receptors is required for the increased mIPSC amplitudes in nrx-1 mutants ? To address this question , we recorded mIPSCs in double mutants . We found that the effect of nrx-1 mutations on mIPSC amplitude was eliminated in nrx-1; nlg-1 double mutants ( Figure 6B ) . By contrast , the nrx-1 mIPSC amplitude defect was not blocked in nrx-1; frm-3 double mutants ( Figure 6B ) . These results suggest that presynaptic NRX-1α inhibits diffusional trapping of NLG-1-stabilized UNC-49 receptors but has little effect on the mobility of FRM-3-associated receptors . The preceding results suggest that UNC-49B synaptic puncta comprise two pools of receptors that are stabilized by different scaffolds . Immobilized UNC-49B receptors fail to undergo diffusional exchange in FRAP experiments and are stabilized by FRM-3 and LIN-2A . Mobile UNC-49B receptors mediate fluorescence recovery in FRAP experiments and are stabilized by NLG-1 and NRX-1α . In this scenario , we expect that double mutants lacking both scaffolds would have additive defects , lacking both receptor pools . Consistent with this idea , UNC-49B puncta fluorescence was significantly reduced ( Figure 5B , C ) while GFP-UNC-49B FRAP was significantly increased ( Figure 7A , B ) in frm-3 nlg-1 double mutants compared to the corresponding single mutants . In frm-3 nlg-1 double mutants , both the immobile pool of synaptic UNC-49B ( Figure 7A , B ) and mIPSCs ( Figure 5F ) were completely eliminated . Given the absence of mIPSCs , we could not measure quantal size in double mutants . As an alternative , we measured mIPSC rate and found that it was dramatically reduced in frm-3 nlg-1 double mutants compared to either single mutant ( Figure 5G ) . Muscimol-activated muscle current in frm-3 nlg-1 double mutants did not significantly differ from wild type controls ( Figure 5I ) , suggesting that the puncta , FRAP , and mIPSC defects were not caused by decreased bulk expression or surface delivery of UNC-49 receptors . These results support the idea that FRM-3 and NLG-1 stabilized UNC-49B receptors represent two distinct pools of synaptic receptors , which together account for all synaptic UNC-49 receptors . 10 . 7554/eLife . 09648 . 013Figure 7 . Both UNC-49 receptor pools contribute to post-synaptic responses . ( A , B ) FRAP analysis suggests that the immobile pool of synaptic GFP-UNC-49B was eliminated in frm-3 nlg-1 double mutants . Representative scatter plots of fluorescence recovery and single exponential fits ( solid lines ) ( A ) and summary data ( B ) are shown . ( C , D ) The FRM-3 and NLG-1 scaffolds increase the diversity of quantal responses . CV of mIPSC amplitudes are shown for the indicated genotypes . CV was significantly decreased in frm-3 ( C ) and in nlg-1 ( D ) mutants . The frm-3 and nlg-1 CV defects were rescued by transgenes restoring expression of the corresponding genes in body muscles ( resc ) . CV was not significantly altered in lin-2 and nrx-1 mutants ( C ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . Values that differ significantly are indicated ( *** , p < 0 . 001; * , p < 0 . 05; ns , not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09648 . 013 Analysis of excitatory and inhibitory synapses in mammals suggests that differences in synaptic receptor abundance contribute significantly to variation in quantal sizes ( Nusser et al . , 1997 , 1998 ) . The coefficient of variation ( CV ) of the mIPSC amplitudes was ∼0 . 6–7 in wild type animals ( Figure 7C , D ) , which is similar to values reported for some mammalian CNS synapses ( Bekkers et al . , 1990; Hanse and Gustafsson , 2001 ) . If quantal variability is dominated by diffusion of UNC-49 receptors in and out of the postsynaptic site , the mIPSC amplitude CV should scale inversely with the square root of receptor number . In this scenario , we expect that mutations decreasing UNC-49B punta fluorescence ( e . g . , frm-3 , lin-2 , and nlg-1 ) would produce corresponding increases in the CV of mIPSC amplitudes . Contrary to this idea , CV was significantly reduced in frm-3 and nlg-1 mutants and was not significantly altered in lin-2 mutants ( Figure 7C , D ) . The frm-3 and nlg-1 mIPSC amplitude CV defects were rescued by transgenes that restore expression of the corresponding genes in body muscles ( Figure 7C , D ) . Changes in CV were also not correlated with changes in mean mIPSC amplitude . Mean mIPSC amplitude was increased in nrx-1 mutants ( Figure 6B ) and decreased in lin-2 mutants ( Figure 3F ) while CV was unaltered in both cases ( Figure 7C ) . Collectively , these results suggest that the FRM-3 and NLG-1 scaffolds increase the diversity of quantal responses , and that this effect cannot be explained by changes in synaptic UNC-49 levels nor by changes in mean mIPSC amplitude . Synaptic UNC-49/GABAA receptors are confined by two scaffolds that function in parallel . The FRM-3/EPB4 . 1 and LIN-2A/CASK scaffold accounts for ∼40% of synaptic receptors while the NLG-1 scaffold accounts for ∼30% of synaptic UNC-49 ( assessed by puncta fluorescence ) . These scaffolds define two pools of synaptic GABAA receptors that function independently . In double mutants lacking both pools , immobilized synaptic UNC-49 receptors and mIPSCs are eliminated . Mobile receptors detected by FRAP in frm-3 nlg-1 double mutants likely correspond to extra-synaptic receptors on the cell surface ( which mediate muscimol-activated currents ) . Thus , the FRM-3 and NLG-1 scaffolds together account for all GABAA receptors at the body wall NMJ . Consequently , the dynamics of GABAA levels at this synapse will be determined by the detailed biochemistry of these two scaffolds . Our results also suggest that synaptic receptors represent a small fraction of total surface UNC-49 receptors . Mutants lacking both scaffolds have no defect in total surface receptors ( assessed by muscimol-activated currents ) . Thus , these scaffolds have no effect on the assembly , anterograde trafficking , and surface delivery of UNC-49 receptors . Instead , these scaffolds play a specific role in concentrating UNC-49 surface receptors at synapses . Our results strongly support the idea that both GABAA pools are required for inhibitory synaptic transmission . Mutations inactivating the FRM-3/LIN-2A scaffold caused similar decreases in immobilized UNC-49B ( 40% decrease ) and mIPSC amplitude ( 30% decrease ) . Inactivating NLG-1 caused corresponding decreases in total synaptic UNC-49B ( 30% decrease ) , mobile UNC-49B ( 40% decrease ) , and mIPSC amplitudes ( 48% decrease ) . Thus , analysis of single mutants suggests that FRM-3 and NLG-1 stabilized receptors contribute equally to post-synaptic currents . We show that FRM-3/EPB4 . 1 and LIN-2A/CASK together comprise a scaffold that immobilizes synaptic GABAA receptors . Other ERM domain containing proteins were previously implicated in neurotransmitter receptor localization . The ERM protein Radixin stabilizes extra-synaptic GABRA5 receptors in mammalian neurons ( Loebrich et al . , 2006 ) , suggesting that different populations of GABA receptors are stabilized by distinct ERM proteins . Mammalian EPB4 . 1 and its Drosophila counterpart ( coracle ) were previously implicated in synaptic localization of glutamate receptors ( Shen et al . , 2000; Chen et al . , 2005 ) . Excitatory and inhibitory synaptic transmission were not significantly altered in mouse CASK knockouts ( Atasoy et al . , 2007 ) ; however , subtle defects ( e . g . , the ∼20% decrease in quantal size reported here ) could have been missed in this study . Drosophila CASK mutants have decreased targeting of glutamate receptors to larval NMJs ( Chen and Featherstone , 2011 ) . A role for CASK in synaptic targeting of GABAA receptors has not been described . We found that frm-3 EPB4 . 1 and lin-2 CASK mutants have decreased synaptic GABAA levels , decreased immobilization of synaptic GABAA receptors , and corresponding decreases in mIPSC amplitudes . Both mutants had significant residual levels of immobile synaptic GABAA receptors and post-synaptic currents . Thus , FRM-3 and LIN-2 define a sub-population of immobilized synaptic GABAA receptors , accounting for ∼40% of total synaptic receptors . For all synaptic phenotypes , lin-2 defects were less severe than those observed in frm-3 mutants , perhaps because LIN-2/CASK regulates UNC-49 targeting by modulating FRM-3 function . For example , prior studies suggest that ERM proteins equilibrate between active ( open ) and inactive ( closed ) conformations ( Pearson et al . , 2000 ) . It is possible that LIN-2/CASK binding stabilizes the active conformation of FRM-3 . Collectively , these results suggest that ERM and CASK proteins play a conserved role in targeting neurotransmitter receptors . In each case , CASK and ERM proteins account for a subset of receptors confined to a synapse . In mammals , Neuroligin 2 works in conjunction with gephyrin and collybistin to localize GABAA receptors to synapses . Knockout mutations in each of these genes reduce but do not eliminate synaptic GABAA receptors ( Mukherjee et al . , 2011; Nair et al . , 2013; Specht et al . , 2013 ) . Gephyrin binding to the TM3-4 loop has been shown to immobilize GABAA receptors at synapses ( Jacob et al . , 2005; Mukherjee et al . , 2011 ) . The impact of Neuroligin 2 mutations on GABAA receptor mobility has not been determined . Our results are largely consistent with these prior studies . Mutations inactivating NLG-1 decreased but did not eliminate synaptic UNC-49 and caused a corresponding decrease in mIPSC amplitudes . Two recent studies reported similar effects of nlg-1 mutations on synaptic UNC-49 levels ( Maro et al . , 2015; Tu et al . , 2015 ) . Thus , as in mammalian neurons , NLG-1 represents one of multiple mechanisms for confining GABAA receptors at synapses . NLG-1 association with UNC-49 could be mediated by binding to an intermediary Gephyrin-like molecule . The C . elegans ortholog of gephyrin is MOC-1; however , its role in synaptic targeting of UNC-49 has not been determined . Our studies provide further insights into how Neurexin and Neuroligin function to target synaptic GABAA receptors . Two results suggest that the NLG-1 stabilized pool consists of both mobile and immobilized UNC-49 receptors . First , in double mutants lacking both scaffolds ( i . e . frm-3 nlg-1 double mutants ) , immobilized UNC-49B receptors and mIPSCs were both eliminated . This result implies that immobile UNC-49B receptors observed in frm-3 single mutants are derived from the NLG-1-stabilized pool and that immobilization of these receptors does not require FRM-3 . Second , mIPSC amplitudes and immobile synaptic UNC-49B levels are significantly increased in mutants lacking presynaptic NRX-1α . Both of these effects were abolished in nrx-1; nlg-1 double mutants . These results suggest that pre-synaptic NRX-1α inhibits immobilization of diffusing NLG-1-stabilized UNC-49 receptors . Trans-synaptic NRX-1α/NLG-1 complexes may confine mobile synaptic receptors via low affinity binding interactions with UNC-49B , or by sterically interfering with UNC-49B lateral diffusion ( Santamaria et al . , 2010 ) . Because NLG-1 stabilizes both forms of synaptic receptors , the contribution of the NLG-1 pool to post-synaptic currents could be mediated by immobilized or diffusing UNC-49 receptors ( or a mixture of the two ) . Synaptic responses are intrinsically noisy , varying considerably even among inputs to the same cell . This noise arises from variability in both pre- and post-synaptic processes ( Lisman et al . , 2007 ) . Prior studies suggested that an important contributor to post-synaptic noise is variation in receptor levels between synapses ( Nusser et al . , 1997 ) , which can be adjusted bidirectionally by activity ( Bannai et al . , 2009; Petrini et al . , 2014 ) . How do FRM-3 and NLG-1 alter quantal size ? Prior modelling studies suggest that the non-homogeneous distribution of synaptic receptors into nanoclusters has profound effects on synaptic transmission ( MacGillavry et al . , 2013; Nair et al . , 2013 ) . Although these studies analyzed glutamatergic synapses , in the following we assume that similar principles apply to GABAergic synapses . Within post-synaptic elements , ∼60% of receptors are localized in immobile nanoclusters ( mean diameter 75 nm , mean receptor number 25 ) while the remaining ∼40% are mobile and diffusely distributed ( MacGillavry et al . , 2013; Nair et al . , 2013 ) . The effect of receptor nanoclusters on transmission results from the fact that a single synaptic vesicle ( SV ) fusion activates receptors in a subsynaptic domain . For glutamate , the domain of activated receptors is estimated to have diameter ∼200 nm ( Raghavachari and Lisman , 2004 ) . The size of the GABA domain has not been calculated , but is likely to be larger ( due to the higher affinity of GABAA receptors ) . SV fusions at C . elegans cholinergic and GABAergic NMJs occur in an active zone with a diameter of 700 nm ( Hammarlund et al . , 2007; Watanabe et al . , 2013 ) . Thus , the spatial extent of GABA in the synaptic cleft will vary depending on the location of the SV fusion event within the active zone ( Barberis et al . , 2011 ) . Consequently , quantal amplitude will vary depending on the proximity of the vesicle fusion site to the receptor nanocluster and the density of receptors in each nanocluster ( Franks et al . , 2003; MacGillavry et al . , 2013; Nair et al . , 2013 ) . By contrast , mobile ( unclustered ) receptors have lower but uniform surface density; consequently , mobile receptors are predicted to mediate smaller quantal responses that have lower CV ( MacGillavry et al . , 2013 ) . Our results are largely consistent with these modelling studies . The FRM-3 and NLG-1 scaffolds increase mIPSC amplitude and CV , presumably due to an increase in the number of UNC-49 nanoclusters at synapses or an increase in the number of receptors in each nanocluster . Changes in synaptic inhibition are proposed to play an important role in the pathophysiology of several neuropsychiatric disorders . Decreased inhibition is implicated in autism spectrum disorders ( ASD ) ( Rubenstein and Merzenich , 2003 ) , whereas excess inhibition has been proposed to occur in mental retardation syndromes , such as Down and Rett Syndromes ( Kleschevnikov et al . , 2004; Dani et al . , 2005 ) . Recurrent mutations in Neurexin , Neuroligin , and CASK are found in ASD ( Sanders et al . , 2012; O'Roak et al . , 2012 ) . We propose that these mutations may directly alter inhibitory transmission by altering the synaptic confinement of GABAA receptors . These results provide additional biochemical links between ASD associated genes and inhibitory transmission . As all of these molecules are conserved , the mechanisms we describe for confining synaptic GABAA receptors are likely to be conserved in other systems , including humans . Strains were maintained at 20° C under standard conditions . OP50 Escherichia coli was used as a food source for all experiments except electrophysiology where HB101 E . coli was utilized . A description of all alleles can be found at www . wormbase . org . The following strains were utilized in this study: KP5330 nlg-1 ( ok259 ) KP7320 nrx-1 ( nu485 ) KP7338 frm-3 ( gk585 ) CB1309 lin-2 ( e1309 ) CB407 unc-49 ( e407 ) MT106 lin-7 ( n106 ) KP7637 lin-10 ( n1508 ) KP7532 nrx-1 ( nu485 ) ; nlg-1 ( ok259 ) KP7367 frm-3 ( gk585 ) nlg-1 ( ok259 ) KP7534 nrx-1 ( nu485 ) ; frm-3 ( gk585 ) KP7514 frm-3 ( gk585 ) lin-2 ( e1309 ) KP5931 nuIs283 [Pmyo-3::unc-49::gfp::unc-54 3′UTR; Punc-25::snb-1::mcherry::unc-54 3′UTR] KP7341 nuIs283; frm-3 ( gk585 ) KP7478 nuIs283; nrx-1 ( nu485 ) KP7133 nuIs283; lin-2 ( e1309 ) KP7596 nuIs283; nlg-1 ( ok259 ) KP7474 nuIs283; frm-3 ( gk585 ) nlg-1 ( ok259 ) KP7340 nuIs283; frm-3 ( gk585 ) lin-2 ( e1309 ) KP7597 nuIs283; unc-49 ( e407 ) ; KP7545 nuIs522 [Pmyo-3::lin-2::mcherry::unc-54 3′UTR]; KP7552 nuIs523 [Pmyo-3::pHluorin::unc-49::unc-54 3′UTR]; KP7614 nuIs523; nlg-1 ( ok259 ) ; KP7615 nuIs523; frm-3 ( gk585 ) ; KP7553 nuIs524 [Pmyo-3::gfp::frm-3::unc-54 3′UTR]; KP7364 nuEx490 [Pfrm-3::gfp::unc-54 3′UTR]; KP7363 nuEx489 [Plin-2::gfp::unc-54 3′UTR]; KP7631 nuIs532 [Pmyo-3::NLG-1::mcherry::unc-54 3′UTR]; EN2630 LGII , krSi2 [Punc-49::unc-49B-RFP; unc-49 ( e407 ) EN3224 LGII , krSi2 [Punc-49::unc-49B-RFP; unc-49 ( e407 ) ; nlg-1 ( ok259 ) KP7894 LGII , krSi2 [Punc-49::unc-49B-RFP; unc-49 ( e407 ) ; frm-3 ( gk585 ) KP7893 nuIs531 [Pmyo-3::frm-3::2flag::unc-54 3′UTR]; nuIs283 The nrx-1 ( nu485 ) null allele was isolated using the mosDEL protocol ( Frokjaer-Jensen et al . , 2010 ) . The C . elegans transposon insertion line ttTi26330 was obtained from the NemaGENETAG consortium . 10 . 416 kb of the nrx-1 locus was replaced with a Cb-unc-119 ( + ) selectable marker ( Frokjaer-Jensen et al . , 2010 ) . The engineered deletion included 982bp of upstream sequence and the first 9434bp of reference transcript C29A12 . 4a . The nu485 mutation deletes exons 1–20 , or 85% , of the nrx-1 coding sequence . Exons 21–27 , downstream of the ttTi26330 insertion site , were left intact . These could potentially encode a 234-residue C-terminal NRX-1 fragment . The EN2630 and EN3224 strains were kindly provided by Dr . Jean-Louis Bessereau . All expression vectors are based on the pPD49 . 26 backbone ( A . Fire ) . Standard methods were utilized to construct all plasmids . A 3 kb myo-3 myosin promoter was used for expression in body muscles , a 1 . 2 kb unc-25 GAD promoter was used for expression in GABAergic neurons . The transcriptional reporter for frm-3 and lin-2 used 5 kb of 5′ flanking sequence . nlg-1 ( C40C9 . 5c ) , nrx-1α ( C29A12 . 4a ) , nrx-1β ( C29A12 . 4f ) , frm-3 ( H05G16 . 1 ) , lin-2 ( F17E5 . 1a ) , and unc-49B ( T21C12 . 1b ) cDNAs were cloned from a cDNA library using primers corresponding to the predicted start and stop codons of the indicated EST . The ERM domain of frm-1 , frm-2 , frm-3 , frm-4 , frm-5 . 1 , frm-8 , frm-9 , frm-10 , max-1 , kin-32 , kri-1 , erm-1 , nfm-1 cDNAs were cloned from a cDNA library using primers corresponding to the predicted start and stop codons of ERM domain . nuIs524 contains GFP tagged FRM-3 cDNA constructs in which GFP was inserted in frame in the C-terminus . nuIs523 contains pHluorin tagged UNC-49B cDNA constructs in which super-ecliptic pHluorin was inserted in-frame immediately after the signal peptide sequence . nuIs522 contains an mCherry-tagged LIN-2 cDNA construct , in which mCherry is inserted in the N-terminus . nuIs532 contains an mCherry-tagged NLG-1 cDNA construct , in which mCherry was inserted 13 residues from the carboxy-terminus ( leaving the predicted PDZ ligand intact ) . nuIs531 contains an FLAG-tagged FRM-3 cDNA construct , containing two copies of the FLAG epitope at the C-terminus . Full descriptions of all plasmids are available upon request . Transgenic animals were prepared by microinjection , and integrated transgenes were isolated following UV irradiation . For yeast two-hybrid screens all the ERM domain containing proteins , Y2HGold yeast cells were co-transformed with pGADT7-UNC-49-TM3-4 loop and all the pGBKT7-ERM constructs respectively ( BD clontech ) . Transformants selected from the SD-Leu-Trp plates were restreaked onto SD-Leu-Trp-His-Ade plates to test interactions . False positive from autoactivation was ruled out by co-transformation of pGBKT7-ERM construct with pGADT7 empty vector alone . Strains for electrophysiology were maintained at 20°C on plates seeded with HB101 . Cyanoacrylate glue was used to immobilize adult worms on the Sylgard-coated coverslip . The dissected adult worms were superfused in an extracellular solution containing 127 mM NaCl , 5 mM KCl , 26 mM NaHCO3 , 1 . 25 mM NaH2PO4 , 20 mM glucose , 1 mM CaCl2 , and 4 mM MgCl2 , bubbled with 5% CO2 , 95% O2 at 20°C . Endogenous GABA IPSC recordings were carried out at 0 mV using an internal solution containing 120 mM CsCH3SO3 , 4 mM CsCl , 15 mM CsF , 4 mM MgCl2 , 5 mM EGTA , 0 . 25 mM CaCl2 , 10 mM HEPES , and 4 mM Na2ATP , adjusted to pH 7 . 2 using CsOH . For muscimol-activated current recordings , 100 μM muscimol was pressure ejected for 1 . 0 s onto body muscles of adult worms . For colocalization studies , images were captured using a 60x objective ( NA 1 . 45 ) on an Olympus FV-1000 confocal microscope at 5× digital zoom . Worms were immobilized with 30 mg/ml 2 , 3-Butanedione monoxamine ( Sigma , St . Louis , MO , United States ) . UNC-49B puncta fluorescence was quantified by wide field fluorescence microscopy in , the dorsal nerve cord of young adults ( midway between the vulva and the tail ) . Images were acquired using a Zeiss Axioskop I , Olympus PlanAPO 100× 1 . 4 NA objective , and a CoolSnap HQ CCD camera ( Roper Scientific , Tuscon , AZ , United States ) . Maximum intensity projections of Z-series stacks were made using Metamorph 7 . 1 software ( Molecular Devices , Sunnyvale , CA , United States ) . Line scans of dorsal cord fluorescence were analyzed in Igor Pro ( WaveMetrics , Lake Oswego , OR , United States ) using custom-written software . Mean fluorescence of 0 . 5 μm FluoSphere beads ( Thermo Fisher , Waltham , MA , United States ) , which was measured during each experiment , was used to control the illumination intensity . All fluorescence values are normalized to wild type controls to facilitate comparison . All p-values indicated were based on ONE-way ANOVA or student t-tests . For FRAP studies , images were captured using a 60x objective ( NA 1 . 45 ) on an Olympus FV-1000 confocal microscope at 5× digital zoom . Worms were immobilized with 0 . 1 μm polystyrene microspheres ( Polysciences ) , and pads composed of 10% agarose in M9 . To image GFP-UNC-49B , pH-UNC-49B , and RAB-3-mCherry , we used 0 . 5% power from a 473 nm ( GFP ) and 559 nm ( mCherry ) solid state diode laser . Five frames of GFP and mCherry signals were recorded prior to photobleaching . Photobleaching was achieved by one scan ( 90% power from the 473 nm laser ) on a square region of interest ( ROI , about 1 . 5 × 1 . 5 μm ) that covered a single GFP punctum . GFP and mCherry signals were further recorded for 5 min at 0 . 2 frames/second . To eliminate motion artifacts , traces with more than 10% changes in mCherry signals were discarded . UNC-49B mobile fractions were calculated in MATLAB by fitting the data with a single exponential function , Ifrap ( t ) = A ( 1-e-τt ) , where A is reported as the mobile fraction . Statistical significance was determined using ONE-way ANOVA or Student's t test and all values reported are means ±SEM . Extracts were prepared from mixed staged worms expressing GFP-UNC-49 or GFP-UNC-49 and FLAG-FRM3 using a microfluidizer in buffer A ( 50 mM HEPES PH7 . 7 , 50 mM KAc , 2mMgAc2 , 250 mM sucrose , 1 mM EDTA and proteinase inhibitors ) . Worm extracts were clarified by centrifugation ( 15 min , 7000g ) . Membranes were isolated from the resulting supernatant by high speed centrifugation ( 40 min , 45 , 000g ) . Membrane proteins were solubilized with a dounce homogenizer ( 5 times ) in IP buffer ( 20 mM HEPES PH7 . 4 , 150 mM NaCl , 2 mM MgCl2 , 0 , 1 mM EDTA , 1% Triton and proteinase inhibitors ) , and incubated with FLAG M2 affinity gel ( A2220 ) overnight at 4o . Flag affinity gel were washed three times with IP buffer , and eluted with loading buffer for Western blot .	Behaviors ranging from movement to memory are dependent on the activity of extensive networks of cells called neurons . Within these networks , neurons communicate across junctions called synapses . The arrival of an electrical signal called an action potential at the ‘presynaptic’ neuron on one side of the synapse triggers the neuron to release chemical neurotransmitter molecules into the synapse . These molecules then bind to receptors on the ‘postsynaptic’ cell on the other side of the synapse . At excitatory synapses , the binding of neurotransmitter to postsynaptic receptors increases the likelihood that the postsynaptic cell will fire its own action potential . By contrast , at inhibitory synapses the binding of neurotransmitters reduces the chances of the postsynaptic cell firing . Most inhibitory synapses use a type of neurotransmitter called GABA , which exerts its effects mainly by binding to a class of receptors called GABA-activated chloride channels ( also known as GABAA receptors ) . GABAA receptors at inhibitory synapses can themselves be divided into two groups: ‘mobile’ receptors , which can move within the cell membrane that surrounds the postsynaptic cell; and ‘immobilized’ receptors that form clusters and cannot move . Recent work in mammalian cells identified a protein complex that anchors GABAA receptors to the cell's internal skeleton to immobilize the receptors . However , there is evidence to suggest that these are not the only proteins that control the location of the receptors . By studying the inhibitory synapses formed between neurons and body muscles in the roundworm species Caenorhabditis elegans , Tong , Hu et al . now show that different groups of proteins maintain the positioning of immobilized and mobile receptors . Specifically , proteins called LIN-2A ( a component of the cell's internal skeleton ) and FRM-3 ( which joins receptors to the cell's skeleton ) immobilize GABAA receptors , whilst the proteins Neuroligin and Neurexin ensure that mobile GABAA receptors remain within the synapse . Disturbances to the activity of inhibitory synapses are often seen in autism spectrum disorders , and so too are mutations in the genes that encode the mammalian equivalents of Neuroligin , Neurexin and LIN-2A . The work of Tong , Hu et al . thus suggests a mechanism by which these mutations might contribute to information processing impairments in people with autism . Further research could now investigate if ( and how ) other genes linked to autism spectrum disorders alter inhibitory synapses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	A network of autism linked genes stabilizes two pools of synaptic GABAA receptors
Khoomei is a unique singing style originating from the republic of Tuva in central Asia . Singers produce two pitches simultaneously: a booming low-frequency rumble alongside a hovering high-pitched whistle-like tone . The biomechanics of this biphonation are not well-understood . Here , we use sound analysis , dynamic magnetic resonance imaging , and vocal tract modeling to demonstrate how biphonation is achieved by modulating vocal tract morphology . Tuvan singers show remarkable control in shaping their vocal tract to narrowly focus the harmonics ( or overtones ) emanating from their vocal cords . The biphonic sound is a combination of the fundamental pitch and a focused filter state , which is at the higher pitch ( 1–2 kHz ) and formed by merging two formants , thereby greatly enhancing sound-production in a very narrow frequency range . Most importantly , we demonstrate that this biphonation is a phenomenon arising from linear filtering rather than from a nonlinear source . In the years preceding his death , Richard Feynman had been attempting to visit the small republic of Tuva located in geographic center of Asia ( Leighton , 2000 ) . A key catalyst came from Kip Thorne , who had gifted him a record called Melody tuvy , featuring a Tuvan singing in a style known as Khoomei , or Xöömij . Although he was never successful in visiting Tuva , Feynman was nonetheless captivated by Khoomei , which can be best described as a high-pitched tone , similar to a whistle carrying a melody , hovering above a constant booming low-frequency rumble . This is a form of biphonation , or in Feynman’s own words , "a man with two voices" . Khoomei , now a part of the UNESCO Intangible Cultural Heritage of Humanity , is characterized as "the simultaneous performance by one singer of a held pitch in the lower register and a melody … in the higher register" ( Aksenov , 1973 ) . How , indeed , does one singer produce two pitches at one time ? Even today , the biophysical underpinnings of this biphonic human vocal style are not fully understood . Normally , when a singer voices a song or speech , their vocal folds vibrate at a fundamental frequency ( f0 ) , generating oscillating airflow , forming the so-called source . This vibration is not , however , simply sinusoidal , as it also produces a series of harmonics tones ( i . e . , integer multiples of f0 ) ( Figure 1 ) . Harmonic frequencies in this sound above f0 are called overtones . Upon emanating from the vocal folds , they are then sculpted by the vocal tract , which acts as a spectral filter . The vocal-tract filter has multiple resonances that accentuate certain clusters of overtones , creating formants . When speaking , we change the shape of our vocal tract to shift formants in systematic ways characteristic of vowel and consonant sounds . Indeed , singing largely uses vowel-like sounds ( Story , 2016 ) . In most singing , the listener perceives only a single pitch associated with the f0 of the vocal production , with the formant resonances determining the timbre . Khoomei has two strongly emphasized pitches: a low-pitch drone associated with the f0 , plus a melody carried by variation in the higher frequency formant that can change independently ( Kob , 2004 ) . Two possible loci for this biphonic property are the source and/or the filter . A source-based explanation could involve different mechanisms , such as two vibrating nonlinear sound sources in the syrinx of birds , which produce multiple notes that are harmonically unrelated ( Fee et al . , 1998; Zollinger et al . , 2008 ) . Humans however are generally considered to have only a single source , the vocal folds . But there are an alternative possibilities: for instance , the source could be nonlinear and produce harmonically-unrelated sounds . For example , aerodynamic instabilities are known to produce biphonation ( Mahrt et al . , 2016 ) . Further , Khoomei often involves dramatic and sudden transitions from simple tonal singing to biophonation ( see Figure 1 and the Appendix for associated audio samples ) . Such abrupt changes are often considered hallmarks of physiological nonlinearity ( Goldberger et al . , 2002 ) , and vocal production can generally be nonlinear in nature ( Herzel and Reuter , 1996; Mergell and Herzel , 1997; Fitch et al . , 2002; Suthers et al . , 2006 ) . Therefore it remains possible that biphonation arises from nonlinear source considerations . Vocal tract shaping , a filter-based framework , provides an alternative explanation for biphonation . In one seminal study of Tuvan throat singing , Levin and Edgerton examined a wide variety of song types and suggested that there were three components at play . The first two ( ‘tuning a harmonic’ relative to the filter and lengthening the closed phase of the vocal fold vibration ) represented a coupling between source and filter . But it was the third , narrowing of the formant , that appeared crucial . Yet , the authors offered little empirical justification for how these effects are produced by the vocal tract shape in the presented radiographs . Thus it remains unclear how the high-pitched formant in Khoomei was formed ( Grawunder , 2009 ) . Another study ( Adachi and Yamada , 1999 ) examined a throat singer using magnetic resonance imaging ( MRI ) and captured static images of the vocal tract shape during singing . These images were then used in a computational model to produce synthesized song . Adachi and Yamada argued that a "rear cavity" was formed in the vocal tract and its resonance was essential to biphonation . However , their MRI data reveal limited detail since they were static images of singers already in the biphonation state . Small variations in vocal tract geometry can have pronounced effects on produced song ( Story et al . , 1996 ) and data from static MRI would reveal little about how and which parts of the vocal tract change shape as the singers transition from simple tonal song to biphonation . To understand which features of vocal tract morphology are crucial to biophonation , a dynamic description of vocal tract morphology would be required . Here we study the dynamic changes in the vocal tracts of multiple expert practitioners from Tuva as they produce Khoomei . We use MRI to acquire volumetric 3D shape of the vocal tract of a singer during biphonation . Then , we capture the dynamic changes in a midsagittal slice of the vocal tract as singers transition from tonal to biphonic singing while making simultaneous audio recordings of the song . We use these empirical data to guide our use of a computational model , which allows us to gain insight into which features of vocal tract morphology are responsible for the singing phonetics observed during biophonic Khoomei song ( e . g . , Story , 2016 ) . We focus specifically on the Sygyt ( or Sigit ) style of Khoomei ( Aksenov , 1973 ) . We made measurements from three Tuvan singers performing Khoomei in the Sygyt style ( designated as T1–T3 ) and one ( T4 ) in a non-Sygyt style . Songs were analyzed using short-time Fourier transforms ( STFT ) , which provide detailed information in both temporal and spectral domains . We recorded the singers transitioning from normal singing into biphonation , Figure 1 showing this transition for three singers . The f0 of their song is marked in the figure ( approximately 140 Hz for subject T2 , 164 Hz for both T1 and T3 ) and the overtone structure appears as horizontal bands . Varying degrees of vibrato can be observed , dependent upon the singer ( Figure 1; see also longer spectrograms in Appendix 1—figure 6 and Appendix 1—figure 7 ) . Most of the energy in their song is concentrated in the overtones and no subharmonics ( i . e . , peaks at half-integer multiples of f0 ) are observed . In contrast to these three singers , singer T4 performing in a non-Sygyt style exhibited a fundamental frequency of approximately 130 Hz , although significant energy additionally appears around 50–55 Hz , well below an expected subharmonic ( Appendix 1—figure 5 ) . If we take a slice , that is a time-point from the spectrogram and plot the spectrum , we can observe the peaks to infer the formant structure from this representation of the sound ( red-dashed lines in Figure 1 and Appendix 1—figure 4 ) . As the singers transition from normal singing to biphonation , we see that the formant structure changes significantly and the positions of formant peaks shift dramatically and rapidly . Note that considering time points before and after the transitions also provides an internal control for both normal and focused song types ( Appendix 1—figure 4 ) . Once in the biphonation mode , all three singers demonstrate overtones in a narrow spectral band around 1 . 5–2 kHz; we refer to this as the focused state . Specifically , Figure 1 shows that not only is just a single or small group of overtones accentuated , but also that nearby ones are greatly attenuated: ±1 overtones are as much 15–35 dB and ±2 overtones are 35–65 dB below the central overtone . Whereas the energy in the low-frequency region associated with the first formant ( below 500 Hz ) is roughly constant between the normal-singing and focused states , there is a dramatic change in the spectrum for the higher formants above 500 Hz . In normal singing ( i . e . , prior to the focused state ) , spectral energy is distributed across several formants between 500 and 4000 Hz . In the focused state after the transition , the energy above 500 Hz becomes narrowly focused in the 1 . 5–2 kHz region , generating a whistle-like pitch that carries the song melody . To assess the degree of focus objectively and quantitatively , we computed an energy ratio eR⁢ ( fL , fH ) that characterizes the relative degree of energy brought into a narrow band against the energy spread over the full spectrum occupied by human speech ( see Materials and methods ) . In normal speech and singing , for [fL , fH]=[1 , 2kHz] , typically eR is small ( i . e . , energy is spread across the spectrum , not focused into that narrow region between 1 and 2 kHz ) . For the Tuvan singers , prior to a transition into a focused state , eR⁢ ( 1 , 2 ) is similarly small . However once the transition occurs ( red triangle in Figure 1 ) , those values are large ( upwards of 0 . 5 and higher ) and sustained across time ( Appendix 1—figure 2 and Appendix 1—figure 3 ) . For one of the singers ( T2 ) the situation was more complex , as he created multiple focused formants ( Figure 1 middle panels and Appendix 1—figure 6 , Appendix 1—figure 8 ) . The second focused state was not explicitly dependent upon the first: The first focused state clearly moves and transitions between approximately 1 . 5–2 kHz ( by 30% ) while the second focused state remains constant at approximately 3–3 . 5 kHz ( changing less than 1% ) . Thus the focused states are not harmonically related . Unlike the other singers , T2 not only has a second focused state , but also had more energy in the higher overtones ( Figure 1 ) . As such , singer T2 also exhibited a different eR time course , which took on values that could be relatively large even prior to the transition . This may be because he took multiple ways to approach the transition into a focused state ( e . g . , Appendix 1—figure 9 ) . Plotting spectra around the transition from normal to biphonation singing in a waterfall plot indicates that the sharp focused filter is achieved by merging two broader formants together ( F2 and F3 in Figure 2; Kob , 2004 ) . This transition into the focused state is fast ( ∼40–60 ms ) , as are the shorter transitions within the focused state where the singer melodically changes the filter that forms the whistle-like component of their song ( Figure 1 , Appendix 1—figure 8 ) . While we can infer the shape of the formants in Khoomei by examining audio recordings , such analysis is not conclusive in explaining the mechanism used to achieve these formants . The working hypothesis was that vocal tract shape determines these formants . Therefore , it was crucial to examine the shape and dynamics of the vocal tract to determine whether the acoustic measurements are consistent with this hypothesis . To accomplish this , we obtained MRI data from one of the singers ( T2 ) that are unique in two regards . First , there are two types of MRI data reported here: steady-state volumetric data Figure 3 and Appendix 1—figure 18 ) and dynamic midsagittal images at several frames per second that capture changes in vocal tract position ( Figure 4A–B and Appendix 1—figure 20 ) . Second is that the dynamic data allow us to examine vocal tract changes as song transitions into a focused state ( e . g . , Appendix 1—figure 20 ) . The human vocal tract begins at the vocal folds and ends at the lips . Airflow produced by the vocal cords sets the air-column in the tract into vibration , and its acoustics determine the sound that emanates from the mouth . The vocal tract is effectively a tube-like cavity whose shape can be altered by several articulators: the jaw , lips , tongue , velum , epiglottis , larynx and trachea ( Figure 4C ) . Producing speech or song requires that the shape of the vocal tract , and hence its acoustics , are precisely controlled ( Story , 2016 ) . Several salient aspects of the vocal tract during the production of Khoomei can be observed in the volumetric MRI data . The most important feature however , is that there are two distinct and relevant constrictions when in the focused state , corresponding roughly to the uvula and alveolar ridge . Additionally , the vocal tract is expanded in the region just anterior to the alveolar ridge ( Figure 4A ) . The retroflex position of the tongue tip and blade produces a constriction at 14 cm , and also results in the opening of this sublingual space . It is the degree of constriction at these two locations that is hypothesized to be the primary mechanism for creating and controlling the frequency at which the formant is focused . Having established that the shape of vocal tract during Khoomei does indeed have two constrictions , consistent with observations from other groups , the primary goals of our modeling efforts were to use the dynamic MRI data as morphological benchmarks and capture the merging of formants to create the focused states as well as the dynamic transitions into them . Our approach was to use a well-established linear "source/filter" model ( e . g . , Stevens , 2000 ) that includes known energy losses ( Sondhi and Schroeter , 1987; Story et al . , 2000; Story , 2013 ) . Here , the vibrating vocals folds act as the broadband sound source ( with the f0 and associated overtone cascade ) , while resonances of the vocal tract , considered as a series of 1-D concatenated tubes of variable uniform radius , act as a primary filter . We begin with a first order assumption that the system behaves linearly , which allows us for a simple multiplicative relationship between the source and filter in the spectral domain ( e . g . , Appendix 1—figure 10 ) . Acoustic characteristics of the vocal tract can be captured by transforming the three-dimensional configuration ( Figure 3 ) into a tube with variation in its cross-sectional area from the glottis to the lips ( Figure 4 and Figure 5 ) . This representation of the vocal tract shape is called an area function , and allows for calculation of the corresponding frequency response function ( from which the formant frequencies can be determined ) with a one-dimensional wave propagation algorithm . Although the area function can be obtained directly from a 3D vocal tract reconstruction ( e . g . , Story et al . , 1996 ) , the 3D reconstructions of the Tuvan singer’s vocal tract were affected by a large shadow from a dental post ( e . g . , see Figure 4 ) and were not amenable to detailed measurements of cross-sectional area . Instead , a cross-sectional area function was measured from the midsagittal slice of the 3D image set ( see Materials and methods and Appendix for details ) . Thus , the MRI data provided crucial bounds for model parameters: the locations of primary constrictions and thereby the associated area functions . The frequency response functions derived from the above static volumetric MRI data ( e . g . , Figure 4D ) indicate that two formants F2 and F3 cluster together , thus enhancing both their amplitudes . Clearly , if F2 and F3 could be driven closer together in frequency , they would merge and form a single formant with unusually high amplitude . We hypothesize that this mechanism could be useful for effectively amplifying a specific overtone , such that it becomes a prominent acoustic feature in the sound produced by a singer , specifically the high frequency component of Khoomei . Next , we used the model in conjunction with time-resolved MRI data to investigate how the degree of constriction and expansion at different locations along the vocal tract axis could be a mechanism for controlling the transition from normal to overtone singing and the pitch while in the focused state . These results are summarized in Figure 5 ( further details are in the Appendix ) . While the singers are in the normal song mode , there are no obvious strong constrictions in their vocal tracts ( e . g . , Appendix 1—figure 11 ) . After they transition , in each MRI from the focused state , we observe a strong constriction near the alveolar ridge . We also observe a constriction near the uvula in the upper pharynx , but the degree of constriction here varies . If we examine the simultaneous audio recordings , we find that variations in this constriction are co-variant with the frequency of the focused formant . From this , we surmise that the mechanism for controlling the enhancement of voice harmonics is the degree of constriction near the alveolar ridge in the oral cavity ( labeled CO in Figure 5 ) , which affects the proximity of F2 and F3 to each other ( Appendix 1—figure 12 ) . Additionally , the degree of constriction near the uvula in the upper pharynx ( CP ) controls the actual frequency at which F2 and F3 converge ( Appendix 1—figure 13 ) . Other parts of the vocal tract , specifically the expansion anterior to CO , may also contribute since they also show small co-variations with the focused formant frequency ( Appendix 1—figure 14 ) . Further , a dynamic implementation of the model , as shown in Appendix 1—figure 14 , reasonably captures the rapid transition into/out of the focused state as shown in Figure 1 . Taken together , the model confirms and explains how these articulatory changes give rise to the observed acoustic effects . To summarize , an overtone singer could potentially ‘play’ ( i . e . , select ) various harmonics of the voice source by first generating a tight constriction in the oral cavity near the alveolar ridge , and then modulating the degree of constriction in the uvular region of the upper pharynx to vary the position of the focused formant , thereby generating a basis for melodic structure . The notion of a focused state is mostly consistent with vocal tract filter-based explanations for biphonation in previous studies ( e . g . , Bloothooft et al . , 1992; Edgerton et al . , 1999; Adachi and Yamada , 1999; Grawunder , 2009 ) , where terms such as an ‘interaction of closely spaced formants’ , ‘reinforced harmonics’ , and ‘formant melting’ were used . In addition , the merging of multiple formants is closely related to the ‘singer’s formant’ , which is proposed to arise around 3 kHz due to formants F3–F5 combining ( Story , 2016 ) , though this is typically broader and less prominent than the focused states exhibited by the Tuvans . Our results explain how this occurs and are also broadly consistent with Adachi and Yamada ( 1999 ) in that a constricted ‘rear cavity’ is crucial . However , we find that this rear constriction determines the pitch of the focused formant , whereas it is the ‘front cavity’ constriction near the alveolar ridge that produces the focusing effect ( i . e . , merging of formants F2 and F3 ) . Further , the present data appear in several ways inconsistent with conclusions from previous studies of Khoomei , especially those that center on effects that arise from changes in the source . Three salient examples are highlighted . First , we observed overtone structure to be highly stable , though some vibrato may be present . This contrasts the claim by Levin and Edgerton ( 1999 ) that “ ( t ) o tune a harmonic , the vocalist adjusts the fundamental frequency of the buzzing sound produced by the vocal folds , so as to bring the harmonic into alignment with a formant’ . That is , we see no evidence for the overtone ‘ladder’ being lowered or lifted as they suggested ( note in Figure 1 , f0 is held nearly constant ) . Further , this stability argues against a transition into a different mode of glottal pulse generation , which could allow for a ‘second source’ ( Mergell and Herzel , 1997 ) . Second , a single sharply defined harmonic alone is not sufficient to get the salient perception of a focused state , as had been suggested by Levin and Edgerton ( 1999 ) . Consider Appendix 1—figure 9 , especially at the 4 s mark , where the voicing is ‘pressed' . Pressed phonation , also referred to as ventricular voice , occurs when glottal flow is affected by virtue of tightening the laryngeal muscles such that the ventricular folds are brought into vibration . This has the perceptual effect of adding a degree of roughness to the voice sound ( Lindestad et al . , 2001; Edmondson and Esling , 2006 ) . There , a harmonic at 1 . 51 kHz dominates ( i . e . , the two flanking overtones are approximately 40 dB down ) , yet the song has not yet perceptibly transitioned . It is not until the cluster of overtones at 3–3 . 5 kHz is brought into focus that the perceptual effect becomes salient , perhaps because prior to the 5 . 3 s mark the broadband nature of those frequencies effectively masks the first focused state . Third , we do not observe subharmonics , which contrasts a prior claim ( Lindestad et al . , 2001 ) that ” ( t ) his combined voice source produces a very dense spectrum of overtones suitable for overtone enhancement’ . However , that study was focused on a different style of song called ‘Kargyraa’ , which does not exhibit as clearly a focused state as in Sygyt . An underlying biophysical question is whether focused overtone song arises from inherently linear or nonlinear processes . Given that Khoomei consists of the voicing of two or more pitches at once and exhibits dramatic and fast transitions from normal singing to biphonation , nonlinear phenomena may seem like an obvious candidate ( Herzel and Reuter , 1996 ) . It should be noted that Herzel and Reuter ( 1996 ) go so far to define biphonation explicitly through the lens of nonlinearity . We relax such a definition and argue for a perceptual basis for delineating the boundaries of biphonation . Certain frog species exhibit biphonation , and it has been suggested that their vocalizations can arise from complex nonlinear oscillatory regimes of separate elastically coupled masses ( Suthers et al . , 2006 ) . Further , the appearance of abrupt changes in physiological systems ( as seen in Figure 1 ) has been argued to be a flag for nonlinear mechanisms ( Goldberger et al . , 2002 ) ; for example , by virtue of progression through a bifurcation . Our results present two lines of evidence that argue against Sygyt-style Khoomei arising primarily from a nonlinear process . First , the underlying harmonic structure of the vocal fold source appears highly stable through the transition into the focused state ( Figure 1 ) . There is little evidence of subharmonics . A source spectral structure that is comprised of an f0 and integral harmonics would suggest a primarily linear source mechanism . Second is that our modeling efforts , which are chiefly linear in nature , reasonably account for the sudden and salient transition . That is , the model is readily sufficient to capture the characteristic that small changes in the vocal tract can produce large changes in the filter . Thereby , precise and fast motor control of the articulators in a linear framework accounts for the transitions into and out of the focused state . Thus , in essence , Sygyt-style Khoomei could be considered a linear means to achieve biphonation . Connecting back to nonlinear phonation mechanisms in non-mammals , our results provide further context for how human song production and perception may be similar and/or different relative to that of non-humans ( e . g . , Doolittle et al . , 2014; Kingsley et al . , 2018 ) . Nevertheless , features that appear transiently in spectrograms do provide hints of source nonlinearity , such as the brief appearance of subharmonics in some instances ( Appendix 1—figure 15B ) . This provides an opportunity to address the limitations of the current modeling efforts and to highlight future considerations . We suggest that further analysis ( e . g . , Theiler et al . , 1992; Tokuda et al . , 2002; Kantz and Schreiber , 2004 ) of Khoomei audio recordings may help to inform the model and might better capture focused filter sharpness and the origin of secondary focused states . Several potential areas for improvement are: nonlinear source–filter coupling ( Titze et al . , 2008 ) ; a detailed model of glottal dynamics ( e . g . , ratio of open/closed phases in glottal flow [Grawunder , 2009; Li and Hou , 2017] , and periodic vibrations in f0 ) ; inclusion of piriform sinuses as side-branch resonators ( Dang and Honda , 1997; Titze and Story , 1997 ) ; inclusion of the 3-D geometry; and detailed study of the front cavity ( e . g . , lip movements ) that may be used by the singer to maintain control of the focused state and to make subtle manipulations . Although this study did not directly assess the percept associated with these vocal productions , the results raise pressing questions about how the spectro-temporal signatures of biphonic Khoomei described here create the classical perception of Sygyt-style Khoomei as two distinct sounds ( Aksenov , 1973 ) . The first , the low-pitched drone , which is present during both the normal singing and the focused-state biphonation intervals , reflects the pitch associated with f0 , extracted from the harmonic representation of the stimulus . It is well established that the perceived pitch of a broadband sound comprised of harmonics reflects the f0 derived primarily from the perceptually resolved harmonics up to about 10⁢f0 ( Bernstein and Oxenham , 2003 ) . The frequency resolution of the peripheral auditory system is such that these low-order harmonics are individually resolved by the cochlea , and it appears that such filtering is an important prerequisite for pitch extraction associated with that common f0 . The second sound , the high-pitched melody , is present only during the focused-state intervals and probably reflects a pitch associated with the focused formant . An open question , however , is why this focused formant would be perceived incoherently as a separate pitch ( Shamma et al . , 2011 ) , when it contains harmonics at multiples of f0 . The auditory system tends to group together concurrent harmonics into a single perceived object with a common pitch ( Roberts et al . , 2015 ) , and the multiple formants of a sung or unsung voice are not generally perceived as separate sounds from the low harmonics . The fact that the focused formant is so narrow apparently leads the auditory system to interpret this sound as if it were a separate tone , independent of the low harmonics associated with the drone percept , thereby effectively leading to a pitch decoherence . This perceptual separation could be attributable to a combination of both bottom-up ( i . e . , cochlear ) and top-down ( i . e . , perceptual ) factors . From the bottom-up standpoint , even if the focused formant is broad enough to encompass several harmonic components , the fact that it consists of harmonics at or above 10 f0 ( i . e . , the 1500 Hz formant frequency represents the 10th harmonic of a 150 Hz f0 ) means that these harmonics will not be spectrally resolved by cochlear filtering ( Bernstein and Oxenham , 2003 ) . Instead , the formant will be represented as a single spectral peak , similar to the representation of a single pure tone at the formant frequency . Although the interaction of harmonic components at this cochlear location will generate amplitude modulation at a rate equal to the f0 ( Plack and Oxenham , 2005 ) , it has been argued that a common f0 is a weak cue for binding low- and high-frequency formants ( Culling and Darwin , 1993 ) . Rather , other top-down mechanisms of auditory-object formation may play a more important role in generating a perception of two separate objects in Khoomei . For example , the rapid onsets of the focused formant may enhance its perceptual separation from the constant drone ( Darwin , 1984 ) . Further , the fact that the focused formant has a variable frequency ( i . e . , frequency modulation , or FM ) while the drone maintains a constant f0 is another difference that could facilitate their perceptual separation . Although it has been argued that FM differences between harmonic sounds generally have little influence on their perceived separation ( Darwin , 2005 ) , others have reported enhanced separation in the special case in which one complex was static and the other had applied FM ( Summerfield and Culling , 1992 ) – similar to the first and second formants during the Tuvan focused state . The perceptual separation of the two sounds in the Tuvan song might be further affected by a priori expectations about the spectral qualities of vocal formants ( Billig et al . , 2013 ) . Because a narrow formant occurs so rarely in natural singing and speech , the auditory system might be pre-disposed against perceiving it as a phonetic element , limiting its perceptual integration with the other existing formants . Research into ‘sine-wave speech’ provides some insights into this phenomenon . When three or four individual frequency-modulated sinusoids are presented at formant frequencies in lieu of natural formants , listeners can , with sufficient training , perceive the combination as speech ( Remez et al . , 1981 ) . Nevertheless , listeners largely perceive these unnatural individual pure tones as separate auditory objects ( Remez et al . , 2001 ) , much like the focused formant in Khoomei . Further research exploring these considerations would help close the production–perception circle underlying the unique percept arising from Tuvan throat song . Recordings were made at York University ( Toronto , ON , Canada ) in a double-walled acoustic isolation booth ( IAC ) using a Zoom H5 24-bit digital recorder and an Audio-Technica P48 condenser microphone . A sample rate of 96 kHz was used . Spectral analysis was done using custom-coded software in Matlab . Spectrograms were typically computed using 4096 point window segments with 95% fractional overlap and a Hamming window . Harmonics ( black circles in Figure 1 ) were estimated using a custom-coded peak-picking algorithm . Estimated formant trends ( red dashed lines in Figure 1 ) were determined using a lowpass interpolating filter built into Matlab’s digital signal processing toolbox with a scaling factor of 10 . From this trend , the peak-picking was reapplied to determine ‘formant’ frequencies ( red 'x's in Figure 1 ) . This process could be repeated across the spectrogram to track overtone and formant frequency/strength effectively , as shown in Appendix 1—figure 1 . To quantify the focused states , we developed a dimension-less measure eR⁢ ( fL , fH ) to represent the energy ratio of that spanning a frequency range fH-fL relative to the entire spectral output . This can be readily computed from the spectrogram data as follows . First take a ‘slice’ from the spectrogram and convert spectral magnitude to linear ordinate and square it ( as intensity is proportional to pressure squared ) . Then integrate across frequency , first for a limited range spanning [fL , fH] ( e . g . , 1–2 kHz ) and then for a broader range of [0 , fm⁢a⁢x] ( e . g . , 0–8 kHz; 8 kHz is a suitable maximum as there is little acoustic energy in vocal output above this frequency ) . The ratio of these two is then defined as eR , and takes on values between 0 and 1 . This can be expressed more explicitly as: ( 1 ) eR⁢ ( fL , fH ) = ( ∫fLfHP⁢ ( f ) ⁢𝑑f∫0fm⁢a⁢xP⁢ ( f ) ⁢𝑑f ) 2where P is the magnitude of the scaled sound pressure , f is frequency , and fL and fH are filter limits for considering the focused state . The choice of [fL , fH]=[1 , 2] kHz has the virtue of spanning an octave , which also closely approximates the ‘seventh octave’ from about C6 to C7 . eR did not depend significantly upon the length of the fast Fourier transform ( FFT ) window . Values of eR for the waveforms used in Figure 1 are shown in Appendix 1—figures 2 and 3 . MRI images were acquired at the York MRI Facility on a 3 . 0 Tesla MRI scanner ( Siemens Magnetom TIM Trio , Erlangen , Germany ) , using a 12-channel head coil and a neck array . Data were collected with the approval of the York University Institutional Review Board . The participant was fitted with an MRI compatible noise-cancelling microphone ( Optoacoustics , Mazor , Israel ) mounted directly above the lips . The latency of the microphone and noise-cancelling algorithm was 24 ms . Auditory recordings were made in QuickTime on an iMac during the scans to verify performance . Images were acquired using one of two paradigms , static or dynamic . Static images were acquired using a T1-weighted 3D gradient echo sequence in the sagittal orientation with 44 slices centered on the vocal tract , TR = 2 . 35 ms , TE = 0 . 97 ms , flip angle = 8 degrees , FoV = 300 mm , and a voxel dimension of 1 . 2 × 1 . 2×1 . 2 mm . Total acquisition time was 11 s . The participant was instructed to begin singing a tone , and to hold it in a steady state for the duration of the scan . The scan was started immediately after the participant began to sing and had reached a steady state . Audio recordings verified a consistent tone for the duration of the scan . Dynamic images were acquired using a 2D gradient echo sequence . A single 10 . 0 mm thick slice was positioned in a sagittal orientation along the midline of the vocal tract , TR = 4 . 6 ms , TE = 2 . 04 ms , flip angle = 8 degrees , FoV = 250 mm , and a voxel dimension of 2 . 0 × 2 . 0×10 . 0 mm . One hundred measurements were taken for a scan duration of 27 . 75 s . The effective frame rate of the dynamic images was 3 . 6 Hz . Audio recordings were started just prior to scanning . Only subject T2 participated in the MRI recordings . The participant was instructed to sing a melody for the duration of the scan , and took breaths as needed . For segmentation ( Figure 3 ) , 3D MRI images ( Run1; see Appendix ) were loaded into Slicer ( version 4 . 6 . 2 r25516 ) . The air-space in the oral cavity was manually segmented using the segmentation module , identified and painted in slice by slice . Careful attention was paid to the parts of the oral cavity that were affected by the artifact from the dental implant . The air cavity was manually repainted to be approximately symmetric in this region using the coronal and axial view ( Figure 3A ) . Once completely segmented , the sections were converted into a 3D model and exported as a STL file . This mesh file was imported into MeshLab ( v1 . 3 . 4Beta ) for cleaning and repairing the mesh . The surface of the STL was converted to be continuous by removing non-manifold faces and then smoothed using depth and Laplacian filters . The mesh was then imported into Meshmixer where further artifacts were removed . This surface-smoothed STL file was finally reimported into Slicer , generating the display in Figure 3B . Measurement of the cross-distance function is illustrated in Figure 4 . The inner and outer profiles of the vocal tract were first determined by manual tracing of the midsagittal image . A 2D iterative bisection algorithm ( Story , 2007 ) was then used to find the centerline within the profiles extending from the glottis to the lips , as shown by the white dots in Figure 4A . Perpendicular to each point on the centerline , the distance from the inner to outer profiles was measured to generate the cross-distance function shown in Figure 4B; the corresponding locations of the anatomic landmarks shown in the midsagittal image are also indicated on the cross-distance function . The cross-distance function , D⁢ ( x ) , can be transformed to an approximate area function , A⁢ ( x ) , with the relation A⁢ ( x ) =k⁢Dα⁢ ( x ) , where k and α are a scaling factor and exponent , respectively . If the elements of D⁢ ( x ) are considered to be diameters of a circular cross-section , k= ( π/4 ) and α=2 . Although other values of k and α have been proposed to account for the complex shape of the vocal tract cross-section ( Heinz and Stevens , 1964; Lindblom and Sundberg , 1971; Mermelstein , 1973 ) , there is no agreement on a fixed set of numbers for each parameter . Hence , the circular approximation was used in this study to generate an estimate of the area function . In Figure 4C , the area function is plotted as its tubular equivalent , where the radii D⁢ ( x ) /2 were rotated about an axis to generate circular sections from the glottis to the lips . The associated frequency response of that area function is shown in Figure 4D and was calculated with a transmission line approach ( Sondhi and Schroeter , 1987; Story et al . , 2000 ) , which included energy losses due to yielding walls , viscosity , heat conduction , and acoustic radiation at the lips . Side branches such the piriform sinuses were not considered in detail in this study . The first five formant frequencies ( resonances ) , F1 , … , F5 , were determined by finding the peaks in the frequency response functions with a peak-picking algorithm ( Titze et al . , 1987 ) and are located at 400 , 1065 , 1314 , 3286 , and 4029 Hz , respectively . To examine changes in pitch , a particular vocal tract configuration was manually ‘designed ( Appendix 1—figure 6 ) such that it included constrictive and expansive regions at locations similar to those measured from the singer ( i . e . , Figure 4 ) , but to a less extreme degree . We henceforth denote this area function as A0⁢ ( x ) , and it generates a frequency response with widely spaced formant frequencies ( F1⁢…⁢5=[529 , 1544 , 2438 , 3094 , 4236]⁢ Hz ) , essentially a neutral vowel . In many of the audio signals recorded from the singer , the fundamental frequency , fo ( i . e . , the vibratory frequency of the vocal folds ) , was typically about 150 Hz . The singer then appeared to enhance one of the harmonics in the approximate range of 8⁢fo⁢…⁢12⁢fo . Taking the 12th harmonic ( 12×150=1800 Hz ) as an example target frequency ( dashed line in the frequency response shown in Figure 5c ) , the area function A0⁢ ( x ) was iteratively perturbed by the acoustic-sensitivity algorithm described in Story ( 2006 ) until F2 and F3 converged on 1800 Hz and became a single formant peak in the frequency response . Additional details on the perturbation process leading into Figure 5 are detailed in the Appendix .	The republic of Tuva , a remote territory in southern Russia located on the border with Mongolia , is perhaps best known for its vast mountainous geography and the unique cultural practice of “throat singing” . These singers simultaneously create two different pitches: a low-pitched drone , along with a hovering whistle above it . This practice has deep cultural roots and has now been shared more broadly via world music performances and the 1999 documentary Genghis Blues . Despite many scientists being fascinated by throat singing , it was unclear precisely how throat singers could create two unique pitches . Singing and speaking in general involves making sounds by vibrating the vocal cords found deep in the throat , and then shaping those sounds with the tongue , teeth and lips as they move up the vocal tract and out of the body . Previous studies using static images taken with magnetic resonance imaging ( MRI ) suggested how Tuvan singers might produce the two pitches , but a mechanistic understanding of throat singing was far from complete . Now , Bergevin et al . have better pinpointed how throat singers can produce their unique sound . The analysis involved high quality audio recordings of three Tuvan singers and dynamic MRI recordings of the movements of one of those singers . The images showed changes in the singer’s vocal tract as they sang inside an MRI scanner , providing key information needed to create a computer model of the process . This approach revealed that Tuvan singers can create two pitches simultaneously by forming precise constrictions in their vocal tract . One key constriction occurs when tip of the tongue nearly touches a ridge on the roof of the mouth , and a second constriction is formed by the base of the tongue . The computer model helped explain that these two constrictions produce the distinctive sounds of throat singing by selectively amplifying a narrow set of high frequency notes that are made by the vocal cords . Together these discoveries show how very small , targeted movements of the tongue can produce distinctive sounds .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems" ]	2020	Overtone focusing in biphonic tuvan throat singing
Speech perception presumably arises from internal models of how specific sensory features are associated with speech sounds . These features change constantly ( e . g . different speakers , articulation modes etc . ) , and listeners need to recalibrate their internal models by appropriately weighing new versus old evidence . Models of speech recalibration classically ignore this volatility . The effect of volatility in tasks where sensory cues were associated with arbitrary experimenter-defined categories were well described by models that continuously adapt the learning rate while keeping a single representation of the category . Using neurocomputational modelling we show that recalibration of natural speech sound categories is better described by representing the latter at different time scales . We illustrate our proposal by modeling fast recalibration of speech sounds after experiencing the McGurk effect . We propose that working representations of speech categories are driven both by their current environment and their long-term memory representations . The way the brain processes sensory information to represent the perceived world is flexible and varies depending on changes in the stimulus landscape . Neurocognitive adaptation to varying stimuli can be driven by an explicit external feedback signal , but might also take place with simple passive exposure to a changing stimulus environment via implicit statistical learning ( Saffran et al . , 1996; Gilbert et al . , 2001; Barascud et al . , 2016; Schwiedrzik et al . , 2014 ) . In the domain of speech perception , neural representations of sound categories are susceptible to stimulus-driven recalibration . Typically , the perception of unclear or ambiguous speech stimuli that have previously been disambiguated by context ( McQueen et al . , 2006; Clarke and Luce , 2005 ) or by a concurrent visual stimulus ( Bertelson et al . , 2003; Vroomen et al . , 2007 ) is biased by the disambiguating percept . Even simple exposure to novel statistics e . g . , a variation in the spread of sensory features characteristic of stop consonants , quickly results in a modified slope in psychometric functions , and changes the way listeners classify stimuli ( Clayards et al . , 2008 ) . Interestingly , acoustic representations are also modified after altered auditory feedback during production ( Nasir and Ostry , 2009; Lametti et al . , 2014; Patri et al . , 2018 ) . These observations illustrate that speech sound categories remain largely plastic in adulthood . Using two-alternative forced choice tasks , studies have shown that changes in acoustic speech categories can be induced by input from the visual modality ( e . g . Bertelson et al . , 2003 ) . Reciprocally , acoustic information can also disambiguate lipreading ( Baart and Vroomen , 2010 ) , resulting in measurable categorization aftereffects . While such effects can be observed after repeated exposure to the adapting stimuli , recalibration can also occur very rapidly , with effects being observable after a single exposure ( Vroomen et al . , 2007 ) . This fast and dynamic process has been modelled as incremental Bayesian updating ( Kleinschmidt and Jaeger , 2015 ) where internal perceptual categories track the stimulus statistics . According to this model , when listeners are confronted with altered versions of known speech categories , the perceived category representation is updated to become more consistent with the actual features of the stimulus . The resulting recalibration weighs all evidence equally , disregarding their recency . The model hence successfully describes perceptual changes observed when listeners are confronted with repeated presentations of a single modified version of a speech sound . However , it cannot appropriately deal with intrinsically changing environments , in which sensory cues quickly become obsolete . Real moment to moment physical changes in the environment are referred to as ‘volatility’ to distinguish them from the trial-by-trial response variability observed in a fixed environment . Inference in volatile environments has been studied mostly in relation to decision making tasks in which participants can use an explicit feedback to keep track of the varying statistics of arbitrary cue-reward associations ( e . g . Behrens et al . , 2007 ) or arbitrarily defined categories ( e . g . Summerfield et al . , 2011 ) . These studies suggest that humans are able to adjust their learning rate to the volatility in the stimulus set , with faster learning rates ( implying a stronger devaluation of recent past evidence ) in more volatile environments . This has led to normative models focussing on the online estimation of volatility , in which task-relevant features are represented at a single variable time scale ( Behrens et al . , 2007; Mathys et al . , 2011 ) . The notion of variable learning rates likely also applies to speech processing . However , we propose that when recalibrating natural speech categories , a normative model should additionally take into account that these categories may themselves change at different time scales . For example , transient acoustic changes within a given speech sound category , such as those coming from a new speaker , must not interfere with the long-term representation of that category that should be invariant for example to speakers . We therefore hypothesize that speech sound categories could be represented by more than a single varying timescale . Although speech category recalibration has not been systematically studied in variable environments , Lüttke and collaborators ( Lüttke et al . , 2016a; Lüttke et al . , 2018 ) found evidence for recalibration in an experiment that included audio-visual McGurk stimuli shown without an explicit adapting condition . The first study involved six different vowel/consonant/vowel stimuli presented in random order , and recalibration was observed even when acoustic stimuli were not ambiguous ( e . g . the /aba/ sound in the McGurk trials ) . The McGurk effect ( the fact that an acoustic stimulus /aba/is mostly perceived as an illusory/ada/ when presented with the video of a speaker producing /aga/ ) was powerful enough to yield observable adaptive effects across consecutive trials . Specifically , the probability of an acoustic/aba/ to be categorized as/ada/ , was higher when the trial was preceded by an audio-visual McGurk fusion . Recalibration effects do not generalize to all phonetic contrasts/categories ( Reinisch et al . , 2014 ) . After participants had recalibrated acoustic sounds in the /aba /- /ada/continuum ( on the basis of acoustic formant transitions ) , recalibration was neither present for /ibi /- /idi/ ( cued by burst and frication ) , nor for /ama/- /ana/or to /ubu/- /udu/continua ( both cued by formant transitions ) . Likewise , in a word recognition task , participants were able to keep different F0/VOT ( fundamental frequency/voice onset time ) correlation statistics for different places of articulation ( Idemaru and Holt , 2014 ) . Based on this failure to generalize , we hypothesized that very short-term changes can modify the internal mapping between sensory features and sublexical speech categories rather than phonemic categories . To test this hypothesis , we simulated Lüttke et al . ’s experiment ( Lüttke et al . , 2016a ) , using an audiovisual integration model based on hierarchical Bayesian inference . The model was a version of a previous model of the McGurk effect ( Olasagasti et al . , 2015 ) that further included an adaptation mechanism using residual prediction errors to update internal representations associated with the perceived category . The model divides the process in two steps: perceptual inference and internal model recalibration . During perceptual inference the model takes the sensory input and infers a perceived speech category , by choosing the category that minimizes sensory prediction error . However , ‘residual’ prediction errors might remain following perceptual inference . This is typically the case after McGurk fusion; since the best explanation for the multisensory input , ‘ada’ , is neither the audio /aba/ nor the visual /aga/ , ‘residual’ prediction errors remain in both acoustic and visual modalities . The best match to the experimental results described above ( Lüttke et al . , 2016a ) was obtained when we considered that adaptation included two different time scales , resulting in 1 ) a transient effect leading to recalibration towards the most recently presented stimulus features , decaying towards 2 ) a longer-lasting representation corresponding to a mapping between category and stimulus features determined within a longer time span . Overall , these findings are consistent with theories that posit that the brain continuously recalibrates generative ( forward ) models to maintain self-consistency ( e . g . , Friston et al . , 2010 ) , and offers a neuro-computationally plausible implementation to resolve cognitive conflicts , which can sometimes appear as irrational behaviors , such as in post-choice re-evaluation of alternatives ( Coppin et al . , 2010; Izuma et al . , 2010; Colosio et al . , 2017; Otten et al . , 2017 ) . In a re-assessment of an existing dataset ( Lüttke et al . , 2016a ) selected participants with high percentage of fused percepts for McGurk stimuli . When presented with acoustic /aba/ together with a video of a speaker articulating /aga/ the most frequent percept was /ada/; we refer to these as fused McGurk trials . These listeners were combining information from acoustic ( A ) and visual ( V ) modalities since the same /aba/ acoustic token was correctly categorized as /aba/ when presented alone . After a fused McGurk trial , participants showed recalibration . They classified acoustic only /aba/ stimuli as ‘ada’ more frequently ( 29% ‘ada’ percepts ) than when the acoustic only /aba/was presented after any other stimulus ( 16% ‘ada’ percepts ) . Our goal was to compare generative models that interpret sensory input and continuously recalibrate themselves to best match the incoming input . Unlike many other studies of speech recalibration , the stimulus generating recalibration in Lüttke et al . ( the McGurk stimulus ) was not presented alone , but as part of a set of six stimuli that were presented in random order , thus making transient effects detectable . The assessment involved three acoustic only stimuli with sounds corresponding to /aba/ , /ada/ or/aga/; and three audiovisual stimuli – congruent /aba/ , congruent /ada/ , and the McGurk stimulus ( acoustic /aba/with video of /aga/ ) . We simulated Lüttke et al . ’s experiment by using a generative model relating the three possible speech categories ( /aba/ , /ada/ and /aga/ ) to the sensory input . We characterized sensory input with a visual feature , the amplitude of lip closure during the transition between the two vowels ( sV ) ; and an acoustic feature , the amplitude of the 2nd formant transition ( sA ) . We use ‘A’ to refer to quantities related to the acoustic feature and ‘V’ to quantities related to the visual feature . The internal generative model that characterizes the participant , described in detail in the methods section , generates sensory inputs ( sA and sV ) for the congruent versions of each of the three possible categories ( k = /aba/ , /ada/ , /aga/ ) . The model has a representation for each congruent category based on a Gaussian distribution in a two-dimensional feature space , itself a product of two univariate Gaussian distributions centered at ( θk , V θk , A ) and with standard deviations ( σk , V , σk , A ) for tokens k = {/aba/ , /ada/ , /aga/} ( Figure 1 , right panel ) . The model assumes that given a speech token k , 2nd formant amplitude CA and lip closure amplitude CV for each individual trial are chosen from the corresponding Gaussian distribution ( Figure 1 , right panel ) . Once values for CA and CV have been determined , sensory lip closure and 2nd formant transitions are obtained by adding sensory noise ( parameterized by σV and σA ) , to obtain the sensory input: sA and sV . During inference , the model is inverted and provides the posterior probability of a token given the noisy sensory input p ( k\|sV , sA ) . In both acoustic and audio-visual trials the listener was asked to report the perceived acoustic stimulus in a three-alternative forced choice task . To simulate the listener’s choice , we calculated the probability of the acoustic token given the stimulus . In our notation , this probability is expressed by p ( kA\|sV , sA ) for audiovisual stimuli and by p ( kA\|sA ) for unimodal acoustic stimuli ( see Materials and Methods for details ) . The percept at a single trial was determined by choosing the category that maximizes the posterior . The model qualitatively reproduces the average performance across participants in the task . We chose parameters that elicit a very high rate of McGurk percepts ( Figure 2 , middle panel of the bottom row ) and assumed that listeners were always integrating the two sensory streams . After the model has made the perceptual decision , it is recalibrated . After each trial , we changed the generative model’s location parameters associated with the perceived category ( θk , V θk , A ) , which represent the categories through their expected sensory feature values in each modality . This was done for both the acoustic and visual parameters after an audio-visual trial , and for the acoustic parameter after an acoustic trial . We assumed that this happens for every trial as part of a monitoring process that assesses how well the internal model matches sensory inputs . The changes are thus driven by residual sensory prediction error , the difference between the expected and observed values for the modulation amplitudes in each modality . When listeners consistently reported the fused percept ’ada’ when confronted with a video of /aga/ and the sound of /aba/ , the presence of the visual stream modified the acoustic percept from ’aba’ to ‘ada’ . Given that the acoustic input did not correspond to the one that was most expected from the perceived token , there was a systematic residual sensory prediction error . Since this residual prediction error was used as a signal to drive the model’s adaptation , the /ada/ representation moved towards the McGurk stimulus parameters after a fused percept ( Figure 3A ) . In the model , the residual prediction error occurs in the transformation from token identity to predicted modulation of the acoustic feature ( CA ) . Sensory evidence drives estimated CA towards the experimentally presented value: /aba/ for McGurk stimuli . Thus , when the percept is /ada/ , there is a mismatch between the top-down prediction as determined by the top-down component p ( CA\|k ) that drives CA towards θ/ada/ , A , and the actual value determined by the bottom-up component . This is evident in the following expression ( 1 ) Ck , A=σA2σA2+σk , A2θk , A+σk , A2σA2+σk , A2SAwith the first term reflecting the prior expectation for category ‘k’ and the second reflecting the sensory evidence . The expression can be rewritten to make the prediction error explicit . ( 2 ) Ck , A=θk , A+σk , A2σA2+σk , A2 ( SA−θk , A ) This highlights how the listener’s estimate of the modulation comes from combining the prediction from the category ( first term ) and the weighted residual prediction error ( in the second term ) . To minimize residual prediction error we consider that the participant recalibrates its generative model , which changes θk , V and θk , A towards sV and sA . If the stimuli are chosen with parameters ‘adapted’ to the listener , as we do , sV ~ θstim , V and sA ~ θstim , A . To drive recalibration we considered three different update rules; one derived from the Bayesian model used by Kleinschmidt and Jaeger ( 2015 ) , which assumes a stable environment and two empirically motivated rules . As a control we also simulated the experiment with no parameter updates . For each recalibration model and parameter value set , we run the experiment 100 times , therefore simulating 100 different listeners that share the same perceptual model . Each of the 100 simulated listeners was presented with a different random presentation of the six stimulus types; each presented 69 times ( as in the original paper ) . This gives a total of 414 trials per listener . To compare with the results from Lüttke et al . , who considered 27 participants , we randomly sampled groups of 27 from the 100 simulated listeners to obtain an empirical sampling distribution for the quantities of interest . We focus on the ‘McGurk contrast’: proportion of /aba/ sounds reported as ‘ada’ 29% when preceded by a fused McGurk trial versus 16% when preceded by other stimuli . We will also consider the ‘/ada/ contrast’: the difference in the proportion of purely acoustic /aba/ categorized as ‘ada’ when the preceding trial was a correctly categorized /ada/ sound ( 17% ) versus other stimuli ( 15% ) ( percentages correspond to the values reported in Lüttke et al . , 2016a ) . Although we did not perform an exhaustive parameter search , we did repeat the simulations for 20 different values of perceptual model parameters and we also varied recalibration parameter values for each update rule ( details in Methods ) . Below , for each recalibration model , we report results based on the parameter values that led to the best fit to the ‘McGurk contrast’ . As a control we simulated the experiment with no recalibration . For the simulation with the closest fit to the McGurk contrast , the percentage of acoustic /aba/ categorized as ‘ada’ was 14% after fused McGurk stimuli and 15% after the control stimuli ( Wilcoxon signed-rank test p=0 . 6 , Figure 4A ) . The 95% CI for the difference between trials preceded or not by a fused McGurk stimulus was [−6 . 48 , 5 . 25]% . For the /ada/ contrast , the percentage of acoustic /aba/ categorized as ‘ada’ was 14% after correctly identified /ada/ sounds and 19% after the control stimuli ( Wilcoxon signed-rank test p=0 . 5 ) . The 95% CI for the difference between the two conditions was [−8 . 24 , 4 . 21]% . We first considered the same updating principle as in Kleinschmidt and Jaeger ( 2015 ) to model changes in speech sound categorization after exposure to adapting stimuli . After each trial the generative model updates the parameters by considering their probability given the sensory input and the categorization p ( θ\|k sV sA ) , leading to sequential Bayesian updating ( Equation 6 in the methods section ) . The closest fit to the McGurk contrast was obtained for simulations with κk , f , 0 = 1 and νk , f , 0 = 1 . The percentage of /ada/ responses to acoustic only /aba/ stimuli was 23% after fused McGurk stimuli and 22% after control stimuli . This difference was not significant ( Wilcoxon signed-rank test p=0 . 7 , Figure 4B ) . The 95% confidence interval for the difference in medians between the two conditions was [−6 . 14 , 7 . 97]% , thus failing to reproduce the effect of interest . This might be due to the fact that Bayesian update rules have the form of a delta rule with a decreasing learning rate . As a consequence , the magnitude of changes in the categories diminishes as the experiment progresses and all stimuli end up being related to the same internal model . The updates did lead to observable effects; there was an overall increase of /ada/ responses to acoustic /aba/ ( 24% vs . the 14% of the control experiment without parameter updates ) . The resulting changes in model parameters are expected to induce an after-effect , that is , the point of subjective equivalence in an /aba /- /ada/ acoustic continuum should be shifted in the direction of /aba/ . The Bayesian update rule used above assumes that the parameters are constant in time and that therefore all samples have equal value , whether they are old or recent . This is equivalent to a delta rule with a learning rate tending to zero . We therefore considered a rule with a constant learning rate , which allows for updates of similar magnitude over the whole experiment . The model’s expected modulation for the perceived category was recalibrated according to:Δθk , f=0 . 2 ( sf−θk ) p ( k\|sA , sV ) onlyfork=perceptwhere f indexes the feature ( f = A , acoustic feature; f = V , visual feature ) . As Figure 4C shows , the percentage of acoustic /aba/ categorized as ‘ada’ was not significantly higher when the preceding trial was a fused McGurk trial compared with any other stimulus ( 27% vs . 24% , Wilcoxon signed-rank test , p = 0 . 08 Figure 4C; median difference at 95% CI [−0 . 14 12]% ) . Note that in this case too , the proportion of /ada/ responses for acoustic /aba/ inputs is increased compared with simulations run without any adaptation ( Figure 4A ) . Although the learning rate is constant , which means that recalibration magnitude does not necessarily decrease during the experiment , recalibration does not decay across trials . As a result , for trials between consecutive /ada/ percepts , stimuli experience a similar /ada/ category and the simulations do not lead to a significant difference in the classification of acoustic /aba/ whether preceded or not by a fused McGurk trial . We also tested an alternative update rule that was expected to better reflect how changes occur in the environment . We considered that they might occur hierarchically , with just two levels in a first approximation , corresponding to keeping ‘running averages’ over different time scales , enabling sensitivity to fast changes without erasing longer-lasting trends . We considered two sets of hierarchically related variables associated with a single category: θk , f ( fast ) and μk , f ( slow ) . The faster decaying one , θk , f , is driven by both sensory prediction error and the more slowly changing variable , μk , f ( more details can be found in Materials and Methods ) . This slowly changing and decaying variable , μk , f , keeps a representation based on a longer term ‘average’ over sensory evidence . In the limit , μk , f is constant; and this is what we consider here for illustrative purposes . Thus the update rules include an instant change in the fast variable due to the sensory prediction error in the perceived category plus a decay term toward the slower variable for every category . The instant change corresponds to the traditional update after an observation; the decay reflects the transient character of this update . The results in Figure 4D were run with the following parameters:Δθk , f=0 . 4 ( sf−θk ) p ( k\|sA , sV ) onlyfork=perceptΔθk , f=0 . 14 ( μk , f−θk , f ) forallkΔμk , f=0forallkwhere subscript f indexes the feature ( f = A for acoustic , f = V for visual ) . All categories decay toward the corresponding long-term stable values ( μk , A , μk , V ) in the inter-trial interval . By comparing the decay contribution Δθk , f = 0 . 14 ( μk , f - θk , f ) with the update expression for a quantity with decay time constant τ in an interval Δt ( here Δt = 5 s , the interval between consecutive trials ) Δθk , f = Δt/τ ( μk , f - θk , f ) , we can derive a rough estimate for the decay time constant τ ~5/0 . 14 s ~ 35 s . The percentage of acoustic /aba/categorized as ‘ada’ after control trials was 18% vs . 29% after fused McGurk ( Wilcoxon ranked-signed test , p=0 . 0004 ) , the median difference being 95% CI: [5 . 8 , 18 . 0]% . Therefore two effects can be observed; the overall increase in acoustic /aba/ categorized as /ada/ and the rapid recalibration effect reflected in the specific increase observed when acoustic /aba/ was preceded by a fused McGurk trial . We have modelled recalibration as the continuous updating of the model parameters that represent each of the speech categories used to guide perceptual decisions , in particular the expected values of sensory features associated with each category θk , V , θk , A ( k indexing the category ) . With this approach , the ideal adapter Bayesian account turned out to be incompatible with the experimental findings , due to the erroneous underlying assumption of a stable environment . Because the model assumes that all the sensory observations are derived from exactly the same non-changing distributions , past observations do not lose validity with time . As a result , the location estimate corresponds to the running average of the feature values in the stimuli that have been associated with each category in the course of the experiment . As the occurrence of a perceived category increases , the size of recalibration decreases , until categorization differences across successive trials are no longer observable ( Figure 5A ) . The ‘delta rule’ and the ‘hierarchical update with decay’ both involve a constant learning rate implying that the parameter changes following each perceptual decision do not decrease as the experiment progresses ( Figure 5B–C ) . Although both models were able to qualitatively reproduce the main result , namely that the rate of acoustic /aba/ categorized as ‘ada’ was higher immediately after a fused McGurk trial , the delta rule without decay did not provide a good fit . The ‘hierarchical update with decay’ provided the best explanation for the experimental results . Specifically , its advantage over the ‘delta rule’ is that the update decays across trials after a perceptual decision towards a less volatile representation of the category , providing an effective empirical prior ( Figure 5C ) . As our interest lies at the computational level , we have not tested other , non-Bayesian models of speech category learning and recalibration ( reviewed in Heald et al . , 2017 ) . Some existing models involve processes that are similar to ours; for example , non-Bayesian abstract models of new speech category learning use Gaussian distributions with parameters that are updated with delta rules similar to that of Equation 7 ( Vallabha et al . , 2007a; McMurray et al . , 2009 ) . On the other hand , connectionist models of speech category learning posit a first layer of units with a topographic representation of the sound feature space and a second layer representing individual speech categories . Learning or recalibration is modelled by changing the connection weights between the two layers . In this way , Mirman et al . ( 2006 ) , modelled the recalibration of established prelexical categories that arises when an ambiguous sound is disambiguated by the lexical context as in the Ganong effect – a sound between /g/ and /k/ that tends to be classified as /g/ when preceding ‘ift’ or as /k/ when preceding ‘iss’ ( Ganong , 1980 ) . This effect shares some similarities with the McGurk effect , although the latter is stronger as it changes the perception of a non-ambiguous sound . The benefit of having two timescales is also illustrated by a connectionist model of the acquisition of non-native speech sound categories in the presence of well-established native ones ( Vallabha and McClelland , 2007b ) . Interference between new and existing categories was avoided by positing a fast learning pathway applied to the novel categories , and a slower learning pathway to the native ones . Finally , connectionist models can also reproduce short-term effects such as perceptual bias and habituation ( Lancia and Winter , 2013 ) . These examples suggest that a connectionist model could provide a physiologically plausible instantiation of our abstract Bayesian model as long as one incorporates two pathways with two different timescales or a single pathway that uses metastable synapses ( Benna and Fusi , 2016 ) . Parallel learning systems working at different temporal scales have previously been proposed in relation to speech; one able to produce fast mappings and heavily relying on working memory , while the other relies on procedural learning structures that eventually results in effortless , implicit , associations ( Myers and Mesite , 2014; Zeithamova et al . , 2008; Maddox and Chandrasekaran , 2014 ) . Our proposal can theoretically be motivated on similar grounds . We argue that the brain implements at least two representations of natural categories; one more flexible than the other . The more flexible one might be used to achieve the agent’s current goal , while the more stable and less precise representation keeps general knowledge about sound categories . We propose the term ‘working’ representation for the more flexible sound category representation , to distinguish it from its more stable ‘episodic’ form . Behaviourally , this can be advantageous when specific instances of a unique category , for example from a single speaker , have less associated uncertainty than the overall prior distribution across all possible instances across speakers . The ‘working representation’ corresponds to an ‘intermediate’ representation that has lower uncertainty and therefore makes the sensory integration process more precise , leading to more confident perceptual decisions at the single trial level . This strategy allows the agent to use a precise ‘working’ category that can quickly change from trial to trial . In this view , the agent needs to infer the distribution ( mean and covariance ) that defines the working representation , and combine sensory evidence with the prior , that is , with the corresponding long-term ‘episodic’ representation . Based on these Bayesian principles we wrote the hierarchical recalibration rule ( Equation 8 ) , which appears whenever there are three quantities informing the current estimate of a variable ( under Gaussian conditions ) . In our model , the current expected value for the working representation is informed by the value derived from the observation in the previous and current trials , and from the episodic representation as schematized in the right panel of Figure 6 . Bayesian inference assuming known volatilities for the two levels in Figure 6 , and under the mean-field approximation , can be calculated analytically resulting in update equations that take the form written at the bottom of the diagram . The coefficients , which determine what is referred to as ‘learning rates’ in the reinforcement learning literature , are functions of the parameters of the model related to the different sources of uncertainty: volatilities at both hierarchical levels , sensory noise and the width of the episodic representation . Our proposed update equation therefore assumes that agents have already estimated the volatilities at the two levels ( as in Behrens et al . , 2007; Nassar et al . , 2010; Mathys et al . , 2014 ) . Finally , from the optimal agent’s perspective , the internal model used for a given trial is the predictive working representation built from 1 ) updated representations after the last observation and 2 ) their expected change in the intervening time . The latter component denotes the uncertainty associated with the representations , for example more volatile representations becoming more uncertain more quickly . This last point is important as it means that , across trials , increased uncertainty associated with the previous estimate of the working representation implies more reliance on the long-term representation . In other words , across trials , the expected sound feature modulation encoded by the working representation ‘decays’ back towards that of the long-term representation . This reflects a form of ‘optimal forgetting’ , that is , the expected loss of relevance of a past observation for the current trial . Whether the two time scales that were needed to explain simultaneous tracking of long and short-term category representations are hierarchically organized or implemented in parallel , and what brain regions or mechanisms might be implicated is an open question . There are at least three scenarios that could support recalibration at different time scales . First , a hierarchy of time scales might exist at the single synapse level . Hierarchically related variables with increasing time scales ( dynamics described by equations as in Figure 6B ) have been used in a modelling study to increase the capacity of memory systems and improve the stability of synaptic modifications ( Benna and Fusi , 2016 ) , and a model of synapses with a cascade of metastable states with increasing stability was able to learn more flexibly under uncertainty ( Iigaya , 2016 ) . In the latter model , the intrinsic decay of synaptic modifications was faster for the more labile memory states , i . e . those that are more sensitive to new evidence . In contrast , the deeper , more stable memory states , showed slower decay , which overall nicely concurs with our proposal . It is thereforepossible that the representations are encoded at synapses with the transient recalibration corresponding to synaptic modification at the more labile states and the long-term component residing in the less labile states . A second option involves prefrontal cortex working in tandem with other brain regions . In perceptual classification tasks under volatile conditions , prefrontal cortex can flexibly combine alternative strategies , such as optimal Bayesian-like learning in stable environments and a working memory model in volatile environments ( Summerfield et al . , 2011 ) . In our setting , to guide the speech classification process , it could conceivably combine a ‘working’ short time scale representation of the speech categories with a long-term ‘episodic’ representation , which might reside in different brain networks . Several fronto-parietal regions have indeed been implicated in controlling the effect of sensory and choice history on perceptual decisions: ‘Sensory evidence , choice and outcome’ could be decoded from ventrolateral prefrontal cortex and predicted choice biases ( Tsunada et al . , 2019 ) . Neuronal responses in fronto-parietal circuits could provide a basis for flexible timescales ( Scott et al . , 2017 ) , as dissociated effects of working memory and past sensory history have been found to involve the prefrontal cortex and posterior parietal cortices respectively ( Akrami et al . , 2018 ) . The observed sensitivity to sensory choice history and sensory evidence is consistent with our model , which uses internal category representations to interpret sensory evidence , with category representations being recalibrated based on choice ( i . e . , the perceived category ) . Finally , the hierarchical nature of perception and action ( Kiebel et al . , 2008; Friston , 2008 ) might be paralleled in the brain , by hierarchical processing in prefrontal cortex ( Badre , 2008; Koechlin and Jubault , 2006; Summerfield et al . , 2006 ) and sensory areas ( Felleman and Van Essen , 1991; Chevillet et al . , 2011 ) . It is hence conceivable that the relation between the two timescales is hierarchical with higher-level representations becoming increasingly abstract and time insensitive . This could happen , for example , if the brain used representations at a speaker level that are drawn from more general representations of speech categories at the population level . Recalibration might work at every level of the temporal hierarchy , with higher levels integrating update information within increasingly longer time windows ( longer timescales ) , making them less and less sensitive to new observations . While the ‘ideal adapter’ account focused on cumulative recalibration ( Kleinschmidt and Jaeger , 2015 ) , our results suggest that shorter-lived effects are also behaviorally relevant . The ideal adapter could be formalized as a simple incremental optimal Bayesian inference in a non-volatile environment ( Figure 6 , left panel ) , whereas our update rule could be cast in a normative framework that explicitly accounts for environmental volatility . A hierarchical model with constant volatility at two levels ( Figure 6 , right panel ) could lead to hierarchical update equations ( Wilson et al . , 2013 ) that can be approximated by constant learning rates ( with higher learning rates - faster ‘forgetting’- being related to stronger volatility ) . The right panel of Figure 6 assumes that the higher level ( μ ) has lower volatility than the intermediate level ( θ ) , hence combining volatility with hierarchy . This combination departs from models used to explain decision making in changing environments ( Behrens et al . , 2007; Summerfield et al . , 2011; Mathys et al . , 2014 ) , which are not hierarchical , and focus on the nontrivial task of inferring the environment volatility . These studies show that human participants adapt their learning rates to the changing volatility , which could be modelled without keeping representations across several time scales . In these tasks , participants need to keep track of short-lived changes in arbitrary cue-reward associations or in arbitrarily defined sensory categories ( Summerfield et al . , 2011 ) , whereas we model overlearned and behaviourally relevant categories , which also requires to maintain long-term estimates as empirical priors . Our modelling results are relevant to continuous speech processing , in particular to account for auditory processing anomalies in dyslexia . Evidence from a two-tone frequency discrimination task suggests that participants’ choices are driven not only by the tones presented at a given trial , but also by the recent history of tone frequencies in the experiment , with recent tones having more weight than earlier ones ( Jaffe-Dax et al . , 2017 ) . It turns out that , when compared with controls , subjects with dyslexia show a decreased reliance on temporally distant tones , suggesting a shorter time constant ( Jaffe-Dax et al . , 2017 ) . Translating this result to the current model , we could hypothesize that in dyslexia the long-term component ( μ in Figure 6B ) has either a shorter time span , or is coupled to the lower representation with a lower weight . In both cases , we would expect a deficit in building long-term stable speech category representations since they would be overly driven by the current context . ASD individuals on the other hand , are optimally biased by long-term tones , but do not show the bias by short-term tones of neurotypical participants ( Lieder et al . , 2019 ) , which suggests a faster decay or an absence of the short-term component in Figure 6B . This would predict a failure of ASD individuals to show the specific effect after McGurk trials in the experiment simulated here . We present a revised ‘ideal adapter’ model for speech sound recalibration that has both transient and cumulative components organized hierarchically . This new model provides evidence for a hierarchy of processes in the recalibration of speech categories , and highlights that after experiencing the McGurk effect , it is not the acoustic features related to the sensory input that are modified , but higher-level syllabic representations . The model implies that the activity changes in sensory cortices are not locally generated but reflect the interaction of bottom-up peripheral sensory inputs and top-down expectations from regions where categorical perception takes place . Considering natural speech processing as the inversion of a continuously monitored and recalibrated internal model can unveil the potential operations and strategies that listeners use when they are confronted with the acoustic volatility associated with speech categories , which by their nature have both rapidly changing ( e . g . speaker specific ) and slowly changing ( e . g . speaker general ) components . Such a model can be implemented by a hierarchy of empirical priors that are subject to changes at different time scales . Although developed in the context of speech processing , our proposal may also apply to other cognitive domains requiring perhaps more nested timescales , such as action planning ( Badre , 2008; Koechlin and Summerfield , 2007; Koechlin et al . , 2003 ) . The goal of inference is to establish which is the speech token that gave rise to the incoming sensory input . We restrict ourselves to three possible tokens: /aba/ , /ada/ and /aga/ ( as in Lüttke et al . , 2016a ) . Although several acoustic and visual features can distinguish between them , we choose to model the 2nd formant transition , which is minimal for /aba/ but increases for /ada/ and /aga/ , and the degree of lip closure , which is maximal for /aba/ and less prominent for /ada/ and /aga/ ( lip closure /aba/>/ada/>/aga/ ) . This choice is based on the fact that what distinguishes between the three speech sounds is the place of articulation . Acoustically the 2nd formant transition is an important cue for place of articulation , particularly within the ‘a’ vowel context ( Liberman et al . , 1957 ) ; visually it is the degree of lip aperture at the time of the vocal cavity occlusion depending on its location ( complete lip closure for the bilabial ( /aba/ ) , and decreasing lip closure for the alveolar ( /ada/ ) and velar ( /aga/ ) ( Campbell , 2008; Varnet et al . , 2013 ) . The generative model has three levels; the higher level encodes the speech token , the speech token in turn determines the expected values for the audiovisual cues , as represented in Figure 1A . The model includes the three possible tokens , each determining the expected distribution of its associated audiovisual features . We also introduce sensory noise to account for sensory variability . The parameters , location and spread of features associated with each token , as well as the parameter associated with the level of sensory noise , define an individual listener’s internal model . That is , the listener models both the variability due to different articulations of the same speech category as well as the variability due to noise in the sensory system . We use ‘k’ as the speech token index k = {/aba/ , /ada/ , /aga/} , ‘f’ as feature index f = {V , A} , where ‘V’ stands for the visual feature ( lip aperture ) and ‘A’ for the acoustic feature ( 2nd formant transition ) . The idea is that the amplitudes of the lip aperture and 2nd formant modulations vary according to the identity of the speech token ( ‘k’ ) . ‘CV’ or ‘CA’ denote hidden states associated with these amplitudes . Finally , ‘sV’ or ‘sA’ stand for the actual features in the audiovisual sensory input . The internal generative model considers ‘sV’ and ‘sA’ the versions of ‘CV’ or ‘CA’ corrupted by sensory noise ( σV , σA ) . There are two sources of variability , one related to sensory noise ( σV , σA ) and the variability of modulation amplitudes across different articulations of the same speech token ( σk , V , σk , A ) , k = {/aba/ , /ada/ , /aga/} . The hierarchical generative model is defined by the following relations: ( 3 ) p ( sf\|Cf ) ∝exp ( − ( sf−Cf ) 22σf2 ) , f=A , V ( 4 ) p ( Cf\|k ) ∝exp ( − ( Cf−θk , f ) 22σk , f2 ) , f=A , V , k=speechtoken While the above defines the generative ( top-down model ) p ( sf\|Cf , k ) , our interest lies in its inversion p ( k\|sA , sV ) , where sA , sV represents the sensory input in a single trial . From the inversion of the model defined by the relations above one obtains: ( 5 ) p ( k\|sA , sV ) ∝exp ( − ( sV−θk , V ) 22 ( σV2+σk , V2 ) − ( sA−θk , A ) 22 ( σA2+σk , A2 ) ) p ( k ) which results from marginalizing over ‘CV’ and ‘CA’-the intermediate stages that encode the visual and acoustic features , explicitly:p ( k\|sA , sV ) =∫dCVdCAp ( k , CV , CA\|sA , sV ) =∫dCVdCAp ( sV\|CV ) p ( CV\|k ) p ( sA\|CA ) p ( CA\|k ) p ( k ) ∝exp ( − ( sV−θk , V ) 22 ( σV2+σk , V2 ) − ( sA−θk , A ) 22 ( σA2+σk , A2 ) ) p ( k ) ∫dCV1σk , Vexp⁡ ( − ( CV−CV , k ) 22σV , k2 ) ∫dCA1σk , Aexp⁡ ( − ( CA−CA , k ) 22σA , k2 ) Cf , kσf , k2≡sfσf2+θk , fσk , f2 , 1σf , k2≡1σf2+1σk , f2 , f=V , A We assume that initial variance and prior probabilities are equal across categories ( p ( k ) =1/3 ) . Alternatively , marginalization over ‘k’ gives the probabilities over the hidden variables ‘CV’ and ‘CA’ , which we associate with encoding of stimulus features ( lip aperture , 2nd formant transition ) in visual and auditory cortex respectively . p ( CV , CA\|sV , sA ) =∑kp ( k\|sV , sA ) 12πσV , kexp⁡ ( − ( CV−CV , k ) 22σV , k2 ) 12πσA , kexp⁡ ( − ( CA−CA , k ) 22σA , k2 ) Cf , kσf , k2≡sfσf2+θk , fσk , f2 , 1σf , k2≡1σf2+1σk , f2 , f=V , A This shows explicitly how internal estimates of sensory features ( lip aperture ‘CV’ and 2nd formant ‘CA’ ) are driven by bottom up sensory evidence ( sV , sA ) and top-down expectations related with each category ‘k’ contributing according to its internal expectations ( θk , V θk , A ) . When there is strong evidence for a given category ‘k’ , the sum can be approximated by a single Gaussian centered at a compromise between θk , V and sV for the visual feature and between θk , A and sA for the acoustic feature . In principle the model could also be made to perform causal inference ( Magnotti and Beauchamp , 2017 ) , that is , decide whether the two sensory streams belong to the same source and therefore should be integrated , or whether the two sensory streams do not belong to the same source , in which case the participant should ignore the visual stream . Since Lüttke et al . explicitly selected the participants that consistently reported /ada/ for the McGurk stimulus , these subjects were fusing the two streams . We hence assume that integration is happening at every audio-visual trial . The model’s percept corresponds to the category that maximizes the posterior distribution: p ( k\|sA , sV ) . The previous section presented how the model does inference in a single trial . We now turn to how the model updates the parameters that encode the internal representation of the three speech categories . This happens after every trial , thus simulating an internal model that continuously tries to minimize the difference between its predictions and the actual observations; we assume that in this process , in which the model tries to make itself more consistent with the input just received , it will only update the category corresponding to its choice . We will present three updating rules . The normative incremental Bayesian update model used by Kleinschmidt and Jaeger ( 2015 ) , and the empirically motivated constant delta rule and hierarchical delta rule with intrinsic decay . The internal representation of the speech categories is determined by six location parameters ( θk , V θk , A ) and six width parameters ( σk , V , σk , A ) . We follow Gelman et al . ( 2003 ) and define the following prior distributions for the internal model parameters ( θk , V , σk , V; θk , A , σk , A ) . For each of the six ( θ , σ ) pairs ( 2 sensory features × 3 categories ) the prior is written as:p ( θk , f , σk , f ) =p ( θk , f\|σk , f ) p ( σk , f ) ∝σk , f2 ( σk , f2 ) − ( vk , f , 0/2+1 ) exp⁡ ( −[κk , f , 0 ( θk , f−μk , f , 0 ) 2+vk , f , 0σk , f , 02]2σk , f2 ) As above , f refers to the sensory feature , either V or A , and k to the speech category , either /aba/ , /ada/ or /aga/ . After a new trial with sensory input ( sV , sA ) the updated prior has the following parameters:κk , f , 0←κk , f , 0+1μk , f , 0←μk , f , 0+1κk , f , 0+1 ( sf−μk , f , 0 ) vk , f , 0←vk , f , 0+1vk , f , 0σk , f , 02←vk , f , 0σk , f , 02+κk , f , 0κk , f , 0+1 ( sf−μk , f , 0 ) 2 After each trial the inference process described in the previous section determines the percept from the posterior probability p ( k\|sV sA ) . Only the feature parameters of the representation corresponding to the percept are subsequently updated . We use the values that maximize the posterior over the parameters given the input and the current estimated category ‘k’ to determine the point estimates that will define the updated model parameters for the next trial ( Equation 6 ) . The updates for the location and spread parameters then take the form: ( 6 ) Δθk , f=1κk , f , 0+n ( k ) ( sf−θk , f ) Δσk , f2=1vk , f , 0+n ( k ) +1[κk , f , 0+n ( k ) −1κk , f , 0+n ( k ) ( sf−θk , f ) 2−σk , f2] Where ‘k’ is the perceived category , n ( k ) the number of times the category has been perceived , f = V , A designates the sensory feature and νk , f , 0 and κk , f , 0 are parameters from the prior distribution . The larger νk , f , 0 and κk , f , 0 are , the more k ‘perceptions’ it takes for the parameter values of category ‘k’ to plateau but also the smaller the updates after each trial . The above update equations implicitly assume that the environment is stable and therefore updated parameters keep information from all previous trials . This is the result of the generative model , which did not include a model for environmental parameter changes . Introducing expectations about environmental changes led us to consider rules with constant learning rates . We restrict ourselves here to updates for the six location parameters ( θk , V θk , A ) of Equation 6 . We first considered a constant delta rule scaled by the evidence in favor of the selected category p ( k\|sV , sA ) , ( 7 ) Δθk , f=A ( sf−θk , f ) p ( k\|sV , sA ) As in the Bayesian case , updates accumulate without decay between trials . The main difference is that θk , f is driven more strongly by recent evidence than by past evidence , implicitly acknowledging the presence of volatility . Finally we consider updates that decay with time . We reasoned that the decay should be towards parameter estimates that are more stable , which we denote by μk , f . We propose a hierarchical relation , with updates in μk , f being driven by θk , f , while updates in θk , f are driven by sensory evidence . At each trial all categories ( k’ ) decay toward their long-term estimates and only the perceived category ( k ) updates both μk , f and θk , f: ( 8 ) Δθk′ , f=D ( μk′ , f−θk′ , f ) Δθk , f=R1 ( sf−θk , f ) p ( k\|sV , sA ) Δμk , f=R2 ( θk , f−μk , f ) p ( k\|sV , sA ) The first equation reflects the decay while the last two equations apply to the perceived category ‘k’ . ‘D’ , ‘R1’ and ‘R2’ are constant parameters . R2 was set to zero since we do not expect the long-term component to change significantly within the experimental session . We simulate the experimental paradigm in Lüttke et al . ( 2016a ) , in which human participants were asked about what they heard when presented with auditory syllables or auditory syllables accompanied with a video of the corresponding speaker’s lip movements . There were six stimulus types: three acoustic only stimuli: /aba/ , /ada/ and /aga/ and three audiovisual stimuli , congruent /aba/ , congruent /ada/ and McGurk stimuli , that is , acoustic /aba/ accompanied by the video of a speaker articulating /aga/ . Each stimulus type was presented 69 times to each participant . In the original experiment three different realizations of each of the six types were used . In our simulations we use a single realization per stimulus that is corrupted by sensory noise . As described above , our model proposes that syllables are encoded in terms of the expected amplitudes and variances of audiovisual features . The expected amplitudes were taken from the mean values across 10 productions from a single male speaker ( Olasagasti et al . , 2015 ) , the amplitudes were then normalized by dividing by the highest value for each feature resulting in the values: ( θ/aba/ , A = 0 . 1 , θ/ada/ , A = 0 . 4 , θ/aga/ , A = 1 ) , ( θ/aba/ , V = 1 , θ/ada/ , V = 0 . 6 , θ/aga/ , V = 0 . 37 ) . For the other parameters defining the perceptual model , variances and sensory noise levels , we explored 20 different combinations . Five possible values for the pair ( σV σA ) : ( 0 . 1 , 0 . 1 ) , ( 0 . 12 , 0 . 12 ) , ( 0 . 12 , 0 . 15 ) , ( 0 . 15 , 0 . 12 ) , ( 0 . 15 , 0 . 15 ) . For each , we used four possible ( σk , A σk , V ) pairs: ( 0 . 1 , 0 . 1 ) , ( 0 . 1 , 0 . 2 ) , ( 0 . 2 , 0 . 1 ) and ( 0 . 2 , 0 . 2 ) . Parameters outside this range typically led to categorization accuracy worse than that from the participants in Lüttke et al . ( 2016a ) . For each of the 20 parameter sets defining the perceptual model , we tested a set of values for the parameters that define each recalibration model . Standard Bayes has two free parameters per category and feature: κk , f , 0 and νk , f , 0 ( Equation 6 ) . We tested the same values for all categories and features and therefore we drop the ‘k’ ( category ) and ‘f’ ( feature ) subscripts; κk , f , 0 = κ0 , νk , f , 0 = ν0 . ( κ0 , ν0 ) = ( 1 , 5 , 10 ) ⊗ ( 1 , 5 , 10 ) ( where ⊗ denotes the tensor product ) . For the constant delta rule , there is a single parameter per category and feature; we use the same for all categories but tested different values for different features . The learning rates for the visual feature ( RV ) and acoustic feature RA tested were ( Equation 7 ) ( RV , RA ) = { ( 0 . 05 , 0 . 1 ) ⊗ ( 0 . 02 , 0 . 04 , 0 . 06 , 0 . 08 ) , ( 0 . 1 , 0 . 2 ) ⊗ ( 0 . 1 , 0 . 2 , 0 . 3 , 0 . 4 , 0 . 5 , 0 . 6 ) , ( 0 . 4 , 0 . 5 , 0 . 6 ) ⊗ ( 0 . 4 , 0 . 5 , 0 . 6 ) , ( 0 . 7 , 0 . 8 ) ⊗ ( 0 . 7 , 0 . 8 ) } . In this case , we tested different values for visual and acoustic to increase the chances of the delta rule to reproduce the experimental results . The hierarchical delta rule with decay has three parameters per category and feature ( Equation 8 ) . For all the simulation we set R2 , V and R2 , A = 0 . For the other two parameters D and R1 ( common across categories and features ) we tested 35 pairs: ( R1 , D ) = ( 0 . 05 , 0 . 1 , 0 . 12 , 0 . 14 , 0 . 16 , 0 . 18 , 0 . 2 ) ⊗ ( 0 . 05 , 0 . 1 , 0 . 2 , 0 . 3 , 0 . 4 ) . Although we did not perform an exhaustive exploration of parameter space , the ranges tested were determined by the informal observation that parameters outside the ranges tested led to excessive recalibration—too many /aba/s categorized as ‘ada’ overall . The stimuli presented to the modelled participant corresponded to the expected acoustic and visual features of their internal model . Thus if the internal model for /aba/ is centered at θ/aba/ , A for the acoustic feature and at θ/aba/ , V for the visual feature , those are the amplitudes chosen for the input stimuli . In other words , stimuli were tailored to the modelled participant . It is worth emphasizing that the modelled agent does not have a fused ‘McGurk’ category; their model only includes congruent expectations . The six stimuli were defined by: Even if the underlying parameters for a given stimulus type were the same for every trial , sensory noise created variability . The input to the model was the pair sV , sA defined by:sV=CV+σVηVsA=CA+σAηAwhere ηV and ηA are sampled from independent Gaussian distributions with zero mean and unit variance . That is , sV sA are noisy versions of the true amplitude modulations in the visual and auditory modality . The six stimulus types were presented to the model in random order with 69 repetitions for each . At the end of the presentation , the model chose a percept based on the posterior distribution over syllable identity ‘k’ given the stimulus . The perceived syllable was then recalibrated by updating its defining parameters ( either both mean and variance or mean alone depending on the specific update rule ) . In the model recalibration step , sometimes θ/ada/ , A became smaller than θ/aba/ , A . This happened mostly for the constant delta rule as we increased the learning rate parameter , which also lead to increases in the McGurk contrast . Despite this modification , the observed McGurk contrast for the constant delta rule was not statistically significant . θ/ada/ , A becoming smaller than θ/aba/ , A constitutes a reversal of the initial relation between these parameters; empirically one finds that 2nd formant modulation is larger for /ada/ than for /aba/ ( θ/ada/ , A > θ/aba/ , A ) . We included a line in our code that made sure that this did not occur . If after recalibration θ/ada/ , A was smaller than θ/aba/ , A , the two were interchanged . This can be interpreted as a prior that incorporates information about the relations between categories . If reversals were accepted , subsequent acoustic /aba/ would be systematically classified as /ada/ . To evaluate the performance of the models , we used the following data from the original experiment . 1 ) The McGurk contrast , defined by two values: pada , Mc the proportion of acoustic only /aba/ categorized as ‘ada’ when preceded by a fused McGurk; ( 29% ) or by other stimuli pada , oth , acoustic /aba/ and /aga/ and congruent /aba/ and /aga/ , ( 16% ) . 2 ) Overall performance: the proportion of the most frequent category for each of the six stimulus types: 80% of ‘aba’ percepts for acoustic only /aba/ , 83% of ‘ada’ percepts for acoustic only /ada/ , 98% of ‘aga’ percepts for acoustic only /aga/; 98% of ‘aba’ percepts for congruent audiovisual /aba/ , 87% of ‘ada’ percepts for incongruent McGurk ( acoustic /aba/ and visual /aga/ ) , and 98% of ‘aga’ percepts for congruent /aga/ . We will represent these values as the six entries of the vector cstim . While the original experiment had 27 participants , we run the experiment 100 times . By drawing 6000 random samples of 27 from the 100 runs , we estimated appropriate sampling distributions . For the quantities of interest listed in the previous paragraph we calculated the medians over the 6000 samples . In a first step , for each update rule , we selected the model with the parameters that lead to the minimum mean squared error for the McGurk contrast 2∆2Mc = ( pada , Mc - 29 ) 2 + ( pada , oth - 16 ) 2 . The coefficients in the update rules appearing in the Results section correspond to those that lead to the minimum ∆Mc for each update rule . In a second step , we concentrated on the model with the best parameters for each update rule . The 6000 random samples of size 27 were used to build a sampling distribution for ∆Mc and a measure of overall performance also based on a mean squared error: 6∆2overall = ( cAb - 80 ) 2 + ( cAd - 83 ) 2 + ( cAg - 98 ) 2 + ( cVbAb - 98 ) 2 + ( cMcGurk - 87 ) 2 + ( cVgAg - 98 ) 2 . Additionaly we calculated the 95% confidence intervals for the size of the McGurk contrast ( the difference pada , Mc - pada , oth ) . We choose as representative for the 6000 samples , the sample with the median value for 6∆2overall+2∆2Mc . For each of the 6000 random samples we tested whether the McGurk contrast paired values pada , Mc and pada , oth were significantly different ( Wilcoxon signed rank test ) . In the results section we report the 95% confidence intervals for the difference pada , Mc - pada , oth , as well as the Wilcoxon signed rank test for the sample with the median value for 6∆2overall+2∆2Mc , ( the representative sample from the 6000 ) . Figure 4 , includes the control or ‘/ada/ contrast’ for the representative sample . This contrast is defined by the proportion of acoustic only /aba/ sounds categorized as ‘ada’ when preceded by an acoustic only /ada/ correctly categorized as ‘ada’ or by other stimuli ( acoustic only /aba/ or /aga/ ) . All simulations and statistical tests were performed using custom scripts written in MATLAB ( Release R2014b , The MathWorks , Inc , Natick , Massachusetts , United States ) . The original MATLAB scripts used to run the simulations are available online ( https://gitlab . unige . ch/Miren . Olasagasti/recalibration-of-speech-categories; copy archived at https://github . com/elifesciences-publications/recalibration-of-speech-categories; Olasagasti , 2020 ) .	People can distinguish words or syllables even though they may sound different with every speaker . This striking ability reflects the fact that our brain is continually modifying the way we recognise and interpret the spoken word based on what we have heard before , by comparing past experience with the most recent one to update expectations . This phenomenon also occurs in the McGurk effect: an auditory illusion in which someone hears one syllable but sees a person saying another syllable and ends up perceiving a third distinct sound . Abstract models , which provide a functional rather than a mechanistic description of what the brain does , can test how humans use expectations and prior knowledge to interpret the information delivered by the senses at any given moment . Olasagasti and Giraud have now built an abstract model of how brains recalibrate perception of natural speech sounds . By fitting the model with existing experimental data using the McGurk effect , the results suggest that , rather than using a single sound representation that is adjusted with each sensory experience , the brain recalibrates sounds at two different timescales . Over and above slow “procedural” learning , the findings show that there is also rapid recalibration of how different sounds are interpreted . This working representation of speech enables adaptation to changing or noisy environments and illustrates that the process is far more dynamic and flexible than previously thought .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Integrating prediction errors at two time scales permits rapid recalibration of speech sound categories
Maintaining a healthy body weight requires an exquisite balance between energy intake and energy expenditure . To understand the genetic and environmental factors that contribute to the regulation of body weight , an important first step is to establish the normal range of metabolic values and primary sources contributing to variability . Energy metabolism is measured by powerful and sensitive indirect calorimetry devices . Analysis of nearly 10 , 000 wild-type mice from two large-scale experiments revealed that the largest variation in energy expenditure is due to body composition , ambient temperature , and institutional site of experimentation . We also analyze variation in 2329 knockout strains and establish a reference for the magnitude of metabolic changes . Based on these findings , we provide suggestions for how best to design and conduct energy balance experiments in rodents . These recommendations will move us closer to the goal of a centralized physiological repository to foster transparency , rigor and reproducibility in metabolic physiology experimentation . Mice are an instructive tool for the study of human metabolism as they can mirror human physiology in their responses to age-related and diet-induced obesity , and their physiological compensations to resist weight loss ( Speakman et al . , 2007 ) . The study of genetic and environmental factors influencing energy balance using laboratory animals has been advanced by the introduction of indirect calorimetry systems ( Even and Nadkarni , 2012 ) . Indirect calorimeters measure metabolic rates using gas sensors to capture rates of change in O2 consumption and CO2 production within an open flow system . Physical activity is monitored by recording infrared beam breaks or using electromagnetic receivers . Food intake is measured using sensitive mass balances . Weight gain results when food intake outpaces metabolic rate . Inversely , when food intake falls below metabolic rate , weight loss ensues . Physiological constraints limit both unrestrained weight gain and weight loss . The ability to promote durable weight loss through increases in metabolic rate or decreases in food intake is a major therapeutic goal in the context of the modern obesogenic environment . It is important to be able to compare the physiological data across papers studying metabolism . However , comparing results from published indirect calorimetry studies has been hampered by inconsistent application of analytical techniques . In studies of obesity , metabolic rates of mice with different body compositions are sometimes subjected to inappropriate normalization attempts ( Arch et al . , 2006; Katch , 1972; Kleiber , 1932; Tanner , 1949 ) . Position papers decrying this situation have been routinely produced , and just as often ignored . A turning point occurred with the publication of a strongly worded commentary accusing authors of incorrectly representing their metabolic analyses ( Butler and Kozak , 2010 ) . A result of the commentary was a new appreciation of the complexity involved in handling the large amount of data resulting from indirect calorimetry experiments and the development of new tools ( Mina et al . , 2018 ) . Many groups have agreed that the optimal data treatment uses the ANCOVA , an ANOVA with body mass or body composition as a covariate ( Arch et al . , 2006; Kaiyala , 2020; Kaiyala , 2014; Kaiyala et al . , 2010; Kaiyala and Schwartz , 2011; Speakman et al . , 2013; Tschöp et al . , 2012 ) . However , the implementation of these recommendations has been uneven ( Fernández-Verdejo et al . , 2019 ) . To address this problem , we developed CalR , which facilitates investigators uploading and analyzing their calorimetry data by automating many of the routine steps of data curation , pre-defining statistical significance cutoffs , and allowing users to automatically perform appropriate statistical tests for interaction effects ( Mina et al . , 2018 ) . CalR is the first step toward standardization of indirect calorimetry analysis across different equipment platforms . However , we realized that the results provided by CalR included measures of statistical significance , but were lacking critical physiological context . To establish the normal range of metabolic rate under standardized experimental conditions , we analyzed two large independent datasets . For our primary training model , we utilized a dataset from the Mouse Metabolic Phenotyping Centers ( MMPCs ) representing longitudinal data from four sites in the United States where groups of male C57BL/6J mice were followed for 12 weeks on either a standard low-fat diet ( LFD ) or an obesogenic high-fat diet ( HFD ) . The results of our analysis from the MMPC experiment were applied to a larger secondary dataset from the International Mouse Phenotyping Consortium ( IMPC ) , an ambitious large-scale project with metabolic data from more than 30 , 000 mice , seeking to determine the genetic contributions to mammalian physiology . Here , we present the results of our analyses from these two datasets , and provide generalized recommendations to facilitate the creation of a centralized metabolic data repository . To understand the different components affecting mouse metabolism , cohorts of genetically identical mice were shipped to four independent US MMPCs and assessed longitudinally over 12 weeks in indirect calorimeters while on LFD or HFD . Weekly body masses were similar among institutions for mice on LFD but diverged significantly among mice on HFD ( Figure 1A and B ) . Regression plots of 24 hr average energy expenditure ( EE ) versus total body mass showed distinct site-specific metabolic rates which could be attributed to differences in body mass for mice on LFD or HFD ( Figure 1C and D ) . Mice at each site and overall had a positive association between mass and EE . This mass effect reflects the biological consequence of Newton’s 2nd law of motion , whereby movement of greater mass requires more energy , and by extension larger animals require more energy for both resting metabolism and to perform work under standard conditions ( Kleiber , 1932; White and Seymour , 2005 ) . For mice on LFD , animals at each institution formed distinct slopes and intercepts indicative of location-specific differences in housing temperature , experimental apparatus , microbiome , and other factors . For animals on 4 or 11 weeks of HFD , we observed identical slopes with different intercepts ( Figure 1D ) . This suggests that the relationship of EE to mass among genetically identical mice is similar despite absolute differences in EE at different locations . Despite an identical genetic background , a single source colony for all mice , as well as a single source of animal diet , an unexpectedly large weight variation was observed by site among the 30 mice randomly assigned to HFD . Body masses and compositions diverged appreciably , both among institutions and within each site’s colony . The mass variation was not observed for mice on LFD , suggesting strong effects driving obesity which are not encoded by genetic variation in C57BL/6 mice ( Figure 1E ) . We sought to quantify the relative contribution of the likely sources for the variation in EE using a widely adopted method ( Grömping , 2006 ) . The R2 of 72% reflects total explained variance and a good overall fit . This analysis reveals the largest source of variation in metabolic rate is body composition . Other sources include activity , time of day ( photoperiod ) , diet and acclimation ( Figure 1F ) . The institutional site of experimentation which can affect these factors accounts for 16 . 3% of the residual variation not explained by these biological factors ( Figure 1G ) . Rodents engage in a behavioral response when challenged with a novel environment ( Archer , 1975 ) . While there are no firm guidelines for how long mice may need to acclimate to indirect calorimeters , age , strain and diet can impinge on acclimation time . Here we defined the acclimation period as the first 18 hr or prior to the start of the first full photoperiod in the calorimeter . Analyses of the pre-acclimation and post-acclimation dark photoperiods were performed for each site . The hourly mean values for EE , energy intake , and respiratory exchange ratio ( RER ) from all four sites are plotted vs time . The strong effects of entrainment to 12 hr light/dark photoperiods produce differences at all sites for all parameters measured . The effects of acclimation were highly variable , differentially affecting mice at each location . At one of the four sites , EE increased following acclimation ( Figure 1—figure supplement 1A and B ) . Similarly , energy intake was increased at only one site in non-acclimated mice ( Figure 1—figure supplement 1C and D ) . The RER value , representing substrate oxidation , was significantly altered in three of four sites ( Figure 1—figure supplement 1E and F ) . In this study , locomotor activity was not significantly impacted by lack of acclimation ( Figure 1—figure supplement 1G and H ) . Overall , lack of acclimation adds a small but unpredictable level of noise and variation to the measurements tested . How HFD feeding contributes to obesity has been a topic of detailed investigation . Both male and female C57BL/6 mice typically gain weight on a diet high in fat ( Johnston et al . , 2007; West et al . , 1992 ) . The degree of weight gain however is highly variable and consequently there is considerable heterogeneity in the underlying mechanisms involving energy intake , metabolic rate , and absorption ( Kohsaka et al . , 2007; Lin et al . , 2000; Mayer and Yannoni , 1956; Storlien et al . , 1986; Yang et al . , 2014 ) . Here , we compare the calorimeter-determined energy intake and EE responses of WT male mice to HFD feeding at four different locations . After 11 weeks on LFD or HFD , as before , the EE was dependent on body mass ( slopes of kJ/hr vs gram body weight ) for all mice at all sites . However , HFD did not significantly alter this relationship except at one center ( Figure 2A ) . We fitted the LFD mouse data at each site to a model including body mass and activity to predict expected values for mice on HFD . The difference from the expected values , or residual values from this fit show an overall decrease in the EE of HFD mice relative to the predicted EE ( Figure 2B ) . We similarly examined how rates of energy intake changed with diet and found a surprising degree of variance among sites ( Figure 2C and D ) . Body weight is ultimately affected by long-term differences in energy balance ( energy intake minus EE ) . When calculating the difference between calories consumed and calories expended , as before , we observed a large variation in responses to HFD , including positive and negative changes in energy balance ( Figure 2E and F ) . However , energy intake was the primary driver of energy balance . These findings serve to underscore the large site-specific effects that contribute to variability of metabolism . To evaluate the contribution of body composition to the variable metabolic rates observed between sites , we used lean mass values as the covariate for our ANCOVA analysis ( Figure 2—figure supplement 1A–1C ) . In three of four sites , HFD altered the dependency of EE on lean mass . At one site , HFD altered the dependency of energy intake on lean mass . We also explicitly calculated the relative contribution of fat mass to EE using multiple linear regression and found an unexpectedly large range of fat mass contributions varying from +33% to −19% ( Figure 2—figure supplement 1D ) . When we examined the time course of the calorimetry data for all the centers , they demonstrated the characteristic circadian patterns of EE , activity and RER in animals on LFD during the 3-day period ( Figure 2G , L and H ) . Moreover , on HFD , EE was increased , and RER was decreased . Total energy intake and cumulative intake were increased by HFD , yet energy balance was unaltered by diet reflecting a neutral energy balance and weight stability ( Figure 2K ) while in the calorimeter . We also plotted change in energy balance vs weight change between the start and end of the first calorimetry experiment at week 0 ( Figure 2—figure supplement 2A ) . The normal circadian patterns seen in these mice , as well as the positive linear relationship between their energy balance and mass balance ( i . e . animals in positive energy balance tend to gain weight ) speak to the quality of the acquired data . To understand the relevance of the initial analysis , we applied this approach to the IMPC dataset ( Rozman et al . , 2018 ) . We similarly plotted the change in energy balance vs weight change between the start and end of the IMPC calorimetry experiments . The data appear to be of high quality as indicated by the expected positive linear relationship . Furthermore , mice in positive energy balance showed a higher RER consistent with carbohydrate oxidation , and mice with weight loss exhibit a lower RER consistent with oxidizing endogenous fat stores ( Figure 2—figure supplement 2B ) . The IMPC data demonstrate a pronounced mass effect , representing that bigger male and female animals expend more energy under standard housing conditions . The IMPC data include results of indirect calorimetry performed on 32 , 748 distinct animals including 9358 WT C57BL/6N mice at 10–11 weeks of age . 69% of these mice were male . The mean body mass of female mice is lower than that for male mice ( at 21 . 7 g and 27 . 4 g respectively; Figure 3A ) . In this analysis , there is a small but significant difference in the slope of the relationship between EE and body mass in WT male and female mice ( Figure 3B and Figure 3—source data 1 ) . We next sought to understand whether it is appropriate to compare mice of different weights . To do this , we examined whether the slope of the relationship between EE and body mass changes when comparing mice over a large range of masses ( Figure 3—figure supplement 1A ) . We grouped all male and female WT mice into quartiles by body mass ( small , 14 . 00–20 . 75 g; medium , 20 . 75–27 . 50 g; large , 27 . 50–34 . 25 g; largest , 34 . 25–41 . 00 g ) , and found that there were significant differences in the slope of the relationship between EE and body mass among the four groups . Somewhat surprisingly , largest mice are not significantly different from any of the other groups , which may be attributed to the fact that there are the fewest mice in the group with the largest masses . Our main finding shows the ‘large’ mice have a significantly shallower slope in the relationship between body mass and EE compared to both small and medium mice . This finding was similar to the relationship observed in the MMPC cohort on LFD and HFD ( Figure 2A ) . This is due to the differential contribution of lean and fat mass to EE which are not equally accumulated in mice with greater body weight . Indeed , when examining only lean body mass as the covariate , there were no significant differences between any of the groups by ANCOVA ( Figure 3—figure supplement 1B ) . This finding suggests that when comparing mice with more than 20% difference in body mass , that body composition data must be added to an ANCOVA model so as to not produce spuriously significant results . There is no indication that metabolic rate of female mice is more variable than that of male mice; a long-held assumption in metabolic research , despite recent contradictory evidence ( Prendergast et al . , 2014 ) . Studies on mice of both sexes were performed at seven IMPC institutions; three institutions examined only male mice ( Figure 3—figure supplement 2A and B ) . To determine the variability in metabolic rate , we fitted EE to the available data for each institution and both sexes . The R2 value of this fit represents the quality of the fit , with larger numbers representing a better fit , with lower unexplained variability . The R2 values varied considerably by site , likely due to six sites reporting locomotor activity values and only three sites reporting accurate ambient temperature data , while all sites reported body mass . Despite the great variation between sites , R2 values were similar between males and females at six of the sites ( +/- 7 . 3% difference ) . The remaining site had a female R2 of less than 0 . 001 , a poor fit , due to a small number of mice with little variation in mass , the sole predictor in this model , and therefore not an instructive example . The Canadian site reported mass , temperature and activity , and despite having the third smallest sample size had the best fit with R2female = 0 . 507 and R2male = 0 . 468 . Combining the data from both sexes at this site further improved the fit to 0 . 581 suggesting only minor differences in metabolic rate among female and male young , chow-fed WT mice . Clearly , the more accurate the covariate information reported , the better the explanatory value of the model . As the sample sizes of the two sexes are unequal in these sites ( 2889 females and 6469 males ) , we also examined the distribution of variance in male and female mice with a modified quantile-quantile ( Q-Q ) plot . Here , the slope of the blue and pink lines represents the theoretical standard deviation ( SD ) for the EE of each sex . Parallel lines indicate similar variability; while points not falling on the line represent variation from the normal distribution . The greater slope for male mice is indicative of greater variability in the data . Overall , this representation suggests that female mice have a qualitatively better fit and lower overall variation ( Figure 3—figure supplement 2C ) . Due to the significant differences in EE observed due to sex among all WT IMPC mice , for subsequent analysis of control and experimental mice , we present results only for male mice henceforth , but we find qualitatively similar metabolic results for WT male and female mice in the IMPC dataset when analyzed separately . The 10 sites represented in the IMPC data each reported between 240 and 1367 male WT mice ( Figure 3C ) . We observed more than a three-fold difference in metabolic rate slopes ( 1 . 18–5 . 32 × 10−2 kJ/hr vs gram body mass ) between sites despite standardized experimental protocols on mice of similar ages and similar diets ( Figure 3D and Figure 3—source data 1 ) . To further understand these differences , we examined other factors which influence metabolic rate including mass , temperature , season , and locomotor activity . The ambient temperature for mouse experiments varied by site . A range of ambient temperatures were reported for experiments performed by the sites in Canada , Germany , and Oxfordshire , UK . However , most sites report one single nominal temperature ( e . g . 21 . 0°C ) which likely does not accurately reflect daily variation in room temperature , nor the actual cage temperature , and may be a major source of unexplained variation ( Figure 3E ) . WT mice housed at low temperatures had the highest EE , and conversely mice housed at warmer temperatures had the lowest EE ( Figure 3F ) . We next examined the effect that time of year ( i . e . season ) might be having on metabolic rate . In the wild , mammals have seasonal differences in metabolic rate . There is evidence that despite the climate-controlled atmosphere of a vivarium , laboratory mice could similarly detect changes in season ( Drickamer , 1977 ) . All sites contributing data ( Canada , China , France , Germany , Japan , Korea , UK , and USA ) are located in the Northern Hemisphere and experience seasonal differences ( Figure 3C , inset ) . Within these data , there were no significant differences in EE by season , suggesting that seasonal variation was not a significant contributor to experimental variability ( Figure 3G ) . Differences in locomotor activity can contribute to whole-body EE , but typically have a negligible effect at non-thermoneutral temperatures ( Virtue et al . , 2012 ) . We find a strong positive correlation between ambulatory locomotor activity and EE in the six sites reporting movement data ( Canada , Oxfordshire UK , Japan , France , Korea , and Cambridgeshire UK ) ( Figure 3H ) . We examined the relative importance of these effects to predict rates of EE ( Figure 3I ) . As with the MMPC dataset , lean mass is one of the top predictors of variability , though temperature ( not included in the MMPC data ) contributes the highest to explaining EE rate . Unlike the MMPC dataset , more than 60% of the variation in the IMPC data was unaccounted for by examining only these reported variables , suggesting that our model is as yet incomplete . Only 9 . 6% of the residual variance can be attributed to unaccounted for institutional differences ( Figure 3J ) . We also examined the contribution of fat mass to overall EE and found a highly variable contribution from +55% to −69% among sites ( Figure 3K ) . The variability is likely due to their low adiposity as these animals are on chow diets . Overall , our analysis of 9358 WT mice identifies housing temperature , body composition , locomotor activity , and sex as contributing variables to EE . To test the applicability of our analysis to the IMPC dataset , we chose the data from the three sites that reported both body composition and ambient temperature to within 0 . 1°C: Toronto Centre for Phenogenomics ( Canada ) , Helmhotlz Munchen ( Germany ) , and MRC Harwell ( Oxfordshire , UK ) . We examined the contribution of total body mass to EE . Within this subset of 1996 male and 1153 female mice we re-examined sex as a mass dependent covariate . We find no significant difference between the EE rates of male and female mice ( Figure 4A ) . Using total body mass as a covariate , we observe nearly identical slopes for EE vs mass at each institution ( Figure 4B ) . When temperature effects were examined at these sites , mice maintained at warmer temperatures ( 22–29°C ) had significantly lower EE rates than mice maintained at colder temperatures ( Figure 4C and Figure 3—source data 1 ) . We find an improved R2 of 67% for this dataset of 3149 mice ( Figure 4D ) . This example accounts for only three sites , but now has accurate ambient temperature , mass , sex and season for all animals . These data do not account for locomotor activity . This greater fit has improved predictive powers over sites without accurate temperature recording . Here , a mere 0 . 078% of residual variation is explained by institution ( Figure 4E ) . After observing the large inter-site variation for WT mice , we next examined whether phenotypic differences due to genetic modulation could be observed reproducibly at multiple sites . In the IMPC dataset , there were six strains examined in at least four locations , Ap4e1-/- , Dbn1+/- , Dnase1l2-/- , Nxn+/- , and Prkab1-/- ( AMPKβ1-/- ) , and Rnf10-/- . We also compared results for the Cdkal1-/- gene from two IMPC sites with our results ( Massachusetts , USA ) . We plotted EE vs body mass for male mice at each of the sites to examine phenotypic variability ( Figure 5A ) . This representation plots each individual mouse at each site but does not convey the difference from the local WT populations and thus is challenging to make direct comparisons . To provide context for the magnitude of these phenotypes , mass , activity ( if available ) , and temperature were fit to a multiple linear regression model for WT male mice . We calculated the deviation from the model at each site and plotted these residual values ( Figure 5B ) . Four of these strains have not previously been published and thus the expected phenotype is not known . The only strain with consistent phenotypic findings was Cdkal1-/- , where EE values similar to WT were observed for mice at all three locations tested ( Palmer et al . , 2017 ) . Unidirectional phenotypes were observed for four strains . Significantly increased EE was observed in Ap4e1-/- , Dbn1+/- , and Rnf10-/- mice at four of nine sites , three of seven sites , and two of four sites , respectively . Nxn+/- mice had significantly lower EE at three of eight sites . Bidirectional changes were observed for the remaining two strains . Dnase1l2-/- mice were similar to WT controls at six sites , with two sites showing significantly altered EE , with one higher and one lower . Prkab1-/- mice were similar to controls at five sites , consistent with results from an independently generated strain of AMPKβ1 deficiency which found no significant EE phenotype ( Dzamko et al . , 2010 ) . Yet at two sites Prkab1-/- mice were statistically different , higher and lower than controls at one site each . Consistently , the largest residual values were from sites that did not report accurate temperature values or locomotor activity , pointing toward incomplete modeling from these locations . These results illustrate the large phenotypic variability among mice with identical genetic perturbations observed at different locations . Our analysis helps to quantify the magnitude by which site , temperature , mass , and sex contribute to metabolic rate in WT mice . However , genetic variations can also impact metabolic rate . One of the goals of the IMPC is to use mouse models to understand the contribution of genetic factors to phenotypic variation . The IMPC dataset contains the metabolic analysis of 2329 gene KO strains . Using the multiple linear regression model fitted from WT male mice , we examined the phenotype of all KO strains . To find the strains of KO mice with the greatest impact on metabolic rate we fitted the effects of mass and ambient temperature . After fitting the mean EE of each KO strain to the model , we plotted the unexplained metabolic effect , or EE residuals . The residuals are normally distributed around 0 for male mice ( Figure 6A ) suggesting the model is appropriately fitted ( Figure 6B ) . The results of this analysis are included in Figure 6—source data 1 . To validate this data treatment , we plotted the IMPC data for both EE vs total body mass ( Figure 6C ) and residual EE values from the regression model vs total body mass ( Figure 6D ) . Using the IMPC data , we plotted the results for 10 genotypes previously shown to affect body weight in mice . These strains affect weight by different mechanisms , including by increasing food intake , limiting metabolic rate , and affecting energy absorption . Food intake drives obesity in strains deficient for melanocortin 2 receptor accessory protein 2 , Mrap2-/- ( Asai et al . , 2013 ) , carboxypeptidase E , Cpe-/- ( Alsters et al . , 2015; Naggert et al . , 1995 ) , proprotein convertase 1 , Pcsk1+/- ( O'Rahilly et al . , 1995; Zhu et al . , 2002 ) , or growth differentiation factor 15 , Gdf15-/- ( Tsai et al . , 2013 ) . Low metabolic rate drives obesity in mice lacking growth hormone , Gh-/- ( Meyer et al . , 2004 ) . Relatedly , at standard room temperatures , mice lacking the thermogenic uncoupling protein 1 have reduced metabolic rate , but obesity is not observed in Ucp1-/- mice until housing under thermoneutral conditions ( Enerbäck et al . , 1997 ) . Mice lacking the biosynthetic enzyme for creatine , Gatm-/- , have also have a reduced metabolic rate , similar to findings in adipocyte specific knockout mice ( Kazak et al . , 2017 ) . We also plotted the metabolic rate of strains which promote leanness due to high metabolic rate , Pparg+/- , Fgfr4+/- , Fgfr4-/- , and Acer1-/- . Peroxisome proliferator activated receptor-γ , Pparg heterozygous mice have increased metabolic rate driven by increased locomotor activity , controlled by the central nervous system ( Lu et al . , 2011 ) . Elevated metabolic rate in mice with fibroblast growth factor receptor 4 , Fgfr4 knockout and antisense inhibition have elevated metabolic rate likely due to increased levels of FGF19 and FGF21 ( Ge et al . , 2014; Yu et al . , 2013 ) . Alkaline ceramidase 1 , Acer1-deficient mice have a skin barrier defect and progressive hair loss , likely accounting for compensatory increases in metabolic rate ( Liakath-Ali et al . , 2016 ) . The uniform phenotyping standards of the IMPC allow a comprehensive comparison of relative phenotypic magnitude . Institutional effects on EE can make the true magnitude of effects between control and experimental strains difficult to discern in regression plots of EE vs mass ( Figure 6C ) . Instead , when plotting against residual values , it is apparent that four of these genes have modest phenotypes residing within 1 SD of the predicted mean , Cpe-/- , Mrap2-/- , Gdf15-/- , and Ucp1-/- ( Figure 6D ) . While the modestly decreased EE may contribute to obesity in the Cpe-/- , and Mrap2-/- mice , food intake and other factors are likely the primary drivers as reported . There was no suggestion of altered EE in Gdf15-/- , consistent with emerging reports that Gdf15 induction is required for altered body weight phenotypes ( Coll et al . , 2020 ) . The −1 . 0 SD phenotype observed in the Pcks1+/- mouse has a large impact on EE to predispose to obesity ( O'Rahilly et al . , 1995; Zhu et al . , 2002 ) . Both Gh-/- and Gatm-/- groups have lower body weights than control mice , yet have lower than predicted EE as significant contributors to obesity in these strains . The hypermetabolic 2 SD and 3 SD effects of Fgfr4+/- , Fgfr4-/- , and Acer1-/- contribute to resistance to obesity , while the effects of Pparg heterozygosity are modest at just over 1 SD . The EE effects have been combined with other assays including skin barrier function or HFD challenge to explain these phenotypes . In concert with comprehensive phenotyping , the IMPC data have confirmed these previously published findings and demonstrated the robustness and utility of the IMPC experiment . We next examined the human genetic loci linked to increased risk for obesity and body weight related traits through genome wide association studies ( GWAS ) . The IMPC strains with a gene in a mapped obesity locus ( n = 42 ) with phenotypes +/- 1 SD from the mean ( n = 7 ) were plotted ( Figure 6E and F ) . We found four strains with lower metabolic rate that might predict obesity susceptibility , Pax5+/- ( Melka et al . , 2012 ) , Chst8-/- ( Tachmazidou et al . , 2017 ) , Tfap2b+/- ( Lindgren et al . , 2009 ) , and Pald1-/- ( Cotsapas et al . , 2009 ) . We also found three KO strains which might protect from obesity , Pepd-/- ( Shungin et al . , 2015 ) , Klf12+/- ( Jiang et al . , 2018 ) , and Pacs1+/- ( Wheeler et al . , 2013 ) . Further characterization of these strains , including exposure to HFD , may reveal insights into the phenotypic effects driving common forms of human obesity . The normal distribution of changes in EE which are not accounted for by site , mass , and temperature is relatively narrow , with 2/3 of all strains ( 1 SD ) having modest differences of +/- 0 . 161 kJ/hr . To place these values into a physiological context we compared the magnitude of examples of the effects of age , voluntary exercise , adrenergic activation , and temperature on metabolism in C57BL/6 mice ( Figure 7A ) . The effects of aging on basal metabolic rate , as reported by Houtkooper et al . ( 2011 ) were modest , with differences at 15 , 55 , and 94 weeks of age differing by less than 0 . 04 kJ/hr on average , all corresponding to less than 1 SD ( Figure 7B ) . In voluntary wheel running , O'Neal et al . , 2017 report a strong ( 3 SD ) induction of EE after one week , which decreases to a 2 SD effect at 2 and 3 weeks of exercise ( Figure 7C ) . In mice housed at thermoneutrality , administration of the β3 adrenergic agonist CL316 , 243 produces a > 3 SD effect over 3 hr ( Figure 7D ) . We also plotted metabolic rates at ambient temperatures varying from 6°C to 30°C ( Figure 7E ) . Using 22°C as the comparator , increasing the temperature just 3 degrees produced a 2 SD effect . Similarly , decreasing temperature 4 degrees has a nearly 3 SD effect . Greater temperature changes produced larger effects on EE except for the transition from 28°C to 30°C which likely reflected a departure from true thermoneutrality . These plots help to provide a frame of reference for the magnitude of effects observed in both the MMPC and IMPC datasets . The past decade has seen a transformation in our understanding of mammalian metabolism . New genetic tools , increased commitment to large-scale experimentation , and greater sharing of data are welcome developments . Large-scale systemic experiments such as the IMPC provide an unprecedented view into the genetic pathways that control energy balance in mammals . Yet , despite information on thousands of mice being deposited annually into the IMPC , data from far greater numbers of experimental animals are produced at institutions worldwide . These smaller datasets remain siloed and unshared at great loss to the scientific community . A critical unmet need exists to create centralized data repositories on metabolism if we are to make sense of often conflicting information on the drivers of obesity . Differences in observed phenotypes may well be due to environmental variances including temperature and diet . Preclinical studies in model organisms should pave the way forward for greater understanding of the role of genetic variation , diet , exercise , macronutrient composition , age , and obesity on human metabolism . A metabolism data repository is the obvious next step given these advances . The necessary prerequisites to achieve this reality are: 1 ) standardization of experimentation and analysis , 2 ) standardization of data formats , and 3 ) consensus approaches to essential metadata collection . The results of experiments in mice vary by strain , age , diet , temperature , site , intervention protocols , and other factors , and there can be useful covariates in explaining variance among experiments . The absence of this information can turn an attempt at experimental replication into part of an ongoing replication crisis in biomedical research . The goal of the present study was to understand the critical variables that would be necessary to make in vivo studies of metabolic rate broadly comparable . The largest source of variation observed in the small-scale pilot MMPC experiment was mass , including body composition data . Locomotor activity , photoperiod , diet and acclimation were smaller contributors . In the IMPC dataset , temperature and mass were the largest contributors to variation . The lack of accurate temperature information reduced the utility of the IMPC data at many of the sites where even reporting average daily temperature would be beneficial . Other details including the type of indirect calorimeter or body composition method ( e . g . DEXA or MRI ) would also help interpretation . We recommend that in future publications , the minimal necessary information to compare and interpret metabolic studies ( Table 1 ) should be included . At the outset of this project , we envisioned determining the range of metabolic rates observed in WT mice on a LFD or HFD . We predicted a high level of concordance between sites due to the experience , strong reputation for excellence , and know-how in metabolic phenotyping at the participating sites . The MMPC and IMPC experiments were both expressly designed to minimize variation between sites by standardizing protocols , and by studying age and diet-matched animals . Despite these safeguards , we saw a large variance in metabolic rate in each experiment ( Figures 1 and 3 ) . For the MMPC , these small-scale studies ( n = 6–8 mice per group ) are typical and can be instructive . In this case , the response of these mice to HFD was highly variable with mice at one site gaining three times more fat mass than mice at another site . Despite the large differences in phenotype , these four MMPC sites produced high-quality data on body mass , locomotor activity , photoperiod , diet , and acclimation which accounted for 80% of the variation . There are no right and wrong phenotypes , only the phenotypic differences in institutional responses , which we have quantified but have yet to fully elucidate . Here temperature and body composition provide the greatest sources of variation . The IMPC data afford the largest evaluation of non-genetic and genetic sources of variation yet described . This dataset is both impressively large and yet still narrowly focused on a single age and genetic background , on a standard chow diet . Expanding the types of experiments , strains , and covariates beyond those studied by the MMPC and IMPC will be essential for greater understanding . Our analysis helps to quantify the relative contributions of well-known environmental variables including mass , temperature , activity , sex , and season . This was further exemplified both by the comparison of 9358 WT mice ( Figure 3B ) and by comparing the same KO strains at up to nine different institutions ( Figure 5 ) . We observe phenotypic differences for specific KO strains at some locations , but not at others . However , for most strains we observe good agreement . Understanding the drivers of this variation can help increase reproducibility . Pooling data from multiple sites can also help to reach consensus . The variation or differences among sites is one component of the reproducibility crisis ( Drucker , 2016 ) and understanding institutional variability is a key to increasing reproducibility of metabolic data . The standardization of the IMPC pipeline is an unparalleled resource for re-assessing phenotypic effects of genes with known effects on energy balance . In the scope of performing standardized phenotyping on knockout strains of all mouse gene , several well-characterized genetic experiments have been reproduced by the IMPC consortium members . These genes include those shown to regulate food intake: Cpe , Mrap2 , and Gdf15; metabolic rate and thermogenesis: Ucp1 , PPARg , Gh , Gatm , and Fgfr4; Skin barrier function: Acer1 , and intestinal absorption Pcsk1 . Our analysis of obesity-related GWAS results specifically focused on strains which may impact EE . Genetic variants identified with these methods are often outside of protein-coding sequences and may affect positive or negative gene regulation ( Buniello et al . , 2019 ) . It is therefore of interest that our analysis predicts KO strains which both increase and decrease metabolic rate . One of the biggest advantages conferred by a repository of metabolic data will be the opportunity for researchers to investigate phenotypes of specific alleles or interventions . In the publication describing the metabolic data from the IMPC , the authors chose a ratio-based reporting system based on a residuals from multiple linear regression ( Rozman et al . , 2018 ) . We have reported the residual as well as standard deviation from the mean . Both approaches are similar , but we find reporting the average mass and deviations of the KO strain can add a dimension of context to the results . The strains we identified have not been well characterized with regard to their roles in metabolic pathways , suggesting follow-up studies are warranted . Strains with lower metabolic rate which might contribute to development of obesity include Chst8 , Pax5 , Pald1 , and Tfap2b . Chst8 , carbohydrate sulfotransferase 8 mediates sulfation of the carbohydrate structures of the sex hormone LH; deletion of Chst8 produces mice with elevated LH levels . Chst8 is also expressed in the hypothalamus and pituitary and is predicted to modify proopiomelanocortin ( POMC ) , the precursor to several key hormones involved in the control of stress and body weight regulation ( Mi et al . , 2008 ) . High-throughput IMPC analysis revealed decreased locomotor activity in Chst8-/- mice . Pald1 , phosphatase domain containing paladin 1 , is a regulator of angiogenesis and vascular function . Female Pald1-/- exhibit an emphysema-like lung disorder ( Egaña et al . , 2017 ) . However , cell culture studies have shown Pald1 to have a role in insulin signaling by negatively regulating insulin receptor expression and phosphorylation ( Huang et al . , 2009 ) . Pax5 , paired box 5 , is a transcription factor controlling B cell development ( Nutt et al . , 1999 ) . IMPC analysis confirmed lower leukocyte counts and also found increased cardiac ejection fraction suggestive of cardiopulmonary phenotypes in the Pax5+/- mice . Tfap2b , transcription factor AP-2 beta , KO mice exhibit a form of patent ductus arteriosus ( Satoda et al . , 2000 ) . However Tfap2b heterozygous mice are largely unaffected ( Zhao et al . , 2011 ) , findings confirmed in IMPC analysis . Strains with increased metabolic rate include Pepd , Klf12 , and Pacs1 . Pepd , peptidase D ( or prolidase ) is an enzyme which cleaves dipeptides including collagen . Patients harboring mutations in PEPD have skin ulcers ( Sheffield et al . , 1977 ) . Pepd KO mice reveal an important role for this protein in bone and hematopoetic development . These mice have defects including alterations in bone and immune development ( Besio et al . , 2015 ) . The Pepd gene affects coat pigmentation , suggesting an interaction with melanocortin signaling pathways ( Cota et al . , 2008 ) . IMPC analysis reveals also numerous behavioral and developmental defects . Klf12 , kruppel like factor 12 is a transcriptional repressor linked by GWAS to EE ( Jiang et al . , 2018 ) . Klf12 is necessary for NK cell proliferation in mice ( Lam et al . , 2019 ) and IMPC studies reveal lower levels of locomotor activity . Pacs1 , phosphofurin acidic cluster sorting protein 1 , is a protein with a putative role in the localization of trans-Golgi network membrane proteins and has been shown to be important for vesicular trafficking of POMC into dense secretary granules in neuroendocrine cells , a pathway essential to its downstream cleavage into its active peptide hormones . Pacs1 may also affect the function of cilia , a structures that when altered are associated with development of obesity ( Schermer et al . , 2005 ) . While Pacs1 homozygous deletion mice were embryonic lethal , heterozygote mice revealed no suggested pathology under unchallenged conditions . In summary , these seven genes may affect EE through roles in POMC or melanocortin signaling , skin barrier function , lung defects , hematopoiesis , and bone formation . Follow-up studies will help to clarify the role of these proteins in mouse biology . Greater understanding in mice would provide testable hypotheses for assignment of causal human genetic variants to clinical phenotypes . Several key limitations of the current study include the currently accepted methodology for analysis which requires reduction of hundreds of data measurements collected for each animal down to a single value for ANCOVA or multiple linear regression-based statistical analysis . The IMPC data provides one data point per metabolic variable ( e . g . oxygen consumption or EE ) per animal . Further research on time-series could potentially extract meaningful information from these discarded values . One possible source of unexplained variation between sites is the different models of indirect calorimetry systems used . This study was unable to directly compare the efficacy of different indirect calorimeter manufacturers due to the large institutional differences in EE even between sites using the same instrumentation . However , a direct comparison of results from mice with two different systems within the same room has recently been reported ( Soto et al . , 2019 ) . Different calorimeter parameters and calibrations , such as how many chambers feed into a single gas analyzer , and humidity , pressure and flow rate of gases , can influence results and should be included when reporting data ( Table 1 ) . It is not common practice for investigators to calibrate the whole mouse calorimetry system ( calorimetry chambers through the gas analyzers ) . However , this whole system calibration can be accomplished by introducing a standard gas of known composition into each chamber at a controlled flow rate and measuring recovery of carbon dioxide and oxygen . Appropriate software modules can be used to ensure the accuracy of calibrations and data accuracy . This type of calibration system should be encouraged to facilitate comparisons of indirect calorimetry apparatus . The process of observing mice in an indirect calorimeter may affect their behavior and fail to accurately reflect normal food intake , locomotion , and EE . The energy balance calculated for the MMPC experiment is inconsistent with the body weight gain over the 11-week period on HFD . This finding emphasizes the difficulty in accurately determining energy balance over a few days and may be better captured by longer measurements . It is also possible that the calorimetry environment altered energy balance compared to home cage conditions . It can be challenging to accurately measure energy intake and systems designed to minimize spillage could also impact food intake by making it more difficult for mice to eat . Observing serial measurements of body weight can help to mitigate interpretation of experiments where data unexpectedly reflects weight loss ( as in Figure 1 ) . Both in-depth phenotyping and large-scale population-based phenotyping reveal the large effect size of institutional variation . Two possible but untested sources of variation in these datasets include changes in gut microbiota and/or epigenetic changes induced by different environmental triggers . Mice have specific microbial flora which change over time and affect energy balance ( Turnbaugh et al . , 2006 ) . The microbiome is likely to be an institutionally local phenomenon which can affect metabolic rates; indeed studies linking the dependence on thermogenesis with the microbiome have reported different results in experiments performed in Boston , New York City , Geneva and Beijing ( Chevalier et al . , 2015; Krisko et al . , 2020; Li et al . , 2019 ) . Knowledge of microbial populations which correlate with EE may help to refine metabolic studies in the future . Attempts to rescue phenotypic variation by microbiome reconstitution may prove fruitful . Similarly , it may soon be possible to interrogate the set of epigenetic modifications to genes controlling metabolic rate and help to predict or modify metabolic rate accordingly . There is a surprisingly large degree of phenotypic variation observed in mouse energy expenditure . The two experimental paradigms analyzed were deliberately designed to create consistent , reproducible results . Contrary to our predictions , there were still unaccounted-for differences among institutions either in the response of WT animals to differentially gain body weight or with intrinsic differences in energy expenditure due to uncontrolled factors such as ambient temperature . This variation was sufficiently large to prevent consistent phenotypic conclusions caused by the same genetic interventions . The experimental location variability strongly affected results in both WT and genetically modified strains . These findings suggest that reporting experimental conditions including body composition , accurate temperature , and activity should be essential to reproducing and comparing calorimetry data , and can explain the majority of institutional differences . Identifying the as-yet unknown sources of experimental variability among sites should also become a priority to foster consistency and reproducibility in experimental results among multiple institutions . However , within any one institution , we can trust that results obtained comparing littermate controls using ANCOVA for analysis are experimentally valid for those conditions . Yet investigators dedicated the reproducibility of their experimental model at other sites will need to report the essential information for interpretation of indirect calorimetry studies ( Table 1 ) . Using this model , individual labs or centers may leverage the big data approach to understanding phenotypes by meticulously creating datasets of non-littermate control mice ( as in the IMPC data , Figure 6 ) . Once validated , these data can be used to regress against smaller cohorts of experimental mice , as well as controls to validate the approach in each instance . This strategy suggests that indirect calorimetry can still be a useful tool to understand metabolic phenotypes . Data from the NIH-funded MMPCs ( RRID:SCR_008997 ) represent longitudinal measurements at 4 sites located at University of California , Davis ( UC Davis ) , University of Massachusetts ( UMass ) , Yale University ( Yale ) and Vanderbilt University ( Vanderbilt ) . Indirect calorimetry measurements were recorded for each mouse for at least 4 days , allowing for analysis of pre-acclimation ( the first 18 hr ) and post-acclimation ( 18–96 hr ) . Unless otherwise noted , our analyses use the post-acclimation data . The 4 geographically distinct MMPC sites used calorimeters from 3 different manufacturers ( Columbus Instruments at UC Davis , California and Yale , Connecticut; TSE Systems at UMass , Massachusetts; Sable Systems International at Vanderbilt , Tennessee ) . Genetically identical C57BL/6J male mice ( n = 60 , 6–7 weeks of age ) originating from the same room in the Jackson Laboratory facility were divided into four groups , and each group was shipped to one site . On arrival , mice were maintained on LFD for one week and then randomized into two groups for all further experiments ( n = 6–8 mice per diet per site ) . To control for the transition from group to single housing , all animals remained singly housed for the duration of the study . Indirect calorimetry was performed on all mice at each site ( week 0 ) after which one group was given HFD for the next 12 weeks , while the other was maintained on LFD . All mice were returned to the calorimeters at 4 and 11 weeks post-diet randomization . Mouse diets were purchased as a uniform lot from a singular production stream for all sites from Research Diets ( LFD: D12450B , 10% energy derived from fat , 15 . 69 kJ/g; HFD: D12452 , 60% energy derived from fat , 21 . 92 kJ/g ) and were delivered to each site simultaneously . Energy intake was calculated by taking the product of food intake in grams with the energy density of the diet in kJ . UC Davis used a PIXIMus DEXA scanner under anesthesia for body composition measurements . Other sites used an NMR-based Bruker minispec without anesthesia . Reported room temperatures for the UC Davis , UMass , Yale , and Vanderbilt MMPC sites were 22 . 0 , 21 . 1 , 21 . 5°C and 22 . 5°C respectively . For indirect calorimetry , the Vanderbilt site used a temperature-controlled environment of 24°C . Beam breaks reported from Columbus Instruments , Sable Systems International , and TSE Systems correspond to different distances . Accordingly , locomotor activity was calculated as beam breaks as a percent of the global maximum per site . IMPC ( RRID:SCR_006158 ) data were collected as described ( Rozman et al . , 2018 ) . In version 10 . 0 of this dataset , the 24 hr averaged metabolic data from more than 30 , 000 mice is publicly available . The IMPC data uses highly similar indirect calorimetry protocols across 10 sites in 8 countries . All indirect calorimetry data were collected from 11-week-old mice , except for body composition data . Lean and fat masses corresponding to week 11 were estimated for each mouse based on body composition percentages collected at week 14 . The calorie content and macronutrient compositions of the chow diets used were provided when requested . The IMPC 10 . 0 sites uniformly provide a single average value reported for oxygen consumption , carbon dioxide release , RER , and EE , per animal although additional hourly data is available through their API . Data were not consistently present at all sites for both males and females , and sites did not uniformly report locomotor activity , precise housing temperatures , or food intake data . Similarly , acclimation was performed only at a subset of sites and pre-acclimation data was not readily available . On initial visualization , the data deposited showed several abnormalities . These included one site where larger animals paradoxically consumed less oxygen than smaller animals , demonstrating an inverse mass effect ( Figure 3—figure supplement 3A ) . When contacted this site quickly identified and corrected a persistent error in their data analysis pipeline which will be rectified in a future IMPC data release . A different site reported erroneous food intake data with each animal consuming exactly 0 . 05 g of food per hour regardless of body mass or genotype . This site also provided an updated dataset with corrected food intake values ( Figure 3—figure supplement 3B ) . We also observed differences in methods to calculate EE from indirect calorimetry data . When calculating EE using the Weir formula ( Weir , 1949 ) we find two sites which produced different values , likely due to implementation of other formulae including the Lusk equation ( Lusk , 1993; Figure 3—figure supplement 3C ) . Lastly , we identified multiple sites providing data from calorimetry experiments shorter than the reported 21 hr minimum duration . At one site , the full run data was erroneously excluded from the IMPC database , and once contacted the complete data was provided for use in this analysis . All other experiments with durations shorter than 18 hr were excluded from our analysis ( Figure 3—figure supplement 3D ) . To improve consistency , we re-calculated data for all mice using the Weir equation . Erroneously keyed data were corrected as described in Figure 3—figure supplement 3 . One strain was excluded due to highly variable values within the genotype in a subset of mice , Fbxl19 . The acronyms describing the IMPC sites have changed in some instances . For simplicity , the following sites are indicated by the country in which they are located . For countries with two sites , the state or county is indicated ( Table 2 ) . All data were collected with Columbus Instruments CLAMS indirect calorimeters in male mice . Data demonstrating the effect of age in chow-fed ( Teklad 2018 ) C57BL/6J mice were graciously contributed by Houtkooper et al . ( 2011 ) . EE values during voluntary wheel running exercise in 6-month-old chow-fed ( Teklad 7022 ) C57BL/6 male mice at 22°C are as reported in the supplemental materials from O'Neal et al . ( 2017 ) . Studies of the effect of β3-adrenergic stimulation and of temperature change were performed in a temperature-controlled chamber enclosing the indirect calorimeter . Both studies were conducted in Boston MA . For β3 stimulation , 6 . 5-month-old chow-fed ( LabDiet 5053 ) C57BL/6J global Cdkal1-/- or WT littermate male mice were maintained at 30°C for 24 hr . Mice received an IP administration of 1 mg/kg CL , 5 hr into the light photoperiod . The mean EE over the 3 hr post-injection period was used in this analysis . No significant differences in EE were observed between genotypes; accordingly , data for all mice were pooled . EE measurements of Cdkal1-/- or WT littermate control male mice maintained at 23°C were used in Figure 5 . The study of the effect of temperature on EE was performed on 10-week-old chow-fed ( LabDiet 5053 ) C57BL/6J male mice with an incremental decrease in temperature . Mice were maintained at 30°C for 24 hr intervals between each 24 hr temperature challenge ranging from 28°C to 6°C . CalR analysis ( RRID:SCR_015849 ) was performed on the MMPC dataset ( Figure 1—figure supplement 1 ) as described ( Mina et al . , 2018 ) . The ‘remove outliers’ feature was enabled to exclude from analysis data that were recorded during momentary cage opening . Previously unpublished studies were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . All the animals were handled according to approved institutional animal care and use committee ( IACUC ) at the site where they were performed . Data input , cleaning , visualization , and statistical analysis were performed in the R programming language ( R Development Core Team , 2019 ) . The relative contribution of covariates were calculated with the relaimpo package ( Grömping , 2006 ) The multiple linear regression model and model to quantify the explained variance for the MMPC experiment after 11 weeks on diet included body composition , locomotor activity , photoperiod , diet , and acclimation . These models were not improved by the addition of equipment manufacturer or temperature as each site reported a single , unique temperature , and only two sites used a similar manufacturer of indirect calorimeter . The IMPC models included body composition , locomotor activity , ambient temperature , sex and season . These models were not improved with addition of equipment manufacturer . The q-q plot was performed with a geom_qq using EE data from WT mice at the seven sites examining both sexes ( Wickham , 2009 ) . All plots were produced with ggplot2 or CalR ( Mina et al . , 2018; Wickham , 2009 ) . Unless otherwise specified , all reported data are based on the daily EE ( DEE ) , 24 hr averaged EE values . Experiments were not performed at thermoneutrality , eliminating the possibility of determining basal EE or resting metabolic rate at thermoneutrality ( Even and Nadkarni , 2012; Meyer et al . , 2015 ) .	Maintaining a healthy weight requires the body to balance energy intake and expenditure . The body converts food to energy through a process called energy metabolism . Genetic and environmental factors can affect energy metabolism and energy balance contributing to conditions like obesity . To better understand metabolism , scientists often study mice in laboratories , but mice from different laboratories appear to convert food to energy at different rates . This makes it hard to determine what is ‘normal’ for mouse metabolism . These discrepancies could be due to small differences between how mice are kept in different laboratories . For example , the temperatures of the mouse cages or how active the mice are might differ depending on the laboratory . Identifying the effects of such differences is essential , but it requires looking at data from hundreds of mice . Corrigan et al . examined data from more than 30 , 000 mice at laboratories around the world to show that room temperatures and the amount of muscle and fat in a mouse’s body have the biggest influence on energy balance . These two factors affected the metabolism of both typical mice and mice with mutations that affect their energy balance . These results suggest that it is important for scientists to report factors like room temperatures , the body make-up of the mice , or the animals’ activity levels in metabolism studies . This can help scientists compare results and repeat experiments , which could speed up research into mouse metabolism . Corrigan et al . also found that other unknown factors also affect mouse metabolism in different laboratories . Further studies are needed to identify these factors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2020	A big-data approach to understanding metabolic rate and response to obesity in laboratory mice
Slow-wave sleep is an optimal opportunity for memory consolidation: when encoding occurs in the presence of a sensory cue , delivery of that cue during sleep enhances retrieval of associated memories . Recent studies suggest that cues might promote consolidation by inducing neural reinstatement of cue-associated content during sleep , but direct evidence for such mechanisms is scant , and the relevant brain areas supporting these processes are poorly understood . Here , we address these gaps by combining a novel olfactory cueing paradigm with an object-location memory task and simultaneous EEG-fMRI recording in human subjects . Using pattern analysis of fMRI ensemble activity , we find that presentation of odor cues during sleep promotes reactivation of category-level information in ventromedial prefrontal cortex that significantly correlates with post-sleep memory performance . In identifying the potential mechanisms by which odor cues selectively modulate memory in the sleeping brain , these findings bring unique insights into elucidating how and what we remember . Only a small fraction of the events that are experienced during wakefulness are stamped into long-term memory . Understanding how memories are formed , and which memories are ultimately retained or forgotten , is a pivotal focus of neuroscience research . The process through which memories are stabilized and integrated for long-term storage , termed memory consolidation , is robustly enhanced during sleep ( Deak and Stickgold , 2010; Diekelmann and Born , 2010 ) . Although the neural mechanisms underlying sleep-based consolidation are far from clear , research increasingly implicates memory replay , whereby the same neural activity that occurs during memory encoding comes back online spontaneously during sleep to facilitate integration of the replayed memory into distributed cortical networks for long-term storage . Direct cellular-level evidence for replay comes from rodent studies ( Skaggs and McNaughton , 1996; Wilson and McNaughton , 1994 ) , while in humans , an indirect measure often termed ‘reactivation’ has also been demonstrated using fMRI , surface EEG , and intracranial EEG techniques ( Bergmann et al . , 2012; Deuker et al . , 2013; Peigneux et al . , 2004; Schönauer et al . , 2017; Zhang et al . , 2018 ) . Intriguingly , it has been shown that external sensory cues , such as odors or sounds , can be presented during sleep to manipulate what information is preserved . After a sensory cue is presented during memory encoding in the wake state , re-presentation of that cue during sleep favors subsequent retrieval for the associated material ( Oudiette and Paller , 2013; Rasch et al . , 2007; Rudoy et al . , 2009; Schouten et al . , 2017; Shanahan and Gottfried , 2017; Spiers and Bendor , 2014 ) . It follows that cue-evoked reactivation of content-specific information during sleep might promote these memory gains , but direct evidence for such mechanisms is highly limited ( Schouten et al . , 2017 ) . Groundbreaking work in rodents showed that auditory cues induce within-sleep replay of spatial sequences in hippocampal place cells , although these effects were not accompanied by an index of learning ( Bendor and Wilson , 2012 ) . Recent human studies have tested the idea that patterns of EEG activity in sleep can be used to decode different forms of auditory-cued memories , including procedural sequence learning ( Belal et al . , 2018 ) and place versus object information ( Cairney et al . , 2018 ) , with the latter study demonstrating a correlation with post-sleep memory performance . However , because the spatial resolution of surface EEG measures is poorly suited for pinpointing the involvement of specific brain regions and networks , there remains a critical gap in understanding how cued reactivation of specific memory content is induced in the sleeping brain , and whether content-specific reactivation holds relevance for behavior . To address this important gap , we designed a novel olfactory fMRI paradigm optimized to investigate the functional links between odor-evoked memory reactivation and declarative memory consolidation . Here , subjects completed an object-location memory task , in which objects belonged to four different categories ( animals , buildings , faces , tools ) ( Figure 1 ) . In turn , each object category was associated with a different odor cue ( e . g . banana , cedar , cinnamon , garlic ) . Critically , the use of an MRI-compatible EEG system enabled us to deliver these odor cues selectively during slow-wave sleep ( SWS ) , such that odor-evoked patterns of fMRI activity in the sleeping brain could be compared to canonical category representations from a preceding wake scan . In this manner , we were able to identify brain regions where fMRI signatures of within-sleep reactivation would have a direct impact on the strength of memory recall , as a function of reactivation strength within subjects . Specifically , we predicted that we would observe this relationship in task-relevant sensory brain areas ( Ji and Wilson , 2007; Rothschild et al . , 2017 ) , and in memory retrieval networks in the hippocampus ( Bendor and Wilson , 2012; Peigneux et al . , 2004; Rasch et al . , 2007 ) and prefrontal cortex ( PFC ) ( Bonnici and Maguire , 2018; Euston et al . , 2012; Jin and Maren , 2015; Preston and Eichenbaum , 2013; Takashima et al . , 2007 ) . In an initial learning phase , subjects performed a visuospatial memory task , in which they learned the locations of objects from four categories: animals , buildings , faces , and tools . There were 32 objects per category , and object images were presented on a 4-x-4 spatial grid during fMRI scanning ( Figure 1b ) . Subjects were instructed to memorize the location associated with each object , each of which appeared three times over the course of the scan . In line with previous studies demonstrating that visual category perception elicits unique ensemble patterns of fMRI activity ( Haxby et al . , 2001; Norman et al . , 2006 ) , a multivoxel pattern-based fMRI searchlight analysis revealed category specificity in widely distributed brain regions , including much of the visual pathway , and substantial parts of parietal and prefrontal cortices ( Figure 1c; Figure 1—figure supplement 1 ) . fMRI data collected from the initial learning phase were then used to define multivoxel representations of stimuli belonging to each object category ( Figure 1d ) , for use as reference templates to identify content-specific fMRI activity that might emerge in subsequent sleep . Importantly , these pattern templates were defined prior to the introduction of odor cues , to ensure that fMRI templates exclusively reflected visual category information . After completing the initial learning phase , subjects were moved from the scanner to a testing room , where they learned to associate four easily-distinguishable odors with each of the four object categories to criteria ( Figure 1e ) . Based on individual ratings , each subject received a unique combination of four odors that were maximally discriminable ( see Materials and methods ) , from a set of eight familiar odors ( banana , cedar , cinnamon , garlic , lemon , mint , rose , vanilla ) . Following odor-category learning , subjects completed another learning session to reinforce odor-category associations and to learn object locations to criteria , and were then fitted with an MRI-compatible EEG electrode cap . In a subsequent memory pretest session , subjects were asked to recall object locations ( without feedback ) . This test took place immediately before the fMRI sleep session , providing a memory baseline for comparison to post-sleep memory performance . Recall accuracy at pretest was robust ( Figure 2a , first column ) , without significant differences between those categories that were later cued during sleep and those that were not ( t ( 17 ) = 0 . 62 , p = 0 . 54; Figure 2—figure supplement 1 ) . Subjects then returned to the scanner , where they were instructed to try to relax and fall asleep during approximately 75 min of fMRI scanning . During sleep stage 2 and SWS , two of the four odors were presented ( 16 s on/16 s off ) via a Teflon tube secured beneath the nose , to cue two of the four object categories ( Figure 1f; Figure 1—figure supplement 2 ) . Selection of the odor cues in sleep was arranged to ensure that the specific categories designated for reinforcement were counterbalanced across subjects . Eighteen subjects reached criteria , in that during sleep they received a minimum of 14 presentations of each odor cue , and during debriefing they did not recall having smelled any odors during the scanning session . On average , these 18 subjects slept for 65 . 36 min , of which 26 . 50 min were spent in sleep stage 2 , and 30 . 17 min were spent in SWS ( Table 1 ) . Analysis of the within-sleep EEG data did not reveal significant spectral differences between odor-on and odor-off periods ( repeated-measures ANOVA , time by cue interaction: F ( 39 , 17 ) = 0 . 31 , p = 1 . 00; Figure 1g ) , suggesting that odor delivery was not associated with physiological arousal . Upon waking , but prior to the final post-sleep memory test , subjects were removed from the scanner and taken to a behavioral testing room , where they underwent a memory interference task to learn new grid locations for the same set of objects . This maneuver has been shown to provide added sensitivity to identifying effects of sleep on memory performance ( Ellenbogen et al . , 2009 ) . Five minutes after completing the interference encoding task , subjects were asked to try to place the objects in their new locations . As predicted , overall recall accuracy on the interference task was lower compared to pre-sleep recall accuracy ( Figure 2a , second column ) . Recall accuracy did not differ for cued compared to non-cued objects ( t ( 17 ) = 0 . 04 , p = 0 . 97 ) . However , odor cues significantly influenced recall speed for the new object locations , whereby response times ( RTs ) for placing objects on the grid were slower for cued versus non-cued objects ( t ( 17 ) = 2 . 76 , p = 0 . 01; Figure 2b ) , probably reflecting increased competition between the new object locations and the original object locations that were reinforced by odor cues during sleep . This finding provides a first indication that odor cues presented in sleep had an impact on memory storage . Finally , after a 30-min break , subjects were asked to recall the original object locations during a posttest session , for direct comparison to pretest performance . Posttest recall ( Figure 2a , third column ) was predictably lower than pretest recall ( t ( 17 ) = 10 . 06 , p < 0 . 001 ) . Critically , subjects demonstrated less forgetting for cued object locations from pretest to posttest , when compared to non-cued object locations ( Z = 1 . 70 , p = 0 . 04; Wilcoxon signed-rank test; Figure 2c ) , with selective memory gains for 15/18 subjects ( p = 0 . 003; Binomial test ) . These findings , as well as the RT data , confirm that delivery of odor cues during SWS selectively biased recall toward object categories previously paired with those odors in the wake state . Of note , although memory performance was significantly above chance during a follow-up memory test 1 week after the main experiment ( Figure 2a , fourth column ) , effects of odor cueing on memory performance were no longer evident ( recall accuracy: t ( 17 ) = 0 . 17 , p = 0 . 43; RT: t ( 17 ) = 0 . 46 , p = 0 . 65 ) . Having established the selective influence of odor cueing on recall performance , we next examined the mechanisms underlying this memory-enhancing effect . Our central hypothesis was that if odor cues induce within-sleep reactivation of associated category information in medial temporal and prefrontal brain areas known to participate in memory consolidation and retrieval ( Bonnici and Maguire , 2018; Euston et al . , 2012; Jin and Maren , 2015; Preston and Eichenbaum , 2013; Takashima et al . , 2007 ) , then the degree of memory reactivation should predict subsequent memory performance in the wake state . To this end , we first correlated odor-evoked fMRI ensemble activity in sleep with each of the four category-specific fMRI template patterns from the initial learning phase of the task , and then compared the degree of pattern overlap between cued and non-cued category templates , yielding a measure of cue-specific reactivation strength for each subject ( Figure 1d , f ) . By regressing this within-sleep reactivation measure onto changes in memory performance from pretest to posttest , we were able to determine whether memory recall varied with reactivation strength across subjects . Note , this analysis was restricted to regions that demonstrated category specificity during prior learning . In this way , we found that greater category reactivation in ventromedial PFC ( vmPFC ) was associated with increased recall accuracy at posttest for cued over non-cued objects ( [−6 , 46 , –12] , t ( 16 ) = 7 . 53 , pFWE = 0 . 01 , Figure 3a ) . These effects were robust across both odor cues , as within-sleep reactivation strength in vmPFC was significantly correlated with post-sleep recall when each cue was considered independently ( r1 ( 16 ) = 0 . 53 , p1 = 0 . 01; r2 ( 16 ) = 0 . 48 , p2 = 0 . 02; Figure 3b ) . In a time-resolved analysis of stimulus-evoked activity during the sleep period , correlations between odor-evoked reactivation in vmPFC and recall accuracy increased at odor onset and persisted over several seconds , with a maximal effect size ( r value ) of 0 . 70 ( 90% confidence interval = 0 . 42 to 0 . 86 ) , returning to baseline prior to odor offset ( Figure 3c ) . Together these findings in vmPFC highlight the categorical and temporal specificity of odor-cued reactivation on memory retrieval ( see Figure 4 for additional analysis ) . Because our paradigm fundamentally involves a visuospatial task , we also reasoned that within-sleep reactivation of cued content might manifest in visual associative brain regions ( Ji and Wilson , 2007; Rothschild et al . , 2017 ) . Across subjects , recall accuracy for cued versus non-cued object categories scaled with the degree of odor-evoked category reactivation in posterior fusiform cortex ( [−24 , –70 , 0] , t ( 16 ) = 6 . 73 , pFWE = 0 . 04; Figure 3d ) . When considering each cued category separately , the correlation between reactivation in fusiform cortex and posttest memory performance was evident , but only for one of the odor cues ( r1 ( 16 ) = 0 . 43 , p1 = 0 . 04; r2 ( 16 ) = 0 . 22 , p2 = 0 . 19 ) ( Figure 3e ) . Similar to the time-course profile in vmPFC , within-sleep reactivation in fusiform cortex also emerged following odor onset , but appeared more sustained throughout the duration of odor presentation ( Figure 3f ) . Parallel analyses in the hippocampus , which has been previously implicated in memory cueing studies ( Bendor and Wilson , 2012; Diekelmann et al . , 2011; Rasch et al . , 2007; van Dongen et al . , 2012 ) , did not reveal any association between within-sleep measures of reactivation and subsequent memory performance ( Figure 3—figure supplement 1 ) . The above analyses were predicated on the idea that among the brain areas that might be reactivated during sleep in response to odor cueing , only those areas that had a systematic effect on post-sleep memory performance would have behavioral relevance . Nonetheless , we also examined whether odor cues evoked reactivation of the associated category per se , without considering memory performance as a covariate . This analysis revealed odor-evoked reactivation of cued category templates in lateral occipital complex ( [−56 , –62 , 4] , t ( 16 ) = 3 . 99 , punc = 0 . 001 ) ( Figure 5a ) and inferior frontal gyrus ( [−50 , 30 , 8] , t ( 16 ) = 3 . 91 , punc = 0 . 001 ) , although neither of these clusters survived correction for multiple comparisons . Moreover , the degree of memory reactivation in these regions was not correlated with recall accuracy for cued versus noncued objects upon waking in lateral occipital complex ( r ( 16 ) = −0 . 33 , p = 0 . 19; Figure 5b ) or inferior frontal gyrus ( r ( 16 ) = 0 . 25 , p = 0 . 32 ) . We also conducted an exploratory analysis to test whether within-sleep reactivation manifests in different brain areas depending on which object category is being cued . By considering each object category separately ( rather than collapsing across all four categories ) , we found that , at a threshold of p < 0 . 001 uncorrected , odor cues promoted reactivation of associated multivoxel patterns in different brain areas ( Figure 5—figure supplement 1 ) . However , it is important to note that our experiment was not designed to test individual category effects , because any given category was only cued for 50% of subjects ( n = 9 ) . As such , this analysis was underpowered , and therefore these findings should be considered as tentative . The finding that an index of odor-cued reactivation in vmPFC and fusiform cortex during sleep predicts memory recall necessarily implies that the olfactory system must communicate with extra-olfactory structures to consolidate visuospatial representations . To define potential pathways by which odors can induce cortical reactivation in sleep , we implemented a univariate analysis to characterize which brain areas were activated in the presence of the odor cues , irrespective of their effects on memory performance . We identified an olfactory-related cluster in the left amygdala , extending into left hippocampus ( [−28 , 0 , –24] , t ( 17 ) = 4 . 18 , pSVC = 0 . 03 ) , which could plausibly serve as a conduit to cortical structures . Interestingly , this brain area overlaps closely with previous findings showing odor-cued reactivation of left anterior hippocampus during SWS ( Rasch et al . , 2007; Diekelmann et al . , 2011 ) . A cluster in right amygdala was also observed in an almost symmetrical location to the left amygdala cluster , but did not survive correction for multiple comparisons ( [26 , 0 , -24] , t ( 17 ) = 3 . 47 , punc = 0 . 001 ) . It is worth pointing out that the identification of olfactory-related regions in sleep could arise because the odor cues were driving downstream activity in the olfactory system , or because they were reactivating previously associated memory content . Ideally , delivery of a control odor that had never been associated with category information would help resolve these two different possibilities . In our study , we opted not to deliver a control odor , to maximize the number of odor trials available for pattern analysis . Therefore , subjects did not receive any odors that were not associated with prior learning during scanning . As such , it was not possible to disentangle contributions of the odor per se , and those related to associated memories . Rather , odor-cued activity in this analysis likely reflects an amalgam of olfactory and reactivation-related influences . To assess whether the amygdala-hippocampal cluster might be preferentially coupled with vmPFC or fusiform cortex in the presence of sleep-based odor cues , we used psychophysiological interaction ( PPI ) analyses to test the functional connectivity between the amygdala-hippocampal cluster and the downstream cortical areas . Significant coupling was observed during odor presentation between this olfactory-related region and vmPFC ( t ( 17 ) = 1 . 95 , p = 0 . 03 ) but was not found between the same region and fusiform cortex ( t ( 17 ) = 0 . 85 , p = 0 . 20 ) . The implication is that cued memory reactivation in the sleeping brain begins with odor-evoked activity in medial temporal brain structures , which may mediate the instantiation of visual categorical content in vmPFC . Our data indicate that an olfactory-prefrontal network may be an important substrate underlying odor-evoked category reactivation . Here , we used EEG-fMRI recordings combined with multivoxel pattern analysis to investigate the neural mechanisms underlying memory outcomes in a novel olfactory cueing paradigm . First , we demonstrated an olfactory cueing effect , namely , that within-sleep odor cues boost memory performance selectively for associated object categories . Critically , we observed that these behavioral effects are robustly correlated with the degree to which odors promote category-specific reactivation in vmPFC and posterior fusiform cortex , and that these effects hold even when considering both cued object categories separately . Finally , we found that during sleep , odors evoke neural activity in brain areas related to olfactory and limbic function , and that these areas are connected to vmPFC during odor presentation . Together , our findings highlight the functional significance of cue-evoked memory reactivation in promoting consolidation of declarative memories within sleep . In auditory cueing studies , a multitude of sound cues are routinely presented in sleep to induce memories that are highly specific , and that are often semantically linked to the cue ( e . g . cat picture paired with ‘meow’ sound cue ) ( Fuentemilla et al . , 2013; Oudiette et al . , 2013; Rudoy et al . , 2009 ) . By contrast , prior olfactory cueing paradigms have largely employed a single arbitrary odor to cue an entire memory task ( Diekelmann et al . , 2012; Diekelmann et al . , 2011; Rasch et al . , 2007 ) . Recent studies that utilized two distinct odors during learning and a single olfactory cue during sleep have established that olfactory stimuli can influence behavior with some specificity ( Hauner et al . , 2013; Rihm et al . , 2014 ) , but to our knowledge ours is the first olfactory study to cue multiple task components during sleep ( i . e . two odor cues associated with different object categories ) . Odors offer unique benefits over sounds as memory cues . Namely , olfactory stimuli are less likely to provoke arousal from sleep ( Carskadon and Herz , 2004 ) , particularly pure odorants lacking a trigeminal component ( Grupp et al . , 2008; Stuck et al . , 2007 ) . In addition , the lack of a requisite thalamic relay for odor stimuli and the relative proximity of olfactory and limbic structures in the brain may confer an anatomical advantage ( Gottfried , 2010 ) . The demonstration that odor cues can promote memory gains more selectively should increase confidence in their utility for more complex memory cueing paradigms . Perhaps more critically , we show that these selective memory benefits are strongly correlated with the degree to which odors drive reactivation of category information in vmPFC and fusiform cortex in the sleep period . Simultaneous recording of EEG and fMRI data during sleep is technically challenging , and for the vast majority of sensory cueing studies , the sleep period is not scanned . Thus , research exploring fMRI correlates of memory cueing is scant . Two previous EEG-fMRI sleep studies have demonstrated that olfactory cues evoke activity in left anterior hippocampus , at similar coordinates as we observed here ( Diekelmann et al . , 2011; Rasch et al . , 2007 ) . However , there was no non-cued condition for behavioral comparison in either study , preventing more nuanced conclusions regarding the relationship between brain activity and behavior . One additional study found that auditory cues elicit fMRI activity in parahippocampal cortex during sleep , but did not demonstrate a behavioral effect of cueing on memory performance , perhaps due to subjects’ reduced ability to process sound cues in the noisy scanner environment ( Berkers et al . , 2018; van Dongen et al . , 2012 ) . Although these studies bring important understanding to the dynamics of cueing memories during sleep , our study uniquely highlights its behavioral benefits , permitting us to relate memory performance to neural reactivation on a subject-by-subject basis . Moreover , by utilizing ensemble pattern-based analysis of fMRI data , we were able to probe the contents of within-sleep memory reactivation with greater specificity than would be possible with more conventional fMRI analyses that have been employed in previous studies . In addition , by measuring re-emergence of category-specific brain activity using fMRI , we could pinpoint brain regions participating in neural reactivation with a level of regional and network specificity that EEG approaches cannot provide . It is worth considering what type of information is being reactivated during sleep . One possibility is that odor cueing leads to reactivation of higher-order representations of object categories that have been established through prior experience with animals , buildings , faces , and tools . An intriguing implication is that if odor cueing in sleep elicits reactivation at the categorical level , then the effects of odor cueing should enhance recall not only for the original group of objects , but also for any other objects in that same category , and possibly even for related semantic categorical content from any sensory modality . Alternatively , category-specific odors may evoke neural reactivation more specifically , by cueing a subset of objects belonging to that category . Our experimental design does not allow for us to disentangle these two possibilities , but future research should define memory templates during learning at both stimulus and category-specific levels prior to within-sleep cueing to address this question . Outside of the context of sensory cueing , consolidation is thought to involve the gradual integration of declarative memory traces within the neocortex , guided by the hippocampus . It has been proposed that the PFC may play a key role in integrating memories across cortical modules ( Frankland and Bontempi , 2005 ) , and indeed frontal lobe damage has been shown to impair recollection ( Simons and Spiers , 2003 ) . Additionally , medial PFC ( mPFC ) has been increasingly implicated in remote memory retrieval ( Bonnici and Maguire , 2018; Euston et al . , 2012; Jin and Maren , 2015; Preston and Eichenbaum , 2013; Takashima et al . , 2007 ) , especially when memory encoding is followed by a period of sleep ( Gais et al . , 2007; Sterpenich et al . , 2009 ) . It has also been suggested that sleep-based integration of memories into neocortical networks is aided by the slow oscillations characteristic of SWS ( Rasch and Born , 2013 ) . In line with this concept , mPFC is thought to be a predominant generator of slow wave activity ( SWA ) ( Murphy et al . , 2009 ) , and cortical volume loss in mPFC has been linked to parallel deficits in SWA and sleep-dependent memory retention ( Mander et al . , 2015; Mander et al . , 2013 ) . Our findings that odors presented during SWS may drive reactivation of associated mnemonic content in vmPFC to support recall meshes well with this prior work , providing robust mechanistic support that cue-evoked cortical reactivation promotes within-sleep consolidation in the human brain . Thirty-two healthy human subjects ( 21 female; mean age = 25 . 25 years; age range = 19–37 years ) gave written informed consent to take part in the study , which was approved by the Institutional Review Board at Northwestern University . All subjects were right-handed non-smokers under 40 years of age who had undergone fMRI scanning at least once prior to the study , and reported that they thought they could feel relaxed and fall asleep during fMRI scanning . Additional criteria for exclusion included history of significant medical or psychiatric illness , history of sleep disorder , use of psychotropic medications , nasal congestion , and frequent snoring . Subjects were required to go to bed at their habitual bedtime the night before the main experiment , and to wake up three hours earlier than their habitual wake time the following morning . Sleep-wake activity was monitored via actigraphy ( Spectrum , Phillips ) during the night prior to the main experiment , and subjects completed an online sleep diary ( adapted from the National Heart , Lung , and Blood Institute ) for 1 week prior to the main experiment . Subjects were also asked to refrain from napping , and from consuming caffeine or alcohol on the day of the main experiment . Eleven subjects were excluded from analysis due to insufficient or fragmented SWS during the nap phase , which prevented the experimenters from presenting each of the two odor cues a minimum of 14 times . An additional three subjects were excluded due to arousal and subsequent odor perception during the nap phase . Eighteen subjects , equal to 56 . 25% of all subjects , were retained for analysis ( 11 female; mean age = 25 . 11 years; age range = 19-37 ) , in line with prior studies reporting subjects’ ability to fall asleep and stay asleep in the fMRI scanner environment ( Diekelmann et al . , 2011; van Dongen et al . , 2012 ) . Visual stimuli consisted of 128 high-resolution portable network graphic ( PNG ) images obtained from the internet . Images were cropped and displayed on a grey background with identifying labels . Images included 32 well-known exemplars from each of four categories: animals ( e . g . camel , zebra ) , buildings ( e . g . Eiffel Tower , Taj Mahal ) , faces ( e . g . Barack Obama , Emma Watson ) , and tools ( e . g . screws , wrench ) . To ensure that exemplars were familiar to subjects , an independent group of 12 subjects provided labels for each image of a larger stimulus set , and the most easily identifiable images were retained for use in the study . Thirty-two additional scrambled images were generated from a subset of category images , and accompanying labels consisted of arbitrary combinations of letters . Scrambled images were included as a localizer , to allow for identification of functionally defined voxels if needed . Olfactory stimuli comprised four easily distinguishable , familiar odorants . To ensure that odor cues could be easily discriminated from each other , four odorants were selected from a larger set of eight well-known odorants ( banana , cedar , cinnamon , garlic , lemon , mint , rose , vanilla ) on an individual subject basis ( see next section ) . Odorants were delivered by a custom 12-channel computer-controlled olfactometer at a flow rate of 6 . 72 L/min via a Teflon tube secured beneath the nose . The day before the main experiment , each subject took part in an odor selection task so that we could determine which four odors were most discriminable . Immediately prior to the task , subjects were familiarized with eight odors and their respective labels . Subjects then made pairwise similarity ratings between all possible pairs ( 28 unique pairs ) . Each odor pair was presented two times , for a total of 56 trials . During each trial , subjects were cued to sniff two consecutive odors presented 4 . 5 s apart and then to make a pairwise similarity rating on a visual analog scale from ‘extremely different’ to ‘extremely similar’ . A ‘pairwise similarity score’ was calculated for each odor pair by taking the average similarity rating across two trials . Based on pairwise similarity scores , four odors were selected to minimize perceptual overlap ( i . e . , minimal pairwise similarity scores for odor pairs included in the final set of four ) . Thus , a ‘total similarity score’ was computed for each possible combination of four odors ( 70 total combinations ) . For instance , for a set of four odors ( A , B , C , and D ) , with all possible pairwise combinations among these odors , the total similarity score would be the sum of similarity ratings for A vs . B , A vs . C , A vs . D , B vs . C , B vs . D , and C vs . D . The set of four odors with the lowest total similarity score was retained for the main experiment . The main experiment lasted approximately 6 . 5 hr and took place at night , such that the nap phase aligned with each subject’s usual bedtime , to increase the likelihood of subjects falling asleep during fMRI scanning . The main experiment began between 7 pm and 10:30 pm , and ended between 1:30 am and 5 am . Upon arriving at the MRI facility , subjects rated the four selected odors in terms of intensity ( from ‘barely detectable’ to ‘extremely strong’ ) and valence ( from ‘extremely unpleasant’ to ‘extremely pleasant’ ) . Odors selected as within-sleep cues versus those not selected did not differ in perceived intensity ( t ( 17 ) = 0 . 60 , p = 0 . 56 ) or valence ( t ( 17 ) = 0 . 18 , p = 0 . 86 ) . Subjects were instructed to learn the locations of visual stimuli ( category images and scrambled images ) on a grey 4-x-4 grid while undergoing fMRI scanning . Each of the images appeared a total of three times over the course of the task , which was divided into 12 2 . 25 min runs . Each run consisted of five blocks ( animals , buildings , faces , tools , scrambled ) presented in a random order , with 12 s between blocks . Each block lasted 14 s , and consisted of a series of eight images presented on the grid for 1 s each , with 0 . 75 s between consecutive image presentations . Grid locations were balanced such that two objects per category were presented in each of the 16-grid spaces . Subjects were then led from the scanner to a testing room , where they learned to associate each of the four odors with each of the four object categories ( e . g . rose odor + building images ) . Odor-category pairs were randomly assigned for each subject . On each trial , subjects were cued to sniff upon odor presentation , and then immediately afterwards one object from each of the four categories appeared in the four different quadrants of the screen . Category objects were identical to those presented during initial learning . Subjects were instructed to select the object from the associated category as quickly and accurately as possible , and then a green box appeared around the correct choice for 2 s , as feedback . Trials were spaced at least 6 s apart , to avoid habituation and odor cross-contamination . The task continued until each odor-category pair ( e . g . rose odor + building images ) was correctly identified 16 times ( number of trials for perfect performance = 64 ) , to ensure robust odor-category associations . Subjects learned these associations rapidly ( mean = 69 . 44 , ±0 . 74 SEM , range = 66–78 trials to reach criterion ) . Subjects then continued to learn the same object locations as in the initial learning phase , with two key differences: ( 1 ) objects appeared in the presence of category-specific odors , and ( 2 ) subjects were actively tested on their knowledge of object locations . During each trial , an object appeared in the center of the screen for 0 . 5 s , and then subjects attempted to select the grid space where the object belonged ( 16 grid spaces , chance = 6 . 25% ) . Then , the object appeared in the correct grid space for 0 . 5 s , as feedback . Objects were presented in category blocks of eight objects per category during continuous presentation of the associated odor . Blocked presentation of objects allowed for efficient delivery of associated odors , as blocks were spaced 12 s apart to avoid habituation and odor cross-contamination . The task continued until subjects placed each object in the correct grid space once ( mean = 250 . 22 , ±12 . 64 SEM , range = 178–367 trials to reach criterion ) . To ensure continued attention to both odor and object category stimuli and associations , subjects performed 16 ‘catch’ trials ( four per odor-category pair ) over the course of the task . During catch trials , subjects selected the category that belonged with the presented odor , and then received feedback ( identical to the odor-category association task above ) . Subjects retained a strong knowledge of odor-category associations during catch trials ( mean = 94 . 79% correct , ±1 . 69% SEM ) . Subjects were fitted with an MRI-compatible EEG cap ( BrainCap MR , Brain Products; see sleep recording section for more details ) . While wearing the EEG cap and immediately prior to the nighttime nap , subjects were tested outside the fMRI scanner on their knowledge of object locations . During this phase , each object appeared in the center of the screen for 0 . 5 s , and subjects selected the grid space where they believed the object belonged , without receiving feedback . As in the previous phase , objects were presented in category blocks of eight objects per category during continuous presentation of the associated odor , and blocks were spaced 12 s apart . Again , subjects performed 16 catch trials ( four per odor-category pair ) over the course of this pretest session . Catch trials were identical to those of the previous task , except that subjects did not receive feedback . During catch trials , subjects continued to demonstrate excellent retention of odor-category associations ( mean = 95 . 83% , ±1 . 13% SEM ) . Subjects were instructed to relax and try to fall asleep during continuous fMRI scanning . At the start of the scan , subjects were given the option to take part in a monotonous reaction time task for approximately 5 min , to help them re-acclimate to the scanning environment . Subjects were instructed to press an MRI-compatible button each time a central crosshair changed color , but to disengage from the task if they felt very drowsy . Subjects were further instructed to press the button during the nap if they perceived an odor . During stages 2 and 3 , two of the four task-related odors were presented in alternating 16 s on/16 s off blocks ( to prevent habituation ) , as category-specific cues ( mean = 50 . 61 , ±4 . 61 SEM , range = 30–105 total odor presentations ) . To decrease the chances of subjects waking during odor presentation , experimenters waited to observe a minimum of 2 min of continuous stage 2 sleep prior to initiating odor delivery . Odors were selected as within-sleep stimuli strategically , to ensure that cueing of categories was counterbalanced across subjects ( six possible category pairs , each presented to three of the 18 subjects ) . After the nap , subjects exited the scanner and took a quick shower , to rinse EEG gel from their hair and to overcome sleep inertia . For the remainder of the main experiment , subjects were not exposed to odorants . Approximately 30 min after waking , subjects learned new grid locations for the same objects as presented in initial learning . Task structure was identical to that of initial learning , except that intervals between category blocks were limited to 4 s , since longer intervals were unnecessary in the absence of odors . Subjects were allowed a 5 min break after interference learning , and then they were tested on their knowledge of the new object locations ( without feedback ) . The interference test was identical to the pretest , except that intervals between category blocks were limited to 4 s , and there were no odors or catch trials . After a 30-min break , subjects completed a posttest to assess their knowledge of the original ( non-interference ) object locations . Task structure was identical to that of the interference test . One week after the main experiment , subjects returned for a series of follow-up memory tests . Although the follow-up visit was scheduled in advanced , subjects were not given information regarding the nature of follow-up tasks . First , subjects completed a free recall test , in which they were given two minutes per category to list as many exemplars from that category as they could remember . Next , subjects were tested on their retention of the original object locations in a task that was identical to the posttest from the main experiment . There was no difference between cued and non-cued categories for free recall ( t ( 17 ) = 0 . 20 , p = 0 . 42 ) , change in spatial recall from pretest baseline ( t ( 17 ) = 0 . 17 , p = 0 . 43 ) , or RTs ( t ( 17 ) = 0 . 46 , p = 0 . 65 ) during follow-up memory tests . Finally , subjects were verbally instructed to sniff each odorant , and to recount which category it was paired with during the main experiment . Seventeen of 18 subjects could correctly recall odor-category associations . During the sleep session , EEG data were collected with an MRI-compatible EEG system ( BrainAmp MR Plus , Brain Products ) in order to restrict odor delivery to stages 2 and 3 of sleep . The 32-channel EEG cap contained 26 scalp electrodes and two electrooculography ( EOG ) electrodes , and wires to connect electrodes for chin electromyography ( EMG ) and electrocardiography ( ECG ) . EEG data were sampled at 5 kHz , scanner and cardioballistic artifacts were removed online using Brain Products software ( Rec View ) , and data were scored in accordance with standard criteria ( Silber et al . , 2007 ) ( Table 1 ) . To compare power spectral density of EEG data across conditions ( i . e . , odor-on versus odor-off; stage 3 sleep versus stage 1 sleep ) , we employed a fast Fourier transform analysis of frontal electrode ‘Fpz’ using the ‘pwelch’ function in Matlab . MRI data were collected with a 3-Tesla scanner ( Siemens PRISMA ) equipped with a 64-channel head coil , using T2-weighted echoplanar imaging . Each volume comprised 40 slices covering the whole brain ( field of view , 210-x-203 mm; matrix size , 124-x-120 voxels; slice thickness , 3 mm; in-plane resolution , 1 . 69-x-1 . 69 mm; repetition time , 2500 ms; echo time , 25 ms; flip angle , 80° ) . An additional whole-brain anatomical T1-weighted MRI scan was acquired for coregistration purposes ( GRAPPA; voxel size , 0 . 8 mm³ ) . MRI data were preprocessed and analyzed using SPM12 ( http://www . fil . ion . ucl . ac . uk/spm/ ) . Functional images were realigned to the mean of the images , motion corrected , and coregistered to the T1-weighted image . For univariate and PPI analyses , images were normalized and then spatially smoothed with a 6 mm Gaussian kernel . Multivariate searchlight analyses were conducted in each subject’s native space , and images were minimally smoothed with a 2 mm Gaussian kernel . Searchlight analyses were restricted to grey matter voxels , by generating a grey matter mask from the SPM12 tissue probability map , and then warping that mask to the subject’s native space using the transformation parameters from the standard T1 template to the subject’s individual T1-weighted image . To identify category-sensitive voxels , a GLM was constructed for each subject from initial learning scans , where category images were modeled as five separate regressors of interest ( animals , buildings , faces , tools , scrambled ) in a blocked design . Nuisance regressors included the six motion parameters generated from realignment , and beta estimates were calculated for each condition . To quantify pattern discrimination of the four categories during initial learning , a whole-brain searchlight-based correlation analysis was implemented . At each search sphere ( radius = 3 . 7 voxels ) , 48 beta pattern vectors were extracted ( four object categories X 12 runs ) . In a leave-one-out approach , beta patterns from 11 ‘training’ runs were averaged , and compared to beta patterns from the remaining ‘test’ run . In each iteration , the mean across conditions was first subtracted from training and test beta patterns separately . This was followed by calculation of the linear correlations between all training and test patterns , and then the resulting correlation coefficients were Fisher’s Z transformed . This procedure was repeated a total of 12 times , so that each run could be left out in turn as a test run . The resulting values were averaged across iterations . As an index of pattern discrimination , within-category correlations ( training patterns versus the category-congruent test pattern ) were compared to the average of between-category correlations ( training patterns versus the three category-incongruent test patterns ) . This procedure was repeated at each search sphere , and the resulting pattern discrimination maps were normalized and then smoothed with a 6-mm Gaussian kernel prior to group comparison . Category-sensitive voxels were designated at the group-level , from the contrast within-category correlations > between-category correlations ( p < 0 . 001 ) . Critically , the next analysis was restricted to brain regions that demonstrated category-selectivity during learning ( see below ) , as this was a prerequisite to identifying reactivation of the same category information during sleep . In a follow-up analysis to determine the selectivity for each object category separately , the same steps were repeated with one exception . Rather than collapsing indices of pattern discrimination across categories , within-category correlations were compared to the average of between-category correlations for each object category separately . This resulted in four group-level pattern selectivity maps , one per object category . To determine whether category information re-emerged in response to odor cues during sleep , the same GLM that was used for selection of category-specific voxels ( see previous section ) was used to construct templates of the four object categories . An additional sleep-based GLM was constructed for each subject , where each of the two odor cues were modeled as separate regressors of interest . Odor onset times were adjusted to align with the first point of inhalation ( i . e . rising slope on the breathing trace ) after odor presentation . To minimize signal contributions induced by head motion during the sleep scan , all volumes prior to the first odor presentation and following the last odor presentation were discarded ( with the exception of six additional volumes on either end ) . Given the large number of volumes remaining ( mean = 948 . 61 , range = 401–1416 ) , there were sufficient degrees of freedom in our GLM to incorporate extra nuisance regressors without overfitting . This allowed us to account for head motion more rigorously given the long duration of the sleep session . ( In contrast , this more rigorous motion correction was not possible for the wake sessions without overfitting , due to the smaller number of fMRI volumes ( 58 per session ) , and thus fewer degrees of freedom . ) Nuisance regressors included the six motion parameters generated from realignment , and their squares , derivatives , and squared derivatives ( 24 total ) . To account for within-scan motion , the signal difference between even and odd slices , the within-volume variance across slices , and derivatives of both parameters were also included as nuisance regressors . Additional nuisance regressors were included when necessary , to capture individual volumes demonstrating excessive head motion . Finally , the respiration trace was post-processed and down sampled to the scanner repetition time frequency to be included as a nuisance regressor . Beta estimates were calculated for each of the two odor conditions . To search for pattern-based reactivation during sleep , a whole-brain searchlight-based correlation analysis was implemented ( as in voxel selection ) . At each searchlight sphere , template images were constructed from the initial learning data for each of the four categories by averaging the extracted beta pattern vectors across runs for each condition ( category ) separately , and then subtracting the mean across conditions from each voxel . Next , two beta pattern vectors were extracted from the sleep period , each corresponding to one of the two odor cues ( mean activity was not subtracted from each voxel here , since doing so across the two cues would have created two anticorrelated vectors , essentially reducing two data points to a single data point ) . Correlation coefficients were calculated between those odor-cued activity pattern vectors and each of the four category templates separately , and the resulting r values were converted to Fisher’s Z scores . As an index of memory reactivation , the cued correlation ( sleep activity pattern versus cued learning template ) was compared to the average of non-cued correlations ( sleep activity pattern versus the three non-cued learning templates ) for each beta pattern vector separately , and then the resulting values were averaged across the two conditions . This procedure was repeated at each search sphere , and the resulting memory reactivation maps were normalized and then smoothed with a 6-mm Gaussian kernel prior to group comparison . A group-level correlation analysis was conducted , in which cued memory benefit was included as a covariate of interest , and category-selective voxels ( identified during prior learning ) were designated as an explicit mask . Cued memory benefit was defined as visuospatial memory performance at posttest ( expressed as a percentage of items remembered from pretest baseline ) for cued minus non-cued categories . In a follow-up analysis , the same steps were repeated for the two reactivated categories separately ( without considering the two non-reactivated categories ) . Memory reactivation maps were constructed , and values were extracted from clusters of interest in vmPFC and fusiform cortex ( identified from the multivoxel correlation analysis , p < 0 . 001 ) , and then averaged . The resulting values were correlated with posttest recall accuracy ( again , expressed as a percentage of items remembered from pretest baseline ) for the two reactivated categories individually . To reveal the time course of reactivation , a separate sleep-based finite impulse response ( FIR ) model was constructed . Ten regressors of interest , spaced apart at 2 . 5 s intervals , were included for each of the two odorants , spanning a 25 s time window . The regressors modeled responses starting three volumes prior to odor onset , a single volume aligned to odor onset , and six volumes following odor onset . The model also included the same nuisance regressors described above . For regressors corresponding to each time point , beta estimates were extracted for each of the two odors . Next , a searchlight analysis was conducted ( as in the main analysis ) , resulting in ten separate memory reactivation maps ( one per time point ) . Again , values were extracted from the resulting reactivation maps for both vmPFC and fusiform cortex clusters , and then correlated with cued memory benefit at each respective time point . Contrary to prediction , searchlight analyses did not reveal a relationship between memory reactivation in the hippocampus and memory performance . To further investigate effects in this brain area , all three analyses outlined above were repeated for an ROI in bilateral hippocampus taken from the AAL Atlas ( as an alternative to the whole-brain searchlight approach ) . Finally , two additional analyses were conducted to elucidate the main effect of within-sleep odors on reactivation of associated object categories , without considering cued memory benefit as a covariate . The first analysis was derived from the same model that included cued memory benefit as a covariate of interest , but here the main effect was considered instead of its relationship with memory performance . Again , results were masked according to category-selective voxels identified from the initial learning phase . A second analysis was conducted to determine whether there were spatial differences in reactivation maps for the four object categories . Here , instead of considering all four object categories across all 18 subjects , four separate second-level models were constructed , and each included reactivation maps for a single object category . Because each of the 18 subjects received two odor cues during sleep , this meant that each category was only cued for nine subjects . Thus , a total of nine reactivation maps were included in each of the four second-level models . Given the limited number of subjects included in each model , the resulting category maps were not constrained by a mask . In an additional analysis , we aimed to identify brain regions activated by odor cues during sleep . The GLM used to identify voxels activated by odor cues was identical to the GLM constructed from the sleep period for the multivoxel analysis described previously , except that the model was applied to images that had been normalized and smoothed with a 6 mm Gaussian kernel . Individual contrast maps for odor onset were computed at the subject level and tested at the group level using a one-sample t test . The gPPI toolbox ( McLaren et al . , 2012 ) was employed to measure connectivity between hippocampus and vmPFC during odor presentation . We estimated a PPI model for each subject using normalized , smoothed images . The physiological factor was defined as fMRI activity in a seed region ( 5-mm sphere surrounding the peak voxel in amygdala/hippocampus identified from the univariate analysis ) , and odor onset was the psychological factor . Voxel-wise connectivity parameters at odor onset were extracted from the same vmPFC and posterior fusiform clusters as previously and averaged across voxels for each subject individually , and then a one-sample t test was implemented for group-level comparison . To test our a priori prediction that memory retention for cued information would be enhanced compared to non-cued information , as in previous cueing studies ( Oudiette and Paller , 2013; Rasch et al . , 2007; Rudoy et al . , 2009; Schouten et al . , 2017; Shanahan and Gottfried , 2017; Spiers and Bendor , 2014 ) , we used a one-tailed Wilcoxon signed-rank test and a binomial test . To analyze RTs from the interference test , trial-by-trial RTs were log transformed , and compared for cued versus non-cued category objects across subjects using a two-tailed paired t test . To determine the effects of odor cues ( odor-on versus odor-off ) and sleep stage ( stage 3 versus stage 1 ) on EEG power spectral density , we performed two-factor ( condition-x-frequency ) repeated-measures ANOVAs ( two-tailed ) . To identify brain regions containing category information , and where reactivation of category information was correlated with cued memory benefit , we conducted one-tailed t tests on normalized and smoothed searchlight maps . p-Values were defined at the peak voxel , and adjusted for the family-wise error rate based on the explicit initial learning mask ( i . e . voxels that contained category information at initial learning ) . In follow-up tests , reactivation values were extracted from searchlight maps for each subject , and one-tailed linear correlations with behavioral recall measures were computed . In the univariate analysis used to identify voxels activated during odor presentation , we again used a one-tailed paired t test of normalized and smoothed contrast maps , and p-values were defined at the peak voxel , corrected for the family-wise error rate based on an olfactory region of interest in left amygdala from the AAL Atlas . Finally , for the PPI analysis , one-tailed t tests were used to compare connectivity values extracted from a cluster of interest in vmPFC to zero .	It may not always feel like it , but we encounter an incredible volume of new information every day . We experience so much that it is not feasible to remember every detail . The brain's process for reorganizing memories – keeping some secure and discarding others – is known as memory consolidation . There are ways of directing consolidation toward certain memories . One of them is to associate a memory – such as an arrangement of objects – with a particular smell . If this odor is then wafted at the person when they sleep , they are better at recalling the associated memory the next day . The neural mechanisms in the brain that support this process are largely unknown . Researchers want to find out exactly how odor cues can alter brain activity while participants are asleep to allow for better recall on awakening . Shanahan et al . used fMRI scans to see how an odor affects the sleeping brain . First , the participants learned the locations of several objects on a four by four grid – including animals , faces , buildings and tools – and then learned to associate each category with a different background odor . Then , the volunteers took a nighttime nap inside the MRI scanner , and were exposed to two of the odors in their sleep . The next morning , they better remembered the locations of the objects from the two categories associated with the odor cues delivered in sleep . Analyses of the brain scans revealed that the extent to which odors reactivated the category information in a part of the brain called the ventromedial prefrontal cortex was predictive of how successful memory recall was after sleep . This brain region is involved in retrieving old memories . Memory disorders are an ever-increasing problem as the average life-span continues to rise . Reliable treatments to slow or prevent memory decline in patients with conditions such as Alzheimer's are still unavailable . Using odor cues during sleep could be one way to enhance memories in patients with memory loss and dementia , but also in healthy individuals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Odor-evoked category reactivation in human ventromedial prefrontal cortex during sleep promotes memory consolidation
About 300 loss-of-function mutations in the IKs channel have been identified in patients with Long QT syndrome and cardiac arrhythmia . How specific mutations cause arrhythmia is largely unknown and there are no approved IKs channel activators for treatment of these arrhythmias . We find that several Long QT syndrome-associated IKs channel mutations shift channel voltage dependence and accelerate channel closing . Voltage-clamp fluorometry experiments and kinetic modeling suggest that similar mutation-induced alterations in IKs channel currents may be caused by different molecular mechanisms . Finally , we find that the fatty acid analogue N-arachidonoyl taurine restores channel gating of many different mutant channels , even though the mutations are in different domains of the IKs channel and affect the channel by different molecular mechanisms . N-arachidonoyl taurine is therefore an interesting prototype compound that may inspire development of future IKs channel activators to treat Long QT syndrome caused by diverse IKs channel mutations . Long QT syndrome ( LQTS ) is a condition of the heart which in most cases is caused by a mutation in cardiac ion channels ( Hedley et al . , 2009; Morita et al . , 2008 ) . In LQTS , the action potential of the heart is prolonged , which is observed as a prolonged QT interval in the electrocardiogram . LQTS patients have an increased risk of developing ventricular tachyarrhythmias called torsades de pointes when exposed to triggers such as adrenergic stress ( Morita et al . , 2008; Cerrone et al . , 2012 ) . These arrhythmias can cause palpitation , syncope or sudden death due to ventricular fibrillation . To improve the clinical outcome of LQTS patients , it is therefore critical to prevent these LQTS-induced life-threatening arrhythmias . Most mutations causing LQTS are located in the KCNQ1 gene ( Hedley et al . , 2009 ) . KCNQ1 codes for the potassium channel KV7 . 1 , which in the heart co-assembles with the beta-subunit KCNE1 to form the slowly-activating , voltage-dependent potassium channel IKs ( Barhanin et al . , 1996; Sanguinetti et al . , 1996 ) . The IKs channel provides one of the important delayed rectifier outward potassium currents that repolarizes the cardiomyocyte and terminates the cardiac action potential ( Nerbonne and Kass , 2005 ) . Reduced IKs function therefore tends to delay cardiomyocyte repolarization , thereby causing prolonged cardiac action potential durations and a prolonged QT interval . The cardiac IKs channel consists of four KV7 . 1 subunits and two to four KCNE1 subunits ( Nakajo et al . , 2010; Plant et al . , 2014; Murray et al . , 2016 ) . Throughout this work , we will refer to the IKs channel as KV7 . 1+KCNE1 . KV7 . 1 has six transmembrane segments named S1-S6 ( Liin et al . , 2015 ) ( Figure 1a ) . S1-S4 of each KV7 . 1 subunit forms a voltage-sensing domain where S4 is the voltage sensor with three positive gating charges . S5 and S6 from all four KV7 . 1 subunits form the pore domain with a putative gate in S6 that needs to move to open the ion-conducting pore of the channel . KCNE1 has a single-transmembrane segment ( Figure 1a ) and is proposed to be localized in the otherwise lipid-filled space between two voltage-sensing domains of neighbouring KV7 . 1 subunits ( Nakajo and Kubo , 2015 ) . Upon cardiomyocyte depolarization , the voltage sensor of KV7 . 1 moves outward in relation to the membrane . It has been proposed that this movement of the voltage sensor is transferred to the pore domain via the S4-S5 linker and induces channel opening by moving the S6 gate ( Liin et al . , 2015 ) . 10 . 7554/eLife . 20272 . 003Figure 1 . Biophysical properties of LQTS and LQTS-like KV7 . 1+KCNE1 channel mutants expressed in Xenopus oocytes . ( a ) Topology of KV7 . 1 and KCNE1 , and position of tested LQTS and LQTS-like mutants . ( b ) G ( V ) midpoints ( V50 ) from the Boltzmann fits for mutants co-expressed with KCNE1 . n = 5–11 . Data as mean ± SEM . The statistics represent one-way ANOVA with Dunnett’s Multiple Comparison Test to compare the mutants to wild-type KV7 . 1+KCNE1; p<0 . 01; ns is p≥0 . 05 . # denotes lowest estimate . Dashed line denotes wild-type V50 . ( c ) Representative example of KV7 . 1/S225L+KCNE1 G ( V ) ( black line and symbols ) compared to wild-type KV7 . 1+KCNE1 ( blue line and symbols , mean ± SEM , n = 5 ) . ( d–e ) Representative example of KV7 . 1/S225L+KCNE1 opening kinetics and KV7 . 1+KCNE1/K70N closing kinetics ( black lines ) compared to wild-type KV7 . 1+KCNE1 ( blue lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 00310 . 7554/eLife . 20272 . 004Figure 1—figure supplement 1 . KV7 . 1/F351S mutant expressed in Xenopus oocytes . The KV7 . 1/F351S mutant does not generate currents when expressed in Xenopus oocytes . The holding voltage is –80 mV , and test voltages range between –80 and +60 mV for 3 s in 10 mV increments . The tail voltage is –20 mV . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 00410 . 7554/eLife . 20272 . 005Figure 1—figure supplement 2 . V50 of LQTS and LQTS-like KV7 . 1 mutants expressed in Xenopus oocytes . G ( V ) midpoints ( V50 ) for LQTS and LQTS-like mutants without co-expression of KCNE1 . Mean ± SEM . n = 5–12 . The statistics represent one-way ANOVA with Dunnett’s Multiple Comparison Test to compare V50 of mutant to V50 of wild-type KV7 . 1; p<0 . 01; ns p≥0 . 05 . Dashed line denotes wild-type V50 . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 00510 . 7554/eLife . 20272 . 006Figure 1—figure supplement 3 . KV7 . 1/R583C mutant expressed in Xenopus oocytes . ( a ) The KV7 . 1/R583C mutant generates currents that inactivate at positive voltages . The holding voltage is –80 mV , and test voltages range between –80 and +40 mV for 3 s in 20 mV increments . The tail voltage is –20 mV . Tail currents are measured at the arrow . Inset: representative current trace at +40 mV for wild-type KV7 . 1 . ( b ) Representative example of G ( V ) curves generated using the protocol in panel a ( filled circles ) or a triple pulse protocol ( open circles ) with a brief hyperpolarizing pulse ( –140 mV for 20 ms ) between the test pulse and the tail pulse to release a fraction of channels from inactivation . The triple pulse protocol generates a G ( V ) that is shifted ~9 mV towards positive voltages ( V50 = ~ –39 mV for the regular protocol and –30 mV for the triple pulse protocol ) , which matches the G ( V ) of the wild-type KV7 . 1 fairly well ( V50 = –29 . 4 mV ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 00610 . 7554/eLife . 20272 . 007Figure 1—figure supplement 4 . Comparison of current amplitude of wild-type KV7 . 1+KCNE1 and LQTS and LQTS-like mutants when expressed in Xenopus oocytes . KV7 . 1 and KCNE1 were co-injected in oocytes for homozygous ( a ) and heterozygous ( b ) expression , as described in Materials and Methods . Current were recorded after two days of incubation at 16°C . The holding voltage is –80 mV , and test voltages range between 0 and +60 mV for 5 s in 20 mV increments . The tail voltage is –20 mV . Current amplitudes at the end of the 5 s test pulse are normalized to the wild-type KV7 . 1+KCNE1 current amplitude at +60 mV recorded in the same batch of oocytes . Dashed line in ( a ) is the wild-type curve shifted +25 mV . ( c–d ) Detailed comparison of current amplitudes at +20 mV ( c ) and +40 mV ( d ) . Mutant current amplitudes are normalized to the wild-type KV7 . 1+KCNE1 current amplitude at indicated voltage . Dashed lines denote relative wild-type KV7 . 1+KCNE1 current amplitude ( =1 ) . The statistics represent one-way ANOVA with Dunnett’s Multiple Comparison Test to compare the current amplitude of mutants to wild-type current amplitudes . p<0 . 05; p<0 . 01; p<0 . 001; ns is p≥0 . 05 . Mean ± SEM . n = 4–12 . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 007 Altogether , about 300 mutations in KCNQ1 and KCNE1 have been identified in patients suffering from LQTS ( Hedley et al . , 2009 ) ( http://www . fsm . it/cardmoc/ ) . These mutations are distributed throughout the channel sequence and are therefore likely to cause channel dysfunction by different mechanisms , which are , however , largely unknown . Potential mechanisms for KV7 . 1+KCNE1 channel loss of function by a mutation could , for example , be interference with voltage sensor movement , gate opening , or membrane expression . LQTS is today treated with drugs that prevent the triggering of arrhythmic activity , such as beta-blockers , or with arrhythmia-terminating implantable cardioverter defibrillator ( Hedley et al . , 2009 ) . A different treatment strategy for LQTS caused by loss-of-function mutations in the KV7 . 1+KCNE1 channel would be to pharmacologically augment the KV7 . 1+KCNE1 channel function of these LQTS mutants , thereby shortening the prolonged QT interval and lower the risk of arrhythmia development . However , there is currently no clinically approved KV7 . 1+KCNE1 channel activator . In this study , we investigate the biophysical properties and potential mechanism of action of LQTS-associated KV7 . 1+KCNE1 channel mutations and test the ability of the fatty acid analogue N-arachidonoyl taurine ( N-AT ) to restore the function of these mutants . We selected eight mutations of residues mutated in patients with LQTS located in different segments of the KV7 . 1+KCNE1 channel and that were previously shown to form active channels ( Bianchi et al . , 2000; Yamaguchi et al . , 2003; Eldstrom et al . , 2010; Henrion et al . , 2009; Yang et al . , 2013; Yang et al . , 2002; Harmer et al . , 2010; Splawski et al . , 1997 ) . We measure the movement of the S4 voltage sensor in selected mutants using voltage clamp fluorometry to further our understanding of the molecular mechanisms underlying the defects caused by the diverse mutations . We find that the eight LQTS-associated mutations affect the voltage dependence and/or closing kinetics , in some cases by different molecular mechanisms . Moreover , we find that N-AT restores much of the channel activity in these eight LQTS-associated KV7 . 1+KCNE1 mutants . This suggests that N-AT may function as a general activator of KV7 . 1+KCNE1 channels with diverse mutational defects . We first study the biophysical properties of six point mutations in KV7 . 1 ( F193L , V215M , S225L , L251P , F351S , R583C ) , and two in KCNE1 ( K70N , S74L ) identified in patients with LQTS ( Yamaguchi et al . , 2003; Yang et al . , 2002; Splawski et al . , 1997; Priori et al . , 1999; Napolitano et al . , 2005; Lai et al . , 2005 ) ( Figure 1a ) . As L251P and F351S did not produce functional channels ( Napolitano et al . , 2005; Deschenes et al . , 2003 ) ( Figure 1—figure supplement 1 ) , we engineered the milder L251A and F351A mutants instead . L251A and F351A will be referred to as 'LQTS-like mutants' . When expressed alone in Xenopus oocytes , all investigated KV7 . 1 mutants , except F193L and V215M , display a shifted conductance versus voltage curve ( G ( V ) ) compared to the wild-type KV7 . 1 channel ( Figure 1—figure supplement 2; Supplementary file 1 ) . S225L , L251A and F351A shift the G ( V ) towards positive voltages compared to wild-type KV7 . 1 . In contrast , R583C shifts the half-maximal activation , V50 , ~10 mV towards negative voltages compared to wild-type KV7 . 1 . This apparent negative shift is likely caused by the pronounced inactivation of the R583C mutant ( Figure 1—figure supplement 3a ) , which is seen to a considerable smaller extent in the other KV7 . 1 mutants and wild-type KV7 . 1 ( inset in Figure 1—figure supplement 3a ) . When a fraction of the channels are released from inactivation , by introducing a brief hyperpolarizing pulse between the test pulse and the tail pulse , R583C has a V50 fairly comparable to wild-type KV7 . 1 ( Figure 1—figure supplement 3b ) . When the KV7 . 1 mutants are co-expressed with KCNE1 , all KV7 . 1 and KCNE1 mutants except KV7 . 1/F193L+KCNE1 have a G ( V ) that is shifted towards positive voltages compared to the wild-type KV7 . 1+KCNE1 channel ( Figure 1b ) . KV7 . 1/F351A causes the most dramatic change by shifting V50 more than +30 mV . We are therefore only able to record the foot of the G ( V ) curve of KV7 . 1/F351A+KCNE1 , and a shift in V50 of +30 mV is a lower estimate of the change in V50 ( ΔV50 ) . One of the other mutants with dramatically shifted G ( V ) is KV7 . 1/S225L+KCNE1 . V50 for KV7 . 1/S225L+KCNE1 is shifted almost +30 mV compared to wild-type KV7 . 1+KCNE1 ( Figure 1c; Supplementary file 1 ) . S225L also slows down KV7 . 1+KCNE1 channel opening kinetics ( p<0 . 01; Figure 1d; Supplementary file 1 ) . All mutations , except for L251A , accelerate channel closing kinetics compared to wild-type KV7 . 1+KCNE1 ( Supplementary file 1 ) . K70N has the most dramatic effect on KV7 . 1+KCNE1 channel closing by accelerating the closing kinetics by approximately a factor of 5 ( Figure 1e; Supplementary file 1 ) . When comparing the amplitude of K+ currents generated by these mutants with the current amplitude of the wild-type KV7 . 1+KCNE1 channel in the same batch of oocytes , we note that all mutants generate smaller currents than wild-type over a large voltage range ( Figure 1—figure supplement 4 ) . Although defective trafficking may contribute to these reduced currents in Xenopus oocytes , the current amplitudes for most mutants matches fairly well with the predicted current amplitude from channels with G ( V ) curves shifted towards positive voltages as observed for these mutants ( Figure 1—figure supplement 4a ) , suggesting that the reduced current amplitudes in Xenopus oocytes are mainly a result of gating defects ( and not trafficking defects ) . To summarize , all mutations change channel function by altering the voltage dependence of opening and/or the kinetics of opening and/or closing . Reduced function of the KV7 . 1+KCNE1 channel induced by these LQTS and LQTS-like mutations may largely be explained by the right-shifted G ( V ) and the faster closing kinetics caused by these mutations . F193L does not alter the G ( V ) , but speeds up KV7 . 1+KCNE1 channel closing by a factor of 2 ( Supplementary file 1 ) . These results are consistent with previous reported findings for some of these mutants ( Bianchi et al . , 2000; Yamaguchi et al . , 2003; Eldstrom et al . , 2010; Henrion et al . , 2009; Yang et al . , 2013; Yang et al . , 2002; Harmer et al . , 2010 ) . Patients with LQTS mutations can be either homozygous or heterozygous for the mutation . To mimic heterozygous expression , we co-inject the mutated KV7 . 1 subunit and KCNE1 subunit together with the wild-type KV7 . 1 subunit ( or wild-type KCNE1 subunit for KCNE1 mutants ) ( cartoon in Figure 2 ) . We refer to this as heterozygous expression . Figure 2a–b compares the homozygous expression ( KV7 . 1wt+KCNE1mut or KV7 . 1mut+KCNE1wt ) with heterozygous expression ( KV7 . 1wt+KV7 . 1mut+KCNE1wt or KV7 . 1wt+KCNE1wt+KCNE1mut ) for KV7 . 1/S225L ( Figure 2a ) and KCNE1/K70N ( Figure 2b ) . Both of these examples show that heterozygous expression generates channels with more wild-type like opening or closing kinetics and G ( V ) compared to homozygous expression of the mutant subunit . A milder biophysical phenotype upon heterozygous expression is generally seen for the LQTS and LQTS-like mutants in terms of G ( V ) , current amplitude , and/or closing kinetics ( Figure 2c–d , Figure 1—figure supplement 4 , Supplementary file 2 ) . This milder phenotype indicates that the wild-type subunit can partly restore KV7 . 1+KCNE1 function . Alternatively , for mutants with a G ( V ) that is very shifted to positive voltages ( e . g . F351A ) , it may be that channel complexes that contain the mutated subunits are largely out of the physiological voltage range and therefore do not contribute substantially to the recorded current . Also , for mutants with low membrane expression ( e . g . possibly F193L [Yamaguchi et al . , 2003] ) , it may be that channels containing the wild-type subunit are favoured so that in most KV7 . 1+KCNE1 channel complexes the majority ( or all ) of the subunits will be wild-type subunits . 10 . 7554/eLife . 20272 . 008Figure 2 . Comparison of homozygous and heterozygous expression of LQTS and LQTS-like mutants . ( a–b ) Representative example of kinetics ( middle panel ) and G ( V ) ( right panel ) for homozygous expression and heterozygous expression of S225L ( a ) and K70N ( b ) . Currents in response to steps from –80 mV to +40 mV ( a , middle pane ) and from +40 mV to –20 mV ( b , middle panel ) . Homozygous expression ( black ) , heterozygous expression ( gray ) , and KV7 . 1+KCNE1 wild-type ( blue ) . n = 7–13 . ( c–d ) Summary of V50 ( c ) and T50 for closing ( d ) for homozygous and heterozygous expression . Data as mean ± SEM . n = 5–13 . The statistics represent one-way ANOVA with pair-wise Bonferroni’s Test to compare homozygous and heterozygous expression; p<0 . 01; **p<0 . 001; ns is p≥0 . 05 . # denotes lowest estimate . Not determined ( nd ) . The statistics was not calculated for F351A . Dashed lines denote corresponding values for wild-type KV7 . 1+KCNE1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 008 Although most of the mutations shift channel voltage dependence and affect channel closing kinetics , the underlying mechanism of mutation-induced changes in KV7 . 1+KCNE1 channel function is most likely different for different mutations . For instance , mutations located in S5 and S6 ( e . g . F351A ) may mainly affect gate movement , while mutations in S1–S4 ( e . g . S225L ) are more likely to affect voltage sensor movement . To explore whether different mutations interfere with different gating transitions , we use voltage clamp fluorometry , in which the movement of the voltage sensor in KV7 . 1 can be tracked by the fluorescence change from the fluorescent probe Alexa-488-maleimide attached to G219C in the S3-S4 loop ( referred to as G219C ) ( Barro-Soria et al . , 2014; Osteen et al . , 2010; Osteen et al . , 2012 ) . Voltage sensor movement ( measured by fluorescence ) and gate movement ( measured by ionic currents ) are then monitored under two-electrode voltage clamp . The KV7 . 1/G219C* construct by itself or co-expressed with KCNE1 gives voltage-dependent fluorescence changes ( Figure 3a ) . As previously reported , the fluorescence versus voltage ( F ( V ) ) curve of KV7 . 1/G219C* correlates well with the G ( V ) curve ( Figure 3a , left panel ) , while the F ( V ) curve of KV7 . 1/G219C+KCNE1 is divided into two components ( Figure 3a , right panel ) ( Barro-Soria et al . , 2014; Osteen et al . , 2010; Osteen et al . , 2012 ) . For KV7 . 1/G219C+KCNE1 , the first fluorescence component ( F1 ) has been suggested to represent the main voltage sensor movement and the second fluorescence component ( F2 ) to be correlated with gate opening ( Barro-Soria et al . , 2014 ) . We introduce G219C into KV7 . 1/S225L and KV7 . 1/F351A . The G ( V ) curves of both KV7 . 1/G219C/S225L and KV7 . 1/G219C/F351A are shifted towards more positive voltages compared to the wild-type channel , but the F ( V ) curves are differentially affected by the two mutations ( Figure 3b–c , left panels ) . For KV7 . 1/G219C/S225L , the F ( V ) curve is shifted to a similar extent as the G ( V ) curve , while for KV7 . 1/G219C/F351A , the F ( V ) curve is shifted to a considerably smaller extent ( Osteen et al . , 2010 ) . When these mutants are co-expressed with KCNE1 , we observe different effects on the voltage dependence of the two fluorescent components F1 and F2 induced by the mutations . The S225L mutation primarily shifts F1 towards positive voltages so that F1 and F2 of KV7 . 1/G219C/S225L+KCNE1 are hardly distinguishable in the F ( V ) curve ( Figure 3b , right panel ) . In contrast , the F351A mutation primarily shifts F2 towards positive voltages so that F1 and F2 are clearly separated ( Figure 3c , right panel ) . Thus , S225L and F351A seem to shift the G ( V ) curve of KV7 . 1+KCNE1 towards positive voltages by interfering with different gating transitions . 10 . 7554/eLife . 20272 . 009Figure 3 . Voltage-clamp fluorometry recordings of wild-type and mutated KV7 . 1+KCNE1 channels . ( a-c ) Representative fluorescence traces and mean F ( V ) /G ( V ) curves for KV7 . 1/G219C ( a ) , S225L ( b ) , and F351A ( c ) . Left panels without KCNE1 and right panels with KCNE1 . The holding voltage is –80 mV , the pre-pulse –120 mV for 2 s ( left panels ) and –160 mV for 5 s ( right panels ) , and test voltages between –140 and +80 mV for 3 s ( left panels ) and between –160 and +80 mV for 5 s ( right panels ) in 20 mV increments . The tail voltage is –80 mV ( left panels ) and −40 mV ( right panels ) . For KV7 . 1/G219C/F351A+KCNE1 , the pre-pulse is –120 mV for 3 s , and test voltages ranging between –160 and +100 mV . The bottom of the fit of the KV7 . 1/G219C/S225L+KCNE1 F ( V ) curve ( which saturates fairly well at negative voltages ) is set to 0 in the normalized F ( V ) curves in the right panels . The F1 amplitude of KV7 . 1/G219C/F351A+KCNE1 is normalized to the F1 amplitude of wild-type . Data as mean ± SEM . n = 4–14 . The dashed lines in ( b ) and ( c ) denote F ( V ) ( red ) and G ( V ) ( black ) for wild-type ( from a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 00910 . 7554/eLife . 20272 . 010Figure 3—figure supplement 1 . Kinetic models for KV7 . 1 and KV7 . 1+KCNE1 channel gating . ( a ) A 10-state allosteric gating scheme for KV7 . 1 channels . Horizontal transitions represent independent S4 movements that increase the fluorescence to an intermediate level ( which generates the F1 component ) . The vertical transition represents concerted channel opening with a concomitant additional fluorescence increase ( which generates the F2 component ) . Cartoon shows KV7 . 1 channel labeled with a fluorophore on S3-S4 with all four voltage sensors in the resting state ( C0 ) , with one ( C1 ) , or four ( C4 ) voltage sensor activated in the closed channel ( top ) or with all four voltage sensors in the resting state ( O0 ) , with one ( O1 ) , or four ( O4 ) voltage sensor activated with the channel opened ( bottom ) . ( b ) A 6-state allosteric gating scheme for KV7 . 1+KCNE1 channels . Horizontal transitions represent independent S4 movements that increase the fluorescence to an intermediate level ( which generates the F1 component ) . The vertical transition represents concerted channel opening with a concomitant additional fluorescence increase ( which generates the F2 component ) . Cartoon shows KV7 . 1 channel labeled with a fluorophore on S3-S4 with all four voltage sensors in the resting state ( C0 ) , with one ( C1 ) , or four ( C4 ) voltage sensor activated without channel opening ( top ) that is followed by a concerted conformational change of all four S4s associated with channel opening ( O4 ) ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 01010 . 7554/eLife . 20272 . 011Figure 3—figure supplement 2 . Simulations of wild-type and mutant KV7 . 1 and KV7 . 1+KCNE1 channels reproduce currents and fluorescence . Simulated G ( V ) ( black ) and F ( V ) ( red ) curves for ( a ) wild-type , ( b ) S225L , and ( c ) F351A KV7 . 1 ( left ) and KV7 . 1+KCNE1 ( right ) channels using the KV7 . 1 and KV7 . 1+KCNE1 models in Figure 3—figure supplement 1 . Parameters for the wild-type models were determined in earlier studies ( see Supplementary file 4 for all rate constants ) . Current and fluorescence traces were simulated using Berkeley Madonna ( Berkeley , CA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 01110 . 7554/eLife . 20272 . 012Figure 3—figure supplement 3 . Voltage-clamp fluorometry recordings of the KV7 . 1/G219C/F351L mutant with and without KCNE1 co-expressed . Mean F ( V ) /G ( V ) curves for KV7 . 1/G219C/F351L ( mean ± SEM ) are shown together with corresponding mean F ( V ) /G ( V ) curves for WT KV7 . 1/G219C ( blue lines ) and KV7 . 1/G219C/F351A ( dashed red/black lines ) . Experiments are performed and data normalized as described in Figure 3 . Note that all data presented in this graph are done on constructs with a Kv7 . 1/C214A/C331A background ( Barro-Soria et al . , 2014; Barro-Soria and Perez , 2015 ) . The F ( V ) /G ( V ) curves are therefore shifted towards negative voltages compared the data presented in Figure 3 ( which are done in WT background ) . n = 4–6 . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 012 To further explore the different effects of S225L and F351A in the voltage-clamp fluorometry experiments , we use two kinetic models previously developed to reproduce the currents and fluorescence from KV7 . 1/G219C ( Osteen et al . , 2012 ) and KV7 . 1/G219C+KCNE1 channels ( Barro-Soria et al . , 2014 ) , respectively . The KV7 . 1/G219C model is an allosteric model with 10 states ( Figure 3—figure supplement 1a ) , where the horizontal transition is the main S4 movement ( which generates the main fluorescence component F1 ) and the vertical transition is channel opening accompanied by an additional smaller S4 movement ( that generates a smaller additional fluorescence component F2 ) ( Osteen et al . , 2012; Zaydman et al . , 2014 ) . The KV7 . 1/G219C* model allows for channel opening after only a subset of four S4s are activated , which thereby generates F ( V ) and G ( V ) that are close in the voltage dependence ( reference ( Osteen et al . , 2012 ) ; and Figure 3—figure supplement 2a ) . The KV7 . 1/G219C+KCNE1 model has 6 states ( Figure 3—figure supplement 1b ) , where the horizontal transition is the main S4 movement ( which generates the main fluorescence component F1 ) and the vertical transition is channel opening accompanied by an additional smaller S4 movement ( that generates a smaller additional fluorescence component F2 ) ( Osteen et al . , 2012; Zaydman et al . , 2014 ) . The KV7 . 1/G219C+KCNE1 model only allows for channel opening after all four S4s are activated , which thereby generates F ( V ) and G ( V ) that are separated in voltage dependence ( reference [Barro-Soria et al . , 2014]; and Figure 3—figure supplement 2a ) . Using these models , we can reproduce the main features of the fluorescence and currents from KV7 . 1/G219C/S225L and KV7 . 1/G219C/S225L+KCNE1 by only shifting the main voltage sensor movement by +50 mV in both models ( Figure 3—figure supplement 2b ) , as if the S225L mutation mainly affects the main S4 movement . In the KV7 . 1 model , shifting the main voltage sensor movement by +50 mV shifts both the G ( V ) and F ( V ) curves by +35–40 mV , similar to the effect induced by the S225L mutation in the experimental data . In the KV7 . 1+KCNE1 model , shifting the main voltage sensor movement by +50 mV results in that the F1 and F2 components overlap in voltage , such that it is hard to distinguish the two components , and that the G ( V ) is shifted by +10 mV . Both effects are similar to the effects induced by the S225L mutation in the experimental data ( cf . Figure 3b ) . We can reproduce the main features of the fluorescence and currents from KV7 . 1/G219C/F351A and KV7 . 1/G219C/F351A+KCNE1 by only shifting the voltage dependence of the opening transition by +140 mV in both models ( Figure 3—figure supplement 2c ) , as if the F351A mutation mainly affects the opening transition . In the KV7 . 1 model , shifting the opening transition by +140 mV shifts the G ( V ) by +100 mV whereas the F ( V ) is shifted less and has a shallower slope , similar to the effects induced by the F351A mutation in the experimental data . In the KV7 . 1+KCNE1 model , shifting the opening transition by +140 mV results in that the F1 and F2 components are further separated in voltage and that the G ( V ) is shifted by +100 mV . Both effects are similar to the effects induced by the F351A mutation in the experimental data ( cf . Figure 3c ) . In summary , our voltage-clamp fluorometry experiments together with kinetic modeling are compatible with a model in which the S225L mutation primarily interferes with the main S4 movement , whereas the F351A mutation interferes with later gating transitions associated with pore opening . One note of caution is that the interpretation of the mutational effects is dependent on the models used for the wild-type channels . Other models for KV7 . 1 and KV7 . 1+KCNE1 channels have been proposed ( Zaydman et al . , 2014; Ruscic et al . , 2013 ) , but these have not been as extensively tested or developed as our models . Although other alternative mechanisms for the effects of these mutations are possible , the different impacts of S225L and F351A on the fluorescence versus voltage relationships suggest that these mutations introduce distinct molecular defects . We previously observed that the effect of regular polyunsaturated fatty acids , such as docosahexaenoic acid , on KV7 . 1 is impaired by co-expression with the KCNE1 subunit ( Liin et al . , 2015 ) . In contrast , we found that the PUFA analogue N-arachidonoyl taurine ( N-AT , structure in Figure 4 ) retained its ability to activate the KV7 . 1 channel also in the presence of KCNE1 . N-AT activated the wild-type KV7 . 1+KCNE1 by shifting the G ( V ) roughly –30 mV ( Liin et al . , 2015 ) ( Figure 4—figure supplement 1 ) . The magnitude of this N-AT-induced shift is comparable to , but in the opposite direction , to the G ( V ) shifts observed for several of the LQTS and LQTS-like mutants . We therefore here test the ability of N-AT to enhance the function of the eight KV7 . 1+KCNE1 mutant channels . Figure 4a–b shows representative effects of 7–70 µM N-AT on KV7 . 1/S225L+KCNE1 . 70 µM N-AT increases current amplitude by a factor of 16 at +20 mV ( Figure 4a ) and shifts the G ( V ) curve by about –50 mV ( Figure 4b , Supplementary file 3 ) . Steady state of N-AT effects is reached within a few minutes ( Figure 4—figure supplement 2 ) . We note a small instantaneous ‘leak’ component in the 70 µM N-AT trace of KV7 . 1/S225L+KCNE1 ( Figure 4a ) . This leak component in KV7 . 1/S225L+KCNE1 is observed also in the absence of N-AT , but at more positive voltages ( Figure 4—figure supplement 3 ) . We do not observe this leak component in wild-type KV7 . 1+KCNE1 upon application of N-AT ( Figure 4—figure supplement 1a ) , which suggests that this phenomenon is associated with the S225L mutation . The human ventricular action potential has a duration of about 300–400 ms and a systolic voltage range of about 0 to +40 mV ( O'Hara et al . , 2011; Piacentino et al . , 2003 ) . To test the behaviour of the S225L mutation during shorter stimulating pulses , we apply repetitive 300 ms pulses to +40 mV at a frequency of 1 Hz and at 28°C ( 37°C was not tolerated by the oocytes ) . In response to this protocol , the KV7 . 1/S225L+KCNE1 channel barely opens and thus generates only minor currents ( Figure 4c ) . In contrast , we observe large KV7 . 1/S225L+KCNE1 currents upon application of 70 µM N-AT ( Figure 4c ) . N-AT also restores the gradual increase in current amplitude during repetitive pulsing seen experimentally ( inset in Figure 4c ) and in computer simulations ( Silva and Rudy , 2005 ) for the wild-type KV7 . 1+KCNE1 channel . 10 . 7554/eLife . 20272 . 013Figure 4 . Effect of N-AT on LQTS and LQTS-like mutants . All these experiments are done in the presence of KCNE1 . Structure of N-AT is shown . ( a–b ) Representative effect of 7–70 µM N-AT on current amplitude ( a ) and G ( V ) ( b ) of KV7 . 1/S225L+KCNE1 . Dashed line in ( a ) denotes 0 µA . ( c ) Representative currents generated by KV7 . 1/S225L+KCNE1 during pulsing at 1 Hz and +28°C in control solution ( black ) and after the cell had been bathed continuously in 70 µM N-AT ( light to dark green , # denotes sweep order ) . Inset: corresponding currents from wild-type KV7 . 1+KCNE1 scaled similarly as KV7 . 1/S225L+KCNE1 . Light grey trace denotes sweep #1 , grey trace denotes sweep #2 , and dark grey trace denotes sweep #20 . ( d ) Summary of V50 for LQTS and LQTS-like mutants before and after 70 µM N-AT application . Dashed line denotes V50 for wild-type KV7 . 1+KCNE1 . ( e–f ) Summary of ΔV50 ( e ) and ΔΔGo ( f ) for LQTS and LQTS-like mutants induced by 70 µM N-AT . # denotes an approximation . Dashed lines denote corresponding ΔV50 and ΔΔGo induced by 70 µM N-AT for wild-type KV7 . 1+KCNE1 . The statistics in ( f ) represent one-way ANOVA with Dunnett’s Multiple Comparison Test to compare the N-AT-induced change in ΔΔGo of mutants to N-AT-induced change in ΔΔGo of wild-type KV7 . 1+KCNE1; p≤0 . 05 . Only significant differences shown in ( f ) , other comparisons have p>0 . 05 . ( g ) Estimate of the ability of 70 µM N-AT to restore LQTS and LQTS-like mutant current amplitude at +40 mV . The mean N-AT induced increase in current amplitude for each mutant ( from Figure 4—figure supplement 4b ) is multiplied with the control amplitude for each mutant ( from Figure 1—figure supplement 4d ) . Not determined ( nd ) . Data as mean ± SEM . n = 5–12 . Dashed line denotes relative wild-type KV7 . 1+KCNE1 current amplitude in control solution ( i . e . without N-AT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 01310 . 7554/eLife . 20272 . 014Figure 4—figure supplement 1 . N-AT effect on wild-type KV7 . 1+KCNE1 expressed in Xenopus oocytes . Representative effect of 70 µM N-AT on current amplitude ( a ) and G ( V ) ( b ) of wild-type KV7 . 1+KCNE1 . The holding voltage is –80 mV and the tail current amplitude in ( b ) measured at –20 mV after 5 s test pulses . Dashed line in ( a ) denotes 0 µA current . G ( V ) curves in ( b ) are normalized to the fitted Gmax ( as described in Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 01410 . 7554/eLife . 20272 . 015Figure 4—figure supplement 2 . The time course of N-AT wash-in on KV7 . 1/S225L+KCNE1 expressed in Xenopus oocytes . Representative example showing that N-AT effects on current amplitude reaches steady state for each concentration within minutes . The holding voltage is –80 mV and current amplitude measured at the end of a 5 s test pulse to +20 mV . Dashed line denotes baseline ( control amplitude ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 01510 . 7554/eLife . 20272 . 016Figure 4—figure supplement 3 . ‘Leak’ component of KV7 . 1/S225L+KCNE1 . Currents generated by the KV7 . 1/S225L+KCNE1 mutant have a small instantaneous ‘leak’ component at positive voltages . Dashed line denotes 0 µA . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 01610 . 7554/eLife . 20272 . 017Figure 4—figure supplement 4 . Effect of N-AT on current amplitude of LQTS and LQTS-like mutants . ( a-b ) Current amplitudes in the presence of 70 µM N-AT measured at the end of a 5 s test pulse to +20 mV ( a ) or +40 mV ( b ) . The currents are normalized to the current amplitude in control solution in the same oocyte . Dashed lines denote N-AT effects on wild-type KV7 . 1+KCNE1 current amplitude ( a factor 2 . 9 ± 0 . 4 and 1 . 9 ± 0 . 3 ( n = 5 ) , respectively ) . ( c ) Ability of 70 µM N-AT to restore LQTS and LQTS-like mutant current amplitude at +20 mV . The mean N-AT-induced fold increase in current amplitude for each mutant ( data from panel a ) is multiplied by the relative current amplitude for each mutant compared to wild-type KV7 . 1+KCNE1 in control solution ( from Figure 1—figure supplement 4 ) . Dashed line denotes relative wild-type KV7 . 1+KCNE1 current amplitude in control solution ( i . e . without N-AT ) . Mean ± SEM . n = 4–12 . nd = not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 017 Further testing of N-AT show that 70 µM N-AT shifts the G ( V ) curve of all tested mutants by 30–50 mV towards more negative voltages ( Figure 4d–e , Supplementary file 3 ) . The G ( V ) curve of wild-type KV7 . 1+KCNE1 is shifted by –27 . 0 ± 2 . 5 mV ( Liin et al . , 2015 ) . Thus , 70 µM N-AT completely corrects the positive G ( V ) shifts induced by the mutations so that in the presence of N-AT the G ( V ) is similar to or shifted negative compared to the G ( V ) of the wild-type KV7 . 1+KCNE1 channel ( Figure 4d , F351A homozygous expression was not included in this analysis because of the very shifted G ( V ) curve of this mutant ) . The G ( V ) of mutants is shifted about equally by N-AT for homozygous and heterozygous expression ( Figure 4e ) . The slope of the G ( V ) curve varies slightly ( 10 . 4 to 16 . 3 ) among the mutants ( Supplementary file 3 ) . To correct for this difference in slope and to better compare the functional effect of N-AT-induced G ( V ) shifts on the different mutants , we also calculate the change in Gibbs free energy for channel opening ( ΔΔGo ) that 70 µM N-AT induces . 70 µM N-AT reduces the energy required to open the channel by 5 . 3–9 . 0 kJ/mol depending on mutant ( 4 . 9 ± 0 . 7 kJ/mol ( n = 5 ) for wild-type ) ( Figure 4f ) . To estimate the functional effect of N-AT on the KV7 . 1+KCNE1 current amplitude of each mutant , we calculate the ratio of the current amplitude at the end of the 5 s test pulse before and after application of N-AT at +20 and +40 mV . The 5 s voltage pulse to +20 mV ( or + 40 mV ) at room temperature was chosen to make the KV7 . 1+KCNE1 channel activate to a similar extent as during a ventricular action potential ( 300–400 ms ) at body temperature ( note that KV7 . 1+KCNE1 channels have a relatively high Q10 of around 5–7 . 5 [Busch and Lang , 1993; Seebohm et al . , 2001] ) . 70 µM N-AT increases the current amplitude of all mutants at these voltages ( Figure 4—figure supplement 4a–b , Supplementary file 3 ) . As expected , current amplitude is most increased for those mutants that have the most shifted G ( V ) curve towards more positive voltages ( e . g . V215M and S225L ) . This is because these mutants are still at the foot of their G ( V ) curve at +20 and +40 mV and a N-AT-induced shift towards more negative voltages results in a relatively larger increase in the current amplitude . By multiplying these relative N-AT-induced increases in current amplitude with the relative current amplitude of each mutant ( compared to wild-type KV7 . 1+KCNE1 channels , from Figure 1—figure supplement 4c–d ) , we observe that 70 µM N-AT compensates fairly well ( or overcompensates ) for the mutation-induced reduction in current amplitude ( Figure 4g , Figure 4—figure supplement 4c ) . Moreover , for all mutant and wild-type KV7 . 1+KCNE1 channels , 70 µM N-AT speeds up the opening kinetics at +40 mV by a factor of 1 . 3–2 . 5 ( Supplementary file 3 ) . 70 µM N-AT also slows down the closing kinetics for most mutants and wild-type KV7 . 1+KCNE1 ( Supplementary file 3 ) . For F351A heterozygous expression and R583C homozygous expression , 70 µM N-AT restores the closing kinetics so that the closing kinetics is not statistically different ( p>0 . 05 ) from wild-type KV7 . 1+KCNE1 closing kinetics ( 737 ± 62 ms and 833 ± 74 ms , respectively compared to 967 ± 47 ms for wild-type ) . In the presence of KCNE1 , channels made with F193L heterozygous expression , L251A homozygous expression , and R583C heterozygous expression have wild-type like closing kinetics already before application of N-AT . We next use voltage clamp fluorometry on KV7 . 1/G219C/S225L+KCNE1 and KV7 . 1/G219C/F351A+KCNE1 to explore the mechanism by which N-AT enhances the activity of two mechanistically different mutants . Surprisingly , N-AT caused a dramatic decrease in the fluorescence from Alexa488-labeled KV7 . 1/G219C+KCNE1 channels ( Figure 5—figure supplement 1a ) . In contrast , N-AT did not decrease the fluorescence from Alexa488-labeled KV7 . 1/G219C* channels nor did high concentrations of taurine decrease the fluorescence from unbound Alexa488 ( even up to concentrations of 0 . 5 M taurine; Figure 5—figure supplement 1b ) , suggesting that N-AT is not a collisional quencher of Alexa488 . The mechanism of the N-AT-induced decrease of fluorescence from Alexa488-labeled KV7 . 1/G219C+KCNE1 channels is not clear , but could be due to N-AT inducing a conformational change in KCNE1 or KV7 . 1 that brings a quenching residue close to Alexa488 . Due to the dramatic decrease in the fluorescence signal from Alexa488-labeled KV7 . 1/G219C+KCNE1 channels , we have to normalize the F ( V ) curves obtained in N-AT to the amplitude of the F ( V ) in control solutions . With this normalization , voltage clamp fluorometry experiments on KV7 . 1/G219C/S225L+KCNE1 indicate that N-AT shifts both the voltage dependence of the first part ( which represents F1 ) and the second part ( which represents F2 ) of the F ( V ) curve towards more negative voltages ( Figure 5—figure supplement 1c ) . However , due to the not completely saturating F ( V ) for KV7 . 1/G219C/F351A+KCNE1 , we are unable to reliably normalize the F ( V ) curves in the presence of N-AT to the control F ( V ) curves . We instead explore the effect of N-AT on the kinetics of the two fluorescence components: F1 , which is seen as a fast fluorescence change at negative voltages , and F2 , which is seen as a slow fluorescence change on top of the F1 component at positive voltages ( Barro-Soria et al . , 2014 ) . F1 correlates with the measured gating currents in KV7 . 1+KCNE1 channels ( and the initial delay in the KV7 . 1+KCNE1 ionic currents ) , whereas F2 correlates with the opening of KV7 . 1+KCNE1 channels ( Barro-Soria et al . , 2014 ) . For both mutants , 70 µM N-AT speeds up F1 kinetics ( Figure 5a , d , measured at –40 mV where virtually no channels open and the fluorescence is mainly composed of F1 ) . Numeric values for N-AT effects on channel kinetics are summarized in Figure 5f . Moreover , N-AT accelerates the channel opening kinetics ( Figure 5b , e ) and both the F1 and F2 fluorescence components at +80 mV for KV7 . 1/G219C/S225L+KCNE1 ( Figure 5f ) . The change in the F2 component is probably larger than what the fits of a double-exponential function suggest , because the slow part of the fluorescence , mainly F2 , overlay nicely on the currents in both the presence and absence of 70 µM N-AT ( Figure 5c , upper panel ) . As a control , we show that the fluorescence in N-AT does not , however , overlay the currents in control solutions and vice versa ( Figure 5c , middle and lower panel ) . For KV7 . 1/G219C/F351A+KCNE1 , the G ( V ) curve and the F2 component are so shifted towards depolarizing voltages that we cannot reliably quantify the F2 component in our fluorescence traces . 70 µM N-AT does , however , speed up KV7 . 1/G219C/F351A+KCNE1 current kinetics ( Figure 5e ) , which suggests that N-AT also speeds up F2 in KV7 . 1/G219C/F351A+KCNE1 . Altogether , these results suggest that N-AT accelerates both conformational changes during the main gating charge movement and channel opening . 10 . 7554/eLife . 20272 . 018Figure 5 . Effect of 70 µM N-AT on S4 movement and gate opening in S225L and F351A mutants . ( a–c ) Representative example of the effect of 70 µM N-AT on F1 kinetics ( a ) , current opening kinetics ( b ) , and F2 kinetics ( c ) in KV7 . 1/G219C/S225L+KCNE1 . Control fluorescence ( red ) and current ( black ) . N-AT fluorescence ( magenta ) and current ( green ) . Top in ( c ) shows an overlay of the later part of the fluorescence ( after most of F1 has occurred ) and the later part of the currents ( after the initial delay ) before and after application of N-AT . Middle and lower ( c ) show that there is not a great overlap of the fluorescence in the presence of N-AT and the current in control solution ( middle ) or the fluorescence in control solution and the current in the presence of N-AT ( lower ) . ( d–e ) Representative example of effect of 70 µM N-AT on F1 kinetics ( d ) and current opening kinetics ( e ) in KV7 . 1/G219C/F351A+KCNE1 . Same colouring as in ( a–b ) . Dashed line in ( b ) and ( e ) denotes 0 µA . Fluorescence traces and all traces in ( c ) have been normalized to better allow temporal comparison . ( f ) Summary of the effect of 70 µM N-AT on the kinetic parameters of KV7 . 1/G219C/S225L+KCNE1 and KV7 . 1/G219C/F351A+KCNE1 . Kinetics of the fast ( F1 ) and slow ( F2 ) fluorescence components were deduced from a double-exponential function fitted to the fluorescence traces . The kinetics of currents were deduced from a single-exponential function fitted to current traces . Ratios of time constants ( τN-AT/τCtrl ) were calculated pair-wise ( control compared to N-AT ) in each oocyte and analysed using two-tailed one sample t-test where ratios were compared with a hypothetical value of 1 . Data as mean ± SEM . n = 4 ( 3for fluorescence kinetics for KV7 . 1/G219C/F351A+KCNE1 ) . p<0 . 05; *p<0 . 01 . nd = not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 01810 . 7554/eLife . 20272 . 019Figure 5—figure supplement 1 . Effect of N-AT on the F ( V ) of KV7 . 1/G219C/S225L mutant co-expressed with KCNE1 in Xenopus oocytes . ( a ) Representative example of the time course of the reduction in fluorescence intensity upon N-AT application . The fluorescence intensities shown is the fluorescence measured at +80 mV ( during repeated applications of the voltage protocol used to measure the complete F ( V ) as in panel c ) , normalized to the fluorescence intensity at +80 mV recorded in the first F ( V ) in control solution . Red symbols denote control ( without N-AT ) and purple symbols denote in the presence of N-AT . The fluorescence signal reduces with time in the presence of N-AT . In contrast , the fluorescence signal is preserved in the absence of N-AT ( red symbol , recorded in another cell ) . ( b ) Summary of fluorescence emission monitored from unbound Alexa488 in control solution and in taurine-supplemented control solution ( 0 . 25 or 0 . 5 M taurine ) . In these experiments , no oocytes or channels were present . A . U . denotes arbitrary units . Data as mean ± SEM . n = 3 . ( c ) Mean F ( V ) curve for KV7 . 1/G219C/S225L+KCNE1 in the absence ( red symbols , data from Figure 3B ) or presence of 70 µM N-AT . The holding voltage is –80 mV , the pre-pulse –160 mV for 5 s , and test voltages between –160 and +100 mV for 5 s in 20 mV increments . The tail voltage is –40 mV . Each F ( V ) curve is normalized between 0 and 1 based on the bottom and top deduced from the double Boltzmann fits for each curve ( see Materials and methods ) . Data as mean ± SEM . n = 3 for N-AT . DOI: http://dx . doi . org/10 . 7554/eLife . 20272 . 019 We show that all studied LQTS and LQTS-like mutations i ) shift the G ( V ) of KV7 . 1+KCNE1 towards more positive voltages , and/or ii ) accelerate KV7 . 1+KCNE1 closing . This suggests that at least part of the mechanism underlying the reduced ability of these mutants to generate K+ currents is by altering these biophysical properties of the KV7 . 1+KCNE1 channel . Using voltage clamp fluorometry in combination with kinetic modeling , we further suggest that these altered biophysical properties in mutants may be caused by interference with different gating transitions . Our experimental data and kinetic modeling are consistent with a model in which KV7 . 1/S225L primarily causes the reduced channel function by altering the main voltage sensor movement , while KV7 . 1/F351A alters later gating transitions associated with pore opening . The different effects of S225L and F351A on the fluorescence versus voltage relationships in KV7 . 1/G219C and KV7 . 1/G219C*+KCNE1 suggest that these mutations cause channel dysfunction via different molecular mechanisms . Note that we used the LQTS-like F351A mutant , because the LQTS mutant F351S did not generate any currents ( Figure 1—figure supplement 1 ) . However , during the review process of this manuscript a new LQTS mutation , F351L , was found ( Vyas et al . , 2016 ) . The current and fluorescence of this LQTS mutant is very similar to the current and fluorescence of F351A ( Figure 3—figure supplement 3 ) , suggesting that our conclusions on the LQTS-like F351A is also relevant for the LQTS mutant F351L . One of the mutations , F193L , has only minor effects on the biophysical properties of KV7 . 1+KCNE1 . This mutant was previously reported to have reduced current amplitude compared to the wild-type KV7 . 1+KCNE1 channel and a mild clinical phenotype ( Yamaguchi et al . , 2003 ) . The F193L mutation may therefore cause loss of function by faster deactivation kinetics and lower current density . Heterozygous expression of mutated subunits and wild-type subunits in equal molar ratios results in general in a milder biophysical phenotype ( more close to the wild-type phenotype ) . This is in line with a milder clinical phenotype generally reported for heterozygous carriers of LQTS mutations compared to individuals with homozygous genotypes ( Priori et al . , 1998; Jackson et al . , 2014; Zhang et al . , 2008 ) . Moreover , for different mutations different biophysical effects of the mutations could be dominant or recessive: For S225L and L251A , heterozygous expression in the presence of KCNE1 partially or completely restores wild-type like V50 , whereas heterozygous expression does not improve closing kinetics compared to homozygous expression . For KCNE1/K70N and KCNE1/S74L , co-expression with wild-type KCNE1 subunits also restores wild-type like V50 , whereas wild-type like closing kinetics is only partially restored . In contrast , for KV7 . 1/R583C , heterozygous expression restores wild-type like closing kinetics , but not wild-type like V50 . However , because of uncertainties regarding the stoichiometry of mutant to wild-type subunits in assembled KV7 . 1+KCNE1 channels ( as mentioned in the Results section ) , further studies will be required to understand the mechanisms underlying these apparent dominant or recessive effects and to evaluate possible physiological impact of these effects . Our results show that all tested mutants respond to N-AT . This is in contrast to previously reported KV7 channel activators on disease-causing KV7 mutants , for which mutants show markedly different sensitivity ( Seebohm et al . , 2003; Xiong et al . , 2007; Leitner et al . , 2012 ) . 70 µM N-AT shifts the G ( V ) curve of the wild-type KV7 . 1+KCNE1 channel and of all LQTS and LQTS-like mutants by approximately ( –50 ) – ( –30 ) mV , accelerates channel opening and slows down channel closing . In the presence of 70 µM N-AT , the V50 of all LQTS and LQTS-like mutants are similar to or more negative than V50 for the wild-type KV7 . 1+KCNE1 channel . For most mutants , 70 µM N-AT overcompensates for the shift in G ( V ) and reduction in current amplitude caused by the mutations , indicating that a lower N-AT concentration or a less potent N-AT analogue could be used to restore wild-type like G ( V ) and current amplitudes . Moreover , KV7 . 1+KCNE1 opening and closing kinetics are partially or completely restored by N-AT . Also , although the disease aetiology of the F193L mutant is likely mainly reduced channel expression , the N-AT induced augmentation caused by a shift in G ( V ) and increased currents may at least in part overcome the reduction in currents caused by the reduced channel expression . This general ability of N-AT to , at least partly , compensate for the reduced function of mutants with mutations in different parts of the KV7 . 1+KCNE1 channel complex and with seemingly different molecular defects , as long as a population of these mutant channels reaches the plasma membrane , suggests that N-AT is an interesting model compound for development of future anti-arrhythmics to treat LQTS caused by diverse KV7 . 1+KCNE1 mutations . Defective trafficking of mutant KV11 . 1 ion channels is a common cause of LQTS type 2 . About 80-90% of LQTS type 2-associated hERG mutants are estimated to suffer from defective trafficking ( Anderson et al . , 2014; Sanguinetti , 2010 ) . The corresponding number for LQTS-associated KV7 . 1 and KCNE1 mutants is not known . Previous studies identify both trafficking defective and trafficking competent KV7 . 1 and KCNE1 mutants , e . g . ( Anderson et al . , 2014; Sanguinetti , 2010 ) . We are mainly interested in understanding the mechanism that underlies abnormal gating of KV7 . 1 and KCNE1 mutants . To avoid mutants with severe trafficking defects , we therefore selected mutants that have previously been shown to localize abundantly enough to the cell membrane to generate detectable K+ currents . Several of the selected mutants have been shown to traffic well in mammalian systems ( KV7 . 1/V215M and KCNE1/S74L [Eldstrom et al . , 2010; Harmer et al . , 2010] ) or generate clearly detectable currents in mammalian cells ( KV7 . 1/R583C [Yang et al . , 2002] ) . Our Xenopus oocyte experiments that compare mutant current amplitudes with wild-type current amplitudes ( Figure 1—figure supplement 4 ) suggest that the reduced ability of the selected mutants to generate currents in Xenopus oocytes may largely be explained by the shifted G ( V ) of mutants . Trafficking defects could be disguised in Xenopus oocytes that are cultured at low temperatures that may rescue some trafficking defects ( Anderson et al . , 2014; Delisle et al . , 2004 ) . These current amplitude experiments should therefore be interpreted with caution until trafficking of specific KV7 . 1 and KCNE1 LQTS mutants in mammalian systems has been explored . Previous studies show that membrane expression of trafficking-defect channel mutants ( e . g . for KV11 . 1 and CFTR ) can be pharmacologically rescued using compounds that are suggested to stabilize channel conformation during folding and trafficking ( Anderson et al . , 2014; Delisle et al . , 2004; Sato et al . , 1996 ) . However , rescue of membrane expression may only partially compensate for mutation-induced loss of function , if these mutants also suffer from defective gating ( Perry et al . , 2016 ) . Our proposed N-AT model for pharmacological correction of ‘G ( V ) ’ LQTS mutants could therefore potentially complement pharmacological correction of trafficking-defect LQTS mutants to improve the outcome of patients suffering from LQTS . We previously suggested that polyunsaturated fatty acids and their analogues ( such as N-AT ) attract the voltage sensor S4 in KV7 . 1 by an electrostatic mechanism and thereby shift the G ( V ) towards more negative voltages and speed up channel opening ( Liin et al . , 2015 ) . We therefore initially hypothesized that N-AT only would restore the function of those LQTS mutations with altered S4 movement . We were pleasantly surprised when N-AT seems to be able to restore the function of many LQTS and LQTS-like mutants , with diverse mutational defects ( such as S225L and F351A ) . Using voltage clamp fluorometry , we have previously shown that both the main gating charge movement and the gate opening of KV7 . 1+KCNE1 channels are accompanied by fluorescence signals from fluorophores attached to S4 ( Barro-Soria et al . , 2014 ) . This suggests that S4 moves both during the main gating charge movement and during the subsequent channel opening in KV7 . 1+KCNE1 channels ( Barro-Soria et al . , 2014 ) , which is similar to observations in Shaker KV channels ( Börjesson and Elinder , 2011; Pathak et al . , 2005; Phillips and Swartz , 2010 ) . Therefore , N-AT could affect both the main gating charge movement and gate opening by acting on the S4 voltage sensor , as has been shown for hanatoxin which targets the voltage-sensing domain in the Shaker KV channel ( Milescu et al . , 2013 ) . This hypothesis is supported by our voltage-clamp fluorometry experiments using KV7 . 1/S225L and KV7 . 1/F351A in which N-AT accelerates the fluorescence components associated with both the main S4 movement ( F1 ) and gate opening ( F2 ) , as well as accelerates the kinetics of channel opening . This proposed mechanism would explain why N-AT can restore the function of mutations that mainly target either the main S4 movement or gate opening . However , the dramatic decrease in the fluorescence signal caused by N-AT makes it hard for us to completely determine the effect of N-AT on the F ( V ) of mutants . Therefore , the complete mechanism of N-AT in the different mutations is not clear . Future studies are required to assess the clinical utility of PUFA analogues in cardiomyocytes and animal models . We see channel specificity of PUFA analogues as one major challenge and recognize the need to improve PUFA analogue affinity to KV7 . 1+KCNE1 to reduce required therapeutic concentrations and minimize potential adverse effects . Despite these challenges , our data show that the magnitude of the N-AT-induced voltage shifts are in a similar range as the shifts induced by several LQTS mutations , thereby serving as proof of concept that this PUFA analogue , at least partly , restores channel function in diverse LQTS and LQTS-like mutants . Fluorescence and currents from the KV7 . 1+KCNE1 models were simulated using Berkeley Madonna ( Berkeley , CA ) . Average values are expressed as mean ± SEM . Mutant parameters ( e . g . V50 and ΔΔGo ) were compared to wild-type parameters using one-way ANOVA with Dunnett’s Multiple Comparison Test . Comparison of homozygous and heterozygous expression was done using one-way ANOVA with pair-wise Bonferroni’s Test . The effects of N-AT on fluorescence and current kinetics were analysed using two-tailed one sample t-test where ratios were compared with a hypothetical value of 1 . p<0 . 05 is considered as statistically significant .	Every heartbeat relies on an electric wave that travels through the heart . This wave must reach different parts of the heart in a specific sequence to ensure that the heart muscle cells contract in a coordinated manner . Such coordinated contractions enable the heart to pump enough blood around the body . By allowing specific ions to flow into or out of the heart muscle cell , proteins called ion channels in the cell membrane generate the electric wave , keep it going and stop it . One such protein called the IKs channel controls the flow of potassium ions , and in doing so stops the electric wave in heart muscle cells . About 300 different mutations in the IKs channel have been shown to cause abnormal heart rhythms in individuals with a disorder called long QT syndrome . People with this condition may suddenly black out because their heart develops prolonged electric waves that prevent blood from being pumped properly . To investigate how mutations in the IKs channel produce heart rhythm abnormalities , Liin et al . genetically engineered the egg cells of African clawed frogs to have one of eight mutant forms of the human IKs channel . Studying these channels revealed that the mutations reduce how well the channels work in a wide variety of ways . However , treating the cells with a particular fatty acid helped to normalize how each of the mutant channels worked . Therefore , variants of the fatty acid could potentially form a useful treatment for people with heart rhythm problems caused by mutations in the IKs channel . More studies are needed to confirm whether the fatty acid is as effective at combating the effects of the mutations in whole hearts and animals . As ion channels related to the IKs channel are found in many types of cells , it is also important to investigate whether treatment with the fatty acid could cause any side effects that affect other organs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2016	Fatty acid analogue N-arachidonoyl taurine restores function of IKs channels with diverse long QT mutations
Intracellular ion channels are essential regulators of organellar and cellular function , yet the molecular identity and physiological role of many of these channels remains elusive . In particular , no ion channel has been characterized in melanosomes , organelles that produce and store the major mammalian pigment melanin . Defects in melanosome function cause albinism , characterized by vision and pigmentation deficits , impaired retinal development , and increased susceptibility to skin and eye cancers . The most common form of albinism is caused by mutations in oculocutaneous albinism II ( OCA2 ) , a melanosome-specific transmembrane protein with unknown function . Here we used direct patch-clamp of skin and eye melanosomes to identify a novel chloride-selective anion conductance mediated by OCA2 and required for melanin production . Expression of OCA2 increases organelle pH , suggesting that the chloride channel might regulate melanin synthesis by modulating melanosome pH . Thus , a melanosomal anion channel that requires OCA2 is essential for skin and eye pigmentation . Ion channels are membrane proteins that regulate the concentration of key signaling ions to control a wide range of cellular functions . While plasma membrane ion channels have been extensively studied , much less is known about the identity and physiology of intracellular channels because they are less accessible to direct electrophysiological characterization . Melanosomes are lysosome-related organelles that produce and store melanin , a natural pigment present in most organisms . Impaired melanin synthesis and storage affects visual system development and pigmentation of the skin , eyes , and hair , leading to reduced protection against ultraviolet radiation and predisposition for skin and eye cancers . A number of genes encoding putative melanosomal ion transport proteins are critical for melanosomal function , as mutations in these genes result in oculocutaneous albinism ( OCA ) ( Montoliu et al . , 2014 ) . This suggests that ionic homeostasis plays an important role in melanin synthesis and storage , yet how ion channels might contribute to melanosome function and pigmentation remains poorly understood ( Bellono and Oancea , 2014 ) . One of the most common forms of albinism is caused by mutations in a highly conserved protein encoded by the oculocutaneous albinism II gene ( OCA2 ) ( Gardner et al . , 1992; Rinchik et al . , 1993; Rosemblat et al . , 1994; Lee et al . , 1994a; Sitaram et al . , 2009 ) ( Figure 1—figure supplement 1 ) . OCA2-deficient animals lack pigment ( Gardner et al . , 1992; Protas et al . , 2006 ) and reduced OCA2 expression leads to decreased melanin underlying blue eye color in humans ( Eiberg et al . , 2008; Sturm et al . , 2008 ) . OCA2 has twelve predicted transmembrane domains ( Gardner et al . , 1992 ) , is localized to melanosomal membranes ( Rosemblat et al . , 1994; Sitaram et al . , 2009 ) , and has been implicated in pH regulation of melanosomes and trafficking of the melanogenic enzyme tyrosinase ( Puri et al . , 2000; Manga and Orlow , 2001; Chen et al . , 2002 , 2004; Ni-Komatsu and Orlow , 2006 ) . Despite its importance , the function of OCA2 and the molecular mechanism by which it regulates melanin are not known . Here we identify and characterize a new intracellular ion channel that resides in the melanosomal membrane and requires OCA2 expression . Using whole-organelle and single-channel patch-clamp recordings we found that OCA2 contributes to an anion channel required for pigmentation . The OCA2-mediated Cl− current was nearly abolished by a mutation identified in patients with oculocutaneous albinism type II . Interestingly , expression of OCA2 in endolysosomes increased organelle pH , providing a potential mechanism for how OCA2 regulates melanogenesis . Thus , a previously uncharacterized OCA2-dependent anion channel is critical for melanosomal function and pigmentation , revealing a novel function for intracellular ion channels . In melanocytes and retinal pigment epithelium ( RPE ) OCA2 is restricted to melanosomes , but when expressed in heterologous systems it localizes to lysosomes and late endosomes ( endolysosomes ) ( Sitaram et al . , 2009 ) . To investigate if OCA2 has ion transport activity , we expressed OCA2 tagged with GFP ( GFP-OCA2 ) or mCherry ( mCherry-OCA2 ) in AD293 cells , where it colocalized with the endolysosomal marker LAMP1 ( Figure 1A ) . Endolysosomes are an ideal melanosome-related heterologous system because they can be enlarged by treating cells with 1 μM vacuolin-1 and mechanically released from the cytoplasm with a glass pipette ( Cerny et al . , 2004; Saito et al . , 2007; Dong et al . , 2008 ) . We recorded whole-organelle currents from endolysosomes expressing GFP-OCA2 and found that voltage pulses evoked a large outwardly rectifying current ( IOCA2 ) that was not present in mock-transfected cells ( Figure 1A ) . 10 . 7554/eLife . 04543 . 003Figure 1 . OCA2 contributes to an endolysosomal chloride current with ion channel properties . ( A ) Heterologous GFP-OCA2 localized to LAMP1-mCherry-positive late endosomes and lysosomes ( endolysosomes ) individually dissected from AD293 cells for patch-clamp experiments ( scale bar = 5 μm ) . Whole-endolysosomal currents elicited by voltage steps between −80 mV and +80 mV in a representative endolysosome expressing GFP-OCA2 or in an endolysosome from a mock-transfected cell . ( B ) Representative whole-endolysosome current–voltage ( I–V ) relationships in response to voltage ramps . Outwardly rectifying IOCA2 was nearly abolished and its reversal potential ( Erev ) shifted in the positive direction when luminal Cl− was substituted for gluconate ( Gluc− ) , similar to currents measured from mock-transfected endolysosomes containing a Cl−-based luminal solution . ( C ) Average current densities ( pA/pF ) measured at 100 mV . Inset: Average Erev ( ±s . e . m . , n = 6–9 endolysosomes for each condition ) . ( D ) In a representative OCA2-expressing endolysosome , Erev was dependent on cytoplasmic [Cl−] ( [Cl−]cyto ) . NMDG-based solutions were used to inhibit endogenous cation permeabilities . ( E ) IOCA2 Erev varied linearly with [Cl−]cyto when [Cl−]luminal was kept at 48 mM . Each point represents one endolysosome at the [Cl−]cyto indicated on the x-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 00310 . 7554/eLife . 04543 . 004Figure 1—figure supplement 1 . OCA2 is a highly conserved transmembrane protein expressed in pigment cells . ( A ) Predicted membrane topology and homology of OCA2 from human , mouse , zebrafish and cavefish . OCA2 sequences were aligned using MegAlign ( DNASTAR ) . Accession numbers of OCA2 sequences used in alignments generated using Clustal W in MegAlign ( DNASTAR ) : NP-000266 . 2 ( human ) , NP-068679 . 1 ( mouse ) , XP_695807 . 5 ( zebrafish ) , ABB29299 . 1 ( cavefish ) . Amino acid percent identity was determined using Protein BLAST ( http://blast . ncbi . nlm . nih . gov ) for the indicated regions . TM = transmembrane domain , AA = amino acid . Inset: OCA2 alignment showing conserved regions with albinism-associated mutations used in this study ( K614E in dark blue , V443I in purple , 5mut in light blue ) . ( B ) OCA2 mRNA expression profile in human tissue . OCA2 expression levels were highest in retina , consistent with OCA2 expression in the densely pigmented retinal pigment epithelium ( RPE ) . Human skin had the second highest levels of OCA2 mRNA , due to high expression in melanocytes and very low levels in keratinocytes and skin fibroblasts , consistent with OCA2 expression in pigment cells . ( n = 4 experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 00410 . 7554/eLife . 04543 . 005Figure 1—figure supplement 2 . IOCA2 is not mediated by K+ flux or regulated by pH . ( A ) In the same endolysosome , IOCA2 was not significantly different when cytoplasmic K+ was substituted with Na+ . ( B ) The IOCA2 amplitude at 100 mV or Erev was not significantly different when cytoplasmic K+ was substituted with Na+ ( ±s . e . m . , n = 3 endolysosomes ) . ( C ) IOCA2 current density ( pA/pF ) measured in endolysosomes expressing OCA2 was similar with a luminal pH of 4 . 6 or 6 . 8 . Inset: Average Erev ( ±s . e . m . , n = 4 endolysosomes per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 00510 . 7554/eLife . 04543 . 006Figure 1—figure supplement 3 . IOCA2 is mediated by Cl− flux . Average IOCA2 current densities ( pA/pF ) were calculated at the indicated voltages . Currents recorded from mock-transfected AD293 endolysosomes ( A ) or endolysosomes expressing OCA2 in the presence of luminal Gluc− ( B ) were subtracted from currents measured from OCA2-expressing endolysosomes in the presence of luminal Cl− . Δ ( OCA2 − mock ) and Δ ( luminal NaCl − NaGluc ) currents have an Erev near ECl ( ECl = Nernst potential for Cl−; ±s . e . m . , n = 5 endolysosomes per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 00610 . 7554/eLife . 04543 . 007Figure 1—figure supplement 4 . Cl− channel blockers DIDS , NFA , and NPPB do not affect IOCA2 . ( A ) In a representative endolysosome IOCA2 was not affected by Cl− channel inhibitors 4 , 4′-diisothiocyanato-2 , 2′-stilbenedisulfonic acid disodium salt ( DIDS ) , niflumic acid ( NFA ) , or 5-Nitro-2- ( 3-phenylpropylamino ) benzoic acid ( NPPB ) ( each at 100 μM ) , compared to control ( DMSO vehicle ) . ( B ) The average current amplitude of IOCA2 normalized to control conditions ( <1% DMSO ) was similar when DIDS , NFA , or NPPB was bath applied at 100 μM or 300 μM ( n = 3–4 endolysosomes for each ) . DIDS blocks several members of the Chloride Channel ( ClC ) family , Ca2+-activated Cl− channels ( CaCC ) , Maxi Cl− channels , and volume regulated Cl− channels ( VRAC ) ; NFA blocks several ClC members , VRAC , and CaCC; NPPB inhibits several ClC members , Maxi Cl− , VRAC , and cystic fibrosis transmembrane conductance regulator ( CFTR ) . Pharmacological profile information was obtained from the International Union of Basic and Clinical Pharmacology ( IUPHAR , http://www . iuphar-db . org ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 007 The measured outward current could result either from K+ , the main cation in the bath , moving into the endolysosome , or Cl− , the main anion in the pipette solution , moving out of the endolysosome . We tested if IOCA2 carries K+ into the lumen by replacing it with Na+ in the bath solution . We detected no significant change in the current amplitude or reversal potential ( Erev ) ( Figure 1—figure supplement 2A , B ) , suggesting that K+ is not required for IOCA2 . We next tested if IOCA2 is mediated by luminal Cl− moving into the cytoplasm by replacing Cl− in the pipette solution with the impermeant anion gluconate ( Gluc− ) . IOCA2 was nearly abolished with Gluc− in the pipette and Erev was significantly shifted in the positive direction , consistent with the endogenous current recorded in endolysosomes from mock-transfected cells ( Figure 1B , C ) . The similar current amplitude and positive Erev measured in mock-transfected and OCA2-expressing endolysosomes with luminal Gluc− suggests the presence of an endogenous cationic permeability . When we subtracted this endogenous current from IOCA2 , the Erev for IOCA2 became similar to ECl predicted by the Nernst equation ( Figure 1—figure supplement 3 ) . The significant reduction in current amplitude in the presence of luminal Gluc− indicates that IOCA2 is mediated by Cl− , consistent with OCA2 homology to bacterial anion transporters ( Rinchik et al . , 1993; Brilliant , 2001; Kobayashi et al . , 2006 ) . We tested whether a range of pharmacological inhibitors of different Cl− channels and transporters affect IOCA2 by bath-applying 4 , 4′-diisothiocyanato-2 , 2′-stilbenedisulfonic acid disodium salt ( DIDS ) , niflumic acid ( NFA ) , or 5-Nitro-2- ( 3-phenylpropylamino ) benzoic acid ( NPPB ) , but none significantly altered IOCA2 amplitude ( Figure 1—figure supplement 4 ) . Thus , IOCA2 has a different pharmacological profile from the channels and transporters that are inhibited by the tested compounds . The Nernst equation predicts that if IOCA2 were due to selective transport of Cl− , Erev will be dependent on the Cl− concentration ( [Cl−] ) gradient across the endolysosomal membrane . To establish whether this is the case , we determined Erev for currents measured at variable cytoplasmic and constant luminal [Cl−] , using NMDG-based solutions to prevent endogenous cationic permeability . Erev for 48 mM luminal and 8 mM cytoplasmic [Cl−] was −42 . 7 ± 2 . 3 mV and shifted to 1 . 6 ± 2 . 5 mV under symmetrical 48 mM [Cl−] ( Figure 1D ) . Erev increased linearly as a function of cytosolic [Cl−] ( [Cl−]cyto ) , with a 10-fold change in [Cl−]cyto corresponding to a shift in Erev of 58 mV , consistent with a highly Cl− selective current ( Figure 1E ) . OCA2 shares little homology with known chloride channels or transporters; OCA2 might be an accessory subunit of a Cl− transporter or form a Cl− channel or carrier protein itself . To determine if specific OCA2 residues are required for Cl− transport , we sought to identify mutations important for ion transport , but not melanosomal localization . We analyzed mCherry-OCA2 variants containing disease-related mutations within highly conserved regions of the protein ( Figure 1—figure supplement 1A ) . We chose mutations identified in patients with OCA type II through human genetic studies: V443I , a common albinism-associated mutation in a predicted luminal loop and important for melanin content in vitro ( Lee et al . , 1994a , 1994b; Sviderskaya et al . , 1997; King et al . , 2003; Garrison et al . , 2004; Hongyi et al . , 2007; Preising et al . , 2007; Hutton and Spritz , 2008; Rimoldi et al . , 2014 ) and K614E , in a predicted cytoplasmic loop ( Passmore et al . , 1999 ) . We also generated an OCA2 variant with five point mutations in the same predicted luminal loop as V443I ( 5mut: V443I , M446V , I473S , N476D , N489D ) ( Lee et al . , 1994a , 1994b; Spritz et al . , 1995; Hongyi et al . , 2007; Preising et al . , 2007 ) ( Figure 2A ) . 10 . 7554/eLife . 04543 . 008Figure 2 . Effect of OCA2 disease-associated mutations on IOCA2 . ( A ) Predicted topology of OCA2 with albinism-associated mutations: K614E ( dark blue ) in a conserved cytoplasmic loop , V443I ( purple ) in a highly conserved luminal loop , and 5mut ( light blue ) consisting of 5 point mutations ( V443I , M446V , I473S , N476D , N489D ) in the same luminal loop . ( B ) mCherry-tagged WT , K614E , and V443I OCA2 localized to LAMP1-GFP-positive isolated endolysosomes ( merged in yellow ) , while 5mut did not ( scale bar = 5 μm ) . ( C ) Representative I–V relationships measured in response to voltage ramps from endolysosomes expressing WT , K614E , V443I , or from endolysosomes isolated from cells expressing 5mut OCA2 . ( D ) Average IOCA2 current densities ( pA/pF ) measured at 100 mV ( ±s . e . m . , n = 4–8 endolysosomes for each condition ) . ( E ) Melan-ink4a melanocytes expressing mCherry-tagged ( red ) WT , V443I , or K614E and stained with anti-tyrosinase-related protein one ( TYRP1 ) antibodies ( green ) . Insets , 3× magnification of boxed regions . Merged images ( right ) show that WT , V443I , and K614E OCA2 variants localize primarily to TYRP1-positive compartments ( scale bar = 10 μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 00810 . 7554/eLife . 04543 . 009Figure 2—figure supplement 1 . Localization of wild-type ( WT ) , K614E , V443I , and 5mut OCA2 . ( A ) HeLa cells expressing mCherry-tagged WT , K614E , V443I , or 5mut OCA2 ( red ) and stained with LAMP1 antibodies ( green ) . All OCA2 variants , except 5mut , localized primarily to LAMP1-positive compartments ( orange in merged images ) ( scale bar = 10 μm ) . ( B ) Melan-ink4a melanocytes expressing 5mut OCA2 tagged with mCherry ( red ) and stained with anti-TYRP1 antibodies show that 5mut OCA2-mCherry does not localize to TYRP1-positive compartments ( green ) in melanocytes ( scale bar = 10 μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 009 Upon expression in HeLa cells , the K614E and V443I mutants colocalized with the endolysosomal marker LAMP1 , similar to WT mCherry-OCA2 ( Figure 2—figure supplement 1A ) , indicating that both mutants traffic efficiently to endolysosomes when expressed in non-melanocytic cells . By contrast , 5mut OCA2 localized primarily to endoplasmic reticulum-like structures rather than endolysosomes ( Figure 2—figure supplement 1A ) , therefore it could be used as a negative control . In pigment cells containing melanosomes ( melan-ink4a ) , expression of OCA2 variants revealed that WT , as well as V443I and K614E mutants colocalized with the melanosomal tyrosinase-related protein 1 ( TYRP1; overlap with TYRP1 = 60 . 1 ± 21 . 3% for WT; 50 . 7 ± 13 . 9% for V433I; and 53 . 6 ± 16 . 1% for K614E mCherry-OCA2 ) ( Figure 2E ) , whereas overlap with the endolysosomal marker LAMP2 was minimal ( 11 . 8 ± 10 . 7% among all variants ) . A portion of K614E localized to vesicular structures without TYRP1 but that contained pigment when imaged by bright field microscopy . To determine if albinism-associated mutations affect IOCA2 , we expressed mCherry-tagged WT , K614E , V443I or 5mut OCA2 in AD293 cells together with LAMP1-GFP ( Figure 2B ) . Currents recorded from AD293 endolysosomes identified by LAMP1-GFP and expressing K614E had similar amplitudes and Erev as WT OCA2 ( Figure 2C , D ) , but those expressing V443I had amplitudes reduced by ∼85% and Erev shifted to more positive values ( Figure 2C , D ) . Endolysosomes of cells expressing 5mut had very small current amplitudes and a positive Erev , similar to mock-transfected cells ( Figure 2C , D ) . Collectively , these data suggest that the V443I-containing luminal loop between transmembrane domains 5 and 6 is critical for the OCA2-mediated Cl− current . The K614E mutation had little effect on localization or OCA2-mediated currents in our experiments . Because K614E was identified in albinism patients with an additional OCA2 mutation ( Passmore et al . , 1999 ) , its associated phenotype might be too mild to detect in our assays or might be masked by overexpression . How does the OCA2-mediated anion conductance affect pigmentation ? Early stage melanosomes are highly acidic ( Raposo et al . , 2001 ) , but because tyrosinase is inactive at pH < 6 . 0 , melanosomal pH is thought to increase in order to allow for melanin synthesis ( Ancans et al . , 2001; Halaban et al . , 2002 ) . We tested the hypothesis that OCA2-mediated anion extrusion from melanosomes regulates luminal pH ( Puri et al . , 2000; Brilliant , 2001; Manga and Orlow , 2001; Chen et al . , 2002 , 2004; Ni-Komatsu and Orlow , 2006 ) , similar to other anionic conductances ( Stauber and Jentsch , 2013 ) . We expressed in endolysosomes of AD293 cells ecliptic pHluorin-LAMP1 , which becomes fluorescent at pH > 6 ( Rak et al . , 2011 ) ( Figure 3A ) . Endolysosomes of control cells were acidic and lacked fluorescence emission at baseline ( Figure 3B ) , but exhibited increased fluorescence upon neutralization ( pH > 6 ) of organellar pH by cellular treatment with the vacuolar H+-ATPase inhibitor bafilomycin A1 ( BafA1 ) ( Figure 3A–C ) . In contrast , coexpression of WT OCA2 with pHluorin-LAMP1 resulted in endolysosomes that were fluorescent at baseline , with only a small increase in fluorescence in response to BafA1 ( Figure 3B , C ) . This indicates that OCA2 expression increased the pH of endolysosomes to > 6 . 10 . 7554/eLife . 04543 . 010Figure 3 . OCA2 expression regulates organellar pH . ( A ) Ecliptic pHluorin-LAMP1 fluoresces when luminal pH > 6 , but not when pH < 6 . The V-ATPase inhibitor bafilomycin A1 ( BafA1 , 2 μM ) was used to neutralize acidic endolysosomes and detect pHluorin-LAMP1 expression . ( B ) Endolysosomes from mock-transfected control ( CTL ) AD293 cells expressing pHluorin-LAMP1 were not fluorescent at baseline ( Fo ) but brightly fluoresced after BafA1 treatment ( FBafA1 ) . Endolysosomes coexpressing pHluorin-LAMP1 and mCherry-tagged WT or K614E OCA2 were fluorescent prior to BafA1 treatment ( Fo ) , indicating that expression increased pH . Endolysosomes that coexpressed pHluorin-LAMP1 and V443I were dim at baseline ( Fo ) , similar to CTL and mislocalized 5mut , representing an acidic lumen , and pHluorin-LAMP1 fluorescence increased upon BafA1 treatment ( FBafA1 ) ( scale bar = 10 µm ) . ( C ) Fluorescence intensity following BafA1 treatment was compared with baseline fluorescence to determine relative baseline acidity in enodolysosomes from CTL cells , expressing WT , K614E , V443I , OCA2 or from cells expressing mislocalized 5mut . Bars represent average normalized change in pHluorin-LAMP1 fluorescence ( ±s . e . m . , n = 3 independent experiments , each experiment calculated as the average of n = 78–214 endolysosomes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 01010 . 7554/eLife . 04543 . 011Figure 3—figure supplement 1 . LysoSensor pH measurements . ( A ) Lysosensor calibration curve determined from the emission ratio W1 ( 417–480 nm ) /W2 ( 490–530 nm ) at 405 nm excitation for pH values between 5 and 7 . Representative curve from one experiment ( ±s . e . m . , n = 231–350 endolysosomes ) . ( B ) Predicted pH values based on LysoSensor calibration in LAMP1-expressing control endolysosomes ( CTL ) compared with endolysosomes expressing WT , K614E , or V443I OCA2 ( ±s . e . m . , n = 3 independent experiments , each experiment calculated as the average of n = 109–635 endolysosomes ) . ( C ) Right: Color scale used for Lysosensor . Acidic endolysosomes are shown in green and more neutral ones in blue . Left: AD293 cells loaded with Lysosensor and expressing LAMP1-mCherry ( used as a control ) or mCherry-tagged WT , K614E , V443I OCA2 . Endolysosomes expressing LAMP1-mCherry are green , suggesting that they have acidic pH . Endolysosomes expressing WT or K614E , but not V443I , shift their pH to more neutral values as shown by the blue color ( scale bar = 10 μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 011 To determine if changes in pH required the OCA2-mediated anion conductance , we examined changes in pHluorin-LAMP1 fluorescence elicited by the expression of albinism-associated OCA2 mutants . Endolysosomes expressing K614E exhibited basal fluorescence and little change in fluorescence following treatment with BafA1 ( Figure 3B , C ) , consistent with its intact conductance and localization to endolysosomes ( Figure 2 , Figure 3—figure supplement 1 ) . However , endolysosomes expressing V443I , which have significantly reduced current amplitudes but intact localization ( Figure 2 , Figure 3—figure supplement 1 ) , had dim fluorescence at baseline that increased dramatically following BafA1 treatment ( Figure 3B , C ) . Endolysosomes from cells expressing mislocalized 5mut lacked basal fluorescence and increased their fluorescence with BafA1 treatment ( Figure 3B , C ) , similar to untransfected cells . Together , these results support the notion that OCA2-mediated ion transport in organelles modulates luminal pH . We confirmed the observed OCA2-associated changes in luminal pH using the ratiometric pH-sensitive dye LysoSensor DND-160 . Based on LysoSensor calibration , the luminal pH of LAMP1-positive compartments in control cells had an approximate pH value of 5 . 12 ± 0 . 03 , while the luminal pH of WT OCA2-expressing endolysosomes was significantly greater ( 6 . 67 ± 0 . 03 , Figure 3—figure supplement 1A–C ) . Expression of the K614E mutant increased pH to 6 . 32 ± 0 . 03 , while V443I only modestly raised the luminal pH of endolysosomes to 5 . 72 ± 0 . 04 , ( Figure 3—figure supplement 1B ) . Melanosomal pH measurements were not possible because fluorescent indicator uptake in melanosomes is impaired and melanin interferes with the emission of fluorescent proteins . Our pH measurements are consistent for the two indicators and show that OCA2-mediated Cl− transport shifts the endolysosomal pH toward neutral values that in melanosomes would be optimal for tyrosinase activity and melanin synthesis . The mechanism by which OCA2 modulates pH is unclear . We hypothesize that OCA2-mediated Cl− efflux from the lumen regulates the organelle membrane potential , reducing vacuolar H+-ATPase activity and resulting in less H+ being pumped in the lumen . Alternatively , the pH modulation by the OCA2-mediated Cl− conductance could be a more complex mechanism involving the contribution of additional channels and transporters . Thus , our model for pH regulation by OCA2 remains tentative and awaits a better understanding of melanosomal membrane conductance and signaling . To investigate endogenous OCA2 , we measured native currents by direct patch-clamp recordings from melanosomes . This is technically challenging due to the intracellular localization and small size ( 300–500 nm ) of melanosomes and because treatment of melanocytes with vacuolin-1 did not significantly enlarge melanosomes . To circumvent these difficulties we exploited a dermal melanocyte cell line derived from mice deficient in ocular albinism 1 ( Oa1 ) ( Palmisano et al . , 2008 ) , in which melanosomes are enlarged to up to ∼1 μm diameter ( Cortese et al . , 2005 ) . Recordings from individual melanosomes dissected from Oa1−/− melanocytes under the same conditions used for endolysosome patch-clamp experiments ( Figure 4A ) revealed a large outwardly rectifying current ( Imelano ) with a negative Erev that was dependent on luminal Cl− , similar to currents recorded from OCA2-expressing endolysosomes ( Figure 4B , C ) . When currents recorded in the presence of luminal Gluc− were subtracted from Imelano measured in the presence of Cl− , Erev was similar to ECl ( Figure 4—figure supplement 1A ) . The Erev for Imelano measured with NMDG-based solutions was found to be −44 . 3 ± 2 . 5 mV for 48 mM luminal and 8 mM cytoplasmic [Cl−] and −2 . 3 ± 1 . 1 mV for symmetrical 48 mM [Cl−] ( Figure 4—figure supplement 1B , C ) , consistent with the values measured for endolysosomes expressing OCA2 ( Figure 1E ) and with the Nernst potential for Cl− selective currents . 10 . 7554/eLife . 04543 . 012Figure 4 . Direct recording from melanosomes identifies an endogenous Cl− current ( Imelano ) . ( A ) Dermal macromelanosomes were individually dissected from Oa1−/− melanocytes and patch-clamped . A NaCl-based luminal solution and KGluc-based cytoplasmic solution were used for recordings ( scale bar = 10 μm ) . ( B ) Representative I–V relationships exhibit a Cl−-dependent outwardly rectifying current ( Imelano ) with a negative Erev . Substituting luminal Cl− for Gluc− reduced the current amplitude and shifted Erev of Imelano . ( C ) Average current densities ( pA/pF ) measured at 100 mV . Inset: Average Erev ( ±s . e . m . , n = 4–5 dermal melanosomes for each condition ) . ( D ) Patch-clamp experiments using freshly isolated bullfrog RPE melanosomes ( scale = 10 μm ) . ( E ) Representative whole-RPE melanosome current–voltage ( I–V ) relationships in response to voltage ramps . Outwardly rectifying Imelano was reduced and Erev shifted in the positive direction when luminal Cl− was substituted for Gluc− . ( F ) Average current densities ( pA/pF ) measured at 100 mV . Inset: Average Erev ( ±s . e . m . , n = 5 RPE melanosomes for each condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 01210 . 7554/eLife . 04543 . 013Figure 4—figure supplement 1 . Imelano Erev is dependent on the Cl− concentration gradient . ( A ) Average current densities ( pA/pF ) at the indicated voltages were subtracted to calculate the Cl− component of Imelano . Δ ( luminal NaCl − NaGluc ) current has an Erev near ECl ( ECl = Nernst potential for Cl−; ±s . e . m . , n = 4 melanosomes per condition ) . ( B ) In a representative dermal melanosome , Erev was dependent on [Cl−]cyto . NMDG-based solutions were used to inhibit endogenous cation permeability . ( C ) Imelano Erev as a function of [Cl−]cyto , when [Cl−]luminal was kept at 48 mM . Each point represents a melanosome at the indicated [Cl−]cyto on the x-axis . Assuming linear Erev dependence on [Cl−]cyto , the slope of the fitted line is 56 mV/10-fold Δ[Cl−]cyto . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 01310 . 7554/eLife . 04543 . 014Figure 4—figure supplement 2 . RPE melanosome dissection . Melanosomes were dissected from freshly isolated bullfrog RPE cells using two glass pipettes . The sequence of images shows one pipette used to hold the non-adherent cell and one to dissect the cell to isolate individual melanosomes ( red circle ) for patch-clamp experiments ( scale = 10 μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 014 In addition to skin melanocytes , melanosomes are also present in the retinal pigment epithelium of the eye . We patch-clamped melanosomes dissected from freshly isolated RPE cells from American bullfrog ( Lithobates catesbeianus ) that are larger than those from mammalian RPE , thus allowing for patch-clamp experiments ( Figure 4E , Figure 4—figure supplement 2 ) . RPE whole-melanosome recordings identified an outwardly rectifying current with a negative Erev that was nearly abolished ( reduced by ∼90% ) by substituting luminal Cl− with Gluc− ( Figure 4E , F ) . Together , these data indicate that a current with similar properties as IOCA2 is present in melanosomes from skin and RPE . To determine if OCA2 contributes to Imelano , we used OCA2-targeted siRNA to reduce OCA2 expression ( Figure 5—figure supplement 1A ) . Consistent with previous findings ( Sviderskaya et al . , 1997 ) , reducing endogenous OCA2 expression markedly decreased melanin content in Oa1−/− melanocytes ( Figure 5A , Figure 5—figure supplement 1B ) . Melanosomes dissected from melanocytes expressing OCA2-targeted siRNA had dramatically reduced current amplitudes and melanin content compared with melanosomes from cells expressing scrambled siRNA ( control ) , indicating that OCA2 is required for Imelano and pigmentation ( Figure 5B , C ) . Expression of WT OCA2 , but not of the V443I mutant , in siRNA-treated cells was sufficient to reconstitute Imelano and rescue melanization ( Figure 5A–C ) , suggesting that the OCA2-mediated current is required for melanin production . 10 . 7554/eLife . 04543 . 015Figure 5 . OCA2 is required for Imelano and pigmentation . ( A ) Representative images of Oa1−/− melanocytes expressing scrambled or OCA2-targeted siRNA , the latter of which significantly decreased melanin content . Pigmentation was restored by transfection with mCherry-tagged WT OCA2 , but not with the mCherry-tagged V443I mutant ( scale bar = 10 μm ) . ( B ) Representative I–V relationships from melanosomes from cells expressing scrambled siRNA , OCA2-targeted siRNA , and rescued with WT or V443I OCA2 . Insets: images of the individual melanosomes used for the shown recordings ( scale bar = 3 μm ) . ( C ) Average Imelano current densities ( pA/pF ) measured at 100 mV in melanosomes from cells expressing scrambled siRNA or OCA2-targeted siRNA ( ±s . e . m . , n = 4 melanosomes per condition ) . ( D ) Average current densities ( pA/pF ) measured at 100 mV in melanosomes expressing WT or V443I OCA2 from cells expressing OCA2-targeted siRNA ( ±s . e . m . , n = 3–4 melanosomes per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 01510 . 7554/eLife . 04543 . 016Figure 5—figure supplement 1 . OCA2-targeted siRNA reduces OCA2 mRNA and melanin content in Oa1−/− melanocytes . ( A ) OCA2 mRNA levels in melanocytes expressing OCA2-targeted siRNA was reduced by ∼90% compared with scrambled siRNA-expressing melanocytes ( ±s . e . m . , n = 3 experiments ) . ( B ) Melanin content of melanocytes expressing OCA2-targeted siRNA was ∼73% lower than that of scrambled siRNA-expressing melanocytes ( ±s . e . m . , n = 3 experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 01610 . 7554/eLife . 04543 . 017Figure 5—figure supplement 2 . Melanin is not required for Imelano . ( A ) Representative images of melanosomes from Oa1−/− melanocytes treated with the tyrosinase inhibitor phenylthiourea ( PTU , 300 μM ) have significantly reduced pigment ( scale bar = 3 μm ) . ( B ) A similar Imelano was measured from melanosomes of PTU-treated and control Oa1−/− melanocytes . ( C ) Average current densities ( pA/pF ) measured at 100 mV . Inset: Average Erev ( ±s . e . m . , n = 5 melanosomes per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 017 Might melanin itself be critical for Imelano ? To address this question , we treated Oa1−/− melanocytes with phenylthiourea ( PTU ) , a potent inhibitor of the key melanogenic enzyme tyrosinase , resulting in melanin depletion ( Figure 5—figure supplement 2A ) . Recording from melanosomes dissected from PTU-treated melanocytes revealed a current with the same properties as in vehicle-treated cells and similar to Imelano , suggesting that melanin does not influence Imelano ( Figure 5—figure supplement 2B , C ) . We thus concluded that OCA2 expression is required for Imelano and melanization , but melanin is not required for Imelano . We next sought to compare the biophysical properties of the endogenous OCA2-mediated melanosome current ( Imelano ) in dermal and RPE melanosomes with those of the OCA2-expressing endolysosome current ( IOCA2 ) . Excised cytoplasmic-side-out patches from partially dissected endolysosomes of OCA2-expressing AD293 cells ( Figure 6A ) exhibited single-channel currents ( iOCA2 ) with greatest activity at positive membrane potentials ( Figure 6B ) , consistent with the outward rectification measured for whole-endolysosome IOCA2 ( Figure 1B ) . Patches from mock-transfected endolysosomes did not exhibit similar single-channel activity ( Figure 6—figure supplement 1A , B ) . The Erev of iOCA2 was −66 mV , close to ECl predicted by the Nernst equation ( −68 mV ) , and the unitary slope conductance was 58 ± 2 pS ( Figure 6C ) . 10 . 7554/eLife . 04543 . 018Figure 6 . Endolyosomal IOCA2 has the same properties as melanosomal Imelano . ( A ) Cytoplasmic-side-out patches were excised from partially dissected organelles to carry out single-channel recordings . ( B ) Representative single-channel currents recorded in response to voltage steps between −80 mV and +80 mV from a patch excised from an OCA2-expressing endolysosome ( O indicates the open state and C the closed state of the channel ) . ( C ) Average single-channel current amplitudes from patches excised from OCA2-expressing endolysosomes ( ±s . e . m . , n = 4 patches ) ( ECl = Nernst potential for Cl− ) . ( D ) Single-channel currents from a representative patch excised from a dermal melanosome that contains at least two channels ( O1 indicates one open channel , O2 indicates two channels open simultaneously , and C both channels closed ) . ( E ) Average single-channel current amplitudes from patches excised from dermal melanosomes ( ±s . e . m . , n = 3 patches ) . ( F ) In a representative OCA2-expressing endolysosome , dermal melanosome , and RPE melanosome , currents were selective for Cl− and Br− , but not I− , F− , or Gluc− . In the same organelle , currents were recorded with 48 mM luminal NMDG-Cl while substituting symmetrical concentrations of cytoplasmic anions in an NMDG-based solution . ( G ) IOCA2 permeability ratios based on Erev measurements ( ±s . e . m . , n = 3–7 organelles for each condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 01810 . 7554/eLife . 04543 . 019Figure 6—figure supplement 1 . OCA2-mediated single-channel current characteristics . ( A ) Top: Representative single-channel currents recorded at 80 mV from cytoplasmic-side-out patches excised from an endolysosome isolated from a mock-transfected AD293 cell , OCA2-expressing endolysosome , and dermal melanosome . Bottom: Associated amplitude histograms . Inset: None of the endolysosome patches from mock-transfected cells exhibited significant single-channel activity ( scale = 10 pA , 250 ms ) . ( B ) Average maximal current amplitudes at 80 mV calculated from the Gaussian fit of amplitude histograms of individual membrane patches excised from mock-transfected , OCA2-expressing endolysosomes , or dermal melanosomes ( ±s . e . m . , n = 4–6 patches for each condition , i = 0 . 18 ± 0 . 18 pA for mock-transfected AD293 endolysosomes , 8 . 59 ± 0 . 46 pA for OCA2-expressing endolysosomes , 9 . 00 ± 0 . 51 pA for dermal melanosomes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04543 . 019 Ion channels are pore-proteins that allow rapid diffusion of ions ( >106 ions/s ) down their concentration gradients , while carriers and pumps typically move ions at slower rates ( <105 ions/s ) . We thus calculated the OCA2-mediated transport rate at 80 mV , where we measured the largest single-channel amplitude , and found a transport rate of ∼5 . 5 × 107 ions/s . The large conductance recorded in isolated patches from OCA2-expressing endolysosomes indicates that OCA2 functions as a critical component of an electrodiffusive anion channel . Similar properties were measured for single-channel currents in excised cytoplasmic-side-out patches of dermal melanosomes from Oa1−/− melanocytes ( imelano ) ( Erev = −67 mV and unitary slope conductance = 60 ± 1 pS ) ( Figure 6D , E ) . Thus , the single-channel properties of this endogenous melanosomal current are nearly identical to those recorded from OCA2-expressing endolysosomes . To determine the anion selectivity of the recorded currents , we estimated the permeability to different anions by determining Erev under symmetrical concentrations ( 48 mM ) of luminal Cl− and cytoplasmic Cl− , Br− , I− , F− or Gluc− . Erev for IOCA2 was close to zero for cytoplasmic Cl− and Br− and shifted to negative voltages for I− , F− , and Gluc− ( Figure 6F ) , suggesting that in endolysosomes OCA2 mediates a current selective for Cl− and Br− , but poorly permeable to I− , F− , and Gluc− . A similar shift in Erev was measured for Imelano in dermal or RPE melanosomes ( Figure 6F ) , indicating that the endogenous currents have the same selectivity profile . Moreover , the calculated permeability ratios ( Px/PCl ) for the heterologously expressed currents were nearly the same as the endogenous ones ( Figure 6G ) . Collectively , these results suggest that IOCA2 and Imelano are mediated by the same anion channel . Our results suggest that OCA2 is an essential component of a melanosome-specific anion channel . Heterologous endolysosomal expression of OCA2 contributes to a chloride channel with biophysical properties similar to an endogenous melanosomal OCA2-mediated channel . Importantly , OCA2 activity controls the melanin content of melanosomes , most likely by regulating organellar pH . We propose that OCA2 contributes to a novel melanosome-specific anion current that modulates melanosomal pH for optimal tyrosinase activity required for melanogenesis . All cells were grown at 37°C and 5% CO2 , and reagents were from Invitrogen/Life Technologies ( Grand Island , NY ) unless stated otherwise . AD293 , HeLa , and NF-SV60 fibroblasts were grown in DMEM , 10% fetal bovine serum ( FBS , Atlanta Biologicals , Flowery Branch , GA ) with or without 1% penicillin/streptomycin ( P/S ) . Immortalized mouse melanocyte cell lines melan-Ink4a ( Sviderskaya et al . , 2002 ) and melan-Oa1 ( Palmisano et al . , 2008 ) were grown in RPMI 1640 , 10% FBS , 1% P/S , 200 nM phorbol 12-myristate-13-acetate ( Sigma , St . Louis , MO ) at 37°C and 10% CO2 . Primary human epidermal melanocytes and keratinocytes were isolated from neonatal foreskin . Human epidermal melanocytes ( Cascade Biologics/Life Technologies ) were cultured in Medium 254 , Human Melanocytes Growth Supplement , and 1% P/S . Human epidermal keratinocytes ( Lifeline Cell Technology , Frederick , MD ) were cultured in Dermalife basal medium supplemented with LifeFactors and 1% P/S . Transfection of cells used for electrophysiology and imaging was carried out using Lipofectamine 2000 ( Invitrogen/Life Technologies ) according to manufacturer's protocol , unless otherwise stated . OCA2 tagged with GFP or mCherry was generated by inserting human OCA2 ( hOCA2 , NM_000275 ) between the BamHI/Xhol sites of pcDNA4/TO ( Invitrogen/Life Technologies ) . K614E and V443I mutations in hOCA2 were made using site-directed mutagenesis and verified by sequencing . 5mut hOCA2 was generated by Genscript ( Piscataway , NJ ) . Melan-ink4a melanocytes were seeded on coverslips coated with Matrigel ( BD Biosciences , San Jose , CA ) at 3–4 × 104 cells/well in a 24-well plate , transfected the next day with 0 . 8 μg of plasmid DNA; HeLa cells were seeded on glass coverslips at 105 cells/well in a six well plate , transfected the next day using GeneJuice ( EMD Millipore ) as recommended by the manufacturer , with 0 . 1 µg of plasmid DNA . Both cell types were analyzed 48 hr post-transfection . Cells were fixed in HBSS ( Invitrogen ) /2% paraformaldehyde ( Sigma ) for 20 min at roo temperature , washed once with PBS and labeled with primary and secondary antibodies diluted in PBS with 0 . 2% ( wt/vol ) saponin , 0 . 1% ( wt/vol ) bovine serum albumin , 0 . 02% ( wt/vol ) sodium azide as described ( Calvo et al . , 1999 ) . Nuclei were labeled with 500 ng/µl Hoescht 33342 ( Sigma ) . Antibodies used were: mouse anti-TYRP1 ( TA99/mel-5 , ATCC ) ; mouse anti-human LAMP1 ( H4A3 ) and rat anti-mouse LAMP2 ( GL2A7; both from Developmental Studies Hybridoma Bank , Iowa City , IA ) . Donkey antibodies specific to mouse or rat immunoglobulin and conjugated to Dylight 594 or 488 were from Jackson Immunoresearch ( West Grove , PA ) . Cells were imaged using a 100× HCX PL APO Lens on a Leica DM IRBE microscope equipped with a Retiga Exi Fast 1394 digital camera ( Qimaging , Surrey , Canada ) and Improvision Openlab software ( Perkin–Elmer , Waltham , MA ) . Sequential z-stack images separated by 0 . 2 µm were acquired and deconvolved using the OpenLab Volume Deconvolution module . Images from single stacks are shown . Final images were generated and insets magnified using Photoshop ( Adobe , San Jose , CA ) . Colocalization analyses were performed using OpenLab software , as described ( Setty et al . , 2007 ) . Briefly , images of individual cropped cells were rendered binary using the ‘Density slice’ module and the densely labeled perinuclear region was excluded from analysis . Pixel overlap between binary red and green images in the remaining cell regions was defined using the ‘Boolean operations’ module and further analyzed using Excel ( Microsoft , Redmond , WA ) . Melanin from Oa1−/− melanocytes expressing scrambled or OCA2-targeted siRNA was quantified as previously described ( Oancea et al . , 2009 ) . Briefly , the soluble and insoluble fractions of melanocytes were separated after cell lysis with 1% Triton X-100 ( Sigma ) in PBS pH 7 . 4 . Total protein was measured in the soluble fraction using a Bradford Assay ( Bio-Rad Laboratories , Waltham , MA ) . The insoluble fraction was dissolved in 1 N NaOH by incubation for 30 min at 80°C and used to quantify melanin by measuring the optical density of each sample at 405 nm , then fit with a standard curve generated using synthetic melanin ( Sigma ) . Average cellular melanin values were quantified as the ratio between total melanin and total protein from the same dish . All data are shown as mean ± s . e . m . Data were considered significant if p < 0 . 05 using unpaired two-tailed Student t test or one-way ANOVA .	Melanin is a pigment found in our skin , eyes and hair . Individuals who are unable to make or store melanin , a condition known as albinism , have unusually pale features and problems with vision . The pigment helps to protect us from harmful UV radiation , and so individuals with albinism also have an increased risk of developing skin and eye cancers . In cells , melanin is made and stored in compartments called melanosomes . The most common type of albinism is caused by defects in a protein called OCA2 , which is found in the membrane that surrounds melanosomes . However the role of OCA2 in melanin production is unclear . It has been proposed that OCA2 may allow charged particles ( or ions ) to enter or leave melanosomes . Here , Bellono et al . used a technique called patch-clamp to study the movement of ions across the membrane of melanosomes from skin and eye cells . The experiments show that a flow of chloride ions out of the melanosome is required for melanin to be produced . OCA2 is involved in the ion movement , and it might alter the acidity of the melanosome when present . Bellono et al . propose that OCA2 is part of an ion channel that allows chloride ions to pass through the membrane , to make the melanosome less acidic and enable melanin to be produced . The next challenge will be to identify other ion channels in the melanosome and understand their roles in producing melanin .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	An intracellular anion channel critical for pigmentation
mir-17-92 , a potent polycistronic oncomir , encodes six mature miRNAs with complex modes of interactions . In the Eμ-myc Burkitt’s lymphoma model , mir-17-92 exhibits potent oncogenic activity by repressing c-Myc-induced apoptosis , primarily through its miR-19 components . Surprisingly , mir-17-92 also encodes the miR-92 component that negatively regulates its oncogenic cooperation with c-Myc . This miR-92 effect is , at least in part , mediated by its direct repression of Fbw7 , which promotes the proteosomal degradation of c-Myc . Thus , overexpressing miR-92 leads to aberrant c-Myc increase , imposing a strong coupling between excessive proliferation and p53-dependent apoptosis . Interestingly , miR-92 antagonizes the oncogenic miR-19 miRNAs; and such functional interaction coordinates proliferation and apoptosis during c-Myc-induced oncogenesis . This miR-19:miR-92 antagonism is disrupted in B-lymphoma cells that favor a greater increase of miR-19 over miR-92 . Altogether , we suggest a new paradigm whereby the unique gene structure of a polycistronic oncomir confers an intricate balance between oncogene and tumor suppressor crosstalk . MicroRNAs ( miRNAs ) are a class of small , non-coding RNAs that regulate post-transcriptional gene repression in a variety of developmental and pathological processes ( Ambros , 2004; Zamore and Haley , 2005; Bartel , 2009; Kim et al . , 2009 ) . Due to their small size and the imperfect nature of target recognition , miRNAs have the capacity to regulate many target mRNAs through translational repression and mRNA degradation , thereby acting as global regulators of gene expression ( Lewis et al . , 2005; Filipowicz et al . , 2008 ) . Unlike mammalian protein-coding genes that follow the one-transcript , one-protein paradigm , many miRNA genes are expressed as polycistronic primary transcripts , generating multiple mature miRNAs under the same transcriptional regulation ( Megraw et al . , 2007 ) . miRNA polycistrons further expand the gene regulatory capacity , since different miRNA components can confer specific yet overlapping biological effects , and their functional interactions can yield unusual complexity . Polycistronic miRNAs often exhibit pleiotropic biological functions with unique gene regulatory mechanisms ( Megraw et al . , 2007 ) . One of the best example is mir-17-92 , a potent oncomir ( i . e . , miRNA oncogene ) , whose genomic amplification and aberrant overexpression have been observed in many human tumors including Burkitt’s lymphoma , diffuse large B-cell lymphoma ( DLBCL ) , and lung cancer ( Lu et al . , 2005; Mendell , 2008 ) . mir-17-92 regulates multiple cellular processes during tumor development , including proliferation , survival , angiogenesis , differentiation , and metastasis ( He et al . , 2007; Uziel et al . , 2009; Conkrite et al . , 2011; Nittner et al . , 2012 ) . As a polycistronic oncomir , mir-17-92 produces a single precursor that yields six individual mature miRNAs ( Figure 1A , Figure1—figure supplement 1A ) ( Tanzer and Stadler , 2004 ) . Based on the seed sequence homology , the six mir-17-92 components are categorized into four miRNA families ( Figure 1A , Figure 1—figure supplement 1A ) : miR-17 ( miR-17 and 20 ) , miR-18 , miR-19 ( miR-19a and 19b ) , and miR-92a ( we will designate miR-92a as miR-92 in the remainder of our paper ) . Interestingly , miR-92 has a more ancient evolutionary history compared to the other mir-17-92 components ( Tanzer and Stadler , 2004 ) . miR-92 is evolutionarily conserved in vertebrates , chordates , and invertebrates , while the remaining mir-17-92 components are only found in vertebrates ( Figure 1—figure supplement 1B , C ) . Conceivably , the distinct mature miRNA sequence of each mir-17-92 component determines the specificity of the target regulation . However , the functional significance of the mir-17-92 polycistronic gene structure remains largely unknown . 10 . 7554/eLife . 00822 . 003Figure 1 . miR-92 negatively regulates the mir-17-92 oncogenic activity in the Eμ-myc B-lymphoma model . ( A ) The gene structure of the mir-17-92 polycistron and its mutated derivatives . Light colored boxes , pre-miRNAs; dark colored boxes , mature miRNAs . Homologous miRNA components are indicated by the same color . ( B ) Schematic representation of the adoptive transfer protocol using Eμ-myc hematopoietic stem and progenitor cells ( HSPCs ) . Eμ-myc/+ HSPCs were extracted from E13 . 5–E15 . 5 mouse embryos , infected with MSCV retroviral vectors overexpressing mir-17-92 and its derivatives , and finally transplanted into lethally irradiated recipient mice . Lymphoma onset of the adoptive transferred mice was monitored to evaluate the oncogenic collaboration between c-Myc and a specific miRNA . ( C ) miR-92 deficiency specifically accelerates the oncogenic activity of mir-17-92 in the Eμ-myc model . Using the Eμ-myc adoptive transfer model , we compared the oncogenic effects between mir-17-92 and mir-17-92Δ92 and observed a significant acceleration of tumor onset in Eμ-myc/mir-17-92Δ92 mice ( p<0 . 0001 , left ) . When the oncogenic effects of mir-17-92 , mir-17-92Δ92 and mir-17-92Mut92 were compared in the same adoptive transfer model , mir-17-92Δ92 and mir-17-92Mut92 similarly accelerated Eμ-myc-induced lymphomagenesis compared to mir-17-92 ( p<0 . 0001 for both comparisons , middle ) . Deficiency of miR-20 failed to affect the oncogenic cooperation between mir-17-92 and Eμ-myc , having minimal effects on tumor onset ( right ) . ( D ) The mutation of miR-92 has minimal effects on the levels of the remaining mir-17-92 components . Eμ-myc B-lymphoma cells were infected with MSCV retrovirus overexpressing mir-17-92 , mir-17-92Δ92 and mir-17-92Mut92 at an MOI ( multiplicity of infection ) of 1 . Expression levels of miR-17 , 18a , 19a , 20a , 19b and 92 were subsequently determined using Taqman miRNA assays . Error bars indicate standard deviation ( n = 3 ) . p<0 . 01 , p<0 . 001 , **p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 00310 . 7554/eLife . 00822 . 004Figure 1—figure supplement 1 . Gene structure and evolutionary conservation of mir-17-92 . ( A ) A diagram represents the gene structure of mir-17-92 and its two mammalian homologs . The six mir-17-92 components are classified into four distinct miRNA families based on the seed sequence conservation . ( B and C ) miR-92 has a more ancient evolutionary history compared to the rest of mir-17-92 components . miR-92 is evolutionarily conserved in Deuterostome , Ecdysozoa and Lophotrochozoa , yet the remaining mir-17-92 components only have vertebrate homologs . ( D ) The mutation of miR-92 or miR-20 in the mir-17-92 retroviral construct has minimal effects on the expression levels of the remaining mir-17-92 components . 3T3 cells were infected with MSCV retrovirus at an MOI ( multiplicity of infection ) of 1 to overexpress mir-17-92 , mir-17-92Δ92 and mir-17-92Mut92 ( left ) , or overexpress mir-17-92Mut20 ( right ) . Expression levels of miR-17 , 18a , 19a , 20a , 19b and 92 were each determined using Taqman miRNA assays . Error bars indicate standard deviation ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 004 The structural analogy to prokaryotic operons has led to the speculation that the co-transcribed mir-17-92 components can collectively contribute to oncogenesis . However , our studies reveal an unexpected functional interaction among mir-17-92 components . In the Eμ-myc mouse B-cell lymphoma model , while the intact mir-17-92 acts as an oncogene , its miR-92 component negatively regulates the oncogenic cooperation with c-Myc . This effect , at least in part , results from the ability of miR-92 to yield aberrant c-Myc dosage , which promotes a strong coupling between oncogene stress and p53-dependent apoptosis . Surprisingly , miR-92 functionally antagonizes miR-19 , a key oncogenic mir-17-92 component , in the context of c-Myc-induced oncogenesis . During B-cell transformation , this miR-19:miR-92 antagonism is disrupted to favor a greater increase of miR-19 than miR-92 . Thus , the polycistronic mir-17-92 employs an antagonistic interaction among its encoded miRNA components to confer an intricate crosstalk between the oncogene and tumor suppressor networks . Since mir-17-92 is overexpressed in human Burkitt’s lymphomas ( Tagawa et al . , 2007 ) , we set out to functionally dissect mir-17-92 components in the Eμ-myc model of Burkitt’s lymphoma ( Figure 1B ) . The Eμ-myc mice carry a c-myc transgene downstream of the immunoglobulin ( Ig ) heavy chain enhancer Eμ ( Langdon , 1986; Adams et al . , 1985 ) , which functionally resembles the Ig-MYC translocations that occur frequently in Burkitt’s lymphomas ( Tagawa et al . , 2007 ) . The resulting B-cell specific , aberrant c-Myc activation promotes excessive proliferation , yet also evokes potent , p53-dependent apoptosis ( Schmitt et al . , 2002; Hemann et al . , 2003 ) . Thus , c-Myc-induced apoptosis enables a self-defense mechanism against malignant transformation , producing B-lymphomas with a late onset ( Lowe et al . , 2004 ) . In our adoptive transfer model ( Olive et al . , 2009 ) , Eμ-myc/+ hematopoietic stem and progenitor cells ( HSPCs ) were transplanted into lethally irradiated recipient mice , generating chimeric mice that faithfully recapitulated the late tumor onset of the Eμ-myc transgenic mice ( Figure 1B ) . When Eμ-myc/+ HSPCs were infected with MSCV ( murine stem cell virus ) retrovirus to overexpress the intact mir-17-92 oncomir , we observed a considerable acceleration in tumor onset compared to the Eμ-myc/MSCV control mice ( p<0 . 01 , Figure 1C ) . Unexpectedly , the oncogenic cooperation between c-Myc and mir-17-92 was significantly stronger when miR-92 was deleted within this oncomir ( Figure 1C ) . The average survival of Eμ-myc/17-92Δ92 mice was 66 days , significantly shorter than that of Eμ-myc/17-92 mice ( 112 days , p<0 . 0001 ) . mir-17-92Δ92 carried a deletion of miR-92 pre-miRNA and its flanking sequences , which might alter the expression of the remaining mir-17-92 components ( Figure 1D , Figure 1—figure supplement 1D ) . We then engineered a 12-nucleotide miR-92 seed mutation within mir-17-92 to abolish the functional miR-92 with minimal disruption to the overall gene structure . The resulting mir-17-92Mut92 phenocopied mir-17-92Δ92 in vivo ( Figure 1C ) , significantly enhancing the oncogenic cooperation with c-Myc without altering the level of any remaining mir-17-92 components ( Figure 1D , Figure 1—figure supplement 1D ) . This unexpected effect was specifically attributable to miR-92 . Mutations of miR-20 or miR-17 failed to affect oncogenesis in the Eμ-myc model ( Figure 1C , Figure 1—figure supplement 1D and data not shown ) , and mutations of both miR-19 miRNAs nearly abolished this oncogenic cooperation ( Olive et al . , 2009 ) . This finding suggests that , although mir-17-92 acted as a potent oncogene as a whole , its miR-92 component confers an internal negative regulation on its oncogenic cooperation with c-Myc . This effect of miR-92 clearly contrasts with that of miR-19 , a key oncogenic mir-17-92 component that promotes c-Myc-induced lymphomagenesis by repressing apoptosis ( Mu et al . , 2009; Olive et al . , 2009; Mavrakis et al . , 2010 ) . In the Eμ-myc model , a strong oncogenic lesion often leads to the B-cell transformation at an earlier developmental stage ( Hemann et al . , 2003 ) . The greater oncogenic activity of mir-17-92Mut92 in comparison with mir-17-92 was consistent with mir-17-92Mut92 preferentially transforming IgM negative progenitor B-cells , and mir-17-92 frequently transforming IgM positive B-cells ( Figure 2A; Table 1 ) . In comparison to Eμ-myc/17-92 mice , both Eμ-myc/17-92Δ92 and Eμ-myc/17-92Mut92 mice developed more aggressive B-lymphomas , characterized by massive lymph node enlargement , splenic hyperplasia , leukemia , and widespread dissemination into visceral organs outside of the lymphoid compartment ( Figure 2B , data not shown ) . 10 . 7554/eLife . 00822 . 005Figure 2 . The miR-92 deficient mir-17-92 cooperates with c-Myc to promote highly aggressive B-lymphomas . ( A ) The percentage of IgM positive and IgM negative B-lymphomas was calculated for each genotype ( Eμ-myc/MSCV , n = 10; Eμ-myc/17-92 , n = 9; Eμ-myc/17-92Δ92 , n = 10; Eμ-myc/17-92Mut92 , n = 10 ) . ( B ) The Eμ-myc/17-92Mut92 and Eμ-myc/17-92Δ92 mice developed high grade B-lymphomas that were frequently disseminated into the liver . When compared to Eμ-myc/MSCV and Eμ-myc/17-92 mice , Eμ-myc/17-92Mut92 and Eμ-myc/17-92Δ92 lymphomas gave rise to more liver dissemination , as indicated by H&E and B220 staining . ( C ) Eμ-myc/17-92Mut92 and Eμ-myc/17-92Δ92 lymphomas exhibited a decreased apoptosis compared to Eμ-myc/MSCV or Eμ-myc/17-92 lymphomas . Representative lymphomas were stained for H&E , cleaved caspase-3 and PCNA . Arrow , ‘starry sky’ feature of apoptotic lymphoma cells; arrowhead , apoptotic cells with positive staining for cleaved caspase-3; scale bar , 50 μm . ( D and E ) Apoptosis was quantitatively measured in representative lymphomas of each genotype using the ‘starry sky’ features ( D ) and cleaved caspase-3 staining ( E ) . p<0 . 05 , p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 00510 . 7554/eLife . 00822 . 006Table 1 . Flow cytometric immunophenotyping of Eμ-myc lymphomas with enforced expression of different mir-17-92 derivativesDOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 006GenotypenPercentage ( % ) ImmunotypeEμ-myc/MSCV440B220+ , IgM− , CD19+ , CD4− , CD8−660B220+ , IgM+ , CD19+ , CD4− , CD8− Eμ-myc/17–92440B220+ , IgM− , CD19+ , CD4− , CD8−550B220+ , IgM+ , CD19+ , CD4− , CD8− †110B220− , IgM− , CD19− , CD4+ , CD8+Eμ-myc/17–92Mut92770B220+ , IgM− , CD19+ , CD4− , CD8−330B220+ , IgM+ , CD19+ , CD4− , CD8− ‡Eμ-myc/1792Δ92880B220+ , IgM− , CD19+ , CD4− , CD8−220B220+ , IgM+ , CD19+ , CD4− , CD8− §1 out of 6 samples predominantly contains IgM+ cells , with a small percentage of IgM− cells . †3 out of 5 samples predominantly contain IgM+ cells , with a small percentage of IgM− cells . ‡1 out of 3 samples predominantly contains IgM+ cells , with a small percentage of IgM− cells . §1 out of 2 samples predominantly contains IgM+ cells , with a small percentage of IgM− cells . During Myc-induced tumorigenesis , aberrant c-Myc dosage yields simultaneous induction of proliferation and apoptosis , imposing a unique selective pressure for pro-survival lesions ( Evan and Vousden , 2001 ) . Thus , we compared the extent of Myc-induced apoptosis in the Eμ-myc/17-92 , Eμ-myc/17-92Δ92 , Eμ-myc/17-92Mut92 , and control Eμ-myc/MSCV lymphomas . The control Eμ-myc/MSCV lymphomas invariably exhibited a high proliferation index accompanied by extensive cell death , as evidenced by the widespread ‘starry sky’ pathology ( Figure 2C , D ) and cleaved caspase 3 staining ( Figure 2C , E ) . The potent oncogenic activity of mir-17-92Δ92 and mir-17-92Mut92 was consistent with the strong reduction of apoptosis in the lymph node tumors . In comparison , the intact miR-92 significantly attenuated the repression of c-Myc-induced apoptosis by mir-17-92 in vivo ( Figure 2C–E ) . We next investigated the effect of miR-92 alone in regulating c-Myc-induced apoptosis . In the Eμ-myc model , miR-92 overexpression significantly enhanced c-Myc-induced apoptosis in vivo ( Figure 3A , B , Figure 3—figure supplement 1A ) , consistent with a rapid depletion of miR-92-infected cells in premalignant Eμ-myc B-cells ( Figure 3—figure supplement 1B ) . Similar miR-92 effects on c-Myc-induced apoptosis were observed in vitro . The R26MER/MER mouse embryonic fibroblasts ( MEFs ) carry a switchable variant of Myc , MycERT2 , downstream of the constitutive Rosa26 promoter , which allows acute activation of the MycER transgene by 4-OHT ( 4-Hydroxytamoxifen ) induced nuclear translocation ( Murphy et al . , 2008 ) . The R26MER/MER MEFs recapitulate c-Myc-induced apoptosis in vitro , as activated MycERT2 induces p53-dependent apoptosis in response to serum starvation ( Murphy et al . , 2008 ) . Enforced miR-92 expression in R26MER/MER MEFs invariably enhanced Myc-induced apoptosis ( Figure 3C , Figure 3—figure supplement 1C ) . 10 . 7554/eLife . 00822 . 007Figure 3 . miR-92 enhances both c-Myc-induced apoptosis and c-Myc-induced proliferation . ( A ) The schematic representation of the adoptive transfer model to evaluate the miR-92 effects on the Eμ-myc premalignant B-cells in vivo . ( B ) miR-92 overexpression enhances the apoptotic response in the premalignant Eμ-myc B-cells in vivo . Using the Eμ-myc adoptive transfer model , we generated well-controlled Eμ-myc/MSCV and Eμ-myc/92 mice reconstituted from donor matched Eμ-myc HSPCs . Premalignant Eμ-myc splenic B-cells were isolated from the Eμ-myc/MSCV and Eμ-myc/92 mice 6 weeks after reconstitution . The in vivo apoptosis was measured by the level of caspase activation using Red-VAD-FMK , a fluorescently labeled caspase inhibitor that specifically bound to cleaved caspases . The percentage of Eμ-myc B-cells positive for cleaved caspases was shown for four independent experiments . ( C ) Enforced miR-92 expression in R26MER/MER MEFs significantly enhanced c-Myc-induced apoptosis . miR-92 overexpressing and the control R26MER/MER MEFs were serum starved , and the MycERT2 transgene was activated by 4-OHT treatment . The level of apoptosis of each MEF was measured using Annexin V staining before ( left ) and after ( middle ) 4-OHT treatment and serum starvation . Quantification of c-Myc-induced apoptosis was performed in three independent MEF lines that overexpressed MSCV or miR-92 ( right panel , error bars represent SEM ) . ( D ) Enforced miR-92 expression in R26MER/MER MEFs significantly enhanced c-Myc-induced proliferation . Proliferative effects of miR-92 was measured by BrdU incorporation in MycERT2 activated R26MER/MER MEFs . miR-92 cooperated with c-Myc to promote BrdU incorporation in both 10% ( left ) and 0 . 2% ( middle ) serum culture conditions . Quantification of BrdU incorporation was performed in two independent experiments ( right ) . ( E ) miR-92 is a potent mir-17-92 component to promote primary B-cell proliferation . The proliferative effects of all mir-17-92 miRNAs were measured individually in primary B-cells using BrdU incorporation . ( F ) The quantification of BrdU incorporation in experiments described in ( E ) was performed in four independent experiments . Error bars represent standard deviation , p<0 . 05 , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 00710 . 7554/eLife . 00822 . 008Figure 3—figure supplement 1 . miR-92 enhances c-Myc-induced apoptosis both in vitro and in vivo . ( A ) miR-92 enhances the apoptotic response in the premalignant Eμ-myc B-cells in vivo . Using the Eμ-myc adoptive transfer model , we generated well-controlled Eμ-myc/MSCV and Eμ-myc/92 mice that were reconstituted from the same Eμ-myc HSPCs . The in vivo apoptosis was measured by the level of caspase activation 6 weeks after the transplantation . The percentage of Eμ-myc B-cells positive for cleaved caspases was shown for four independent experiments . ( B ) miR-92 infected , premalignant Eμ-myc B-cells is significantly depleted in the Eμ-myc adoptive transfer model . We generated well-controlled Eμ-myc/MSCV and Eμ-myc/92 mice reconstituted from the same Eμ-myc HSPCs . We measured the percentage of retrovirally infected cells ( GFP+ ) before reconstitution ( left ) , and demonstrated similar infection efficiency in Eμ-myc/MSCV and Eμ-myc/92 mice . At day 33 post adoptive transfer , we isolated white blood cells from the peripheral blood of these mice , and measured the percentage of retrovirally infected , Eμ-myc B-cells ( B-220-positive; GFP-positive cells ) using FACS . Error bars indicate standard deviation , n = 4 , p<0 . 01 . ( C ) Enforced miR-92 expression in R26MER/MER MEFs significantly enhanced c-Myc-induced apoptosis . The miR-92 effect was most evident when the infected R26MER/MER MEFs were serum starved and treated with 4-OHT . ( D ) miR-92 is required for the potent proliferative effect of mir-17-92 in primary B-cells . miR-92 deficient mir-17-92 miRNA polycistrons exhibited a reduced BrdU incorporation in primary B-cell culture in vitro . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 008 In addition to promoting c-Myc-induced apoptosis , miR-92 unexpectedly enhanced c-Myc-induced cell proliferation . A significant increase of BrdU incorporation was observed in R26MER/MER MEFs overexpressing miR-92 , both under normal culture conditions and , more evidently , under serum starvation ( Figure 3D ) . The same proliferative effect of miR-92 was also observed in primary B-cells . Comparison of the proliferative effect of each mir-17-92 component in bone marrow derived primary B-cells revealed that the miR-92 component yielded one of the strongest effects ( Figure 3E , F ) . In addition , miR-92 deficiency significantly compromised the ability of mir-17-92 to promote cell cycle progression in B-cells ( Figure 3—figure supplement 1D ) . Interestingly , strong proliferative effects have been reported for nearly all mir-17-92 components , yet the exact cell type and biological context can select specific components as the predominant drivers for cell proliferation . Taken together , our data suggest that miR-92 is a unique mir-17-92 component that functionally couples c-Myc-induced cell proliferation and c-Myc-induced apoptosis in the B-cell compartment . To investigate the molecular mechanism underlying miR-92 functions , we performed microarray analyses comparing gene expression profiles of R26MER/MER MEFs overexpressing miR-92 or the control MSCV vector . These MEFs were serum starved and 4-OHT treated to trigger strong Myc-induced apoptosis . miR-92-upregulated genes were significantly enriched for the cell cycle pathway , including ccnd1 , ccnb1 , ccnb2 , cdc25b , cdc25c , and cdk4 ( Figure 4A , B ) , consistent with the ability of miR-92 to promote Myc-induced cell proliferation . Genes upregulated by miR-92 were also enriched for the p53 pathway , including the classic p53 target mdm2 , as well as the pro-apoptotic p53 targets—noxa , bax , puma , perp , and bid ( Figure 4A , B , Figure 4—figure supplement 1A ) . Since aberrant c-Myc activation triggered a p53-dependent apoptotic response ( Lowe et al . , 2004 ) , our observation is consistent with miR-92 further enhancing p53 activation downstream of c-Myc . Interestingly , p21 , a canonical p53 target , was not induced by miR-92 in the MycERT2 activated R26MER/MER MEFs ( Figure 4—figure supplement 1A ) . It is likely that the transcriptional repression of p21 by c-Myc renders p21 irresponsive to p53 activation under this biological context ( Heasley et al . , 2002 ) . Using real-time PCR , we validated the ability of miR-92 to induce cell cycle genes and activate p53 targets in both R26MER/MER MEFs , as well as primary B-cells ( Figure 4C , Figure 4—figure supplement 1A , B ) . Hence , the molecular signature imposed by miR-92 overexpression is consistent with its functional readout . 10 . 7554/eLife . 00822 . 009Figure 4 . miR-92 induces apoptosis through the activation of the p53 pathway . ( A ) The genes upregulated by miR-92 were enriched for the cell cycle pathway and the p53 pathway . Microarray analyses compared gene expression profiles of serum starved and 4-OHT treated R26MER/MER MEFs overexpressing either miR-92 or a control MSCV vector ( n = 3 ) . The differentially expressed genes were defined as those with at least 1 . 5-fold expression level change using SAM ( Significance analysis of microarrays , false discovery rate <1% ) . Pathway analyses were performed on upregulated and downregulated genes using the KEGG database . ( B ) The heatmaps of the miR-92 upregulated genes enriched for the cell cycle and p53 pathways . ( C ) Components of the cell cycle and p53 pathways were upregulated upon miR-92 overexpression in both MEFs ( left ) and primary B-cells ( right ) . The quantitation of gene expression was performed using real time PCR . ( D ) miR-92 overexpression induces the accumulation of Arf and p53 proteins in MEFs and primary B-cells from bone marrow . Western analyses were performed on the R26MER/MER MEFs ( left ) and primary B-cells ( right ) that overexpressed miR-92 or a control MSCV vector in two independent experiments . The infected R26MER/MER MEFs were assayed at 6 hr after serum starvation and 4-OHT treatment; the infected primary B-cells were collected 72 hr post infection . ( E ) The apoptotic effect of miR-92 requires an intact p53 pathway . We infected R26MER/MER MEFs with two MSCV retrovirus , MSCV-p53shRNA and MSCV-92 , to obtain doubly infected cells . Knocking down p53 in R26MER/MER MEFs abolished the ability of miR-92 to enhance c-Myc-induced apoptosis , as measured by Annexin V staining ( two left panels ) . The percentage of apoptotic MEFs of each experimental condition was quantitatively measured ( right ) . ( F ) The induction of the p53 pathway components by miR-92 is dependent on an intact p53 . Knocking down p53 in R26MER/MER MEFs abolished the ability of miR-92 to induce pro-apoptotic p53 targets and other canonical p53 targets , including noxa , perp and mdm2 . Error bars represent standard deviation , p<0 . 05 , p<0 . 01 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 00910 . 7554/eLife . 00822 . 010Figure 4—figure supplement 1 . miR-92 overexpression triggers the activation of the p53 pathway . ( A ) miR-92 overexpression in R26MER/MER MEFs induced several p53 target genes in addition to those described in Figure 3C , including mdm2 , Gtse1 and Bid , but not p21 . ( B ) Induction of p53 targets by miR-92 in R26MER/MER MEFs with and without MycERT2 activation . ( C ) miR-92 overexpression alone enhanced Arf and p53 protein level in R26MER/MER MEFs with and without 4-OHT treatment . ( D ) miR-92 overexpression did not affect p53 mRNA levels in either primary B-cells or in R26MER/MER MEFs . Error bars indicate standard deviation , n = 3 , p<0 . 05; *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 010 The activation of the p53 pathway by c-Myc is essential for the induction of the apoptotic response in the Eμ-myc model ( Schmitt et al . , 2002 ) . A major mechanism that governs Myc-induced p53 activation is the transcriptional induction of the gene encoding Arf , which inhibits Mdm2-mediated p53 ubiquitination and degradation ( Lowe et al . , 2004; Campaner and Amati , 2012 ) . The ability of miR-92 to enhance c-Myc-induced apoptosis and to increase the expression of p53 targets raised the possibility that miR-92 overexpression activates p53 possibly through elevated Arf . In both R26MER/MER MEFs and wild-type primary B-cells , miR-92 overexpression alone caused significant accumulation of Arf mRNA and protein ( Figure 4C , D , Figure 4—figure supplement 1C ) , consistent with the rapid stabilization of the p53 protein ( Figure 4D , Figure 4—figure supplement 1C ) without alteration of p53 mRNA ( Figure 4—figure supplement 1D ) . Notably , the ability of miR-92 to induce p53 activation occurred not only in 4-OHT treated R26MER/MER MEFs with MycERT2 activation , but also in untreated R26MER/MER MEFs with normal c-Myc level . This was clearly demonstrated by the elevation of p53 protein level , as well as the increased p53 target expression ( Figure 4—figure supplement 1B , C ) . The induction of p53 by miR-92 prompted us to investigate the functional importance of p53 in miR-92-induced apoptotic response . Knockdown of p53 in R26MER/MER MEFs not only led to a suppression of c-Myc-induced apoptosis , but also completely abolished the effect of miR-92 to enhance c-Myc-induced apoptosis ( Figure 4E ) . These findings suggested that an intact p53 pathway is required for the apoptotic effect of miR-92 . Consistently , the miR-92 induction of the pro-apoptotic genes , including noxa , perp , and mdm2 , also was mediated by the intact p53 ( Figure 4F ) . Thus , aberrant c-Myc activation triggers an apoptotic response through p53 activation; and co-expression of miR-92 with c-Myc leads to an even stronger p53 activation , and subsequently apoptotic response . Our findings suggest parallels between c-myc and miR-92: both are potent oncogenes that promote excessive cell proliferation coupled with p53-dependent apoptosis , and both are capable to induce expression of cell cycle genes ( ccnb1 , ccnd1 , cdk4 , and cdc25 ) ( Lowe et al . , 2004; Campaner and Amati , 2012 ) and p53 pathway components ( Arf , puma , noxa , perp , and mdm2 ) ( Lowe et al . , 2004; Campaner and Amati , 2012 ) . The functional analogy between c-Myc and miR-92 , as well as the molecular overlap between their downstream pathways , led us to investigate the effect of miR-92 on c-Myc . Intriguingly , miR-92 expression significantly enhanced c-Myc protein level both in MEFs and in primary B-cells ( Figure 5A ) , without affecting the c-myc mRNA level ( Figure 5—figure supplement 1A , data not shown ) . Consistent with the stabilization of endogenous c-Myc , miR-92 overexpression in R26MER/MER MEFs stabilized the MycERT2 protein ( Figure 5B ) . The dosage of c-Myc protein is crucial for its biological readout ( Murphy et al . , 2008 ) . While c-Myc dosage determines the extent of cell cycle gene induction and cell proliferation , it also regulates the degree of p53 activation and subsequent apoptosis ( Murphy et al . , 2008 ) ( Figure 5—figure supplement 1B ) . Thus , the ability of miR-92 to induce aberrant c-Myc accumulation likely constitutes the molecular basis for its ability to promote both cell proliferation and p53-dependent apoptosis . 10 . 7554/eLife . 00822 . 011Figure 5 . miR-92 promotes the accumulation of c-Myc protein through repressing Fbw7 . ( A ) miR-92 enhances the accumulation of c-Myc protein in synchronized R26MER/MER MEFs ( upper ) , as well as primary B-cells ( lower ) . The miR-92 overexpression and the control R26MER/MER MEFs were synchronized by serum starvation and were collected 12 hr after being released into serum culture conditions to determine the c-Myc protein level . This synchronization approach in R26MER/MER MEFs has provided us with the most consistent measurement for c-Myc protein level , because it is regulated in a cell-cycle-dependent manner . ( B ) miR-92 overexpression decreases the turnover of c-Myc protein . Serum-synchronized R26MER/MER MEFs that overexpress either miR-92 or the control MSCV vector were released into the serum for 6 hr , treated with cycloheximide , collected at the indicated time points , then analyzed by western blot to determine the levels of MycER and the endogenous c-Myc protein . ( C ) Schematic representation of the two miR-92 binding sites in the murine fbw7 3′UTR . Additionally , a luciferase reporter and a FLAG tagged fbw7 ORF were each placed upstream of a wild-type fbw7 3′UTR , or a mutated fbw7 3′UTR that abolished the predicted miR-92 binding . ( D ) The expression of Luc-fbw7-3′UTR was specifically repressed by miR-92 in Dicer−/− HCT116 , while mutations of the two putative miR-92 binding sites within the fbw7-3′UTR ( Luc-fbw7-3′UTRMut ) abolished this repression . ( E ) The endogenous fbw7 gene was downregulated by miR-92 post-transcriptionally . Both the endogenous fbw7 mRNA ( left ) and the endogenous Fbw7 protein ( right ) were repressed upon miR-92 overexpression in R26MER/MER MEFs . Due to the lack of a proper antibody to detect endogenous Fbw7 in regular western analysis , we demonstrated the downregulation of endogenous Fbw7 by miR-92 using immunoprecipitation followed by immunoblotting with a polyclonal anti-Fbw7 antibody . ( F ) miR-92 enhances the accumulation of Cyclin E protein . Overexpression of miR-92 increased the accumulation of Cyclin E protein , which was further confirmed by the increased Cyclin E-dependent kinase activity . ( G ) The knockdown of fbw7 resembles the effect of miR-92 to enhance c-Myc-induced apoptosis . Knocking down fbw7 in R26MER/MER MEFs enhanced c-Myc-induced apoptosis , partially recapitulating the phenotype caused by miR-92 overexpression . Apoptosis was quantitatively measured by Annexin V staining in two independent lines of R26MER/MER MEFs upon serum starvation and 4-OHT treatment . ( H ) Overexpression of fbw7 abolished the apoptotic effects of miR-92 in R26MER/MER MEFs . R26MER/MER MEFs were doubly infected by pRetro-fbw7αΔ3′UTR-IRES-dsRed and MSCV-miR-92 . The c-Myc-induced apoptosis was quantitatively measured by Annexin V staining in doubly infected R26MER/MER MEFs upon serum starvation and 4-OHT treatment . Error bars represent standard deviation ) . p<0 . 05; p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 01110 . 7554/eLife . 00822 . 012Figure 5—figure supplement 1 . miR-92 overexpression enhances c-Myc protein level by repressing Fbw7 . ( A ) miR-92 overexpression did not affect c-myc mRNA levels in two independent primary B-cells . ( B ) The c-Myc dosage determines the degree of c-Myc-induced apoptosis in R26MER/MER MEFs . When R26MER/MER MEFs were compared with R26MER/+ MEFs , a twofold increase in the MycERT2 dosage significantly enhanced the c-Myc-induced apoptotic response upon serum starvation . This effect was observed in R26MER/MER MEFs either with or without miR-92 overexpression . ( C ) Negative regulators of c-Myc that contain a putative miR-92 binding site ( s ) were screened for miR-92-mediated repression in R26MER/MER MEFs that overexpress miR-92 or a control MSCV vector . Only fbw7 exhibited a miR-92–mediated repression . ( Error bars indicate standard deviation , n = 3 , p<0 . 01 ) . ( D ) The expression of FLAG-fbw7-3′UTR was significantly repressed by miR-92 in Dicer−/− HCT116 cells . ( E ) fbw7 is downregulated in Eμ-myc lymphomas that overexpress miR-92 . A panel of Eμ-myc/MSCV ( n = 9 ) , Eμ-myc/17-92 ( n = 7 ) , Eμ-myc/17-92Δ92 ( n = 6 ) and Eμ-myc/17-92Mut92 ( n = 5 ) lymphomas were compared for their expression level of endogenous fbw7 . Eμ-myc/17–92 lymphomas exhibited a specific decrease of fbw7 compared to the other genotypes , possibly due to the miR-92 overexpression . ( F ) The c-MYC upregulation by miR-92 requires an intact fbw7 . The effect of miR-92 to upregulate c-MYC protein level was observed in wild-type Hct116 cells , but largely absent in FBW7−/− Hct116 cells . ( G ) fbw7 knockdown by RNAi in R26MER/MER MEFs recapitulated the c-Myc upregulation by miR-92 . ( H ) fbw7 expression level in R26MER/MER MEFs infected with pRetroX-fbw7-IRES-DsRedExpress . Error bars indicate standard deviation , p<0 . 05; p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 012 Based on our findings , we speculated that miR-92 targets could include negative regulators of c-Myc protein accumulation . Therefore , we searched genes known to negatively regulate c-myc for the presence of putative miR-92 binding sites . Using the Targetscan and RNA22 algorithms ( Lewis et al . , 2005; Miranda et al . , 2006; Bartel , 2009 ) , we identified eight candidate miR-92 targets , each of which contained one or more predicted miR-92 binding sites in the 3′ untranslated region ( 3′UTR ) . Real-time PCR analysis of these candidate genes confirmed fbw7 ( F-box and WD repeat domain-containing 7 ) as a likely target of miR-92 ( Figure 5C , Figure 5—figure supplement 1C ) . fbw7 , which contains two miR-92 target sites within its 3′UTR ( Figure 5C ) , is the substrate recognition component of an SCF-type E3 ubiquitin ligase that mediates the degradation of several proto-oncoproteins , including Myc , Cyclin E , c-Jun , and Notch ( Welcker and Clurman , 2008; Crusio et al . , 2010; Wang et al . , 2012 ) . A luciferase reporter or a FLAG-tagged fbw7-encoding ORF ( open reading frame ) , when fused to the wild-type fbw7 3′ UTR , were both significantly repressed in a miR-92 dependent manner ( Figure 5D , Figure 5—figure supplement 1D ) . Yet enforced miR-92 expression failed to repress the luciferase reporter that contained an fbw7 3′UTR with two mutated miR-92 binding sites ( Figure 5D ) , suggesting that miR-92 binding to fbw7 3′UTR is required for this repression . Furthermore , miR-92 effectively repressed endogenous Fbw7 protein level , as demonstrated by the decreased fbw7 mRNA level and Fbw7 immunoprecipitation ( Figure 5E ) . Consistent with fbw7 as an important target for miR-92 , enforced miR-92 expression upregulated multiple Fbw7 substrates at their protein levels , including c-Myc and cyclinE ( Figure 5F ) . Observations from our in vivo experiments also supported this post-transcriptional regulation of Fbw7 by miR-92 , as we observed an inverse correlation between the level of miR-92 and fbw7 when comparing Eμ-myc/17-92Δ92 and Eμ-myc/17-92 lymphoma cells ( Figure 5—figure supplement 1E ) . fbw7 has previously been postulated as a potential miR-92 target based on the presence of miR-92 target sites ( Mavrakis et al . , 2011 ) , yet it remains unclear how fbw7 mediated the pro-apoptotic effects of miR-92 , given its well-characterized functions as a tumor suppressor . Recent findings indicate that the acute inactivation of tumor suppressor Fbw7 imposes a strong oncogenic stress to induce p53-dependent apoptosis , conferring a selective advantage to cells with deficient p53 function ( Minella et al . , 2007; Onoyama et al . , 2007; Matsuoka et al . , 2008; Grim et al . , 2012 ) . This p53-dependent apoptosis is , at least in part , due to an aberrant increase of c-Myc dosage ( Onoyama et al . , 2007; Matsuoka et al . , 2008 ) . These findings suggested that a major mechanism through which miR-92 enhanced the c-Myc protein level , and subsequently , c-Myc-induced apoptosis , could be through its direct repression of Fbw7 . In support of this hypothesis , miR-92 overexpression significantly increased the c-Myc protein level in wild-type Hct116 cells , but not in FBW7−/− Hct116 cells ( Figure 5—figure supplement 1F ) , suggesting FBW7 was essential for miR-92 to induce c-MYC increase . Functionally , acute fbw7 knockdown in R26MER/MER MEFs partially phenocopied the effect of miR-92 to enhance c-Myc-induced apoptosis ( Figure 5G , Figure 5—figure supplement 1G ) ; while overexpression of an fbw7α open reading frame ( ORF ) , albeit above its physiological level , completely abolished this apoptotic effect of miR-92 ( Figure 5H , Figure 5—figure supplement 1H ) . Nevertheless , it is still likely that additional mechanisms downstream of miR-92 also promote its apoptotic effects , because fbw7 knockdown largely recapitulated the extent of c-Myc upregulation by miR-92 ( Figure 5—figure supplement 1G ) , yet only partially phenocopied its pro-apoptotic effects ( Figure 5G ) . In addition , overexpression of fbw7 above its physiological level might amplify the extent of functional interactions between fbw7 and miR-92 in regulating apoptosis . Despite these caveats , our results strongly argue that the miR-92-Fbw7 axis constitutes a major mechanism underlying the pro-apoptotic effects of miR-92 . Downregulation of Fbw7 by miR-92 significantly enhanced the protein level of c-Myc in R26MER/MER MEFs and in primary B-cells ( Figure 5A ) . It is conceivable that the ability of miR-92 to repress Fbw7 in vivo could similarly enhance the c-Myc accumulation in the Eμ-myc/92 premalignant B-cells , promoting rapid cell proliferation and a p53-dependent apoptotic response . Unfortunately , due to technical limitations , we were not able to demonstrate an increased c-Myc protein level as a result of miR-92 overexpression in the Eμ-myc premalignant B-cells . There was a significant depletion of the Eμ-myc/92 premalignant B-cells due to excessive apoptosis ( Figure 3—figure supplement 1B ) , making it difficult to collect enough cells to analyze the c-Myc protein level by western analyses . Similarly , we could not obtain enough cells to compare the protein level of c-Myc in the premalignant B-cells from the Eμ-myc/17-92 , Eμ-myc/17-92Δ92 , and Eu-myc/MSCV animals . Nevertheless , our functional studies in cell culture , combined with the inverse expression correlation between fbw7 and miR-92 in vivo , strongly argue the importance of the miR-92-Fbw7-Myc axis to promote the pro-apoptotic effects of miR-92 . In the context of the c-Myc cooperation , mir-17-92 encodes miRNA components with opposing biological functions . While miR-19 miRNAs repress c-Myc-induced apoptosis to promote Eμ-myc lymphomagenesis ( Mu et al . , 2009; Olive et al . , 2009 ) , miR-92 enhances c-Myc-induced apoptosis to attenuate the tumorigenic effects . Consistent with the opposing effects of miR-19 and miR-92 , co-expression of these two miRNAs as a dicistron attenuated the apoptotic effect of miR-92 in premalignant Eμ-myc B-cells in vivo ( Figure 6A , B , Figure 6—figure supplement 1A ) . A similar antagonistic interaction was also observed in R26MER/MER MEFs; and introducing a miR-19b mutation within mir-19b-92 dicistron abolished this interaction ( Figure 6C ) . Since miR-19 represses pten to promote the PI3K/AKT pathway , the activation of AKT signaling would lead to increased phosphorylation of Mdm2 , thus destabilizing p53 to dampen the apoptotic response induced by miR-92 ( Gottlieb et al . , 2002; Ogawara et al . , 2002 ) . Consistent with this hypothesis , we observed a decreased p53 induction and an unaltered c-Myc level when miR-92 was co-expressed with miR-19 ( Figure 6D , Figure 6—figure supplement 1B ) . 10 . 7554/eLife . 00822 . 013Figure 6 . The antagonistic interaction between miR-19 and miR-92 regulates the balance between proliferation and apoptosis . ( A ) The schematic representation of the Eμ-myc adoptive transfer model to evaluate the functional interaction between miR-92 and miR-19 in vivo . Light colored boxes , pre-miRNAs; dark colored boxes , mature miRNAs . ( B ) miR-19 antagonizes the apoptotic effects of miR-92 in vivo . miR-92 overexpression in the Eμ-myc adoptive transfer model enhanced apoptosis in premalignant Eμ-myc splenic B-cells , while the mir-19b-92 dicistron expression abolished this apoptotic effect ( left three panels ) . A quantitative analysis of apoptosis by FACS was shown for three independent , well-controlled experiments ( right ) . ( C ) miR-19b dampens the miR-92-induced apoptosis in MycERT2 activated R26MER/MER MEFs . R26MER/MER MEFs were infected by miR-92 , mir-19b-92 , mir-19b-92Mut19b and the MSCV control vector , and were subsequently serum starved and treated with 4-OHT to activate MycERT2 . Apoptosis in these samples was measured quantitatively using Annexin V staining ( left four panels ) . The extent of apoptosis induced by MSCV , miR-92 , mir-19b-92 , mir-19b-92Mut19b was normalized to that of MSCV infected R26MER/MER MEFs and then averaged from four independent experiments ( right ) . ( D ) miR-19b dampens the miR-92-induced p53 activation . R26MER/MER MEFs that overexpress the indicated constructs ( miR-92 , mir-19b-92Mut19b and mir-19b-92 ) were collected 48 hr after infection and then analyzed by western blot to determine the level of p53 protein . ( E ) miR-92 and miR-19 exhibit antagonistic effects to regulate hydroxyurea ( HU ) -induced cell death in Xenopus embryos . Representative images of HU-treated Xenopus embryos that were co-injected with human Ago2 and the indicated miRNA mimics ( left ) . Co-injection of miR-92 dampened the cell survival effects of miR-19 on HU-induced apoptosis ( right , n = 3 , with >20 embryos in each group ) . ( F ) miR-92 exhibits a specific antagonistic interaction with miR-19 . Injection of miR-19a or miR-19b rescued HU-induced apoptosis in Xenopus embryos . Co-injection of miR-92 , but not a mutated miR-92 , or other mir-17-92 components , dampened the cell survival effect of miR-19 ( n = 3 , with >20 embryos in each group ) . Error bars represent standard deviation , p<0 . 05; *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 01310 . 7554/eLife . 00822 . 014Figure 6—figure supplement 1 . Functional antagonism between miR-19:miR-92 regulates the balance between proliferation and apoptosis . ( A ) miR-19 antagonizes the apoptotic effects of miR-92 in vivo . miR-92 overexpression enhanced apoptosis in premalignant Eμ-myc bone marrow B-cells in vivo , while co-expression of miR-19 and miR-92 as a dicistron ( mir-19b-92 ) abolished this apoptotic effect ( left three panels ) . A quantitative analysis of apoptosis by FACS was shown for three independent experiments ( right ) . ( B ) miR-19 has no effects on the level of c-Myc protein . While miR-92 overexpression significantly enhanced the level of c-Myc in R26MER/MER MEFs , co-expression of miR-19b and miR-92 did not reverse the increase in c-Myc expression . In addition , miR-19b expression alone did not impact the dosage of c-Myc protein . ( C ) Xenopus fbw7 contains one predicted target site for miR-92 . This predicted miR-92 binding site is conserved between Xenopus and mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 014 This miR-19:miR-92 antagonism appears to be conserved evolutionarily . In Xenopus laevis , miR-19 and miR-92 have identical sequence to their mammalian orthologs ( Figure 1—figure supplement 1B ) . Based on Targetscan and RNA22 miRNA target prediction algorisms ( Lewis et al . , 2003 , 2005; Miranda et al . , 2006; Grimson et al . , 2007 ) , their target specificity is also conserved for key miRNA targets , although the exact binding sites may or may not be conserved ( Olive et al . , 2009 ) ( Figure 6—figure supplement 1C ) . In addition , the biological functions of miR-19 and miR-92 exhibit evolutionary conservation between Xenopus laevis and mammals . Individual injection of miR-19 promoted cell survival of hydroxyurea-treated Xenopus embryos , while co-injection of miR-19a and miR-92 significantly attenuated this pro-survival effect ( Figure 6E ) . This functional antagonism was specific for miR-92 and miR-19 , since co-injection of other mir-17-92 components or a mutated miR-92 did not yield any functional interactions in combination with miR-19 ( Figure 6F ) . Given the opposing biological effects of miR-19 and miR-92 during c-Myc-induced lymphoma development , differential regulation of these two miRNA families could determine the oncogenic activity of mir-17-92 . Under normal physiological conditions , this miR-19:miR-92 antagonism could attenuate the detrimental oncogenic signaling by inducing apoptosis in cells with inappropriate mir-17-92 induction . During malignant transformation , and particularly during c-Myc-induced oncogenesis , this miR-19:miR-92 antagonism could be disrupted to favor cell survival . Using real time PCR analyses , we compared the relative abundance of miR-19a , miR-19b , and miR-92 in normal splenic B-cells , premalignant Eμ-myc B-cells , and Eμ-myc lymphomas ( Figure 7A ) . Comparing to normal splenic B-cells , the levels of all three mature miRNA species were elevated in both premalignant and malignant Eμ-myc B-cells , possibly due to transcriptional activation of mir-17-92 by c-Myc ( Donnell et al . , 2005 ) . However , the miR-19 to miR-92 ratios significantly increased during c-Myc-induced lymphomagenesis ( Figure 7B ) . In other words , when normalized to the respective miRNA levels in normal splenic B-cells , mature miR-19 ( including miR-19a and miR-19b ) exhibited a greater increase in premalignant and malignant Eμ-myc B-cells than mature miR-92 ( Figure 7A–C ) . This differential increase was most evident in premalignant Eμ-myc B-cells; the fully transformed Eμ-myc B-lymphoma cells exhibited a lesser difference ( Figure 7A , B ) . This observation is consistent with premalignant Eμ-myc B-cells having an intact p53-dependent apoptotic response , thus a stronger selective pressure for a greater miR-19:miR-92 ratio . In comparison , most Eμ-myc B-lymphomas have a defective p53 response , hence a less strong selective pressure to maintain a high miR-19:miR-92 ratio . We also validated this observation using northern analysis . Comparing normal splenic B-cells and multiple Eμ-myc lymphoma cells , the levels of the mature miR-19a , miR-19b and miR-92 were all elevated in transformed B-cells; however , the degree of increase for miR-19a and miR-19b was significantly higher than that of miR-92 ( Figure 7C ) . This differential increase of miR-19 and miR-92 was also observed in human Burkitt’s lymphoma cell lines when compared to normal B-cells isolated from the periphery blood ( Figure 7D ) . More importantly , this phenomenon was not limited to c-myc driven B-lymphomas . In the LT2-MYC murine model of hepatocellular carcinoma ( HCC ) , where tumor development was initiated by tetracycline-inducible c-Myc expression , miR-19a and miR-19b also exhibited a stronger increase than miR-92 when comparing tumor cells and the normal counterpart ( Figure 7E ) . 10 . 7554/eLife . 00822 . 015Figure 7 . The miR-19:miR-92 antagonism is disrupted during malignant transformation . ( A and B ) Compared to normal splenic B-cells , premalignant and malignant Eμ-myc B-cells favored a greater increase in mature miR-19 ( miR-19a and miR-19b ) than miR-92 . The purified normal splenic B-cells , premalignant Eμ-myc bone marrow B-cells and malignant Eμ-myc B-lymphoma cells were subjected to Taqman miRNA assays to determine the expression level of miR-19a , miR-19b and miR-92 . Comparing premalignant/malignant Eμ-myc B-cells vs normal splenic B-cells , all three miRNAs exhibited an increased level , although the increase in miR-19a or miR-19b was significantly higher than that of miR-92 ( A ) . In the same experiment , the relative ratios for miR-19a:miR-92 and miR-19b:miR-92 were measured for all normal splenic B-cells and Eμ-myc B-cells ( B ) . ( C ) Mature miR-19 and miR-92 are differentially expressed in normal splenic B-cells and Eμ-myc B-lymphoma cells . The normal splenic B-cells , immortalized human B-cells , premalignant Eμ-myc/+ B-cells , and Eμ-myc/+ B-lymphoma cells were subjected to Northern analysis . Compared to normal splenic B-cells , both malignant and premalignant Eμ-myc/+ B-cells favored a greater increase of miR-19 than miR-92 . ( D ) Compared to normal B-cells isolated from peripheral blood , human Burkitt’s lymphoma cell lines favor a greater increase in mature miR-19 than miR-92 . ( E ) Compared to normal livers ( LT2 ) , mouse hepatocellular carcinomas caused by the inducible c-Myc over-expression ( LT2-myc ) favor a greater increase in mature miR-19 than miR-92 . ( F ) A diagram describes our proposed model to explain the functional interactions between miR-92 and miR-19 in c-Myc-induced B-lymphomagenesis . Aberrant c-Myc expression couples rapid proliferation and p53-dependent apoptosis . miR-92 overexpression further increases c-Myc dosage to strengthen this coupling , at least in part by repressing Fbw7 . This miR-92 effect ensures a potent mechanism to eliminate premalignant c-Myc overexpressing cells . Interestingly , miR-92 and can be antagonized by the survival effects of the miR-19 miRNAs encoded by the same mir-17-92 miRNA polycistron . Taken together , while miR-19 miRNAs repressed c-Myc-induced apoptosis to promote the oncogenic cooperation between mir-17-92 and c-Myc , miR-92 exhibits a negative regulation . Thus , the antagonistic interactions between miR-92 and miR-19 confer an intricate crosstalk between proliferation and apoptosis . Error bars represent standard deviation , p<0 . 05; p<0 . 01 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 00822 . 015 These observations were consistent with a previous finding , where the inducible c-myc activation in a human Burkitt’s lymphoma cell line induced both miR-19a and miR-19b to a greater extent than miR-92 ( Donnell et al . , 2005 ) . Although miR-19 and miR-92 are co-transcribed from the mir-17-92 precursor , the differential increase of miR-19 vs miR-92 occurs in multiple c-Myc-driven tumor types . Thus , the relative abundance of miR-19 and miR-92 could constitute an important molecular basis to regulate the initiation and progression of c-Myc-induced tumor development . The unique polycistronic structure of mir-17-92 constitutes the basis for its pleiotropic functions and the complex mode of interactions among its miRNA components . A high level of mir-17-92 in normal or premalignant cells could lead to suboptimal consequences that are counter-balanced through an intrinsic negative regulation by miR-92 ( Figure 7F ) . As we demonstrated in vitro where miR-92 , by directly downregulating Fbw7 , enhances c-Myc protein level to promote apoptosis , the ability of miR-92 to repress Fbw7 in vivo could similarly constitute a major mechanism to enhance c-Myc-induced apoptosis . This effect of miR-92 is a double edged sword in c-Myc driven tumors , as its overexpression gives rise to a strong and obligated coupling between excessive proliferation and a potent , p53-dependent apoptosis ( Figure 7F ) . This coupling is consistent with the previous observation that a lower level of constitutive c-Myc acts more effectively to promote tumor initiation , while a higher level of c-Myc is selected by the terminal tumors with defective apoptosis machinery ( Murphy et al . , 2008 ) . Therefore , mir-17-92 encodes an internal component to confer a negative regulatory feedback on its oncogenic activity , imposing a strong selection for anti-apoptotic lesions to shape the path of malignant transformation . More interestingly , c-Myc transcriptionally activates mir-17-92 that encodes miR-92 ( Hemann et al . , 2005 ) , which in turn enhances c-Myc dosage , at least in part , by repression Fbw7 . It is possible that aberrant c-Myc activation triggers a positive feedback loop to further increase c-Myc dosage to strengthen the apoptotic response and to eliminate cells with oncogenic potential . It is worth noting that the miR-92 apoptotic effect described in this study depends on an intact p53 response . Consequently , in terminal Eμ-myc B-lymphoma cells that often carry a defective p53 response , miR-92 failed to enhance c-Myc-induced apoptosis ( Mu et al . , 2009 ) . The functional readout of miR-92 heavily depends on cell types and biological contexts . It is important to recognize that miR-92 is not a tumor suppressor miRNA . Like c-Myc , miR-92 elicits potent oncogene stress to engage tumor suppressor response , at least in part , by activating p53 . In the premalignant Eμ-myc/92 B-cells , the effect of miR-92 to repress Fbw7 most likely results in an increase of c-Myc level , which coupled with the intact p53 response to strongly sensitize the cells to miR-92-induced apoptosis . Under other contexts when proliferation becomes a rate-limiting event for oncogenesis , or when p53-dependent apoptosis is compromised , miR-92 could render a pro-proliferative effect that is strictly oncogenic ( Tsuchida et al . , 2011 ) . Likewise , the functional readout of other mir-17-92 components also heavily depends on cell types and biological contexts . miR-19 promotes c-Myc-induced B-lymphomas by repressing apoptosis ( Mu et al . , 2009; Olive et al . , 2009 ) , yet has little effects in promoting Rb-deficient retinoblastomas ( Conkrite et al . , 2011 ) ; miR-17 allows the bypass of Ras-induced senescence by promoting proliferation ( Hong et al . , 2010 ) , yet fails to affect c-Myc-induced lymphomas , possibly due to its functional redundancy with c-Myc . Both cooperative and antagonistic interactions operate among subsets of mir-17-92 components . The miR-19:miR-92 antagonism constitutes a novel mechanism to confer an intricate balance between oncogene signaling and innate tumor suppressor responses ( Figure 7F ) . This balance can be disrupted in premalignant and malignant cells that exhibit c-Myc overexpression , as an increase in the miR-19:miR-92 ratio is likely to favor the suppression of c-Myc-induced apoptosis and to promote oncogenesis . Although all mir-17-92 components are co-transcriptionally regulated , different changes of miR-19 vs miR-92 during oncogenesis could be a result of differential miRNA biogenesis and/or turn-over . It has been shown that specific RNA-binding proteins , such as hnRNP A1 , promote the processing of a specific mir-17-92 component , miR-18 ( Guil and Cáceres , 2007 ) . Future studies are likely to reveal important mechanisms underlying cell type- and context-dependent differential regulation of mir-17-92 components , which will generate important insights on the biology of polycistronic miRNAs . Our current study mostly focuses on the antagonistic interaction between miR-19 and miR-92 in c-Myc driven oncogenesis , yet it reveals a more general mechanism underlying the structural function relationship of polycistronic miRNAs . It is likely that the complex interactions among polycistronic miRNA components can coordinate and balance a multitude of cellular and molecular processes during normal development and disease . Interestingly , in the case of mir-17-92 , miR-92 has a different evolutionary history compared to the other mir-17-92 components . miR-92 is evolutionary conserved in Deuterostome ( including vertebrates and chordates ) , Ecdysozoa ( including flies and worms ) , and Lophotrochozoa , yet the remaining mir-17-92 components are only found in vertebrates ( Figure 1—figure supplement 1C ) . The functional antagonism between the more ancient miR-92 and the newly evolved mir-19 might result from the convergence of these two separate evolutionary paths at the origin of vertebrates . This antagonism could evolve to regulate cell proliferation and cell death downstream or independent of c-Myc in both normal development and disease . Thus , our studies suggest a novel mechanism by which a crosstalk between oncogene and tumor suppressor pathways has been hardwired through evolution into the unique gene structure of a polycistronic oncomir . mir-17-92 Δ92 and mir-17-92 were amplified by PCR and subsequently cloned into the XhoI and EcoRI sites of the MSCV retrovirus vectors . In these vectors , miRNAs were placed downstream of the LTR promoter , which is followed either by a SV40-GFP cassette ( for all in vivo experiments ) , a PGK-Puro-IRES-GFP cassette , or a SV40-CD4 cassette ( for in vitro experiments ) ( Hemann et al . , 2005 ) . To construct MSCV-17-92Mut92 , MSCV-17-92Mut20 , and MSCV-17-92Mut19b vectors , a 12-nucleotide mutation was introduced into the seed region of the mature miR-92 , miR-20 , or miR-19b using the Quikchange XL mutagenesis kit ( 200521; Stratagene ) and the following primers: Mut20 primers: GACAGCTTCTGTAGCACTAAtaaacaataatcGCAGGTAGTGTTTAGTTATC and GATAACTAAACACTACCTGCGATTATTGTTTATTAGTGCTACAGAAGCTGTC . Mut92 primers: CAATGCTGTGTTTCTGTATGGTtaacattaacatCCGGCCTGTTGAGTTTG and CAAACTCAACAGGCCGGATGTTAATGTTAACCATACAGAAACACAGCATTG . Mut19b primers: CTGTGTGATATTCTGCTGacatttaagtacCAAAACTGACTGTGGTAGTG and CACTACCACAGTCAGTTTTGGTACTTAAATGTCAGCAGAATATCACACAG . The loss of miR-92 , miR-20 or miR-19b expression and the intact expression level of the remaining mir-17-92 components were validated using the TaqMan MicroRNA Assays ( 4427975; Applied Biosystems , Foster City , CA ) . mir-19b-92 , mir-19bMut92 , and mir-19b-92Mut19b were similarly amplified by PCR ( ACTGCTCGAGAGCTTCGGCCTGTCGCCC and GTAGAATTCATGTATCTTGTAC ) from the mir-17-92 , mir-17-92Mut92 , and mir-17-92Mut19b construct described above and subsequently cloned into the XhoI and EcoRI sites of the MSCV retrovirus vectors . To construct the MSCV-Shp53 vector , shRNA against p53 was placed downstream of the LTR promoter of the MSCV-SV40-HuCD4 retroviral vector ( Xue et al . , 2007 ) . MSCV-Shfbw7 construct was kindly provided by Dr Hans Guido Wendel ( Mavrakis et al . , 2011 ) . To construct the pRetroX-fbw7-IRES-DsRedExpress ( Xu et al . , 2010 ) , fbw7α ORF was placed downstream of the LTR promoter followed by an IRES-DsRed cassette . The hematopoietic stem and progenitor cells ( HSPCs ) were isolated from E13 . 5-E15 . 5 Eμ-myc/+ mouse embryos and were transduced with MSCV alone or MSCV vectors expressing various mir-17-92 derivatives . The MSCV retroviral vector used in our adoptive transfer model contains a SV40-GFP cassette that allows us to monitor transduced HSPCs both in vitro and in vivo . Infected HSPCs were subsequently transplanted into an 8- to 10-week-old , lethally irradiated C57BL/6 recipient mice . Tumor onset was subsequently monitored by weekly palpation , and tumor samples were either collected into formalin for histopathological studies , or prepared as single cell suspension for FACS analysis and for cell culture studies . Both the Eμ-myc/+ mice and the recipient mice are on C57BL/6 background . The LT2-MYC mouse model for human hepatocellular carcinoma ( HCC ) is a double transgenic mouse model , in which the tetracycline transactivator protein ( tTA ) is driven by the hepatocyte-specific promoter , the liver activator protein ( LAP ) promoter , while the human c-MYC gene is driven by the tetracycline response element ( TRE ) . The LT2-MYC model exhibits ‘dox-off’ regulation , where c-Myc expression is turned on in hepatocytes in the absence of doxycycline . LT2-MYC mice taken off doxycycline-containing food , between 3–5 weeks of age , develop distinct tumor nodules around 8–12 weeks on an average ( Kistner et al . , 1996; Shachaf et al . , 2004 ) . Total RNA was extracted from liver tumor samples from three independent mice , as well as normal livers from the doxycycline treated LT2 mice . Total RNAs were prepared using Trizol ( 15596018; Invitrogen ) and subjected to real time PCR analyses as described below . Primary murine B-cells were prepared from bone marrows of 4- to 6-week-old mice and were cultured in RPMI with 10% fetal bovine serum ( FBS ) , 50 μM beta-mercaptoethanol ( M3148; Sigma ) and 2 ng/ml Il-7 ( 407-ML-005; R&D ) . R26MER/MER and R26MER/+ MEFs were kindly provided by Gerald Evan’s laboratory . MEFs were cultured in DMEM with 10% fetal bovine serum . Eμ-myc tumor cells were derived from lymphomas from the terminal-stage Eμ-myc animals . Eμ-myc lymphoma cells overexpressing various mir-17-92 derivatives were cultured in 45% DMEM , 45% IMDM with 10% fetal bovine serum , and 50 μM β-mercaptoethanol ( M3148; Sigma ) on irradiated NIH-3T3 feeder cells . Immortalized human B-cell lines were cultured in RPMI with 10% FBS and 90 μM beta-mercaptoethanol . Dicer-deficient Hct116 cells , kindly provided by Dr Bert Vogelstein ( Cummins et al . , 2006 ) , and Fbxw7-deficient Hct116 cells ( Grim et al . , 2012 ) were cultured in McCoy’s 5A media with 10% fetal bovine serum . Human Burkitt’s lymphoma cell lines , including BL41 , BL2 , MutuI , Daudi , Raji ( provided by Dr Terry Rabbitts ) , Manca , and Jiyoje were cultured in RPMI with 10% FBS . Mouse primary B-cell cultures or MEFs were infected by MSCV retroviruses expressing various mir-17-92 derived miRNA clusters , shRNA against p53 ( Xue et al . , 2007 ) , shRNA against fbw7 ( Mavrakis et al . , 2011 ) , or fbw7 cDNA ( pRetroX-fbw7-IRES-DsRedExpress ) . In Figure 4E , F , double infection was performed to obtain R26MER/MER MEFs that co-expressed shRNA p53 and miR-92 . In this experiment , MEFs were initially infected with an ecotropic MSCV-p53shRNA-SV40huCD4 retrovirus to a nearly 100% infection efficiency , as validated by FACS analysis using huCD4 antibody . The second infection was achieved using an amphotropic MSCV-miR-92-PGK-Puro-IRES-GFP retrovirus . Doubly infected cells were then selected using puromycin . In Figure 5H , double infection of R26MER/MER MEFs with pRetroX-fbw7-IRES-DsRedExpress and MSCV-miR-92-PGK-Puro-IRES-GFP were similarly performed . For all experiments with primary murine B-cells , bone marrow cells were cultured for 48 hr before retroviral infection and collected or analyzed 72 hr after infection . After 5 days in culture , the percentage of B220-positive cell is 100% . In Figure 3E , Figure 3—figure supplement 1 , B-cells were infected with MSCV retrovirus containing a PGK-Puro-IRES-GFP cassette . FACS analysis was performed after gating on the GFP-positive population . In Figure 5A , the collected B-cells were infected with retrovirus containing SV40-CD4 cassette . Infected cells were purified with Human CD4 Micro-beads ( 130-045-101; Miltenyi Biotec ) using MACS Purification Columns MS ( 130-042-201; Miltenyi Biotec ) . Normal mouse B-cells were isolated from the spleen or the bone marrow of 4- to 6-week-old C57B/6J mice , using CD19 Micro-Beads ( Miltenyi Biotec ) or by negative selection ( Easysep 19754; STEMCELL ) . Similarly , premalignant Eμ-myc B-cells were extracted from the bone marrow of 5- to 6-week-old Eμ-myc transgenic mice . Malignant Eμ-myc B-cells were extracted from the lymph node tumors of terminal-stage Eμ-myc mice . In addition , the normal human B-cells from peripheral blood were FACS sorted from the peripheral blood of healthy donors . Mouse tissue samples were fixed in formalin ( SF100-4; Fisher ) , embedded in paraffin ( AC41677-0020; Fisher ) , sectioned into 5 µm tissue samples , and stained with hematoxylin and eosin ( 7211 & 7111 , Fisher ) . For caspase-3 ( AF835 , 1:200; R&D Systems ) , PCNA ( MS-106P , 1:200; Lab Vision Corp . ) , and B220 ( 14-0452-85 , 1:100; eBioscience ) detection , representative sections were deparaffinized and rehydrated in graded alcohols before subjected to antigen retrieval treatment with 10 mM sodium citrate buffer 10 min in a pressure cooker . Detection of antibody staining was carried out following standard procedures from the avidin-biotin immunoperoxidase methods . Diaminobenzidine ( 002014 , Invitrogen ) was used as the chromogen and hematoxylin as the nuclear counter stain . Quantitation of apoptosis was evaluated by counting the number of starry sky foci in three fields ( 40X ) from seven representative animals of each genotype , as well as by counting the number of caspase-3 positive cells in three fields ( 40X ) from five representative animals of each genotype . To determine the cell surface markers of the lymphoma cells harvested from the animals , cells were resuspended in 10% FBS/PBS to reach a concentration of 107 cells/ml . 20 μl of this cell suspension was stained with antibodies diluted in 10% FBS/PBS for 1 hr . Subsequently , cells were washed with 2% FBS/PBS and resuspended in 10% FBS/PBS for flow cytometry analysis . Antibodies used for FACS analyses include PE anti-mouse IgM ( 12-5790 , 1:200; eBioscience ) , APC-Cy7 anti-mouse B220 ( 552094 , 1:200; BD Pharmingen ) , APC-Cy7 anti-mouse CD4 ( 552051 , BD Pharmingen , 1:200 ) , PE anti-mouse CD8 ( 553032 , 1:200; BD Pharmingen ) , PE anti-mouse CD25 ( 553866 , 1:200; BD Pharmingen ) , and APC anti-mouse CD19 ( 115511 , 1:100; Biolegend ) . Subconfluent MSCV- or miR-92-infected R26MER/MER MEFs were induced and serum starved by incubating the cells with 100 nM of 4-hydroxytamoxifen ( H6278; Sigma ) in DMEM with 0 . 2% fetal bovine serum for 12–24 hr before harvesting the cells for apoptosis analyses using APC-Annexin V antibody ( 550475 , 1:50; BD Pharmingen ) and 7AAD staining solution ( 559925; BD Pharmingen ) . To evaluate the apoptotic effects of miR-92 in our adoptive transfer model in vivo , we collected premalignant Eμ-myc B-cells from spleen or bone marrow of well-controlled Eμ-myc/92 and Eμ-myc/MSCV mice at 5 weeks after adoptive transfer and measured the extent of apoptosis by FACS . Apoptosis in GFP-positive B220-positive premalignant B-cells was measured using the Caspase Detection Kit ( Calbiochem , Red-VAD-FMK ) following the manufacturer’s instructions . To quantitate cell proliferation , 10 μM of BrdU was used to label primary B-cells for 4 hr and MEFs for 30 min . The percentage of BrdU-positive cells was determined using the Flow BrdU kit ( 552598; BD Pharmingen ) . TaqMan MicroRNA Assays ( Applied Biosystems ) were used to measure the level of mature miRNAs , including miR-17 , 18 , 19a , 20 , 19b , and 92 ( 4427975; ABI ) . mRNA level for perp ( GACCCCAGATGCTTGTTTTC , GGGTTATCGTGAAGCCTGAA ) , noxa ( GGAGTGCACCGGACATAACT , TGAGCACACTCGTCCTTCAA ) , puma ( GCGGCGGAGACAAGAAGA , AGTCCCATGAAGAGATTGTAC ) , p21 ( ACGGTGGAACTTTGACTTCG , CAGGGCAGAGGAAGTACTGG ) , bax ( GTTTCATCCAGGATCGAGCAG , CCCCAGTTGAAGTTGCCATC ) , mdm2 ( CTCTGGACTCGGAAGATTACAGCC , CCTGTCTGATAGACTGTCACCCG ) , p53 ( AACCGCCGACCTATCCTTAC , TCTTCTGTACGGCGGTCTCT ) , ccnb1 ( AAGGTGCCTGTGTGTGAACC , GTCAGCCCCATCATCTGCG ) , ccnb2 ( GCCAAGAGCCATGTGACTATC , CAGAGCTGGTACTTTGGTGTTC ) , cdc20 ( AGACCACCCCTAGCAAACCT , GACCAGGCTTTCTGATGCTC ) , cdc25b ( ATTCTCGTCTGAGCGTGGAC , GCTGTGGGAAGAACTCCTTG ) , fbw7 ( CGGCTCAGACTTGTCGATACT , CTTGATGTGCAACGGTTCAT ) , gtse1 ( GCTTTGCCTGTGAGAGGAAG , CACTCTGGGATCCCTTTTCA ) , bid ( CTGCCTGTGCAAGCTTACTG , GTCTGGCAATGTTGTGGATG ) , pten ( CACAATTCCCAGTCAGAGGCG , GCTGGCAGACCACAAACTGAG ) , bim ( ACCACTATCTCAGTGCAATGGCTTCC , CGGTAATCATTTGCAAACACCCTCCTTG ) , cdk4 ( TGGTACCGAGCTCCTGAAGT , GTCGGCTTCAGAGTTTCCAC ) , c-myc ( GTGCTGCATGAGGAGACACCGCC , GCCCGACTCCGACCTCTTGGC ) , Pirh2 ( TGCAGTGCATCAACTGTGAA , CAAACAGGTGGCAAATACTGC ) , Ppp2r5d ( CCGTGATGTTGTCACTGAGG , ACTCTGCTCCTGTGGGATTC ) , Dyrk2 ( CCAGCAACGCTACCACTACA , AACAGCTGCTGAACCTGGAT ) , Romo1 ( ATTCGGAGTGAGACGTCGAG , TGACGAAGCCCATCTTCAC ) , Pak2 ( TTGGCTTTGATGCTGTTACG , CACTGCCTGAGGGTTCTTCT ) , Trpc4ap ( CGCAAATGTCCTTCCTCTTC , GCCAGCATCAGGATTACCAG ) , and Axin1 ( AGGACGCTGAGAAGAACCAG , CTGCTTCCTCAACCCAGAAG ) were determined using real time PCR analyses with SYBR ( KK4605; Kapa Biosystems ) . Actin ( GATCTGGCACCACACCTTCT , GGGGTGTTGAAGGTCTCAAA ) was used as a normalization control in all our real time PCR analysis with SYBR . U6 snRNA assay ( 4427975; ABI ) was used as a normalization control in all our TaqMan MicroRNA Assays ( Applied Biosystems ) . For western analyses , all samples were directly collected into Laemmli buffer . p53 ( 1C12; Cell Signaling ) , Arf ( 5-C3-1; Novus ) , and c-Myc ( 1472-1; Epitomics ) antibodies were used at 1:1000 dilution . FLAG ( M2; Sigma ) and Tubulin ( 12G10 ) were used at 1:2500 dilution . HRP conjugated secondary antibodies ( Santa Cruz Biotechnology , sc-2004 sc-2005 and sc-2006 ) were used at 1:5000 . Three independent R26MER/MER MEF lines were infected by MSCV vector alone or by MSCV vector encoding miR-92 . These MEFs were induced and serum starved by incubating the cells with 100 nM of 4-hydroxytamoxifen ( H6278; Sigma ) in DMEM with 0 . 2% fetal bovine serum for 12 hr before harvesting the cells for RNA preparation . Total RNAs were prepared using Trizol ( 15596018; Invitrogen ) , and subjected to microarray analysis using Affymetrix chip Mouse 430_2 . To identify differentially expressed genes that could be regulated by miR-92 , we used gcRMA in the bioconductor package ( Wu et al . , 2004 ) and SAM ( Significance Analysis of Microarrays ) ( Tusher et al . , 2001 ) for statistical analysis of our microarray data . Gene expression signals were estimated from the probe signal values in the CEL files using statistical algorithm gcRMA . This data processing at the probe level includes background signal subtraction and quantile normalization to facilitate the comparison among microarrays . SAM was then used to identify the genes with significant expression level alterations between miR-92 overexpressing MEFs and the control MEFs . The genes with at least 1 . 5-fold expression level change and FDR <1% were regarded as differentially expressed genes . Pathway analyses were performed on upregulated and downregulated genes using the KEGG database ( Dennis et al . , 2003 ) . Xenopus laevis eggs were collected , fertilized , and embryos cultured by standard procedures . The miR-19b mimics were produced from the annealing products of 5′UGUGCAAAUCCAUGCAAAACUGA3′ and 5′AGUUUUGCAGGUUUGCAUCCAUU3′ ( IDT ) . The miR-17 mimics were produced from the annealing products of 5′CAAAGUGCUUACAGUGCAGGUAGU3′ and 5′UACUGCAGUGAAGGCACUUGUAG3′ ( IDT ) . The miR-18 mimics were produced from the annealing products of 5′UAAGGUGCAUCUAGUGCAGAUAG3′ and 5′ACUGCCCUAAGUGCUCCUUCUG3′ ( IDT ) . The miR-19a mimics were produced from the annealing products of 5′AGUUUUGCAUAGUUGCACUA3′ and 5′UGUGCAAAUCUAUGCAAAACUGA3′ ( IDT ) . The miR-20 mimics were produced from the annealing products of 5′UAAAGUGCUUAUAGUGCAGGUAG3′ and 5′ACUGCAUAAUGAGCACUUAAAGU3′ ( IDT ) . The miR-92 mimics were produced from the annealing products of 5′UAUUGCACUUGUCCCGGCCUG3′ and 5′AGGUUGGGAUUUGUCGCAAUGCU3′ ( IDT ) . The annealing of miRNA mimics were performed by combining two complimentary RNA oligos at a stock concentration of 1 μg/μl , heating the oligos to 80°C for 1 min , and then cooling down to room temperature to allow duplexes to form . The same was done for generating the mutated miR-19 mimics ( Mut-miR-19 ) , by annealing 5′UCAGGUAAUCCAUGCAAAACUGA3′ and 5′AGUUUUGCAGGUUACCUUCGAUU3′ , and mutated miR-92 mimics ( Mut-miR-92 ) by annealing 5′UUAUCGACUUGUCCCGG3′ and 5′GGUUGGGAUUGGUUCGA 3′ . Xenopus embryos were injected into both cells at the two-cell stage with 2 ng of each RNA ( Walker and Harland , 2009 ) . The pcDNA3-myc-AGO2 vector , kindly provided by Dr Greg Hannon , was cut using ScaI; and the synthetic hAGO2 mRNAs were transcribed using mMessage mMachine T7 kit ( Ambion ) . When indicated , a total of 0 . 5 ng hAGO2 mRNA ( Liu et al . , 2004 ) was injected into two-cell stage embryos either alone or with 2 ng of each miRNA ( Lund et al . , 2011 ) . The embryos were then treated with hydroxyurea ( H8627; Sigma ) at a final concentration of 5 mM from stage 3 until stage 10 . Apoptotic embryos were scored as those containing any apoptotic cells based on morphological changes . A luciferase reporter fused with the fbw7 3′UTR was kindly provided by Dr Hans-Guido Wendel ( Mavrakis et al . , 2011 ) . In this psiCHECK-2 based reporter , the fbw7 3′UTR was cloned downstream of the Renilla luciferase reporter , and a separate firefly luciferase cassette was used as a transfection control . Because the two predicted miR-92 binding sites are close to each end of the 3′UTR , we mutated the miR-92 binding sites by PCR using the following primers: 3′UTR-Fbw7-Mut-Xho1-F ( GATCTCGAGCAAGACGACTCTCTAAATCCAACTATTCTTT ) and 3′UTR-Fbw7-mut-Not1-R ( ATGCGGCCGCAACACATTTAGTTATAAGAAAATAAAATTT ) . The PCR fragment was subsequently cloned into the XhoI and Not1 sites of the psiCHECK-2 vector . The reporter construct , together with 50 nM miR-92 mimics , was transfected into Dicer-deficient Hct116 cells ( Cummins et al . , 2006 ) , with transfection of miR-17 or miR-18 as negative controls . Luciferase activity of each construct was determined by dual luciferase assay ( E19100; Promega ) 48 hr post-transfection following the manufacturer’s instructions . The miR-17 mimics were produced by annealing 5′CAAAGUGCUUACAGUGCAGGUAGU3′ and 5′UACUGCAGUGAAGGCACUUGUAG3′ . The miR-18 mimics were produced by annealing 5′UAAGGUGCAUCUAGUGCAGAUAG3′ and 5′ACUGCCCUAAGUGCUCCUUCUG3′ . The miR-92 mimics were produced by annealing 5′UAUUGCACUUGUCCCGGCCUG3′ and 5′AGGUUGGGAUUUGUCGCAAUGCU3′ . Because Fbxw7-substrate degradation was regulated in a cell-cycle-dependent manner , we used serum starvation synchronized MEFs to study Fbw7 regulation by miR-92 during cell cycle progression . MEFs were made quiescent by serum starvation; then Fbw7 expression was examined following release into serum . Cells were lysed in NP-40 buffer supplemented with protease inhibitors . Lysates were normalized and immunoprecipitated with polyclonal anti-Fbw7 antibody kindly provided by Dr Bruce Clurman ( Grim et al . , 2008 ) , followed by immunoblotting with polyclonal anti-Fbw7 antibody ( A301-720A; Bethyl Laboratories ) . Wild-type and FBW7−/− Hct116 cells were used , respectively , as positive and negative controls . The construction of the pFLAG-Fbw7α-3′UTR plasmid was previously described ( Xu et al . , 2010 ) . The construct was transfected into the Dicer-deficient Hct116 cells together with 50 nM of miR-92 mimics or siRNA against GFP as indicated . Anti-FLAG ( M2; Sigma ) antibody was used to detect the FLAG-Fbw7α by western blot 48 hr after transfection . Cyclin E-CDK complexes were immunoprecipitated from MSCV or miR-92 infected Rosa26MER/MER MEFs extracts using affinity-purified polyclonal antibody , provided by Dr Bruce Clurman ( Minella et al . , 2008 ) . Cyclin E immunoprecipitates were then incubated with purified histone subunit H1 ( Sigma ) and ( gamma-32P ) ATP to measure cyclin E-dependent kinase activity . The anti-Grb2 monoclonal ( BD Biosciences ) antibody was used a normalization control .	The role of genes , in very simple terms , is to be transcribed into messenger RNA molecules , which are then translated into strings of amino acids that fold into proteins . Each of these steps is extremely complex , and a wide range of other molecules can speed up , slow down , stop or otherwise disrupt the expression of genes as protein products . Genes can also code for nucleic acids that are not translated into proteins , such as microRNAs . These are small RNA molecules that can reduce the production of proteins by repressing the translation step and/or by partially degrading the messenger RNA molecules . mir-17-92 is a gene that exemplifies much of this complexity . It codes for six different microRNAs in a single primary transcript , and has been implicated in a number of cancers , including lung cancer , Burkitt’s lymphoma and other forms of lymphomas and leukemia . One of six microRNAs has a longer evolutionary history than the remaining five: mir-92 is found in vertebrates , chordates and invertebrates , whereas the other five are only found in vertebrates . However , it is not known how or why the mir-17-92 gene evolved to code for multiple different microRNAs . Olive et al . have studied how these mir-17-92 microRNAs functionally interact in mice with Burkitt’s lymphoma , a form of cancer that is associated with a gene called c-Myc being over-activated . Mutations in this gene promote the proliferation of cells , and in cooperation with other genetic lesions , this ultimately leads to cancer . mir-17-92 is implicated in this cancer because it represses the process of programmed cell death ( which is induced by the protein c-Myc ) that the body employs to stop tumors growing . Olive et al . found that deleting one of the six microRNAs , miR-92 , increased the tendency of the mir-17-92 gene to promote Burkitt’s lymphoma . By repressing an enzyme called Fbw7 , miR-92 causes high levels of c-Myc to be produced . While this leads to the uncontrolled proliferation of cells that promotes cancer , it also increases programmed cell death , at least in part , by activating the p53 pathway , a well-known tumor suppression pathway . The experiments also revealed that the action of miR-92 and that of one of the other microRNAs , miR-19 , were often opposed to each other . These findings have revealed an unexpected interaction among different components within a single microRNA gene , which acts to maintain an intricate balance between pathways that promote and suppress cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2013	A component of the mir-17-92 polycistronic oncomir promotes oncogene-dependent apoptosis
Chemical modification of the gRNA and donor DNA has great potential for improving the gene editing efficiency of Cas9 and Cpf1 , but has not been investigated extensively . In this report , we demonstrate that the gRNAs of Cas9 and Cpf1 , and donor DNA can be chemically modified at their terminal positions without losing activity . Moreover , we show that 5’ fluorescently labeled donor DNA can be used as a marker to enrich HDR edited cells by a factor of two through cell sorting . In addition , we demonstrate that the gRNA and donor DNA can be directly conjugated together into one molecule , and show that this gRNA-donor DNA conjugate is three times better at transfecting cells and inducing HDR , with cationic polymers , than unconjugated gRNA and donor DNA . The tolerance of the gRNA and donor DNA to chemical modifications has the potential to enable new strategies for genome engineering . The CRISPR/Cas9 system uses a gRNA to target and cleave DNA sequences with specificity , and can result in precise genome editing via homology directed repair ( HDR ) if a donor DNA template is simultaneously delivered along with the CRISPR/Cas9 system ( Jinek et al . , 2012; Cong et al . , 2013; Cho et al . , 2013; Mali et al . , 2013a; DeWitt et al . , 2016; Long et al . , 2014 ) . Modifications to the gRNA and the donor DNA are effective methods to engineer CRISPR/Cas9 to enhance editing and develop new applications . For example , extension of the gRNA with aptamer sequences can be used to recruit RNA binding counterparts within the cell that enable either the visualization or transcription of specific DNA sequences in the genome ( Mali et al . , 2013b; Chen et al . , 2013 ) . In addition , incorporation of modified RNA bases into the gRNA results in reduced immunogenicity , increased stability , and enhanced gene editing efficiency ( Rahdar et al . , 2015; Hendel et al . , 2015 ) . Finally , chemical modification of the donor DNA with phosphorothioates dramatically increases HDR efficiency by increasing donor DNA stability ( Renaud et al . , 2016 ) . Thus , modification of either the guide RNA or the donor DNA has great potential for enhancing the biotechnological applications of the CRISPR/Cas9 system . However , outside of base modifications and backbone modifications , very little is known about the chemical modifications that are tolerated by the gRNA of the CRISRP/Cas9 system and the donor DNA . In particular , it is uncertain if the gRNA of Cas9 and the donor DNA tolerate modifications with chemical moieties that are structurally unrelated to nucleotide bases , and how big these modifications can be without affecting the functionality of Cas9 or donor DNA . In addition , nothing is known about the tolerance of the Cpf1 gRNA to chemical modifications . Information about the chemical modifications tolerated by the gRNA and donor DNA can provide new ways to engineer the CRISPR/Cas9 system for high efficiency editing and also enable new applications of the CRISPR/Cas9 system not related to DNA cleavage . In this report , we investigated the types of synthetic modifications that the gRNA and donor DNA can tolerate in the CRISPR gene editing systems , and demonstrate that the gRNA and the donor DNA can be modified at their terminal positions with large modifications ranging from planar hydrophobic molecules to 35 kDa nucleic acids , without losing activity . We exploited the tolerance of the donor DNA to 5’ modifications to develop a method for enriching cells that had been edited via HDR ( Figure 1a ) . In addition , we also exploited the tolerance of the donor DNA and gRNA to terminal modifications to develop a method for increasing the delivery and HDR efficiency of the CRISPR/Cas9 system and donor DNA in cells ( Figure 1b ) . 10 . 7554/eLife . 25312 . 003Figure 1 . gRNA and donor DNA engineering enables the development of new strategies for enriching gene edited cells and for improving their delivery into cells . ( a ) Fluorescently labeled donor DNA can be used as a marker for rapidly enriching HDR edited cells . Cells that internalize fluorescent donor DNA have a high probability of being gene edited via HDR , and can be isolated via FACS based on fluorescence intensity . ( b ) The gRNA and donor DNA can be conjugated together to generate a single molecule ( gDonor ) . Cas9 complexed to gDonor is more efficient at inducing HDR in cells , after transfection with cationic polymers , than free gRNA complexed to Cas9 and donor DNA . The gDonor/Cas9 complex binds polycations , and the resulting nanoparticles have both gRNA and donor DNA in a single nanoparticle , leading to efficient HDR in cells ( left ) . In contrast , Cas9 RNP + donor DNA forms heterogenous complexes with polycations , which are unable to encapsulate Cas9 RNP and donor within the same nanoparticle , leading to low efficiency HDR in cells ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 00310 . 7554/eLife . 25312 . 004Figure 1—figure supplement 1 . Synthesis of DBCO-crRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 00410 . 7554/eLife . 25312 . 005Figure 1—figure supplement 2 . Synthesis of DNA-crRNA . Azide modified DNA was reacted with DBCO-crRNA via copper free click chemistry . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 00510 . 7554/eLife . 25312 . 006Figure 1—figure supplement 3 . The synthesis of DNA-crRNA was confirmed with gel electrophoresis . The low molecular weight band is crRNA . The middle band is Donor DNA . The high molecular weight band is DNA-crRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 006 Very little is known about the tolerance of the gRNAs of Cas9 and Cpf1 towards chemical modifications . Without this information , it is challenging to rationally engineer gRNAs for biotechnological applications . We therefore generated a small library of 8 chemically modified CRISPR targeting RNAs ( crRNAs ) , which had modifications at their 5’ or 3’ ends , and evaluated their ability to cleave genomic DNA , after complexation with Cas9 , in cells expressing blue fluorescent protein ( BFP ) . The chemical modifications we synthesized or purchased are shown in Figure 2a . The library consisted of crRNAs targeting the BFP sequence , which had an amine , azide , fluorescent dye , strained alkyne , disulfide , and a short single stranded DNA ( 87 nucleotides ) , at their 5’ or 3’ position . These modifications were chosen because of their importance in performing conjugation reactions . crRNAs with the chemical modifications shown in Figure 2a were complexed with tracrRNA and Cas9 and electroporated into BFP-HEK cells . The percentage of BFP negative cells was determined via flow cytometry 5 days after the electroporation . Figure 2b demonstrates that the crRNA for Cas9 tolerates large modifications at its 5’ end , and is less tolerant to modifications on the 3’ end . For example , 5’ modified crRNAs had a similar non-homologous end joining ( NHEJ ) frequency in BFP-HEK and BFP-K562 cells ( Figure 2b and Figure 2—figure supplement 1 ) as control unmodified crRNA . In contrast , crRNA with 3’ modifications had a 50% reduction in NHEJ efficiency in cells . The sensitivity of crRNA to 3’ modifications is anticipated as the 3’ of the crRNA in Cas9 hybridizes with tracrRNA and is in close proximity to the Cas9 protein , whereas the 5’ end has a looser interaction with Cas9 ( Jiang et al . , 2015; Yamano et al . , 2016; Jinek et al . , 2014; Fu et al . , 2014 ) . 10 . 7554/eLife . 25312 . 007Figure 2 . The gRNAs for Cas9 and Cpf1 and donor DNA tolerate large chemical modifications at their terminal ends . ( a ) Chemical structure of modified gRNAs . gRNAs with 5’ or 3’ modifications were purchased or synthesized . ( b ) Cas9 crRNAs with 5’ or 3’ modifications and Cas9 were electroporated into BFP-HEK cells , and their activity was quantified by determining the amount of NHEJ they generated in cells ( % BFP negative cells ) . crRNAs tolerate 5’ modifications but are sensitive to modifications at the 3’ end . DNA-crRNAs are crRNAs conjugated to an 87nt scrambled DNA oligonucleotide . One way ANOVA , post-hoc Tukey test , significant difference from control , p<0 . 05 , p<0 . 01 . ( c ) Cpf1 crRNAs with 5’ or 3’ modifications and Cpf1 were electroporated into BFP-HEK cells , and their ability to generate NHEJ was investigated . One way ANOVA , post-hoc Tukey test , significant difference from control crRNA , p<0 . 05 . ( d ) Chemical structures of modified donor DNA . Donor DNA with 5’ or 3’ modifications were purchased or synthesized . ( e ) Donor DNA with 5’ or 3’ modifications and Cas9 RNP were electroporated into BFP-HEK cells and their ability to induce HDR was investigated . The donor DNA tolerates both 5’ and 3’ modifications . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 00710 . 7554/eLife . 25312 . 008Figure 2—figure supplement 1 . The NHEJ frequency of BFP-K562 cells transfected with chemically modified crRNAs . The NHEJ frequency was quantified by performing flow cytometry on BFP-K562 cells , 7 days after nucleofection with Cas9 and chemically modified crRNAs/TracrRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 008 We also investigated the tolerance of the Cpf1 guide RNA to chemical modifications . Cpf1 is a recently discovered RNA-guided endonuclease of the class 2 CRISPR-Cas , and has the potential to be an alternative to Cas9 as it recognizes non-classical PAM sequences ( Yamano et al . , 2016; Zetsche et al . , 2015; Kleinstiver et al . , 2016; Zetsche et al . , 2017 ) . Unlike Cas9 , which requires both crRNA and trans-activating crRNA ( tracrRNA ) , Cpf1 requires only crRNA , and therefore it is an even more attractive target for chemical modifications . crRNA targeting the BFP gene and Cpf1 were electroporated into BFP-HEK cells , and the percentage of BFP negative cells was quantified by flow cytometry . Figure 2c demonstrates that the crRNA from AsCpf1 ( from Acidaminococcus ) tolerates chemical modifications such as amine , azide , and DBCO at its 3’ end; in contrast , bulky 5’ end modifications such as azide and DBCO are less tolerated . For example , BFP-HEK cells electroporated with 3’ amine-crRNA and Cpf1 had a similar NHEJ frequency as cells electroporated with Cpf1 and unmodified crRNA . In contrast , BFP-HEK cells electroporated with crRNAs with 5’ modifications and Cpf1 had a reduction in NHEJ frequency and generated only 50–80% of the NHEJ levels as cells treated with unmodified crRNA . The 5’ nucleotide of the crRNA has more interaction points with the Cpf1 protein than the 3’ nucleotide , and the sensitivity of the 5’ end to chemical modifications corresponds with the crystal structure of Cpf1 ( Yamano et al . , 2016 ) . Thus , the gRNAs for Cas9 and Cpf1 can be modified at one of their terminals with minimal loss of activity , and this tolerance should enable a variety of new strategies for gRNA engineering . In addition , we also investigated the tolerance of the donor DNA to chemical modifications . We investigated the HDR efficiency of donor DNA with 5’ and 3’ modifications composed of an azide modification , an amine modification , and an Alexa 647 modification ( Figure 2d ) . For these experiments , a donor DNA which converts the BFP gene to the GFP gene was electroporated in cells along with Cas9 RNP targeting the BFP gene ( Richardson et al . , 2016 ) . Figure 2e demonstrates that the donor DNA tolerates modifications at both its 5’ and 3’ ends , and BFP-HEK cells electroporated with chemically modified donor DNA were efficiently converted to GFP expressing cells via HDR . The tolerance of the donor DNA to chemical modifications at the 5’ and 3’ ends is anticipated based on the role of the donor DNA in the HDR mechanism . The donor DNA is rarely incorporated directly into the genome after a double stranded break , and instead , acts as a template for the DNA repair . Therefore , modifications of the 5’ and 3’ ends are likely to be tolerated ( Holloman , 2011; Durai et al . , 2005 ) . Thus , the donor DNA can be modified at its 5 or 3’ position without losing activity , and this tolerance should enable a variety of new strategies for donor DNA engineering . Chemically modified gRNAs and donor DNA can potentially be used to address technological challenges limiting progress in the field of genome editing , such as the lack of methods available for rapidly identifying cells genetically edited via HDR . Current methods for identifying gene edited cells requires deep sequencing of randomly selected clones or the introduction of new additional proteins for magnetic bead pull down or fluorescence labeling ( Lee et al . , 2016; Kim et al . , 2013 ) . However , neither of these methods is ideal for selecting gene edited cells and alternatives are greatly needed . For example , randomly choosing clones from cell transfection experiments and sequencing them to identify gene edited cells requires significant levels of investment and requires several weeks to complete . Similarly , introducing additional proteins into cells for the purpose of purification causes numerous additional complications , such as cell perturbation and unnecessary foreign protein expression . In addition , most cell identification and enrichment methods take several days to complete , which are too long for many primary cells because they rapidly change their phenotype after culturing in vitro . The inability to rapidly identify gene edited cells currently limits the development of gene edited cell therapies and the development of engineered cell lines ( Dever et al . , 2016 ) . Therefore , the development of a fast and non-invasive method for enriching gene-edited cells has the potential to accelerate progress in multiple areas of genome engineering . The donor DNA concentration in the nucleus is a key factor that determines the HDR efficiency after a double stranded break . We therefore investigated if donor DNA that was fluorescently labeled could be used as a beacon for FACs sorting , to identify cells that had a high probability of undergoing HDR , after transfection with Cas9 RNP ( Figure 1a ) . Enriching cells for gene editing , non-invasively via FACS , has the potential to significantly accelerate gene editing cell culture protocols , because it will reduce the number of clones that need to be isolated and sequenced . A donor DNA was labeled with Alexa 647 , termed trackable Donor ( tDonor ) , and was electroporated into BFP-HEK cells along with Cas9 RNP . Sixteen hours after the electroporation , cells that internalized high levels of the tDonor and low levels of the tDonor were sorted using fluorescence activated cell sorting ( FACS ) . After three days of culture , the HDR frequency was determined using flow cytometry and compared to bulk unsorted cells . Figure 3 demonstrates that tDonor can be used as a selectable marker to enrich for cells that have a high probability of being edited via HDR . Cells that had internalized high levels of the donor DNA had a high rate of HDR , and the HDR rate in these cells increased by a factor of 2 . Likewise , sorted cells that had low levels of tDonor showed significantly lower levels of HDR editing , demonstrating that the amount of Donor DNA in a cell is an important factor for HDR ( Figures 3b , c and d ) . 10 . 7554/eLife . 25312 . 009Figure 3 . Fluorescently labeled donor DNA can be used to enrich for cells that have been edited via HDR . ( a ) Overview of the cell enrichment process . Cells are transfected with Alexa647-donor DNA ( tDonor ) and Cas9 RNP , and sorted based on their intracellular levels of Alexa647-donor DNA via FACS . Cells with high levels of Alexa647-donor DNA are enriched for HDR . ( b ) BFP-HEK cells were electroporated with Cas9 RNP and Alexa647-Donor , and were sorted via FACS based on Alexa647 fluorescence . The histogram shows the FACS analysis of Alexa647 fluorescence , control untreated cells are in black , and cells electroporated are in red . The yellow box shows the gating used to identify the Alexa 647 negative cells ( bottom 20% gating ) , and the red box shows the gating used to identify the Alexa 647 positive cells ( top 20% gating ) . Fluorescent images and histograms from flow cytometry of the sorted cells demonstrates that cells with high amounts of Alexa647-Donor had higher levels of HDR ( bar: 100 μm ) . ( c ) The HDR rate in BFP-HEK cells was determined by quantifying GFP expression . The bulk population of transfected cells ( without sorting ) , cells with low levels of Alexa647-Donor , and cells with high levels of Alexa647-Donor were analyzed by flow cytometry . Alexa647 based sorting enriches for cells that have a high probability of being edited via HDR . One way ANOVA , post-hoc Tukey test , significant difference from control , p<0 . 05 , p<0 . 01 . ( d ) The HDR rate in BFP-K562 cells was determined by quantifying GFP expression . The bulk population of transfected cells ( without sorting ) , cells with low levels of Alexa647-Donor , and cells with high levels of Alexa647-Donor were analyzed by flow cytometry . Alexa647 based sorting enriches for cells that have a high probability of being edited via HDR . One way ANOVA , post-hoc Tukey test , significant difference from control , p<0 . 05 , *p<0 . 01 . ( e ) Primary myoblasts from mdx mice were transfected with Cas9 RNP and Alexa647-Donor using lipofectamine , and were sorted via flow cytometry based on Alexa647 fluorescence . The correction of the dystrophin mutation in these cells via HDR was quantified by restriction enzyme analysis of the dystrophin gene . Flow sorted cells that internalized high amounts of the Alexa647 donor had more than a 2-fold increase in HDR frequency than unsorted cells . One way ANOVA , post-hoc Tukey test , significant difference from control , p<0 . 05 , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 00910 . 7554/eLife . 25312 . 010Figure 3—figure supplement 1 . BFP-K562 cells with high levels of fluorescently labeled donor DNA are enriched for HDR edited cells . BFP-K562 cells were electroporated with Cas9 RNP and Alexa647-Donor , and were sorted via flow cytometry based on the Alexa 647 fluorescence . Cells that internalized high amounts of Alexa 647 donor were enriched for HDR ( GFP expressing ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 010 In addition , we investigated if sorting cells based on the amount of donor DNA internalized would help in identifying primary cells that had been edited via HDR . Primary myoblasts from the Duchenne muscular dystrophy mouse model ( mdx mice ) , which have a mutation in their dystrophin gene , were used for these experiments Primary myoblasts were transfected with Cas9 RNP and a fluorescently labeled tDonor designed to correct the dystrophin mutation , using lipofectamine . The transfected cells were sorted via flow cytometry , using the fluorescence of the tDonor for gating , cultured , and analyzed for gene editing via restriction enzyme analysis . Figure 3e demonstrates that the HDR rate in primary myoblasts with high levels of tDonor is two fold higher than unsorted cells . Fluorescently labeled donor DNA represents an easy and fast method for enriching gene edited cells , and could find numerous applications in biology and medicine . There is great interest in developing therapeutics based on the CRISPR/Cas9 system; however , delivery problems have limited their clinical progress . Although viral vectors based on AAV can efficiently deliver Cas9 ( Long et al . , 2016; Tabebordbar et al . , 2016; Nelson et al . , 2016 ) , their toxicity is problematic . Therefore there is great interest in developing non-viral delivery vehicles that can deliver Cas9 , gRNA and donor DNA into cells ( Yin et al . , 2016 ) . Cationic polymers are a promising delivery vehicle for Cas9 , as they avoid several of the problems associated with using AAV , such as immunogenicity and off-target DNA damage caused from the sustained expression of Cas9 . However , simultaneous delivery of donor DNA and Cas9 RNP into cells is challenging with cationic vectors , because the donor DNA and the Cas9 RNP have different charge densities , and it is therefore unlikely that a single nanoparticle will contain both Cas9 RNP and donor DNA . As a result , it is challenging to deliver Cas9 RNP and donor DNA simultaneously into the same cell and efficiently induce HDR . To circumvent this obstacle , we performed experiments to determine if the gRNA and donor DNA could be combined into a single molecule ( termed gDonor ) . gDonor should complex polycations even after binding the Cas9 protein because the gDonor has a much higher charge density than either the gRNA or donor DNA , due to its longer length ( Figure 4a ) . In addition , gDonor should significantly increase the transfection efficiency of polycationic vectors because every cell that has internalized gRNA will have also have internalized donor DNA . Finally , using gDonor for inducing HDR ensures that there is donor DNA in the vicinity of the DNA cleavage site , and this may further promote efficient HDR ( Holloman , 2011; Durai et al . , 2005 ) . 10 . 7554/eLife . 25312 . 011Figure 4 . A gRNA-donor DNA conjugate ( gDonor ) transfects cells with higher efficiency than free gRNA and donor DNA . ( a ) The proposed mechanism of gene editing with gDonor/Cas9 complexes in cells . ( b ) Synthesis of gDonor . gDonor was synthesized via click chemistry and gel analysis confirms the synthesis of gDonor . ( c ) gDonor efficiently generates NHEJ in BFP-HEK cells after electroporation . The NHEJ frequency depends on the amount of gDonor . ( d ) gDonor with Cas9 can efficiently induce DNA cleavage and repair via HDR . BFP-HEK cells electroporated with gDonor/Cas9 had a similar HDR frequency as BFP-HEK cells electroporated with Cas9 RNP and donor DNA . ( e ) gDonor has a similar DNA cleavage pattern in cells as free gRNA and donor DNA ( control ) . Deep sequencing analysis of BFP-HEK cells edited with gDonor/Cas9 and comparison to cells edited with Cas9 RNP and donor DNA ( control ) . Cas9 with gDonor has an almost identical DNA cleavage profile as the unmodified control . The targeted Cas9 cleavage site for these experiments was at 64 locus ( position of mutation ) , which is where most of the mutations were observed . ( f ) The gDonor/Cas9 complex was delivered into cells with cationic polymers , and the delivery efficiency was compared against cationic polymers complexed to unconjugated gRNA and donor DNA . gDonor/Cas9 complexed to PAsp ( DET ) was three times more efficient at generating HDR in BFP-HEK cells than PAsp ( DET ) complexed to Cas9 RNP and donor DNA . An additional control composed of a scrambled DNA conjugated to the gRNA did not increase the transfection efficiency of PAsp ( DET ) . Student-t-test , significant difference from gDonor/Cas9 , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 01110 . 7554/eLife . 25312 . 012Figure 4—source data 1 . Raw data from deep sequencing analysis . Genome DNA of BFP-HEK cells edited with gDonor/Cas9 and compared to cells edited with Cas9 RNP and donor DNA ( control ) was analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 01210 . 7554/eLife . 25312 . 013Figure 4—figure supplement 1 . Synthesis and purification of gDonor . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 01310 . 7554/eLife . 25312 . 014Figure 4—figure supplement 2 . Mutation patterns made by gDonor and Cas9 are similar to that of free gRNA and Cas9 . Deep sequencing analysis was conducted for BFP-HEK cells edited with gDonor/Cas9 and compared to cells edited with Cas9 RNP and donor DNA ( control ) . Top images show deletion size ( bp ) from gene editing and bottom images show insertion size ( bp ) from gene editing from both Control and gDonor . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 01410 . 7554/eLife . 25312 . 015Figure 4—figure supplement 3 . PAsp ( DET ) polymer nanoparticles complex both gDonor and Cas9 . The gDonor/Cas9 complex was mixed with PAsp ( DET ) and centrifuged at 17 , 000 rpm to spin-down polymer nanoparticles that were formed . Gel electrophoresis was performed on the isolated pellets and supernatants; the gels were then stained for either protein or nucleic acids , to determine if the gDonor-Cas9 complex was encapsulated within PAsp ( DET ) nanoparticles . Samples were prepared in duplicate , and duplicate samples of supernatants and pellets were run on the gel . The PAsp ( DET ) pellet contains both Cas9 and gDonor , demonstrating efficient complexation . Before polymer: sample without polymer addition , Polymer particle: complete polymer nanoparticle , Supernatant: supernatant collected from the centrifuged polymer nanoparticle , Pellet: pellet collected from the centrifuged polymer nanoparticle . SyBr safe stained gel was imaged first and the same gel was stained with Coomassie blue to visualize Cas9 protein . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 01510 . 7554/eLife . 25312 . 016Figure 4—figure supplement 4 . Dynamic light scattering analysis of PAsp ( DET ) complexes with Cas9 and gDonor . The particle size was measured after 5 min of incubation and had a size of approximately 150 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 25312 . 016 A gRNA-donor DNA conjugate ( gDonor ) was synthesized by conjugating an azide terminated donor DNA with an alkyne modified crRNA , and hybridizing the resulting conjugate with tracrRNA . The gRNA was designed to cut the BFP gene and the donor DNA was designed to convert the BFP gene into the GFP gene . The gDonor was purified via gel extraction , and was synthesized with a 40% yield ( Figure 4b ) . The activity of the gDonor was investigated by determining its ability to induce NHEJ or HDR in BFP-HEK cells , after electroporation with the Cas9 RNP . In addition , the DNA cleavage pattern of the gDonor in cells was also compared against cells treated with Cas9 RNP and donor DNA to determine whether conjugation to the donor DNA affected the function of the gRNA . Figure 4c demonstrates that the gDonor was able to convert the BFP gene to the GFP gene via HDR with an efficiency similar to unmodified gRNA and Donor DNA , and thus both the gRNA and donor DNA of the gDonor are active . Figure 4d demonstrates that the NHEJ frequency induced by gDonor is dose dependent . In addition , deep sequencing analysis of the electroporated cells demonstrates that the gDonor cleaved its target sequence in cells with specificity and induced a similar pattern of indel mutations as unmodified gRNA control ( Figure 4e and Figure 4—figure supplement 2 ) . These results demonstrate that the gDonor can efficiently function as both a gRNA and a donor DNA . We investigated if the gDonor could efficiently induce HDR in cells after delivery with cationic polymers . The cationic polymer , PAsp ( DET ) , was selected as the initial polymer to deliver the gDonor because of its well established ability to deliver siRNA into cells and in vivo ( Miyata et al . , 2008; Kim et al . , 2010 , 2014 ) . The gDonor was mixed with Cas9 and complexed with PAsp ( DET ) , and generated nanoparticles 150 nm in diameter that contained the Cas9-gDonor complex ( Figure 4—figure supplements 3 and 4 ) . The polymer nanoparticles were added to BFP-HEK cells and the HDR efficiency was determined by flow cytometry . Figure 4f demonstrates that gDonor significantly improves the ability of cationic polymers to simultaneously deliver Cas9 , gRNA and donor DNA into cells . For example , the Cas9-gDonor complexed with PAsp ( DET ) induced an 8% HDR frequency in BFP-HEK cells , which was three times higher than that of the free gRNA and donor DNA complexed to PAsp ( DET ) . Additional control cell experiments were conducted with a scrambled DNA conjugated gRNA , which had the same charge density as the gDonor . Cells were treated with the scrambled DNA-crRNA/Cas9 complexed with PAsp ( DET ) and a separate complex of donor DNA/PAsp ( DET ) , and the HDR efficiency was measured . Figure 4f demonstrates that the scrambled DNA-crRNA conjugate did not improve the transfection efficiency of PAsp ( DET ) , suggesting that the gDonor’s ability to enhance the efficacy of PAsp ( DET ) is not related to stronger complexation . The gDonor represents a new reagent for improving the delivery of both Cas9 RNP and donor DNA into cells and has great potential for accelerating the development of Cas9 based therapeutics . In this report , we demonstrate that the gRNAs and donor DNA can be chemically modified at their terminal positions without losing activity . The tolerance of the donor DNA and gRNA to 5’ modifications was exploited to develop a method for enriching cells that have a high chance of undergoing HDR . In addition , we synthesized a gRNA-donor DNA conjugate ( gDonor ) that enabled the efficient delivery of Cas9 RNP and donor DNA into cells . We anticipate numerous applications of chemically modified gRNA and donor DNA for gene engineering given the wide variety of chemical modifications they tolerate . Unmodified crRNA , 5’ Amine-crRNA , 5’ Azide-crRNA , 5’ Thiol-crRNA , 3’ Amine-crRNA , 5’ Amine-Donor , 3’ Amine-Donor , 5’ Azide-Donor , and various DNA sequences were purchased from Integrated DNA Technology ( IDT ) . Phusion High-fidelity DNA Polymerase was purchased from NEB ( Ipswich , MA ) . The Megascript T7 kit , the Megaclear kit , the PageBlue solution , the propidium iodide , and the PureLink genomic DNA kit were purchased from Thermo Fisher ( Waltham , MA ) . Mini-PROTEAN TGX Gels ( 4–20% ) were purchased from Bio-Rad ( Hercules , CA ) . 4- ( 2-hydroxyethyl ) piperazine-1-ethanesulfonate ( HEPES ) were purchased from Mandel Scientific ( Guelph , ON ) . Sodium silicate was purchased from Sigma Aldrich ( St . Louis , MO ) . Matrigel was purchased from BD Biosciences ( San Jose , CA ) . DMEM media , non-essential amino acids , penicillin-streptomycin , DPBS and 0 . 05% trypsin were purchased from Life Technologies ( Carlsbad , CA ) . EMD Millipore Amicon Ultra-4 100 kDa and 300 kDa were purchased from Millipore ( Germany ) . Streptococcus pyogenes Cas9 ( spCas9 ) and Acidaminococcus sp . Cpf1 ( AsCpf1 ) were purchased from the QB3 Macrolab from UC Berkeley . PAsp ( DET ) polymer was a generous gift from Dr . Kataoka’s group ( Miyata et al . , 2008; Kim et al . , 2010 , 2014 ) . TracrRNA and sgRNAs were synthesized using the in vitro transcription method with the MEGAscript T7 kit ( Thermo Fisher ) ( DeWitt et al . , 2016; Richardson et al . , 2016 ) . Purification of gRNAs was conducted using the MEGAclear kit , following the manufacturer’s protocol . Dibenzocyclooctyne ( DBCO ) -crRNA was synthesized according to the synthetic scheme described in Figure 1—figure supplement 1 . Amine-crRNA ( either 5 or 3’ ) ( 100 µM ) was suspended in a 100 μL of DMSO and mixed with a 100-fold molar excess of Compound 1 ( 10mM ) . The reaction was incubated at room temperature for 16 hr and then purified with a desalting column ( Micro Bio-Spin 30 , Bio-rad ) . The concentration of the purified DBCO-crRNA was measured with a Nanodrop spectrophotometer . The synthesis of DBCO-crRNA was verified by conjugating it to either an azide modified DNA or azide-Cy5 . DBCO-crRNAs ( 1 pmole ) were incubated with a three times molar excess of 5’ azide-DNA ( 3 pmole ) or a 100 fold molar excess of azide-Cy5 ( 100 pmole ) in 10 μL HEPES buffer to test the DBCO conjugation yield . The reaction mixture was run on a polyacrylamide gel and analyzed for crRNA content via densitometry or fluorescence intensity , and the reaction yield was determined by comparing the conjugate intensity to free DBCO-crRNA . The DBCO-crRNA samples with higher than 85% of reaction yield were used in further experiments . NHS ester-Rhodamine and Alexa 647 NHS ester ( Alexa647 ) were purchased from Thermo Fisher ( Waltham , MA ) . One hundred times molar excess of NHS ester-Rhodamine or Alexa 647 ( 1 mM ) were added to 5’ amine-crRNA or 5’ amine-DNA ( 10 μM ) in pH 8 . 5 PBS ( 100 μL ) . After 4 hr of incubation at room temperature , a desalting column ( Micro Bio-Spin 30 , Bio-rad ) was used to purify the conjugates . The conjugation yield was determined via fluorescence and samples with higher than an 85% reaction yield were used for further experiments . The conjugation of donor DNA and crRNA ( gDonor ) was conducted using copper-free click chemistry ( Figure 1—figure supplement 2 ) . 5’ Azide-DNA ( 86nt ) was purchased from IDT and 5’ Azide-DNAs with different lengths were synthesized from 5’ amine-DNAs . 5’ Azide-donor DNA ( 10 μM ) was mixed with 5’ DBCO-crRNA ( 10 μM ) in DI water ( 50 μL ) . The solution was incubated at room temperature overnight . The sample was analyzed via gel electrophoresis using a polyacrylamide gel ( 4–20% Mini-protean TGX Precast gel , Biorad ) . PAGE gel extraction was conducted to purify the gDonor conjugate . The DNA-crRNA band was cut with a sharp knife and eluted using the crush and soak method in nuclease-free water for 16 hr , and purified via ethanol precipitation additionally . 200 ng of crRNA , DNA , Azide-DNA , crRNA + DNA , and DNA-crRNA were analyzed via gel electrophoresis using a polyacrylamide gel ( 4–20% Mini-protean TGX Precast gel , Biorad ) ( Figure 1—figure supplement 3 ) . BFP-HEK293T cells and BFP-K562 cells were generated by infecting HEK293T and K562 cells with a BFP-containing lentivirus , followed by FACS-based enrichment , and clonal selection for cells expressing BFP with no silencing after 2–4 weeks ( Richardson et al . , 2016 ) . The cells were STR profiled and mycoplasma contamination test result was negative . Cells were cultured in DMEM with 10% FBS , 1× MEM non-essential amino acids , and 100 μg/mL Pen Strep . BFP-HEK293T cells were plated at a density of 105 cells per well in a 24-well plate , one day before editing experiments were performed . BFP-HEK cells were detached by 0 . 05% trypsin or gentle dissociation reagent , spun down at 600 g for 3 min , and washed with PBS . BFP-K562 cells were collected and washed with PBS . Nucleofection was conducted using an Amaxa 96-well Shuttle system following the manufacturer’s protocol , using 10 µL of Cas9 RNP ( Cas9 - 50 pmole , crRNA and TracrRNA - 60 pmole unless the amount is specified ) ( Richardson et al . , 2016; Lin et al . , 2014; Schumann et al . , 2015 ) . Nucleofection with Donor DNA had an additional 100 pmole of Donor DNA . After the nucleofection , 500 μL of growth media was added to the cells and they were incubated at 37°C in their tissue culture plates . The cell culture media was changed 16 hr after the nucleofection , and the cells were incubated until flow cytometry analysis was performed . The NHEJ frequency was quantified by determining the BFP negative population , and the HDR frequency was quantified by determining the GFP positive population . For the chemically modified gRNA activity tests , chemically modified crRNAs ( 30 pmole ) were nucleofected with TracrRNA ( 30 pmole ) , and Cas9 ( 25 pmole ) into BFP-HEK cells ( 105 cells ) . AsCpf1 requires a TTTN PAM sequence instead of the NGG sequence of SpCas9 ( Zetsche et al . , 2015 ) . We designed a crRNA for AsCpf1 that targets the BFP gene , and crRNA was purchased from IDT and additional modifications were conducted as described above . Nucleofection was conducted using the methods described above with crRNAs ( 30 pmole ) and Cpf1 ( 25 pmole ) into BFP-HEK cells ( 105 cells ) . Flow cytometry was used to quantify the expression levels of BFP . The BFP-HEK and BFP-K562 cells were analyzed 5 days after Cas9 treatment . The BFP-HEK cells were washed with PBS and detached by trypsin . BFP and GFP expression were quantified using a BD LSR Fortessa X-20 and Guava easyCyte . Alexa647 labeled 127 nt donor ssDNA was used for the enrichment experiments . BFP-HEK and BFP-K562 cells were nucleofected using an Amaxa 96-well Shuttle system following the manufacturer’s protocol , using 10 µL of Cas9 RNP ( Cas9 - 50 pmole , BFP sgRNA - 60 pmole , and 647-Donor - 50 pmole in 105 cells ) . Cells were sorted using a BD influx cell sorter ( BD Biosciences ) in the Berkeley flow cytometry facility , 16 hr after the nucleofection . Positive gating captured the top 20% of Alexa647 positive cells , and negative gating captured the bottom 20% of cells , which had the least amount of Alexa647 fluorescence . The cells were cultured for an additional two to five days and then analyzed with flow cytometry . Fluorescence images were taken using a Zeiss inverted microscope and analyzed with Zen 2015 software ( Figure 3b and Figure 3—figure supplement 1 ) . Primary myoblasts were isolated from C57BL/10ScSn-Dmdmdx/J ( mdx ) mice . The detailed procedure can be found in Rando et al . ( Rando and Blau , 1994; Conboy and Conboy , 2010 ) . Lipofection of Cas9 RNP and Alexa647 labeled 127nt donor ssDNA was conducted following previously published work ( Zuris et al . , 2015 ) . Cells were sorted using a BD influx cell sorter ( BD Biosciences ) in the Berkeley flow cytometry facility , 16 hr after the nucleofection . Positive gating captured the top 20% of Alexa647 positive cells and negative gating captured the bottom 20% of cells , which had the least amount of Alexa647 fluorescence . DNA extraction was conducted 3 days later and PCR amplification of the mdx target region was conducted using the primer set ( GAGAAACTTCTGTGATGTGAGGACATATAAAG and CAATATCTTTGAAGGACTCTGGGTAAAATATC ) . The HDR efficiency in cells was determined via restriction enzyme digestion of the PCR amplified target genes ( Fu et al . , 2014; Lin et al . , 2014; Schumann et al . , 2015 ) . The donor DNA for these experiments contained a ClaI restriction enzyme site . The PCR amplicon of the mdx locus was incubated with 10 units of ClaI . After 16 hr of incubation at 37°C , the products were analyzed by gel electrophoresis using a 4–20% Mini-PROTEAN TGX Gel ( Bio-rad ) and stained with SYBR green ( Thermo Fisher ) . Individual band intensity was quantified using ImageLab and the HDR efficiency was calculated using the following equation , where ( a ) is the uncleaved PCR amplicon and ( b ) and ( c ) are the cleavage products: ( b+c ) ( a+b+c ) x 100 . 5’ Azide-donor DNA was purchased from IDT . 5’ Azide-donor DNA ( 10 μM ) was mixed with 5’ DBCO-crRNA ( 30 μM ) in DI water ( 50 μL ) . The solution was incubated at room temperature overnight and the unreacted crRNA was removed by running the reaction solution through a 30k concentrator ( Amicon Ultra , EMD Millipore ) . The gDonor reaction solution was analyzed via gel electrophoresis using a polyacrylamide gel ( 4–20% Mini-protean TGX Precast gel , Biorad ) 200 ng of the reaction mixture was loaded into the gel . The yield of the gDonor was calculated by dividing the band intensity of the gDonor with the combined band intensities of the gDonor + unreacted Donor DNA . The intensity of the gDonor was multiplied by 8/13 to account for its higher molecular weight . PAGE gel extraction was conducted to purify the gDonor conjugate . The gDonor band was cut with a sharp knife and eluted using the crush and soak method in nuclease-free water for 16 hr , and isolated via ethanol precipitation . The purified gDonor was analyzed via gel electrophoresis using a polyacrylamide gel ( 4–20% Mini-protean TGX Precast gel , Biorad ) , 200 ng of crRNA , donor DNA , unpurified gDonor , and purified gDonor were loaded onto the gel . The genomic region of the Cas9 target sequence was amplified by PCR using Phusion high-fidelity polymerase according to the manufacturer’s protocol . Target genes were amplified first with primer sets ( ATGGTGAGCAAGGGCGAGGAGC and TCGATGCCCTTCAGCTCGATGC ) . After PCR clean-up , the target gene was amplified with deep sequencing primers . The amplicons were purified using the ChargeSwitch PCR clean-up kit ( Thermo Fisher ) prior to the deep sequencing PCR . After barcoding and purification , the Berkeley Sequencing facility performed DNA quantification using a Qubit 2 . 0 Fluorometer ( Life Technologies , Carlsbad , CA ) . A BioAnalyzer was then used for size analysis and qPCR quantification . The library was sequenced with the Illumina HiSeq2500 in the Vincent Coates Genomic Sequencing Laboratory at UC Berkeley . The analysis was conducted using the CRISPR Genome Analyzer ( Güell et al . , 2014 ) . gDonor ( 5 µg in 10 µL ) , and TracrRNA ( 2 µg in 10 µL ) were mixed in 80 µL of Cas9 buffer ( 50 mM Hepes ( pH 7 . 5 ) , 300 mM NaCl , 10% ( vol/vol ) glycerol , and 100 µM TCEP ) , and hybridized by incubating at 60°C for 5 min at RT for 10 min . Cas9 ( 8 µg in 10 µL ) was added and incubated for 5 min at RT , and this solution was then added to the PAsp ( DET ) ( 10 µg in 20 µL ) and incubated for 5 min at RT to generate polymer nanoparticles . The particles were added to BFP-HEK cells ( 105 cells ) at a Cas9 concentration of 16 µg/mL in 500 µL volume of culture medium for 16 hr . crRNA-TracrRNA/Cas9 + donor DNA were complexed with PAsp ( DET ) as a control and scrambled DNA-crRNA-TracrRNA/Cas9 and donor DNA were complexed with PAsp ( DET ) as a second control . Cell transfections with the two control nanoparticles were conducted following the same protocol used for transfecting cells with gDonor and TracRNA . Flow cytometry analysis was conducted 3 days after the nanoparticle treatment . PAsp ( DET ) nanoparticles with gDonor and Cas9 were formulated as described above . The polymer nanoparticles were centrifuged at 17 , 000 g for 10 min , and the supernatant and pellet were collected . Each sample was mixed with a 100 µg of heparin for particle dissociation . The collected supernatant and pellets were run on a gel , and analyzed for the Cas9 and gDonor content in the polymer nanoparticles . Gel electrophoresis was performed using a 4–20% Mini-PROTEAN TGX Gel ( Bio-rad ) in Tris/SDS buffer , with a loading dye containing 5% beta-mercaptoethanol . PageBlue solution ( Thermo Fisher ) staining was conducted and imaged with ChemiDoc MP using ImageLab software ( Bio-rad ) . For particle size measurements , a dynamic light scattering study was conducted using a Zetasizer Nano ZS instrument ( Malvern Instruments Ltd . , Worcestershire , UK ) and a folded capillary cell ( DTS 1060 , Malvern Instruments ) . The reported particle size was measured 5 min after particle mixing . All replicates were biological replicates . Statistical analysis was conducted using Prism7 software . Bold: T7 promoter sequence	There are several different technologies that can be used to make specific changes to particular genes in cells . These “gene editing” approaches have the potential to help humans in a variety of different ways , for example , to treat diseases that presently have no cure , or to improve the nutritional quality of crop plants . One such gene editing approach is known as CRISPR . To edit a specific gene , a molecule called a guide ribonucleic acid ( or guide RNA for short ) binds to a section of the gene and recruits an enzyme to cut the DNA encoding the gene in a particular location . Adding a “donor” DNA molecule that contains the desired “edit” can lead to the cell repairing the broken gene in a way that incorporates the desired change . Modifying the guide RNA or the donor DNA can enhance CRISPR editing . For example , extending the guide RNA molecules by adding “aptamer” sequences can enable researchers to specifically activate the genes that have been edited . It is also possible to add chemical tags to RNA and DNA , but it is not clear how chemical modifications to the guide RNA and donor DNA could affect CRISPR . Here Lee et al . investigated whether adding chemical tags to the guide RNA and/or donor DNA could enhance gene editing . The experiments show that the modified guide RNAs and donor DNAs were still active and could edit DNA in mouse and human cells . Adding a fluorescent molecule to the donor DNA allowed Lee et al . to track which cells contained donor DNA and separate them from other cells . The fluorescent cells had twice as much editing compared to groups of unsorted cells . In further experiments , the guide RNA and donor DNA were fused together and supplied to cells together with a DNA cutting enzyme . Cells containing this combined molecule had three times more editing than cells exposed to the original CRISPR system . This change may aid the development of new uses for CRISPR because it simplifies the system from three components ( an enzyme , guide RNA and donor DNA ) to just a cutting enzyme and the combined molecule . Overall , the findings of Lee et al . show that chemical modifications to guide RNA and donor DNA can make the CRISPR system more versatile . It opens up the possibility of new applications such as adding a targeting group that would direct the CRISPR Cas9 system to a specific cell type or tissue .	[ "Abstract", "Introduction", "Results", "and", "discussions", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "short", "report" ]	2017	Synthetically modified guide RNA and donor DNA are a versatile platform for CRISPR-Cas9 engineering
Genetic regulation of gene expression underlies variation in disease risk and other complex traits . The effect of expression quantitative trait loci ( eQTLs ) varies across cell types; however , the complexity of mammalian tissues makes studying cell-type eQTLs highly challenging . We developed a novel approach in the model nematode Caenorhabditis elegans that uses single-cell RNA sequencing to map eQTLs at cellular resolution in a single one-pot experiment . We mapped eQTLs across cell types in an extremely large population of genetically distinct C . elegans individuals . We found cell-type-specific trans eQTL hotspots that affect the expression of core pathways in the relevant cell types . Finally , we found single-cell-specific eQTL effects in the nervous system , including an eQTL with opposite effects in two individual neurons . Our results show that eQTL effects can be specific down to the level of single cells . Gene expression differences have a strong genetic basis , and genome-wide studies have identified thousands of regions affecting gene expression , termed expression quantitative trait loci ( eQTLs ) ( Albert and Kruglyak , 2015 ) . eQTLs have been found to underlie genetic associations with complex traits and diseases ( Albert and Kruglyak , 2015; Gusev et al . , 2014; Hormozdiari et al . , 2018 ) , and genome-wide eQTL mapping holds great promise for uncovering the molecular underpinnings of phenotypic variation . Studies in purified blood cell populations ( Raj et al . , 2014; Fairfax et al . , 2012; Ishigaki et al . , 2017 ) and computational analyses in human tissues ( Donovan et al . , 2020; Kim-Hellmuth et al . , 2020 ) indicate that many eQTLs are cell-type specific . Studying eQTLs in relevant tissues and cell types is thus important for understanding their role in trait variation ( Yao et al . , 2020 ) . However , major challenges remain for cell-type-specific eQTL mapping . First , studying multiple tissues or cell types is laborious and expensive , and has mostly been limited to large consortia ( GTEx Consortium , 2020 ) . Additionally , diseases can be associated with specific subsets of cells that may not be readily separable from the rest of the tissue , limiting the insight that can be gleaned from studying bulk tissue samples ( van der Wijst et al . , 2020 ) . The recently developed technology of single-cell RNA sequencing ( scRNA-seq ) has the potential to significantly mitigate both of the above issues . In scRNA-seq , gene expression is profiled at the level of individual cells , and different cell populations are identified based on their gene expression profiles . This approach allows simultaneous measurement of expression in multiple tissues and cell types in a ‘one-pot’ experiment , reducing the number of samples that need to be processed and profiled separately . Furthermore , distinct cell types can be interrogated directly , facilitating eQTL mapping in disease-relevant but rare cell types . The size and complexity of mammalian tissues has so far limited the ability to use scRNA-seq for eQTL mapping , and studies have focused on purified cell populations and cell lines ( Cuomo et al . , 2020; van der Wijst et al . , 2018 ) . To pilot a whole-organism approach , we therefore turned to the model nematode Caenorhabditis elegans . One of the mainstays of modern genetic research , C . elegans has an invariant cell lineage that leads to each individual having the same number and identity of cells ( Sulston and Horvitz , 1977 ) . Decades of research have uncovered the functions of many of those cells , as well as expression markers that uniquely identify them ( Hall and Altun , 2007; Hobert et al . , 2016 ) . These features and resources make C . elegans exceptionally well-suited for using scRNA-seq to map eQTLs across cell types in the natural physiological context of a whole animal , down to cellular resolution . Genome-wide eQTL mapping involves acquiring genotypes and gene expression profiles for a genetically diverse cohort . We recently developed a method , C . elegans extreme quantitative trait locus ( ceX-QTL ) mapping , for genetic analysis of complex traits in extremely large populations of segregants ( Burga et al . , 2019 ) . The method takes advantage of a mutation in the gene fog-2 that forces the normally hermaphroditic C . elegans to reproduce via obligate outcrossing , allowing us to propagate a large crossing experiment for multiple generations . Here , we build on ceX-QTL by combining it with scRNA-seq to carry out eQTL mapping at cellular resolution in a single one-pot experiment ( Figure 1A ) . In this approach , a large heterogeneous pool of cells from thousands of genetically distinct individuals is profiled using scRNA-seq , cell types are inferred by clustering scRNA-seq profiles and studying known cell-type markers , and genotype information is reconstructed using expressed genetic variants , enabling eQTL mapping in multiple cell types simultaneously . We propagated a cross between the laboratory strain N2 and a highly divergent isolate from Hawaii , CB4856 , for four generations , generating a pool of 200 , 000 genetically distinct F4 segregants . We dissociated the segregant pool to single cells at the L2 larval stage and profiled the cells with scRNA-seq . We identified clusters in a Uniform Manifold Approximation and Projection ( UMAP ) of the dataset ( McInnes et al . , 2018; Traag et al . , 2019; Trapnell et al . , 2014 ) and determined their cell-type identities using known markers ( Cao et al . , 2017; Packer et al . , 2019 ) . Our final dataset comprises 55 , 508 cells classified into 19 different cell types ( Figure 1B; Supplementary file 1—Table S1 ) . The observed number of cells of each type was strongly correlated with the known cell-type abundance in L2 larvae ( Spearman’s ρ=0 . 87 , p=2 . 2×10−6 , Figure 1—figure supplement 1 ) . Most of the cells in our sample were expected to carry unique genotypes ( Materials and methods , Supplementary file 1—Table S1 ) . This design is advantageous for eQTL mapping because it maximizes the sample size ( Mandric et al . , 2020 ) , but it requires de novo genotype calling because the genotype of each cell is unknown beforehand . Previous studies have shown that in addition to providing a readout of global gene expression , transcriptomic data can also be used for genotyping by leveraging variants in transcribed sequences ( Ronald et al . , 2005; West et al . , 2006; Kang et al . , 2018 ) . To that end , we built a HMM for genotyping cells based on scRNA-seq data . The model calculates the posterior probability of the underlying genotypes for each individual based on three components: ( 1 ) prior probabilities for each of the possible genotypes , ( 2 ) emission probabilities for observing variant-informative reads given each of the possible genotypes , and ( 3 ) transition probabilities for recombination events occurring or not occurring between adjacent genotype-informative sites . Emission probabilities were calculated as previously described for low-coverage sequencing data under the assumption that the observed counts of reads for each possible allele at a genotype-informative site arise from a random binomial sampling of the alleles present and that sequencing errors occur independently ( Dodds et al . , 2015; Bilton et al . , 2018 ) . Transition probabilities at each variant position were derived from the genetic map we previously published for the N2 × CB4856 cross ( Rockman and Kruglyak , 2009 ) . Posterior probabilities of each genotype for each individual were calculated using the forward-backward algorithm ( Figure 2; see Materials and methods for a full mathematical description of the model ) . To minimize loss of power due to genotyping errors , we used the posterior probabilities directly to map eQTLs in a negative binomial modeling framework instead of assigning deterministic genotype calls ( Materials and methods ) . We mapped 1718 cis eQTLs in 1294 genes and 451 trans eQTLs in 390 genes at a false discovery rate ( FDR ) of 10% across the different cell types ( Figure 3A , B , Supplementary file 1—Table S2 ) . The number of eQTLs detected in each cell type was strongly correlated with the number of cells of that type ( Spearman’s ρ=0 . 91 , p<2 . 2×10−16 ) . In cell types with >1000 cells , we mapped between 52 and 415 eQTLs ( Supplementary file 1—Table S1 ) . For 1071 of the 1294 genes with a cis eQTL ( 83% ) , the eQTL was detected in only one cell type . For 208 of the remaining 223 genes ( 93% ) , the direction of the eQTL effect was the same in all cell types in which it was detected . We studied to what degree our cis eQTL results were concordant with gene expression differences between the parents . We generated a scRNA-seq dataset from 6721 N2 and 3104 CB4856 cells and used a classifier trained on the segregant dataset to identify cell types in the parental scRNA-seq dataset . We then carried out a differential expression analysis in each cell type . We found 870 differentially expressed genes ( at a greater than twofold change and FDR of 10% ) , of which 201 ( 23% ) had a cis eQTL in the same tissue ( odds ratio [OR] = 18 . 8 , p<2 . 2×10−16 , Fisher’s exact test ) . In total , 191 of these cis eQTL ( 95% ) showed the same direction of effect as the parental difference . Further , the effect sizes of the significant cis eQTLs were strongly correlated with the sizes of the parental differences ( Spearman’s ρ=0 . 66 , p<2 . 2×10−16 ) ( Figure 3—figure supplements 1 and 2 ) . These results provide independent support for our cis eQTL mapping and show that for a sizable fraction of the genes those cis eQTLs are a major cause of differential gene expression between the strains . To investigate the relationship between single-cell and bulk eQTL mapping , we compared our single-cell eQTLs to those previously identified in a panel of 200 recombinant inbred lines ( RILs ) generated from crossing N2 and CB4856 ( Rockman et al . , 2010 ) . In the bulk study , a large population of whole worms from each RIL was recovered at a late larval stage , L4 , and profiled on expression microarrays . We reanalyzed data for 11 , 535 genes expressed in both datasets and identified 981 cis eQTLs in the bulk dataset ( at an FDR cutoff of 10% ) . Despite major differences in experimental design , including the developmental stage of the worms , the overlap with the single-cell cis eQTLs was highly significant , with 335 cis eQTLs shared between the studies ( OR = 7 . 2 , p<2 . 2×10−16 , Fisher’s exact test ) ( Figure 3C ) . These shared loci represented 34% of the bulk cis eQTLs and 32% of the single-cell cis eQTLs . Furthermore , the bulk and single-cell eQTL effect sizes were highly correlated ( Spearman’s ρ=0 . 64 , p<2 . 2×10−16 ) ( Figure 3D ) . Lastly , single-cell eQTLs detected in multiple cell types were more likely to also be seen in the bulk study: 50% of the genes with cis eQTLs detected in multiple cell types were also identified in bulk compared to 28% of the eQTLs detected in only one cell type ( OR = 2 . 58 , p=2 . 1×10−8 ) ( Figure 3C ) . This observation suggests that the single-cell eQTL mapping approach improves the power to detect cell-type-specific effects . We observed that 90 of the 451 trans eQTLs clustered at five hotspots , each containing 12–31 eQTLs ( Figure 4 , Supplementary file 1—Table S3 ) . A hotspot on Chr . I was identified independently in both neurons and seam cells; the top associated variant ( Chr . I:10890182 ) was the same for both cell types ( Figure 4A , B ) . The other hotspots were identified in the body wall muscle ( on Chr . I ) ( Figure 4C ) , the intestine ( on Chr . V ) ( Figure 4D ) , and neurons ( two distinct hotspots on Chr . III ) ( Figure 4B ) . To test whether the target genes of these five hotspots are involved in coherent biological processes , we relaxed the FDR threshold to 20% , which increased the number of genes linked to each hotspot to 21–42 , and performed Gene Ontology ( GO ) enrichment analysis ( Supplementary file 1—Table S4 ) . For three of the hotspots , we found significant enrichments that were consistent with the cell-type specificity of the hotspot . The targets of the hotspot detected in intestinal cells were weakly enriched for genes involved in the innate immune response ( FDR-corrected p=0 . 042 ) , a major role of that tissue ( Pukkila-Worley and Ausubel , 2012 ) . The targets of the hotspot detected in the body wall muscle were enriched for genes associated with the term myofilament ( FDR corrected p=6 . 4×10−8 ) , actin cytoskeleton ( FDR-corrected p=4 . 2×10−6 ) , and related terms . The enrichment was driven by the genes mup-2 , tni-1 , tnt-2 , mlc-2 , mlc-3 , lev-11 , and act-4 . mlc-2 and mlc-3 encode a myosin light chain , and act-4 encodes an actin protein . lev-11 encodes a tropomyosin , and mup-2 , tni-1 , and tnt-2 encode three of the four proteins in C . elegans that are expressed in the body-wall muscle and form troponin complexes , highly conserved regulators of muscle contraction ( Ono and Ono , 2004; Figure 4—figure supplement 1 ) . The targets of the neuronal hotspot on the right arm of Chr . III were enriched for genes involved in vesicle localization ( FDR-corrected p=7 . 5×10−3 ) , as well as for BMP receptor binding genes ( FDR-corrected p=2 . 8×10−3 ) . The latter enrichment was driven by dbl-1 and tig-2 , orthologs of human bone morphogenetic protein ( BMP ) genes BMP5 and BMP8 and ligands of the transforming growth factor beta ( TGF-β ) pathway ( Gumienny , 2013 ) . Notably , dbl-1 was discovered as a gene that regulates body size in C . elegans ( Suzuki et al . , 1999 ) , the hotspot peak marker is located <300 kb from the peak of a QTL we previously identified for body size ( Andersen et al . , 2015 ) , and the corresponding confidence intervals overlap ( Supplementary file 1—Table S3 ) , suggesting that differential regulation of the TGF-β pathway is involved in variation in body size between N2 and CB4856 . C . elegans is a premier model for studying neurobiology at the cellular level , which is aided by its invariant cell lineage and the diverse functions associated with specific individual neurons . Importantly , many of the neurons are highly variable in their gene expression and express specific gene markers ( Hobert et al . , 2016 ) . To identify specific subtypes of neuronal cells , we separately clustered the 12 , 467 cells identified as neurons and compared the clusters to previous C . elegans scRNA-seq datasets , including the recently published C . elegans Neuronal Gene Expression Map and Network ( CeNGEN ) ( Cao et al . , 2017; Packer et al . , 2019; Hammarlund et al . , 2018; Taylor , 2019; Supplementary file 1—Table S5 ) . The neurons fell into 81 distinct clusters , ranging from 17 to 872 cells . We mapped these clusters onto 100 ( 83% ) of the 120 neuronal clusters identified in CeNGEN ( Figure 5—figure supplement 1 ) . We also identified CEM neurons , which are male specific and absent from CeNGEN , based on the expression of the marker cwp-1 ( Portman and Emmons , 2004 ) . We mapped cis eQTLs in each of the single neuronal subtypes ( sn-eQTLs ) and identified a total of 163 sn-eQTLs in 132 genes at an FDR of 10% ( Figure 5A , Supplementary file 1—Table S6 ) . Of these , 117 ( 88% ) were identified in only a single neuronal subtype . Functional annotation of sn-eQTLs identified 25 genes involved in signaling ( FDR-corrected p=0 . 047 ) , including 12 genes involved in G-protein coupled receptor signaling ( FDR-corrected p=0 . 047 ) and eight genes involved in neuropeptide signaling ( FDR-corrected p=9 . 9×10−3 ) . We compared the sn-eQTLs to those identified when all neurons were analyzed jointly ( ‘pan-neuronal mapping’ ) and found that a sizable fraction of the sn-eQTLs did not have evidence for a pan-neuronal signal: 92 were not identified pan-neuronally at an FDR of 10% and 69 were not identified even at a highly permissive FDR of 50% , suggesting that they exert their effects only in specific neuronal subtypes ( Figure 5A ) . Regardless of statistical significance , pan-neuronal eQTLs should have consistent effect directions across neuronal subtypes , while subtype-specific eQTLs should not . We therefore compared the direction of effect of each sn-eQTL in the subtype in which it was detected with its direction of effect in the set of all neurons excluding that subtype . Among the 69 sn-eQTLs with no signal in the pan-neuronal mapping even at the permissive FDR , the direction of the effect was concordant for 33 and discordant for 36 , not significantly different from chance ( p>0 . 5; binomial test ) , as would be expected if these effects are truly subtype-specific ( Figure 5B ) . In contrast , among the 94 that had a pan-neuronal signal at an FDR of 50% , the direction of the effect was concordant for 88 and discordant for only 6 ( p<0 . 000001; binomial test ) , consistent with differences in detection arising from limited statistical power . In a striking case , we observed an sn-eQTL in the neuropeptide gene nlp-21 that showed significant and opposing effects in two neurons ( Figure 5C , D ) . In the RIC neuron , higher nlp-21 expression was associated with the CB4856 allele ( β = 4 . 4 , FDR-corrected p=0 . 03 ) , while in the RIM neuron , higher nlp-21 expression was associated with the N2 allele ( β = −5 . 4 , FDR-corrected p=9 . 8×10−7 ) . In the pan-neuronal mapping , no significant effect is observed for this gene . We identified the RIC and RIM neurons in the parental dataset , and although the small number of cells in each group ( 35 and 27 , respectively , with only 9 and 5 of them from CB4856 ) was insufficient for statistical testing , the directions of the differences agreed with the eQTL effects ( Figure 5E ) . These results provide direct evidence that eQTLs can be specific down to the cellular level . We used scRNA-seq to map eQTLs in C . elegans across cell types in a single one-pot experiment . Earlier scRNA-seq eQTL mapping studies were limited in sample size to at most ~100 individuals , but nevertheless highlighted the potential of this approach to identify cell-type ( van der Wijst et al . , 2018 ) and developmental ( Cuomo et al . , 2020 ) eQTLs , as well as loci affecting expression variance ( Sarkar et al . , 2019 ) . Our novel approach allowed us to map eQTLs in tens of thousands of genotypically distinct individuals and enabled detection of both cis and trans eQTLs , as well as resolution of their effects down to the level of specific cells . One of the major factors affecting gene expression studies is variation resulting from uncontrolled environmental differences between individuals that are grown or processed separately . By using scRNA-seq , we were able to process all individuals jointly . After the initial parental cross , all subsequent steps carried out over the course of five C . elegans life cycles ( 3 weeks ) were performed in bulk , limiting any confounding environmental factors . To minimize the influence of genotype on development , we synchronized the worms at the first larval stage , L1 , and collected samples at the L2 stage , limiting the time for differences to accumulate post synchronization . Even careful synchronization is not expected to completely remove the effects of genetic variation on developmental timing , and such variation can be combined with gene expression time-course data collected during development to increase the power of eQTL mapping and study the developmental dynamics of eQTLs ( Francesconi and Lehner , 2014 ) . This raises the possibility that future scRNA-seq studies of C . elegans across developmental stages would open the door to a similar analysis of our single-cell eQTL dataset . Previous work suggested the existence of cell-type-specific eQTL hotspots in C . elegans based on the expression patterns of hotspot targets ( Francesconi and Lehner , 2014 ) . We discovered three hotspots that are cell-type specific , with targets that are involved in core functions performed by these cell types . Recently , eQTL hotspots have been identified in human blood cells ( Yao et al . , 2017; Kolberg et al . , 2020 ) , as well as in cell lines ( Brynedal et al . , 2017 ) . These results suggest that hotspot and trans eQTL discovery is facilitated by expression studies that can distinguish cell types and point to a larger role of hotspots in the genetics of gene expression in animals . The extent to which eQTL hotspots play a role in determining biological phenotypes is unknown , although several examples have been described in animals and plants ( Albert and Kruglyak , 2015; Brynedal et al . , 2017; Pang et al . , 2019; Andersen et al . , 2014; Orozco et al . , 2012 ) . One of the neuronal hotspots we identified in this study affects the expression of members of the TGF-β pathway , which has been linked to the control of body size in C . elegans ( Gumienny , 2013 ) . Remarkably , our previous study mapped a QTL for body size in a cross between N2 and CB4856 to the same region ( Andersen et al . , 2015 ) . Ultimately , resolving the physiological impact of eQTL hotspots , including the ones in this study , will require fine-mapping to identify the underlying causal variants . A systematic approach for fine-mapping eQTL hotspots has recently been developed in yeast ( Lutz et al . , 2019 ) ; future advances may enable this to be carried out in animals as well . A comparison of the single-cell cis eQTLs to those mapped in a previous whole-worm eQTL study from our laboratory showed a highly significant overlap despite major differences in experimental design . These results join accumulating evidence that a sizable fraction of cis eQTLs have robust , consistent effects ( GTEx Consortium , 2020; Smith and Kruglyak , 2008 ) and show that many of the effects are conserved across worm development . The strong overlap of cis eQTLs mapped by scRNA-seq and whole-worm analysis also suggests that the effect of many cis eQTLs is conserved across cell types . This observation suggests that modeling strategies that borrow power among cell types could increase the power to detect eQTLs . In addition , our approach for cis eQTL discovery could be supplemented with tests for allele-specific expression ( ASE ) . Looking forward , our method enables scaling up the number of studied individuals , and hence increasing statistical power , simply by sequencing a larger number of cells . Thus , the increasing throughput of single-cell technologies and sequencing platforms will enable future work to study cell-type specificity of cis and trans eQTLs in greater detail . Lastly , we discovered cis eQTLs that act in single subtypes of C . elegans neurons , including many that were not found when all neurons were analyzed jointly . Importantly , we also discovered an eQTL that influences expression of the gene nlp-21 in opposing directions in two different neurons . Such antagonism could result from context-dependent effects of the same variant or from different variants acting in each cell . Our study joins a growing body of work that identified eQTLs that are specific to cell types ( Fairfax et al . , 2012; Ishigaki et al . , 2017; Donovan et al . , 2020; Kim-Hellmuth et al . , 2020; van der Wijst et al . , 2018; Zhang et al . , 2018; Westra et al . , 2015 ) , as well as developmental ( Cuomo et al . , 2020; Francesconi and Lehner , 2014 ) and environmental ( Orozco et al . , 2012; Fairfax and Knight , 2014; Kim-Hellmuth et al . , 2017 ) contexts . Distinct genetic effects are found across cell types and conditions , and large-scale study of such effects is therefore crucial for gaining a comprehensive understanding of regulatory variation . C . elegans strains were cultured at 20°C using standard conditions with the exception that the agar in the nematode growth media ( NGM ) was replaced with a 4:6 mixture of agarose and agar ( NGM+agarose ) to prevent burrowing of the CB4856 strain . Parental strains used were QX2314 ( N2 fog-2 ( q71 ) V; hsp-90p::GFP II ) and PTM299 ( CB4856 fog-2 ( kah89 ) ) . Large segregant panels were generated as before ( Burga et al . , 2019 ) . Briefly , 500 L4 males from PTM299 and 500 L4 hermaphrodites from QX2314 were seeded on a plate for 30 hr , and gravid worms and eggs were collected and bleached . Eggs were synchronized to L1 larvae for 24 hr and seeded on 10 cm NGM+agarose plates . In each generation , gravid worms were bleached , their progeny synchronized for 24 hr , and seeded . The entire process was repeated up to F4 , with 3–4 days per generation . In total , 192 , 000 F4 were seeded on four 10 cm NGM+agarose plates seeded with OP50 . L2 were recovered after 24 hr , and staging was validated under a stereomicroscope . L2 cell dissociation was carried out as previously described ( Zhang et al . , 2011 ) , implementing modifications from a later study ( Kaletsky et al . , 2016 ) , as well as our own . Worms were recovered off the plates and washed three times in M9 . Lysis was then done with an SDS-DTT solution ( 200 mM DTT , 0 . 25% SDS , 20 mM HEPES , pH 8 . 0 , 3% sucrose ) in a hula mixer set on low speed to prevent worms from settling . The lysate was observed under the stereoscope every 2 min , and lysis was stopped when a blunted head shape appeared in the majority of worms ( Kaletsky et al . , 2016 ) , after ~4 min . Worms were then washed quickly three times in 1 ml of M9 , and two additional times in 1 ml of egg buffer ( 118 mM NaCl , 48 mM KCl , 2 mM CaCl2 , 2 mM MgCl2 , 25 mM HEPES , pH 7 . 3 , osmolarity adjusted to 340 mOsm with sucrose ) . Worms were then resuspended in 0 . 5 ml of 20 mg/ml Pronase E that was freshly prepared in L15 media supplanted with 2% fetal bovine serum ( L15-FBS ) and adjusted to 340 mOsm with sucrose . Worm dissociation was done by continuous pipetting on the side of the tube and monitored every 2–3 min on a microscope equipped with a ×40 phase contrast objective lens . Dissociation was stopped when few intact worms remained and a high density of cells was visible . Then , 0 . 5 ml of L15-FBS was added to stop the reaction , and the lysate was spun for 6 min at 500 g at 4°C . The cell pellet was resuspended in PBS ( PBS was adjusted to 340 mOsm with sucrose ) . Cell suspension was spun for 1 min in 100 g at 4°C to remove remaining undigested worms , counted , and diluted to 1 M cells/ml in osmolarity-adjusted PBS , and loaded directly onto five lanes of 3′ Chromium single-cell RNA-sequencing flow cells ( 10x Genomics ) , targeting 10 , 000 cells on each lane . Library prep was carried out according to manufacturer's protocol . Prepared libraries were sequenced together on an S4 lane of Novaseq 6000 . A paired-end 2 × 150 run was done to maximize the recovery of single-nucleotide variants ( SNVs ) . In all downstream processing , each of the five 3′ Chromium lanes processed concurrently was treated as a separate ‘batch’ , and lane identity corresponds to the ‘batch’ identity for the rest of the methods . Raw sequencing reads were analyzed using CellRanger ( version 3 . 0 . 2 ) . We used a gene transfer format file that was corrected for misannotation of 3′ untranslated regions that was generated in a previous study ( Packer et al . , 2019 ) . C . elegans cell types differ widely in the number of UMIs that are recovered using scRNA-seq . Therefore , a simple UMI cutoff , as is commonly used , may be biased for cell types with more UMIs . We therefore implemented an iterative pipeline to recover clusters of bona fide cells and remove cell doublets as well as degraded cells . We took 20 , 000 cells with the most UMIs in each cluster ( twice the targeted number of cells , 100 , 000 overall ) , and processed them in Monocle ( version 3 ) ( Qiu et al . , 2017 ) . Default parameters were used , with the exception that 100 dimensions were used for reduction , and batch was added as a covariate . Leiden clustering identified a total of 154 clusters , and we used the top_markers function in Monocle to identify the genes upregulated in each . We then removed clusters whose top genes included any ribosomal genes or the mitochondrial genes ndfl-4 , nduo-6 , atp-6 , ctc-2 , ctc-3 , ctc-1 , which we noticed were usually found together as the most upregulated genes in clusters that did not specifically express any known markers for C . elegans cell types . This removed a total of 30 , 980 cells ( 31% ) . For the remaining 69 , 020 cells , raw counts were processed using the R package SoupX to reduce ambient RNA contamination ( Young and Behjati , 2020 ) . We then normalized , reduced dimensions , and clustered the background-corrected cell profiles in Monocle using the same parameters as above . To annotate cell types , we used the markers described in a previous study ( Table S12 in Packer et al . , 2019 ) that reanalyzed a previous L2 single-cell dataset ( Cao et al . , 2017 ) . Our cell-type annotation corresponds to the ‘UMAP’ column in that table , with the following exceptions: ( 1 ) we separated hypodermis from seam cells , somatic gonad from sex muscle cells , and glia from excretory cells since those groups were not clustering together in our data . ( 2 ) Cells identified as ‘Miscellaneous’ in that table were annotated as individual cell-type identifications in our data , with the exception of the sphincter and anal muscles , which were not differentiated from each other in our data . Finally , we re-evaluated our cell-type identifications and filtered cell doublets as well as dead cell or debris that may still contaminate bona fide cell-type clusters . We trained a classifier using our manually curated cell-type classifications with a L2-penalized multinomial logistic regression framework , as implemented in the Scikit-learn Python package ( v0 . 22 ) ( Pedregosa , 2011 ) . We read the raw gene expression matrices into Python using scanpy ( v1 . 4 . 2 ) . We removed 2582 genes that were expressed in less than 10 cells . The gene expression levels of each cell were corrected so that the total gene expression counts added up to 10 , 000 . Per gene , these corrected counts were normalized using a log ( 1+x ) transformation . To speed up the computation of the multinomial logistic regression , we only used the 2037 genes with a mean expression between 0 . 0125 and 3 , and a minimum dispersion of 0 . 5 . We scaled the gene expression matrix so that the expression level of each gene across cells had a mean of 0 and a variance of 1 , after scaling expression values over 20 were set to 20 . We fit a multinomial logistic regression model using the scaled gene expression values for the 2037 highly variable genes from the complete set of 69 , 020 cells to obtain an estimate for the inverse regularization strength ( C ) . Using the estimated C of 7 . 74 × 10−04 , we performed fivefold cross-validation to estimate the probability that each cell belongs to one of the manually curated cell-type classifications . Any cell with a probability higher than 0 . 2 of belonging to two or more cell types ( 9198 ) was classified as a doublet . Any cell that did not belong to a cell type with probability ≥0 . 4 ( 5547 ) was classified as low quality . In total , we removed 11 , 398 cells that were classified as a doublet or low quality . We removed an additional 2114 cells classified as Neurons as described in the section ‘Neuronal cell-type classification’ . For the remaining final list of 55 , 508 cells , we used the output of the classifier as the final cell-type classification . The final classification is shown in Figure 1 . For display purposes , the plot in Figure 1 was generated by rerunning umap on the finalized dataset with euclidean distance metric and umap . min_dist = 0 . 5 , resulting in a more compressed visualization of the dataset . We estimated the number of unique genotypes in two approaches . First , we noted that calculating the number of expected unique genotypes is akin to the well-known ‘Birthday problem’ in statistics . Given C cells sampled from I individuals , the expected number of cells with a unique genotype is C ( 1–1/I ) C-1 . Assuming 50–90% of worms were successfully dissociated ( a conservative range ) , we expect 31 , 134–40 , 257 unique genotypes . We also calculated an empirical measurement for collisions . We calculated the genotype correlation matrix of the cells and counted the cells with a maximal correlation with another cell is higher than 0 . 9 . This is presented for each cell type in Supplementary file 1—Table S1 . We used a list of SNVs we previously curated for CB4856 compared to the N2 reference ( Burga et al . , 2019 ) . We derived genotype informative UMI counts for N2 and CB4856 variants using Vartrix version 1 . 0 ( https://github . com/10XGenomics/vartrix ) directly on the output of CellRanger . To reduce SNV counts that result from SNVs in the ambient RNA background , we only kept SNVs that resided in genes with positive counts in the background-corrected matrix . We set up a HMM to infer the genotypes of the recombinant progeny ( Broman , 2005; Arends et al . , 2010 ) . The HMM is used to calculate the probability of underlying genotypes for each individual and requires three components: ( 1 ) prior probabilities for each of the possible genotypes , ( 2 ) emission probabilities for observing variant-informative reads given each of the possible genotypes , ( 3 ) and transition probabilities – the probabilities of recombination occurring between adjacent genotype-informative sites . For the autosomal chromosomes , we defined prior genotype probabilities as 0 . 25 for homozygote N2 , 0 . 5 for heterozygote CB4856/N2 , and 0 . 25 for homozygote CB4856 . For the sex chromosome , we defined prior genotype probabilities as 0 . 44 for homozygotes N2 , 0 . 44 for the heterozygote CB4856/N2 , and 0 . 11 for homozygote CB4856 . These values were chosen because to generate the segregant population N2 hermaphrodites were crossed to CB4856 males and thus contributed twice as many X chromosomes to the progeny as CB4856 . Emission probabilities were calculated as previously described for low-coverage sequencing data ( Dodds et al . , 2015; Bilton et al . , 2018 ) under the assumption that the observed counts of reads for both possible variants ( Y ) at a genotype-informative site ( g ) arise from a random binomial sampling of the alleles present at that site and that sequencing errors ( e ) occur independently between reads at a rate of 0 . 002:p ( Y\|g=NN ) = ( Dr ) ( 1−e ) r ( 1− ( 1−e ) ) D−rp ( Y\|g=NC ) = ( Dr ) 12Dp ( Y\|g=CC ) = ( Dr ) ( e ) r ( 1−e ) D−rwhere ( D ) is the total read depth at a genotype-informative site for a given individual , ( r ) is the total read depth for the N2 variant at that site , and N represents the N2 variant and C represents the CB4856 variant . Transition probabilities were derived from an existing N2 × CB4856 genetic map ( Rockman and Kruglyak , 2009 ) . We linearly interpolated genetic map distances from the existing map to all genotype-informative sites in our cross progeny . We scaled these genetic map distances , multiplying them by a factor of 0 . 4 , to account for the fact that the previous genetic map was built using 10 generations of intercrossing , whereas progeny from our cross are derived from four generations of intercrossing ( Rockman and Kruglyak , 2008 ) . For QTL mapping , we used an additive coding , summing the probability that the genotype was homozygote CB4856 with one half the probability that the genotype was heterozygote N2/CB4856 . Genotype probabilities were standardized , and markers in very high LD ( r > 0 . 9999 ) were pruned . This LD pruning is approximately equivalent to using markers spaced 5 centimorgans ( cm ) apart . For each transcript , we counted the number of cells for which at least one UMI count was detected in each cell type . Transcripts with non-zero counts in at least 20 cells in a cell type were considered expressed in that cell type and used for downstream analyses . As has been previously described for droplet scRNA-seq , counts of UMIs can be adequately parameterized by a gamma-Poisson distribution , which is also known as the negative binomial distribution ( Svensson , 2020 ) . Thus we used a negative binomial regression framework for eQTL mapping here . We also note that simpler approaches using log ( counts+1 ) with ordinary least squares behave pathologically , especially in regard to behavior with multiple partially correlated covariates , and simulations ( not shown ) showed such models lead to inflated false-positive rates . For each expressed transcript in each cell type , we first fit the negative binomial generalized linear model: ( 1 ) E[Y]=μ ( 2 ) Var ( Y ) =μ+1θμ2 ( 3 ) μ=exp ( βi+Xtβt+Xbβb+Xcβc ) which has the following log-likelihood: ( 4 ) l ( β , θ ) =−∑n=1N[ ( yn+θ ) log ( μn+θ ) −ynlog ( μn ) +log ( \|Γ ( yn+1 ) \| ) −θlog ( θ ) +log ( \|Γ ( θ ) \| ) −log ( \|Γ ( θ+yn ) \| ) ]where Y is a vector of UMI counts per cell , Xt is a vector of the log ( total UMIs per cell ) and controls for compositional effects , Xb is an indicator matrix assigning cells to batches , and Xc is the vector of standardized genotype probabilities across cells for the closest genotypic marker to each transcript from the pruned marker set . In addition , β is a vector of estimated coefficients from the model , µn is the expected value of Y for a given cell n , N is the total number of cells in the given cell type , and θ is a negative binomial overdispersion parameter . Model parameters were estimated using iteratively reweighted least squares as implemented in the negbin . reg function in the Rfast2 R package . If the model did not converge , model parameters were estimated with the gam function in the mgcv package ( Wood , 2017 ) , which opts for certainty of convergence over speed . We note that due to the computational burden of fitting so many generalized linear models ( GLMs ) in the context of sc-eQTL mapping , we chose to estimate θ once for each transcript in each cell type for this model and use that estimate of θ in the additional models for that transcript within the cell type , as described below . This approach is conservative as the effects of unmodeled factors ( e . g . , trans eQTLs ) will be absorbed into the estimate of overdispersion , resulting in larger estimated overdispersion ( 1θ ) and lower model likelihoods . Computational approaches that re-estimate θ for each model , jointly model all additive genetic effects , or regularize θ across models and transcripts ( McCarthy et al . , 2012 ) may further increase statistical power to identify linkages . To evaluate the statistical significance of cis eQTLs , a likelihood ratio statistic , -2 ( lnc-lfc ) , was calculated comparing the log-likelihood of this model described above ( lfc ) to the log-likelihood of the model , where β is re-estimated while leaving out the covariate Xc for the cis eQTL marker ( lnc ) . A p-value was derived under the assumption that this statistic is X2 distributed with one degree of freedom . This p-value was used for the evaluation of significance of cis eQTLs for the neuronal subtypes . Within each neuronal subtype , FDR-adjusted p-values were calculated using the method of Benjamini and Hochberg , 1995 . For the other cell types ( with typically much larger cell numbers ) and the genome-wide scans for eQTLs , a permutation procedure was used to calculate FDR-adjusted p-values and is described further below . For each expressed transcript in each cell type , we also scanned the entire genome for eQTLs , enabling detection of trans eQTLs . A similar procedure was used as for cis eQTL except that Equation ( 3 ) was replaced with ( 5 ) μ=exp ( βi+Xtβt+Xbβb+Xgβg ) where Xg is a vector of the scaled genotype probabilities at the gth genotypic marker , and the model is fit separately , one at a time , for each marker across the genome for each transcript . A likelihood ratio statistic for each transcript , within each cell type , for each genotypic marker is calculated by comparing this model to the model where β is re-estimated while leaving out the covariate Xg . The likelihood ratio statistic was transformed into a logarithm of the odds ( LOD ) score by dividing it by 2loge ( 10 ) . We also used functions in the fastglm R package for this scan , again re-using estimates of θ obtained as described above for each transcript for each cell type . For each transcript and each chromosome , QTL peak markers were identified as the marker with the highest LOD score . The 1 . 5 LOD-drop procedure was used to define approximate 95% confidence intervals for QTL peaks ( Dupuis and Siegmund , 1999 ) . FDR-adjusted p-values were calculated for QTL peaks . They were calculated as the ratio of the number of transcripts expected by chance to show a maximum LOD score greater than a particular LOD threshold vs . the number of transcripts observed in the real data with a maximum LOD score greater than that threshold , for a series of LOD thresholds ranging from 0 . 1 to 0 . 1+the maximum observed LOD for all transcripts within a cell type , with equal-sized steps of 0 . 01 . Per chromosome , the number of transcripts expected by chance at a given threshold was calculated by permuting the assignments of segregant identity within each batch relative to segregant genotypes , calculating LOD scores for all transcripts across the chromosome as described above , and recording the maximum LOD score for each transcript . In each permutation instance , the permutation ordering was the same across all transcripts . We repeated this permutation procedure 10 times . Then , for each of the LOD thresholds , we calculated the average number of transcripts with maximum LOD greater than the given threshold across the 10 permutations . We used the approxfun function in R to interpolate the mapping between LOD thresholds and FDR and estimate an FDR-adjusted p-value for each QTL peak ( Albert et al . , 2018 ) . The same procedure was performed for cis eQTL analysis , with the difference being that the expected and observed number of transcripts at a given LOD threshold were calculated only at the marker closest to the transcript . We note that , as expected , Benjamini and Hochberg-adjusted p-values and FDR-adjusted p-values from this permutation procedure for cis eQTLs were nearly identical . The parental QX2314 and PTM299 strains were grown separately for four generations on 10 cm plates , with recurrent cycles of bleaching and synchronization as was done for the segregant population . For single-cell preparation , synchronized L1 from both strains were seeded together in equal numbers on 10 cm plates , and they were processed together from that point onwards , to limit any environmental effects . We believe that differences in efficiency of the cell preparation procedure between N2 and CB4856 could explain the imbalanced representation in the final dataset ( 6721 N2 and 3104 CB4856 cells ) . We took advantage of the different parental genotypes when processing the cells and called cells as those with at least 50 SNV counts supporting one genotype and less than 50 supporting the other . Cell-type identification was automated by using the logistic regression model trained on the segregants that is discussed above . Differential expression analysis was carried out using the DEsingle R package in each cell type , as well as globally in all cells combined ( Miao et al . , 2018 ) . To compare differential expression results with our cis eQTL results , we first normalized the effect size of each cis eQTL by its standard error . Those were used directly in comparisons done within each cell type . To compare with global differential expression , those standardized effects were combined across all cell types in which an eQTL was identified using Stouffer’s weighted-Z method ( Whitlock , 2005 ) . Microarray genotype and gene expression data for our published expression QTL data were acquired from the gene expression omnibus ( Rockman et al . , 2010 ) . Probe sequences were realigned to the WBcel235 transcriptome using BWA , and uniquely mapping probes were used . Expression probes that were present in less than 2/3 of the sample were removed . The genotype and expression matrices were standardized . To map eQTLs , we calculated the Pearson correlation between each probe and every genotype . Correlation coefficients were transformed to LOD scores using −n⋅ln ( 1−R2 ) 2ln ( 10 ) . To assess significance and account for multiple testing , we permuted the sample identities 100 times and calculated the average number of transcripts with an identified eQTL at different LOD scores . We compared these results to the unpermuted LOD scores to estimate the FDR and selected a cutoff corresponding to a rate of 10% ( LOD = 4 . 2 ) , equivalent to the single-cell mapping . cis eQTLs were derived by calculating the Pearson correlation between transcript expression and the normalized genotypes in the variant nearest to a given transcript , transforming to LOD score and comparing against the global threshold . To discover hotspots , we split the genome into 130 bins of 5 cm each . We then counted the number of eQTLs in each bin identified in each cell type ( applying a 10% FDR significance threshold ) , after removing all cis linkages . Cis linkages were defined here as those where the transcribed gene falls within the 95% eQTL confidence interval range extended by 1 MB on both sides . A bin was considered to have an excess of linkages if the number of linkages exceeded the number expected by chance from a Poisson distribution , given the average number of linkages per bin for that cell type and a Bonferroni correction for the total number of bins ( p<3 . 8e-4 ) ( Smith and Kruglyak , 2008 ) . The findpeaks function in the pracma R package was used to identify peak hotspot bins and prevent identifying sets of adjacent bins as hotspots . For GO analysis , we identified hotspot targets using the same procedure above , but relaxed the significance threshold to 20% FDR . We then used the R package topGO to identify enriched terms , with the genes expressed in the cell type used as background . Neuronal classification was carried out using a combination of available C . elegans scRNA-seq datasets , including a published L2 dataset ( Cao et al . , 2017; Packer et al . , 2019 ) , and the CeNGEN project ( Taylor , 2019 ) . Neuronal cells were processed separately using monocle3 with default parameters , with the exception that 100 dimensions were specified for the preprocess_cds step . The analysis was carried out in two passes . In the first pass , we processed all cells identified by our classifier as neurons . Following Leiden clustering , we removed 2114 cells that were in clusters whose top genes were mostly mitochondrial and ribosomal genes , similar to the analysis described above for the global dataset . We then processed the pruned dataset in monocle3 as described above . To annotate the final neuronal clusters , we first used the list of marker genes from two previous publications ( Packer et al . , 2019; Hammarlund et al . , 2018 ) to derive candidate clusters that uniquely express marker genes . We next used the top_markers function in monocle3 to identify upregulated genes in each cluster compared to the rest . These were compared with the data available in the online SCeNGEN Shiny application ( https://cengen . shinyapps . io/SCeNGEA ) for the candidate cluster . The full list of genes used for classification is found in Supplementary file 1—Table S5 . In the final dataset , clusters Unknown_1 - Unknown_4 are of unknown identity and do not correspond to the clusters of the same name in CeNGEN , while the clusters Unknown_touch and Unknown_glut_2 do correspond to cell clusters of the same names in CeNGEN . sn-eQTL analysis sn-eQTL mapping is described above ( section ‘eQTL mapping’ ) . GO annotation of genes with sn-eQTL was done in topGO , with the genes expressed in neurons ( determined using the criteria for inclusion in eQTL mapping ) used as background . A heatmap was plotted using the ComplexHeatmap package ( Gu et al . , 2016 ) . To determine the consistency in effect direction between the sn-eQTL neuron and the rest of the neurons , we repeated the eQTL mapping , aggregating cells from all neuronal cell types but omitting the neuron with the sn-eQTL . The RIC and RIM neurons in the parental datasets were identified using the same gene markers used in the segregant eQTL dataset , as described in Supplementary file 1—Table S5 .	DNA sequences that differ between individuals often change the activation pattern of genes . That is , they change how , when , or why genes switch on and off . We call such DNA sequences 'expression quantitative trait loci' , or eQTLs for short . Many of these eQTLs affect biological traits , but their effects are not always easy to predict . In fact , these effects can change with time , with different stress levels , and even in different types of cells . This makes studying eQTLs challenging , especially in organisms with many different types of cells . Standard methods of studying eQTLs involve separately measuring which genes switch on or off under every condition and in each cell . However , a technology called single cell sequencing makes it possible to profile many cells simultaneously , determining which genes are switched on or off in each one . Applying this technology to eQTL discovery could make a challenging problem solvable with a straightforward experiment . To test this idea , Ben-David et al . worked with the nematode worm Caenorhabditis elegans , a frequently-used experimental animal which has a small number of cells with well-defined types . The experiment included tens of thousands of cells from tens of thousands of genetically distinct worms . Using single cell sequencing , Ben-David et al . were able to find eQTLs across all the different cell types in the worms . These included many eQTLs already identified using traditional approaches , confirming that the new method worked . To understand the effects of some of these eQTLs in more detail , Ben-David et al . took a closer look at the cells of the nervous system . This revealed that eQTL effects not only differ between cell types , but also between individual cells . In one example , an eQTL changed the expression of the same gene in opposite directions in two different nerve cells . There is great interest in studying eQTLs because they provide a common mechanism by which changes in DNA can affect biological traits , including diseases . These experiments highlight the need to compare eQTLs in all conditions and tissues of interest , and the new technique provides a simpler way to do so . As single-cell technology matures and enables profiling of larger numbers of cells , it should become possible to analyze more complex organisms . This could one day include mammals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2021	Whole-organism eQTL mapping at cellular resolution with single-cell sequencing
A defining feature of mitochondria is their maternal mode of inheritance . However , little is understood about the cellular mechanism through which paternal mitochondria , delivered from sperm , are eliminated from early mammalian embryos . Autophagy has been implicated in nematodes , but whether this mechanism is conserved in mammals has been disputed . Here , we show that cultured mouse fibroblasts and pre-implantation embryos use a common pathway for elimination of mitochondria . Both situations utilize mitophagy , in which mitochondria are sequestered by autophagosomes and delivered to lysosomes for degradation . The E3 ubiquitin ligases PARKIN and MUL1 play redundant roles in elimination of paternal mitochondria . The process is associated with depolarization of paternal mitochondria and additionally requires the mitochondrial outer membrane protein FIS1 , the autophagy adaptor P62 , and PINK1 kinase . Our results indicate that strict maternal transmission of mitochondria relies on mitophagy and uncover a collaboration between MUL1 and PARKIN in this process . In most animals , including mammals , mitochondria are inherited strictly through the maternal lineage . Because sperm deliver mitochondria into the egg during fertilization , mechanisms likely exist to eliminate paternal mitochondria from the early embryo . Uniparental inheritance of mitochondria ensures that only one haplotype of mitochondrial DNA ( mtDNA ) exists in the offspring , a phenomenon with considerable biomedical implications . It underlies the maternal inheritance of diseases caused by mutations in mtDNA ( Carelli and Chan , 2014 ) and enables the use of mtDNA sequences to track human migrations during evolution . Mouse studies suggest that extensive heteroplasmy , the co-existence of more than one haplotype of mtDNA , is genetically unstable and associated with physiological abnormalities ( Sharpley et al . , 2012 ) . Although uniparental inheritance is a defining characteristic of mitochondria , there is much speculation about its mechanism in vertebrates ( Carelli , 2015 ) . Most of our knowledge has come from invertebrate model organisms . The phenomenon has been most decisively dissected in Caenorhabditis elegans , where paternal mitochondria are eliminated by mitophagy ( Al Rawi et al . , 2011; Sato and Sato , 2011; Zhou et al . , 2011 ) , a process in which mitochondria are engulfed by autophagosomes and delivered to lysosomes for destruction . In Drosophila melanogaster , paternal mitochondrial elimination involves autophagic components but occurs independently of PARKIN ( Politi et al . , 2014 ) , a Parkinson’s disease-related E3 ubiquitin ligase that is central to the most heavily studied mitophagy pathway ( Pickrell and Youle , 2015 ) . However , it is unclear to what extent these insights from invertebrate model organisms extend to mammals . Consistent with a role for autophagy , sperm mitochondria from mice are ubiquitinated ( Sutovsky et al . , 1999 ) and , after fertilization , are immuno-positive for P62 and the ATG8 homologs LC3 and GABARAP ( Al Rawi et al . , 2011 ) . However , a subsequent study in mouse disputed the role of autophagy in elimination of paternal mitochondria ( Luo et al . , 2013 ) . The association of LC3 with paternal mitochondria was observed to be transient and occurred well before paternal mitochondrial elimination . In addition , it was found that paternal mitochondria were segregated unevenly to blastomeres during early embryonic cell division . Based on these results , the authors rejected the role of autophagy and advocated a passive dilution mechanism whereby murine paternal mitochondria are stochastically lost due to uneven segregation to the cells of the embryo ( Luo et al . , 2013 ) . This mechanistic uncertainty highlights the need to move beyond correlative studies relying on co-localization of autophagy markers with paternal mitochondria , and instead to perform functional studies that directly test the role of autophagy . In C . elegans , the functional role of autophagy was revealed by the persistence of paternal mitochondria in embryos depleted for core autophagy genes , such as the ATG8 homologs LGG-1 and LGG-2 ( Al Rawi et al . , 2011; Sato and Sato , 2011; Zhou et al . , 2011 ) . A similar approach is not feasible in mouse , however , because disruption of basal autophagy results in embryonic arrest at the four-cell stage ( Tsukamoto et al . , 2008 ) , well before paternal mitochondria are normally eliminated . To circumvent this technical hurdle , we reasoned that a functional test for the role of mitophagy might be possible by focusing on mitophagy-specific genes , whose depletion would be less likely to arrest early embryonic development compared to core autophagy genes . To obtain a set of candidate mitophagy genes , we first characterized the requirements for mitophagy in cultured cells . These experiments led to the realization that two E3 ubiquitin ligases , PARKIN and MUL1 , synergistically function in degradation of mitochondria . We then used a gene disruption approach in early embryos to show that mitophagy mediates the degradation of paternal mitochondria . To develop an assay to track paternal mitochondria in the early mouse embryo , we utilized male PhAM mice , in which all mitochondria , including those in the sperm midpiece , are labeled with a mitochondrially-targeted version of the photoconvertible Dendra2 fluorescent protein ( Pham et al . , 2012 ) ( Figure 1A ) . When male PhAM mice were mated with wild-type females , the resulting embryos contained brightly fluorescent paternal mitochondria . At 12 hr post-fertilization ( Figure 1B ) , the paternal mitochondria were found in a linear cluster , reflecting their original , compact organization in the sperm midpiece . At 36 hr after fertilization ( Figure 1C ) , this cluster began to disperse in cultured embryos , and thereafter , well-separated individual mitochondria were visible within blastomeres . Over the next 2 days , paternal mitochondrial content progressively decreased ( Figure 1D–F ) . At 84 hr after fertilization , the majority of embryos had lost all paternal mitochondria ( Figure 1F ) . Quantification of these results showed a reproducible and progressive loss of paternal mitochondria between 60 and 84 hr post-fertilization ( Figure 1G ) . To determine whether this pattern is specific to paternal mitochondria , we additionally mated PhAM female mice with wild-type males , resulting in embryos with fluorescent maternal mitochondria . In these embryos , there was no reduction in the maternal mitochondrial content between 60 and 84 hr post-fertilization ( Figure 1H , Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 17896 . 003Figure 1 . Paternal mitochondria are degraded by 84 hr after fertilization . ( A ) Fluorescence of mito-Dendra2 in a live sperm cell isolated from the cauda epididymis of a PhAM mouse . ( B–F ) Mito-Dendra2 in a 12 hr ( B ) , 36 hr ( C ) , 60 hr ( D ) , 72 hr ( E ) , and 84 hr embryo ( F ) . In ( B ) , note that mito-Dendra2 is circumscribed to a distinct rod-like structure . The mitochondria disperse in later embryos and are lost by 84 hr . ( G ) Quantification of the mito-Dendra2 signal ( see Materials and methods ) at 36 , 60 , 72 , and 84 hr after fertilization . Each data point represents the mean of 15 embryos . Error bars indicate SD . ( H ) Representative maximum intensity projection images of maternal mitochondrial content versus paternal mitochondrial content over time . Embryos with mito-Dendra2-labeled maternal mitochondria were derived from crosses of wildtype males with homozygous PhAM females . Embryos with labeled paternal mitochondria were derived from crosses of wild-type females with homozygous PhAM males , whose sperm donate Dendra2-labeled mitochondria to the embryo upon fertilization . Embryos were cultured in vitro and imaged at the indicated time . Note that paternal Dendra2 signal decreases with time , whereas maternal Dendra2 signal does not . ( I ) Schematic of paternal mitochondrial elimination assay . Wildtype females are mated with PhAM males . One-cell embryos are microinjected in the perivitelline space with concentrated lentivirus targeting candidate genes . During in vitro culture , embryos are periodically imaged live and monitored for their ability to eliminate paternal mitochondria . ( J ) Representative images of embryos injected with lentivirus carrying nontargeting shRNA . The left three images show mito-Dendra2 , phase-contrast , and mCherry signals at 60 hr; the right three images show the same as 84 hr . ( K ) Embryos injected with lentivirus carrying Atg3 shRNA . ( L ) Embryos treated with bafilomycin A1 . ( M ) Embryos injected with lentivirus carrying Parkin shRNA . All scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 00310 . 7554/eLife . 17896 . 004Figure 1—figure supplement 1 . Persistence of maternal versus paternal mitochondria after fertilization . ( A ) Quantification of Figure 1H . Embryos were collected from crosses of homozygous mito-Dendra2 females with unlabeled males , or homozygous mito-Dendra2 males with unlabeled females . The total Dendra2 signal in the embryos were quantified from z-stacks captured at 36 , 60 , 72 , and 84 hr after fertilization . Values plotted are normalized to the measurement at 36 hr . Error bars indicate SD . p=3 . 422E-16 ( 72 hr ) , 7 . 208E-08 ( 84 hr ) ( Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 004 We used a lentiviral approach to functionally probe the role of autophagy genes in this process ( Figure 1I ) . We microinjected one-cell stage zygotes with lentivirus encoding mCherry and control shRNA or shRNA targeting the core autophagy gene Atg3 . In embryos injected with lentivirus , the mCherry reporter was expressed within 48 hr of injection ( 60 hr post-fertilization ) . When nontargeting shRNA was expressed , development of the embryo was unaffected , and Dendra2-positive mitochondria were eliminated by 84 hr with the usual kinetics ( Figure 1J ) . In embryos injected with shRNA against Atg3 ( Figure 1K ) , however , embryo development was arrested at the four-cell stage , consistent with a previous report using Atg5-null oocytes ( Tsukamoto et al . , 2008 ) . Similarly , treatment of embryos with bafilomycin , an autophagy inhibitor , arrested embryonic development ( Figure 1L ) . In both cases , the treated embryos showed persistence of paternal mitochondria at 84 hr . However , due to the early disruption of embryonic development , it was not possible to conclude if autophagy has a specific role in elimination of paternal mitochondria . This result indicated that disruption of core autophagy genes in this system is not a viable experimental approach . We therefore decided to focus on mitophagy-specific genes . We injected embryos with lentivirus encoding shRNA against Parkin ( Park2 ) , an E3 ubiquitin ligase that is central to the most studied pathway for mitopahgy ( Durcan and Fon , 2015; Pickrell and Youle , 2015 ) . Such embryos show loss of paternal mitochondria by 84 hr after fertilization , suggesting that the process occurs in the absence of PARKIN ( Figure 1M ) . Given the negative results with PARKIN , we turned to cultured cells , where the role of specific proteins in mitophagy could be more readily analyzed . Our strategy was to identify , in cultured cells , a small set of mitophagy genes , which could then be re-analyzed in early embryos . To monitor mitophagy , we constructed a dual color fluorescence-quenching assay based on an EGFP-mCherry reporter localized to the mitochondrial matrix . Normal mitochondria are yellow , having both green and red fluorescence in the matrix , whereas mitochondria within acidic compartments show red-only fluorescence , due to the selective sensitivity of EGFP fluorescence to low pH . A similar approach using a mitochondrial outer membrane EGFP-mCherry reporter has been effective for monitoring mitophagy ( Allen et al . , 2013 ) . When mouse embryonic fibroblasts ( MEFs ) were cultured with a moderate concentration ( 10 mM ) of glucose , a condition in which their metabolism relies largely on glycolysis , they showed few red-only mitochondria ( Figure 2A ) . We previously defined a glucose-free , acetoacetate-containing culture formulation that induces MEFs to substantially upregulate OXPHOS activity ( Mishra et al . , 2014 ) . When cells were cultured for 4 days in this OXPHOS-inducing medium , many cells exhibited numerous red puncta ( Figure 2A ) . This observation is consistent with a study showing that glucose-free conditions promote increased turnover of mitochondria ( Melser et al . , 2013 ) and likely reflects the higher turnover of mitochondria when the activity of the respiratory chain is elevated . Atg3 knockout MEFs did not form red puncta under the OXPHOS-inducing condition ( Figure 2B–C ) , indicating that formation of red puncta is dependent on the core autophagy machinery . Consistent with this idea , the level of lipidated LC3 , another core component of the autophagy pathway , was elevated ( Figure 2D ) . Moreover , the red-only puncta co-localized extensively with mTurquoise2-LC3B , suggesting that they represent mitochondrial contents within the autophagosome pathway ( Figure 2E , arrows ) . In addition , a subset of the red puncta co-localize with LAMP1 , likely indicating later intermediates that have progressed to lysosomes ( Figure 2F ) . In contrast , in glycolytic medium , mTurquoise2-LC3B did not co-localize with mitochondria ( Figure 2E ) . In addition , we found that p62 ( SQSTM1 ) , a protein implicated in autophagy ( Pankiv et al . , 2007 ) and mitophagy ( Seibenhener et al . , 2013 ) , localized to mitochondria only under the OXPHOS-inducing condition ( Figure 2G ) . Unlike LC3B and LAMP1 , however , P62 was localized to both red punctate mitochondria and elongated yellow mitochondria . These results indicate that the OXPHOS-inducing condition results in an increase in mitophagy intermediates . 10 . 7554/eLife . 17896 . 005Figure 2 . Induction of mitophagy by OXPHOS-inducing medium . Mitophagy was examined in cells stably expressing Cox8-EGFP-mCherry . Wild-type ( A ) or Atg3 knockout mouse embryonic fibroblasts ( MEFs ) ( B ) were grown in Glucose ( Glu ) or Acetoacetate ( Ac ) containing medium for 4 days and then imaged by fluorescence microscopy . The red puncta in the bottom panel of ( A ) represent mitochondrial contents within acidic compartments . ( C ) Quantification of red-only puncta . Error bars indicate SD of three biological replicates , *p<0 . 01 , p=0 . 0039 ( Atg3+/+ Glu vs . Ac ) , p=0 . 0052 ( Atg3+/+ vs . Atg3 -/- ) ( Student’s t-test ) . ( D ) Western blot analysis of LC3B expression in MEFs cultured in the indicated medium . The lower band is lipidated LC3B . Actin is a loading control . ( E ) Co-localization of LC3B with red puncta . MEFs expressing cox8-EGFP-mCherry and mTurquoise2-LC3B were grown in the indicated medium and imaged by fluorescence microscopy . Arrows indicate examples of mTurquoise2-LC3B co-localization with red mitochondrial puncta . ( F ) Co-localization of LAMP1 with red puncta . MEFs stably expressing cox8-EGFP-mCherry were grown in acetoacetate-containing medium and immunostained with anti-Lamp1 antibody ( blue ) . Arrows indicate red mitochondrial puncta that co-localized with LAMP1 . Scale bar in ( A ) is 10 µm and applies to ( A–F ) . ( G ) Co-localization of p62 with mitochondria . MEFs were grown in the indicated medium and immunostained with anti-p62 ( green ) and anti-HSP60 ( red , mitochondrial marker ) . Error bars indicate SD . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 005 With this cellular system , we sought to identify genes required for induced mitophagy . Previous studies suggested that mitochondrial dynamics , particularly mitochondrial fission , is important for efficient mitophagy ( Mao et al . , 2013; Tanaka et al . , 2010 ) . To explore this idea , we examined the efficiency of OXPHOS-induced mitophagy in a panel of MEF cell lines deficient in mitochondrial fusion or fission genes: Mitofusin 1 ( Mfn1 ) , Mitofusin 2 ( Mfn2 ) , both Mfn1 and Mfn2 ( Mfn-dm ) , Optic atrophy 1 ( Opa1 ) , Mitochondrial fission factor ( Mff ) , Dynamin-related protein 1 ( Drp1 ) , and Mitochondrial fission 1 ( Fis1 ) ( Figure 3A ) . MEFs deficient in mitochondrial fusion were competent for mitophagy . In fact , Mfn-dm cells and Opa1-/- cells showed substantial mitophagy even under glycolytic culture conditions , consistent with the findings that mitochondrial fusion protects against mitophagy ( Gomes et al . , 2011; Rambold et al . , 2011 ) and that Mfn-dm cells have constitutive localization of Parkin to mitochondria ( Narendra et al . , 2008 ) . Among cell lines deficient in mitochondrial fission , Drp1-/- and Mff-/- cells showed normal levels of mitophagy under OXPHOS conditions ( Figure 3A ) . 10 . 7554/eLife . 17896 . 006Figure 3 . Mitophagy under OXPHOS-inducing conditions requires FIS1 , TBC1D15 , and p62 . ( A ) Mitophagy in cells with mutations in mitochondrial dynamics genes . MEFs of the indicated genotype were cultured in glucose or acetoacetate medium , and mitophagy was quantified using the Cox8-EGFP-mCherry marker . Neither Mfn-dm cells nor Opa1-/- cells were viable in acetoacetate-containing medium . Error bars indicate SD of three biological replicates , p=0 . 0078 ( Student’s t-test . ( B ) MEFs stably expressing Cox8-EGFP-mCherry were grown in acetoacetate containing medium and then imaged by fluorescence microscopy . p62 and Tbc1d15 shRNAs were introduced by retroviral infection . ( C ) Co-localization of mTurquoise2-LC3B with mitochondria . MEFs were grown in acetoacetate containing medium . Note that mTurquoise2 puncta localize to mitochondrial puncta ( arrows ) only in WT cells . ( D ) Co-localization of P62 with mitochondria . MEFs were grown in acetoacetate containing medium and immunostained with anti-P62 ( green ) and anti-HSP60 ( red ) . ( E ) Quantification of red-only puncta in WT cells and cells containing shRNA against Tbc1d15 or p62 cultured in glucose ( Glu ) or acetoacetate ( Ac ) medium . Error bars indicate SD of three biological replicates , p=0 . 0048 ( Tbc1d15 ) , p=0 . 0053 ( p62 ) ( Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 00610 . 7554/eLife . 17896 . 007Figure 3—figure supplement 1 . p62 knockout cells have defective OXPHOS-induced mitophagy . ( A ) Quantification of red-only puncta in wild-type or p62 knockout cells grown in medium containing glucose ( Glu ) or acetoacetate ( Ac ) . Error bars indicate SD , three biological replicates , p=0 . 0163 ( Student’s t-test ) . ( B ) Representative image of p62 knockout cell expressing cox8-EGFP-mCherry . Cells were grown in medium containing acetoacetate and imaged by fluorescent microscopy . Scale bar , 10 μm . ( C ) Rescue of mitophagy by p62 replacement . p62 knockout cells stably expressing cox8-EGFP-mCherry were transduced with mTurquoise2-p62 , grown in acetoacetate ( Ac ) containing medium , and imaged by fluorescence microscopy . Arrow indicates mitochondrial localization of mTurquoise2-P62 . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 007 In contrast , Fis1-/- cells had dramatically reduced mitophagy under OXPHOS conditions ( Figure 3A–B ) , and a failure of both LC3 and P62 to co-localize with mitochondria ( Figure 3C–D ) . Although FIS1 is a central player in yeast mitochondrial fission , it does not play a prominent role in mammalian mitochondrial fission ( Losón et al . , 2013; Otera et al . , 2010 ) . Instead , recent studies implicate FIS1 and its interacting protein TBC1D15 ( Onoue et al . , 2013 ) in mitochondrial degradation , specifically in PARKIN-dependent mitophagy ( Shen et al . , 2014; Yamano et al . , 2014 ) . Similar to Fis1 deletion , Tbc1d15 knockdown efficiently blocked mitophagy and decreased LC3 and p62 localization to mitochondria ( Figure 3C–E ) . Expression of shRNA-resistant Tbc1d15 in these cells restored red puncta formation ( Figure 4—figure supplement 1B , C ) . Because depletion of either FIS1 or TBC1D15 blocked mitophagy and abolished P62 localization to mitochondria , we tested whether P62 is required for mitophagy . Cells knocked down for p62 , as well as p62 knockout cells , were deficient for OXPHOS-induced mitophagy and showed reduced mTurquoise2-LC3B localization to mitochondria ( Figure 3C , E; Figure 3—figure supplement 1A–B ) . Expression of mTurquoise2-p62 restored red puncta formation in p62 knockout cells , and expression of shRNA-resistant p62 restored red puncta formation in p62 shRNA expressing cells , consistent with a role for P62 in OXPHOS-induced mitophagy ( Figure 3—figure supplement 1C , Figure 4—figure supplement 1B ) . Taken together , these results place FIS1 and TBC1D15 upstream of P62 in promoting autophagic engulfment of mitochondria . Because PINK1 and PARKIN are central components of the most widely studied pathway for mitophagy ( Pickrell and Youle , 2015 ) , we tested the role of these molecules in our mitophagy assay . Pink1-/- cells showed a substantial reduction in OXPHOS-induced mitophagy ( Figure 4A–B ) . However , Parkin knockout MEFs had normal mitophagy ( Figure 4A–B ) , a surprising observation given that PINK1 is known to operate upstream of PARKIN ( Clark et al . , 2006; Park et al . , 2006; Yang et al . , 2006 ) . This observation suggests that another molecule may compensate for the loss of PARKIN . Recently , the mitochondrial E3 ligase MUL1 ( MULAN/MAPL ) , has been shown to act parallel to the PINK1/PARKIN pathway in ubiquitination and proteasomal degradation of mitofusin ( Yun et al . , 2014 ) . We hypothesized that MUL1 might work in parallel with PARKIN in OXPHOS-induced mitophagy , such that its presence would maintain mitophagy in the absence of PARKIN . Indeed , knockdown of Mul1 by either of two independent shRNAs in the Parkin knockout cell abolished mitophagy ( Figure 4A–B; Figure 4—figure supplement 1A–C ) . In contrast , knockdown of Mul1 alone did not inhibit mitophagy . Inhibition of mitophagy due to loss of PINK1 or PARKIN/MUL1 prevented co-localization of LC3 with mitochondria ( Figure 4C ) . These results reveal that MUL1 and PARKIN have redundant functions in mitophagy . We found a similar redundancy of MUL1 and PARKIN function in mitophagy induced by depolarization of mitochondria with CCCP ( Figure 4—figure supplement 1D ) 10 . 7554/eLife . 17896 . 008Figure 4 . MUL1 and PARKIN have redundant functions in OXPHOS-induced mitophagy . ( A ) Quantification of red-only puncta in cells grown in acetoacetate-containing medium . Presence ( + ) or absence ( - ) of Pink1 , Parkin , or Mul1 is indicated . Error bars indicate SD of three biological replicates , p=0 . 015 ( Pink1 ) , p=0 . 0011 ( Parkin-/- Mulan shRNA ) ( Student’s t-test ) . ( B ) Mitophagy in wild-type and mutant cells . Cells stably expressing Cox8-EGFP-mCherry were grown in acetoacetate-containing medium and imaged by fluorescence microscopy . ( C ) Co-localization of LC3B with mitophagy intermediates . Wild-type and mutant cells were retrovirally transduced with mTurquoise2-LC3B , grown in acetoacetate-containing medium and imaged by fluorescence microscopy . Examples of LC3B co-localization with mitophagy intermediates are indicated by arrows . ( D ) Accumulation of polyubiquitinated proteins in mitochondria . Cells were grown in the indicated medium , and mitochondria were isolated by differential centrifugation . Mitochondrial lysates were analyzed by Western blot for pan-Ubiquitin . HSP60 is a loading control . ( E ) Quantification of polyubiquitinated proteins in mitochondria . Three independent experiments were quantified by densitometry and averages are shown . Ubiquitin level was normalized to HSP60 . Error bars indicate SD , p=0 . 0003 ( WT Glu vs . Ac ) , p=0 . 0011 ( Pink1-/- ) , p=0 . 0016 ( Parkin-/- Mulan shRNA ) , p=0 . 0206 ( Parkin-/- ) ( Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 00810 . 7554/eLife . 17896 . 009Figure 4—figure supplement 1 . Defective mitophagy in Parkin/Mul1-deficient cells . ( A ) Requirement for Parkin/Mul1 in mitophagy . Quantification of red-only puncta in cells grown in medium containing acetoacetate . Wild-type ( + ) or Parkin knockout ( - ) cells were transduced with one of two independent Mulan shRNAs as indicated . Error bars indicate SD from two biological replicates . p values are from the Student’s t-test . ( B ) Western blot analysis of shRNA knockdowns . Cells were transduced with the indicated shRNA ( + ) or not ( - ) and blotted with corresponding antibody . For analysis of the Parkin shRNA , MEFs overexpressing EGFP-Parkin were used , because endogenous Parkin is below the detection limit of the antibody . Actin was used as a loading control . ( C ) Rescue of mitophagy with shRNA-resistant cDNA expression . Cells expressing the indicated shRNA were transduced with shRNA-resistant cDNA constructs as indicated ( + ) . The cells were grown in acetoacetate containing medium , and mitophagy was quantified as described in Figure 4 . Error bars indicate SD from three biological replicates . p values are from the Student’s t-test . ( D ) Dependence of CCCP-induced mitophagy on Parkin and Mul1 . MEFs of the indicated genotypes were incubated with ( + ) or without ( - ) CCCP ( 10 μM ) for 6 hr , and mitophagy was quantified . Error bars indicate SD from three independent experiments . p values are from the Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 009 Mitochondria from cells grown in OXPHOS media are ubiquitinated ~six-fold more than cells grown in glycolytic media ( Figure 4D , E ) . Loss of MUL1 or PARKIN alone had modest or no effect on the induction of mitochondrial ubiquitination under OXPHOS conditions . However , loss of both MUL1 and PARKIN , or PINK1 alone , substantially reduced the ubiquitination of mitochondria ( Figure 4D , E ) . Taken together , these data suggest that MUL1 and PARKIN act in concert to ubiquitinate mitochondrial substrates , and that a threshold level of ubiquitination may be required to trigger mitophagy under OXPHOS conditions . The level of mitochondrial ubiquitination is known to dynamically regulate mitophagy ( Bingol et al . , 2014; Cornelissen et al . , 2014 ) . With these molecular insights from the cellular assay , we re-visited the embryonic system to test whether the same pathway is involved in elimination of paternal mitochondria . We found that embryos expressing shRNA against p62 , Tbc1d15 , or Pink1 showed strong suppression of paternal mitochondrial loss , compared to embryos expressing a non-targeting shRNA ( Figure 5A ) . When these mitophagy genes were knocked down , the majority of embryos retained substantial paternal mitochondria at 84 hr post-fertilization ( Figure 5B ) . In contrast , less than 20% of embryos containing non-targeting shRNA retained significant paternal mitochondria , with the majority of embryos showing complete loss of paternal mitochondria . Depletion of either Parkin or Mul1 alone modestly reduced paternal mitochondrial elimination , but depletion of both had a severe and highly significant effect . Over 60% of Parkin/Mul1-depleted embryos showed retention of paternal mitochondria at 84 hr ( Figure 5A–B , Figure 5—source data 1 ) . 10 . 7554/eLife . 17896 . 010Figure 5 . Clearance of paternal mitochondria in preimplantation embryos requires mitophagy genes . ( A ) Impaired elimination of paternal mitochondria upon inhibition of mitophagy genes . Embryos were injected with lentivirus expressing shRNA against the indicated genes . The mitochondrial Dendra2 signal is shown for live embryos at 60 , 72 , and 84 hr after fertilization . Images are maximum intensity projections . Scale bar , 10 μm . ( B ) Quantification of paternal mitochondrial elimination at 84 hr post-fertilization . Maximum intensity z-projection images were analyzed encompassing the full embryo with z-slices overlapping . Embryos were scored as having no paternal mitochondria ( black bar ) , less than five mitochondrial objects ( white bar ) , or five or more mitochondrial objects ( grey bar ) . Averages of at least three independent injection experiments are shown with 32–200 embryos quantified . Error bars indicate SD , p<0 . 05; p<0 . 01; p<0 . 001 ( Chi-squared test ) . p-Values compare experimental embryos to control embryos with non-targeting shRNA . Chi-squared values: 75 . 386 ( Tbc1d15 ) , 155 . 784 ( p62 ) , 58 . 064 ( Parkin shRNA , Mulan shRNA ) , 1 . 484 ( Mulan shRNA ) , 8 . 074 ( Parkin shRNA ) . ( C ) Clearance of paternal mitochondria in embryos expressing mCherry ( control ) or Fis1-DN . Same scale as ( B ) . ( D ) Quantification of 84 hr results from ( D ) . Error bars indicate SD . p<0 . 001 ( Chi-square test ) . p-Values compare experimental embryos to mCherry control embryos . Chi-squared value: 125 . 584 . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 01010 . 7554/eLife . 17896 . 011Figure 5—source data 1 . Source data for Figure 5B and D . Excel file containing source data for the plots in Figure 5B and D . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 01110 . 7554/eLife . 17896 . 012Figure 5—figure supplement 1 . Inhibition of OXPHOS-induced mitophagy by dominant negative FIS1 . ( A ) Quantification of red-only puncta in wildtype ( WT ) cells or cells transduced with MYC-FIS1-DN retrovirus . Cells were grown in medium containing glucose ( Glu ) or acetoacetate ( Ac ) . Error bars indicate SD from two biological replicates , p=0 . 00066 ( Student’s t-test ) . ( B ) Imaging of cox8-EGFP-mCherry in WT cells or cells transduced with MYC-FIS1-DN retrovirus . Cells were grown in medium containing acetoacetate . Scale bar , 10 μm . ( C ) Diffuse cytosolic localization of MYC-FIS1-DN . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 012 Although FIS1 is a key molecule in the OXPHOS-induced mitophagy pathway , the relevant FIS1 molecules are likely to be contributed by the sperm and not the egg . Our shRNA approach can only knockdown proteins synthesized within the embryo . To circumvent this issue , we developed a dominant negative version of FIS1 ( FIS1-DN ) that lacks the C-terminal transmembrane domain that is essential for mitochondrial outer membrane localization . Retroviral overexpression of the cytosolic FIS1-DN protein in MEFs strongly inhibits OXPHOS-induced mitophagy ( Figure 5—figure supplement 1A–C ) . When FIS1-DN was expressed in embryos , we found that loss of paternal mitochondria was strongly inhibited , with the majority retaining substantial paternal mitochondria ( Figure 5C–D , Figure 5—source data 1 ) . The signal for selective degradation of paternal mitochondria in mammals is unknown , but some other forms of mitophagy are triggered by loss of mitochondrial membrane potential . Using the cationic dye TMRE ( tetramethylrhodamine ethyl ester ) , we found robust staining of sperm isolated from the caudal epididymis of PhAM male mice , indicating intact mitochondrial membrane potential ( Figure 6A ) . At 18 hr after fertilization , paternal mitochondria remained in a linear cluster in the embryo and stained robustly with TMRE . However , over the next 36 hr , paternal mitochondria gradually lost TMRE staining , such that at 48 hr and later , nearly all paternal mitochondria failed to stain with TMRE ( Figure 6B–C ) . In the same experiment , maternal mitochondria always maintained TMRE staining , indicating that there is selective loss of membrane potential in paternal mitochondria10 . 7554/eLife . 17896 . 013Figure 6 . Loss of membrane potential in paternal mitochondria after fertilization . ( A ) Mitochondrial membrane potential in live sperm cell . Spermatozoa were isolated from the cauda epididymis of a PhAM mouse , stained with 20 nM TMRE , washed , and imaged by fluorescent microscopy . Red signal is TMRE; green signal is mito-Dendra2 . The boxed region is enlarged below . Scale bar , 10 μm . ( B ) Membrane potential of paternal mitochondria in early embryos . Embryos , generated by mating wildtype females with PhAM males , were collected at 12 hr after fertilization and cultured in vitro . At 18 , 48 , or 72 hr after fertilization , the embryos were incubated in 20 nM TMRE , washed , and imaged by fluorescent microscopy . Dashed box indicates region enlarged below . Arrows indicate examples of mito-Dendra2-positive spots lacking TMRE signal . Scale bar , 10 μm . ( C ) Fluorescence line analysis of the boxed regions in ( A ) and ( B ) . Each plot measures the TMRE and mito-Dendra2 signals along a one-pixel width line through the center of the boxed region . Note that the mito-Dendra2 and TMRE signals are co-incident at 18 hr after fertilization but not at 48 or 72 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 01310 . 7554/eLife . 17896 . 014Figure 6—figure supplement 1 . Fusion activity of maternal mitochondria versus paternal mitochondria in the early embryo . ( A ) Monitoring mitochondrial fusion in embryos . Embryos were collected from crosses of homozygous mito-Dendra females with unlabeled males ( top panel ) , or homozygous mito-Dendra male with unlabeled female ( bottom panel ) . A subset of mitochondria was photo-converted ( red ) at 36 hr after fertilization . Embryos were again imaged at 60 hr after fertilization . Representative maximum intensity z-projections of the entire embryo are shown . Note the diffusion of the photoconverted Dendra signal ( red ) in maternal but not paternal mitochondria . ( B ) Quantification of experiment in ( A ) . The sum of all red pixel intensities over entire embryo was measured at 36 and 60 hr . The first two bars show the ratios at 60 hr/36 hr ( total intensity ) . The last two bars show the same analysis for average red pixel intensity ( mean intensity ) . Note that mean intensity of maternal photo-converted mitochondria decreases ( indicating fusion ) , whereas that of paternal photo-converted mitochondria does not . Error bars indicate SD of embryos pooled from three females , p=0 . 4156 ( total intensity ) , p=0 . 00013 ( mean intensity ) ( Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17896 . 014 We also examined whether paternal mitochondria fused with maternal mitochondria . To assess mitochondrial fusion , we utilized photo-conversion of Dendra2 . We generated embryos in which either maternal mitochondria or paternal mitochondria were labeled with Dendra2 . At 36 hr after fertilization , we photo-converted a subset of mitochondria and tracked their fate by confocal microscopy . At 60 hr post-fertilization , the photo-converted signal in maternally labeled embryos had spread widely into other mitochondria , resulting in a dramatic reduction in mean pixel intensity ( Figure 6—figure supplement 1A–B ) . In contrast , the photo-converted signal from paternal mitochondria did not diffuse and clearly did not undergo fusion with other mitochondria in the embryo . This segregation of paternal mitochondria is likely to be important for their eventual degradation . Our results provide two major insights about mitophagy in mammals . First , we find in two biological systems—OXPHOS-induced mitophagy in cultured cells and paternal mitochondrial elimination in pre-implantation embryos—that PARKIN and MUL1 work synergistically to promote degradation of mitochondria by autophagy . Previous work showed that PARKIN and MUL1 have partially redundant roles in controlling the ubiquitin-dependent degradation of mitofusin ( Yun et al . , 2014 ) . Our results show that this collaboration extends to the process of mitophagy . In MEFs , we find that both PARKIN and MUL1 regulate the levels of ubiquitin on mitochondria in response to OXPHOS conditions , and removal of both is necessary to cause a substantial reduction of ubiquitination . In a mitophagy assay where PARKIN is overexpressed , polyubiquitination of mitochondrial outer membrane proteins leads to their proteasomal degradation , which in turn is required for turnover of mitochondria by autophagy ( Chan et al . , 2011 ) . The redundant function of MUL1 likely explains why PARKIN knockout mice show surprisingly mild and inconsistent mitochondrial phenotypes ( Palacino et al . , 2004; Perez and Palmiter , 2005 ) . Similarly , the redundant role of MUL1 may also explain why , in Drosophila , PARKIN is dispensable for paternal mitochondrial elimination ( Politi et al . , 2014 ) . In future work , this insight may help to uncover the in vivo functions of PARKIN . Second , we show that mitophagy is likely to be the mechanism underlying the elimination of paternal mitochondria in the early mouse embryo . It is unclear whether MEFs cultured under OXPHOS conditions bear any physiological relation to the early embryo . Nevertheless , we find that the genetic requirements for removal of paternal mitochondria in the embryo mirror those of MEFs undergoing mitophagy in response to OXPHOS induction . Although previous studies had shown that paternal mitochondria in mouse embryos co-localized with autophagy markers ( Al Rawi et al . , 2011 ) , the functional relevance of these localization studies has been challenged and a passive mechanism for loss of paternal mitochondria has been proposed ( Carelli , 2015; Luo et al . , 2013 ) . By identifying several molecules necessary for paternal mitochondrial elimination , our studies provide functional evidence for the role of mitophagy in this process . Because we find that paternal mitochondria lose membrane potential shortly after entering the oocyte , it is tempting to speculate that this membrane depolarization may be the trigger for mitochondrial degradation . Previous studies indicate that PARKIN is recruited to mitochondria upon membrane depolarization ( Narendra et al . , 2008 , 2010 ) , and our results also suggest that PARKIN and MUL1 work together to degrade mitochondria that are depolarized ( Figure 4—figure supplement 1D ) . However , we do not have direct evidence that membrane depolarization has a functional role in paternal mitochondrial degradation . Although uniparental inheritance of mitochondria is nearly universal in animals , its physiological function is mysterious and difficult to address . One recent idea is that uniparental inheritance of mitochondria ensures that offspring contain only one haplotype of mtDNA . When mice with approximately equal proportions of two wild-type haplotypes of mtDNA were generated , they were found to have behavioral and cognitive abnormalities compared to homoplasmic counterparts ( Sharpley et al . , 2012 ) . However , it is unclear to what extent this experimental result is relevant for a case in which paternal mitochondria were not eliminated . Sperm contain many fewer mitochondria ( at least a thousand fold ) compared to the oocyte , and therefore , the ensuing heteroplasmy levels would be very low . The identification of molecules essential for paternal mitochondrial elimination may facilitate further examination of this issue . The following commercially available antibodies were used: anti-Actin ( Mab1501R , Millipore ) , anti-HSP60 ( SC-1054 , Santa Cruz Biotech ) , anti-LAMP1 ( 1D4B , Developmental Studies Hybridoma Bank ) , anti-P62 ( PM045 , MBL ) , anti-LC3B ( 2775 s , Cell Signaling ) , anti-c-Myc ( C3956 , Sigma ) , anti-Ubiquitin ( P4D1 , Cell Signaling ) , anti-PINK1 ( 75488 , Abcam ) , anti-TBC1D15 ( 121396 , Abcam ) , anti-PARKIN ( 15954 , Abcam ) , and anti-MUL1 ( HPA017681 , Sigma ) . For Western analysis , densitometry was done using ImageJ . The intensity of the ubiquitin signal was normalized to that of HSP60 , and the average of three separate experiments was taken . For immunofluorescence experiments , cells were fixed with 10% formalin , permeabilized with 0 . 1% Triton X-100 and stained with the primary antibodies listed above and with the following secondary antibodies: goat anti-mouse Alexa Fluor 633 , donkey anti-goat Alexa Fluor 546 , goat anti-rabbit Alexa Fluor 488 , goat anti-rabbit Alexa Fluor 633 ( Invitrogen , Carlsbad , CA ) . When used , DAPI ( d1306 , Invitrogen ) was included in the last wash . For experiments in MEFs , the retroviral vector pRetroX-H1 , which contains the H1 promoter , was used to express shRNAs . shRNAs were cloned into the BglII/EcoRI sites . For embryo injection experiments , a third-generation lentiviral backbone was used to express shRNAs . The lentiviral vector FUGW-H1 ( Fasano et al . , 2007 ) was modified by replacing the GFP reporter gene with mCherry and changing the shRNA cloning sites from Xba/SmaI to BamHI/EcoRI , generating FUChW-H1 . For dual knockdown experiments in embryos , a second H1 promoter was added , along with XbaI/NheI cloning sites 3’ to the original H1 promoter , generating FUChW-H1H1 . The shRNA target sequences were: p62: TGGCCACTCTTTAGTGTTTGTGT Tbc1d15: GTGAGCGGGAAGATTATAT Mul1 sh1: GAGCTAAGAAGATTCATCT Mul1 sh2: GAGCTGTGCGGTCTGTTAA Pink1: GGCTGACAGGCTGAGAGAGAA Parkin: CCTCCAAGGAAACCATCAA Non-targeting: GACTAGAAGGCACAGAGGG Lentiviral vectors were cotransfected into 293T cells with plasmids pMDLG/pRRE , pIVS-VSVG , and pRSV-Rev . Retroviral vectors were cotransfected into 293T cells with plasmids pVSVG and pUMVC . All transfections were done using calcium phosphate precipitation . For microinjection , virus was collected , filtered , concentrated by ultracentrifugation at 25 , 000 rpm for 2 hr , resuspended in PBS and stored at −80°C as described previously ( Lois et al . , 2002; Pease and Lois , 2006 ) . Viral titers were measured by infecting MEFs with serial dilutions of viral preparations , followed by flow cytometric analysis after 48 hr . Virus was used at 1×107 transducing units/μL . All mouse work was done according to protocols approved by the Institutional Animal Care and Use Committee at the California Institute of Technology . For each experiment , four C57/Bl6J wild-type female mice at 21–25 days old were superovulated by hormone priming as described previously ( Pease and Lois , 2006 ) , and then each was caged with a PhAM male ( Pham et al . , 2012 ) ( RRID:IMSR_JAX:018397 ) . After euthanization of females by CO2 asphyxiation , the embryos were harvested and placed in M2 medium ( MR-015-D , Millipore ) at 12 hr after fertilization as described in ( Pease and Lois , 2006 ) . Approximately 60 to 100 embryos were collected per experiment . Embryos were divided into two equal groups and microinjected with 10 to 100 pl of viral stock into the perivitelline space as described in ( Lois et al . , 2002; Pease and Lois , 2006 ) . Embryos were washed with KSOM+AA medium ( MR-106-D , Millipore ) and cultured in that medium covered by oil ( M8410 , Sigma ) at 37°C and 5% CO2 . For each construct , at least three separate microinjection sessions were performed . In preparation for imaging , embryos were transferred to 10 μl droplets of KSOM+AA medium on glass-bottom dishes ( FD35-100 , World Precision Instruments ) . All images were acquired with a Zeiss LSM 710 confocal microscope with a Plan-Apochromat 63X/1 . 4 oil objective . All live imaging was performed in an incubated microscope stage at 37°C and 5% CO2 . The 488 nm and 561 nm laser lines were used to excite cox8-EGFP-mCherry and imaging was done in line mode to minimize movement of mitochondria between acquisition of each channel . The 405 nm laser line was used to excite mTurquoise2 and DAPI . Alexa 488 , Alexa 546 , and Alexa 633 , conjugated dyes were excited by the 488 nm laser , 561 nm laser , and the 633 nm laser , respectively . In experiments tracking paternal mitochondrial degradation , all viable embryos from each experiment were imaged . Only embryos that were fragmented , lysed , or developmentally delayed were not imaged . The top and bottom of the embryo was set as the top and bottom z slices for z-stack image acquisition . Optical slices were acquired at 1 . 1 μm thickness , and z stacks were oversampled at 0 . 467 μm to ensure that all mitochondria were captured . Maximum intensity projections were created with Zen 2009 software and used for quantification . For quantification of paternal mitochondria , control and experimental embryo images were randomized and counted blind . The number of mitochondria within each embryo was counted manually . In cases where two or more mitochondria were clustered together and could not be definitely resolved as distinct objects with separable borders , the cluster was counted as one object . Each maximum intensity z-projection was categorized as having either no mitochondria , less than five mitochondrial objects , or five or more mitochondrial objects . Embryos from four females were pooled per experiment , and three or more independent replicate experiments were averaged . For photo-conversion of Dendra2 , a region of interest was illuminated with the 405 nm line ( 4% laser power ) for 90 bleaching iterations . The 488 nm laser line ( 5% laser power ) and the 561 nm laser line ( 6 . 5% laser power ) were used to excited Dendra2 in the unconverted state and photo-converted state , respectively . The pinhole used was 159 microns . Bandpass filters were used for detection of unconverted and photo-converted Dendra2 from 494 to 547 nm and 566 to 735 nm , respectively . The mercury lamp was not used to avoid inadvertent photoconversion . For quantification of photo-converted Dendra2 , maximum intensity z-stacks encompassing the entire embryos were analyzed in Matlab . For total intensity measurement , positive pixels were defined as those having an intensity greater than 10 ( a low threshold designed to remove background ) , and the sum of these pixel intensities was calculated . For mean intensity measurement , this sum was divided by the total number of positive pixels . Images were cropped when appropriate , and image contrast and brightness were globally adjusted in Photoshop ( Adobe ) . Replicates are as indicated in figure legends . Sperm were isolated from 4-month-old PhAM male mice . Longitudinal cuts were made in the cauda epididymis , and the tissue was incubated in PBS at 37°C to enable motile , mature sperm to swim out . TMRE fluorescence was used to monitor mitochondrial membrane potential in spermatocytes and embryos . Samples were loaded with 20 nM TMRE for 20 min at 37°C and then washed into PBS ( spermatocytes ) or KSOM+AA ( embryos ) . Samples were imaged live . Line analysis was performed using ImageJ . Mitochondria were isolated by differential centrifugation . Cells were washed in PBS , collected by scraping in isolation buffer ( 220 mM mannitol , 70 mM sucrose , 80 mM KCl , 5 mM MgCl2 , 1 mM EGTA , 10 mM K+HEPES , pH7 . 4 , and HALT protease inhibitors ) , and lysed on ice . Lysates were cleared of cell debris and nuclei with four 600 g spins . A crude mitochondrial fraction was isolated with a 10 , 000 g spin for 10 min and washed three times in isolation buffer . The Cox8-EGFP-mCherry retroviral vector ( kindly provided by Drs . Prashant Mishra and Anh Pham ) consists of the Cox8 mitochondrial targeting sequence placed N-terminal to an EGFP-mCherry fusion . To clone mTurquoise2 fusion proteins , mTurquoise2 was amplified from pmTurquoise2-Mito ( Addgene plasmid # 36208 , Dorus Gadella , [Goedhart et al . , 2012] ) . Human LC3B was amplified from pFCbW-EGFP-LC3 . Mouse p62 was amplified from pMXs-puro GFP-p62 ( Addgene plasmid # 38277 , Noboru Mizushima , [Itakura and Mizushima , 2011] ) . mTurquoise2 fusion proteins were cloned into the retroviral vector , pBABEpuro . The FIS1 dominant negative construct was cloned into pBABEpuro and consists of amino acids 1–121 of mouse FIS1 , with 9 Myc tags at the N-terminus . The corresponding control construct consists of mCherry cloned into the pBABEpuro vector . All plasmids were verified by DNA sequence analysis . Stable cell lines were generated by retroviral infection followed by selection with 2 μg/μl puromycin . MEFs were maintained in Dulbecco’s Modified Eagle’s Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) , 100 U/mL penicillin and 100 U/mL streptomycin at 37°C and 5% CO2 . Glucose and acetoacetate containing media were made as previously described ( Mishra et al . , 2014 ) . For mitophagy experiments , cells were plated on Nunc Lab-Tek II Chambered Coverglass slides ( 155409 , Thermo ) in DMEM-based media . After cells had adhered , they were washed with PBS and glucose- or acetoacetate-containing medium was applied , after which cells were allowed to grow for 4 days and then imaged . Because cells grow more slowly in acetoacetate medium , a four-fold excess of cells was plated relative to glucose medium so that both samples were at the same density on the day of imaging . The cells used included: Atg3-null MEFs ( Sou et al . , 2008 ) ( kindly provided by Yu-Shin Sou and Masaaki Komatsu ) , p62-null MEFs ( Ichimura et al . , 2008 ) ( kindly provided by Shun Kageyama and Masaaki Komatsu ) , Pink1-null , Parkin-null ( both kindly provided by Clement Gautier and Jie Shen ) , and Drp1-null ( Ishihara et al . , 2009 ) ( kindly provided by Katsuyoshi Mihara ) . Mfn1-null ( ATCC Cat# CRL-2992 , RRID:CVCL_L691 ) , Mfn2-null ( ATCC Cat# CRL-2993 , RRID:CVCL_L693 ) , Mfn-dm ( ATCC Cat# CRL-2994 , RRID:CVCL_L692 ) , Opa1-null ( ATCC Cat# CRL-2995 , RRID:CVCL_L694 ) , Mff-null , Fis1-null MEFs have been described previously ( Chen et al . , 2005; Losón et al . , 2013 ) . The identity of MEF cell lines was authenticated by PCR genotyping of the relevant gene . Cell lines were negative for mycoplasma by DAPI staining .	Mitochondria are commonly referred to as the 'powerhouses' of animal cells because these structures provide the majority of the energy in most cells . People inherit their mitochondria from their mother , and not their father . This is because the father's mitochondria , which are delivered by sperm to the egg , are degraded early on when the embryo starts to develop . Previous studies with model organisms , like nematode worms , showed that mitochondria delivered via sperm ( also known as 'paternal mitochondria' ) were delivered to structures called lysosomes and broken down by the enzymes contained within . However , it remained controversial whether this process , named mitophagy , also occurred in mammalian cells , and the molecules involved were unknown . Now , Rojansky et al . have identified key molecules that are essential for the degradation of mitochondria in mouse cells and show that these same molecules are needed to degrade paternal mitochondria in early mouse embryos . These results indicate that paternal mitochondria are indeed degraded by mitophagy in mice . In addition , Rojansky et al . also note that one of the key molecules is a protein called PARKIN , which is mutated in many inherited cases of Parkinson's disease , a major neurodegenerative disorder . Even though these new findings provide a clearer idea as to how paternal mitochondria are degraded , the question of why remains unanswered . As a result , it is likely that this topic will continue to be heavily debated . Nevertheless , having identified the key molecules involved in degrading paternal mitochondria , it may now be possible to address this question more directly – for example by interfering with this process and then examining the consequences .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	Elimination of paternal mitochondria in mouse embryos occurs through autophagic degradation dependent on PARKIN and MUL1
Habit formation is a behavioral adaptation that automates routine actions . Habitual behavior correlates with broad reconfigurations of dorsolateral striatal ( DLS ) circuit properties that increase gain and shift pathway timing . The mechanism ( s ) for these circuit adaptations are unknown and could be responsible for habitual behavior . Here we find that a single class of interneuron , fast-spiking interneurons ( FSIs ) , modulates all of these habit-predictive properties . Consistent with a role in habits , FSIs are more excitable in habitual mice compared to goal-directed and acute chemogenetic inhibition of FSIs in DLS prevents the expression of habitual lever pressing . In vivo recordings further reveal a previously unappreciated selective modulation of SPNs based on their firing patterns; FSIs inhibit most SPNs but paradoxically promote the activity of a subset displaying high fractions of gamma-frequency spiking . These results establish a microcircuit mechanism for habits and provide a new example of how interneurons mediate experience-dependent behavior . Habit formation is an adaptive behavioral response to frequent and positively reinforcing experiences . Once established , habits allow routine actions to be triggered by external cues . This automation frees cognitive resources that would otherwise process action-outcome relationships underlying goal-directed behavior . The dorsolateral region of the striatum has been heavily implicated in the formation and expression of habits through lesion and inactivation studies ( Yin et al . , 2004; Yin et al . , 2006 ) , in vivo recordings ( Tang et al . , 2007; Jog et al . , 1999 ) , and changes in synaptic strength ( Shan et al . , 2015 ) . More recently , properties of the dorsolateral striatum ( DLS ) input-output transformation of afferent activity to striatal projection neuron firing were found to predict the extent of habitual behavior in individual animals ( O'Hare et al . , 2016 ) . Despite these observations , the cellular microcircuit mechanisms driving habitual behavior have not been identified . DLS output arises from striatal projection neurons ( SPNs ) , which comprise ~95% of striatal neurons and project to either the direct ( dSPNs ) or indirect ( iSPNs ) basal ganglia pathways . The properties of evoked SPN firing ex vivo linearly predict behavior across the goal-directed to habitual spectrum in an operant lever pressing task ( O'Hare et al . , 2016 ) . Specifically , habitual responding correlates with larger evoked responses in both the direct and indirect pathways as well as a shorter latency to fire of dSPNs relative to iSPNs . To identify a microcircuit mechanism for habitual behavior , we manipulated the striatal microcircuitry to identify local circuit elements that modulated these habit-predictive SPN firing properties ( Figure 1A , B ) . Glutamatergic corticostriatal synapses express dopamine-dependent forms of long-lasting synaptic potentiation and depression ( Shen et al . , 2008 ) , making these connections a fitting site for experience-dependent adaptation of striatal output . Although such plasticity accompanies changes in behavior , including the formation of habits ( Shan et al . , 2015; Nazzaro et al . , 2012 ) , it does not readily explain the finding that increased gain in the direct and indirect SPNs in habitual mice was balanced ( O'Hare et al . , 2016 ) since synaptic strengthening would occur separately on the two SPN classes through dichotomous mechanisms ( Shen et al . , 2008 ) . In addition , within the DLS , habit-predictive SPN firing properties were distributed uniformly rather than in discrete subpopulations of SPNs ( O'Hare et al . , 2016 ) . Because interneurons are often anatomically suited to tune SPN activity in a similarly broad manner through extensive axonal arbors ( Kawaguchi et al . , 1995; Tepper et al . , 2010 ) , we hypothesized that plasticity of striatal interneurons might underlie the habit-associated changes in striatal output . Among the various interneuron types resident to the striatum ( Tepper et al . , 2010 ) , parvalbumin-positive , fast-spiking interneurons ( FSIs ) provide the strongest source of local modulation , exerting strong , feedforward inhibition of SPNs via perisomatic GABAergic contacts onto virtually all SPNs ( Gittis et al . , 2010; Koós and Tepper , 1999; Koos et al . , 2004; Mallet , 2005; Taverna et al . , 2007; Straub et al . , 2016; Szydlowski et al . , 2013 ) . Notably , FSIs are expressed in the dorsal striatum on a mediolateral gradient with the most residing in DLS ( Gerfen , 1985 ) . FSIs also preferentially innervate dSPNs relative to iSPNs ( Gittis et al . , 2010 ) , suggesting a potential mechanism by which FSI-mediated inhibition could allow iSPNs to fire before dSPNs in response to coincident excitatory input . Based on these considerations , we hypothesized that FSIs might drive the habit-predictive circuit features through a disinhibitory mechanism that would promote SPN firing and a preferentially earlier activation of the direct pathway . Striatal FSI plasticity has been demonstrated through experimenter-induced activity and genetic manipulations ( Mathur et al . , 2013; Winters et al . , 2012; Orduz et al . , 2013; Gittis et al . , 2011a ) , but it remains unknown whether dorsal striatal FSIs undergo plasticity normally in the context of experience-dependent adaptive behavior . Using pharmacological and optogenetic manipulations , we found that striatal FSIs modulate the pathway-specific properties of DLS output that predict habitual behavior . Surprisingly though , silencing FSIs produced the opposite directionality for each habit-predictive circuit feature , suggesting that an increase , rather than decrease , in FSI activity might drive habitual behavior . Indeed , when FSI firing was evoked ex vivo by stimulation of cortical afferents , FSIs from habitual mice fired more readily than FSIs from goal-directed mice . To test the significance of this plasticity for the expression of habitual behavior , we acutely inhibited FSIs in DLS chemogenetically . Inhibiting FSIs in habit-trained mice blocked habit expression , but not lever-pressing per se , while identically-trained control subjects displayed robust habitual behavior . In vivo recordings revealed that the effects of FSI activity on striatal output appear to be more selective than previously appreciated . While FSIs exert the expected strongly inhibitory influence over DLS output , they also promote activity in a subset of SPNs that can be identified a priori based upon individual SPN firing patterns . Our results identify a mechanism for habit by which FSI strengthening reconfigures DLS output and promotes the expression of habitual behavior . To manipulate FSI activity , the calcium-permeable AMPA receptor ( CP-AMPAR ) antagonist IEM-1460 , which predominantly weakens excitatory synaptic inputs onto FSIs in striatum ( Gittis et al . , 2011b ) , was used . Striatal FSIs express AMPARs lacking the GluA2 subunit , rendering them permeable to calcium ( Hollmann et al . , 1991 ) , whereas SPNs do not typically express CP-AMPARs . Consistent with this difference in AMPAR subunit expression , IEM-1460 does not affect excitatory synaptic currents in SPNs but strongly decreases excitatory transmission onto FSIs ( Gittis et al . , 2011b ) . Cell-attached FSI recordings before and after exposure to IEM-1460 ( 50 μM ) confirmed the drug’s efficacy to reduce synaptically-evoked AP firing in our acute parasagittal DLS preparation ( Figure 1—figure supplement 1 ) . To first approximate how FSIs modulate the habit-predictive properties of evoked striatal output , the same ex vivo population calcium imaging approach that identified the behavior-predictive properties ( O'Hare et al . , 2016 ) was used on tissue prepared from untrained animals ( Figure 1A , B ) . Firing responses evoked by electrical activation of cortical afferents were measured in dozens of pathway-defined SPNs of both types simultaneously using the calcium indicator dye fura-2AM , the Drd1a-tdTomato ( Ade et al . , 2011 ) reporter , and vector-mode two-photon laser scanning microscopy ( 2PLSM ) ( Figure 1A; see Materials and methods and O'Hare et al . , 2016 ) . Action potential responses were detected by cross-correlation analysis with a template waveform that was obtained from single-action potential responses during simultaneous cell-attached electrophysiological recordings for each SPN subtype ( see Materials and methods ) . Contamination of dSPN and iSPN datasets by interneurons was minimized by selection criteria and monitoring datasets for outliers ( See Materials and methods for further details ) . Firing properties of SPNs were compared within-cell before and after wash-in of IEM-1460 . IEM-1460 decreased the amplitude of evoked calcium transients in both dSPNs ( t ( 86 ) = 3 . 42 , p=0 . 001 , n = 87 ) and iSPNs ( t ( 51 ) = 2 . 11 , p=0 . 040 , n = 52 ) . IEM-1460 also changed the relative latency to fire between direct and indirect pathway SPNs by increasing the pre-existing bias in relative pathway timing whereby iSPNs tend to respond to cortical excitation more quickly than dSPNs ( Figure 1F ) ( mean absolute latency values for dSPNs: 144 . 03 ± 7 . 08 ms ACSF , 154 . 33 ± 7 . 92 ms IEM-1460 , N = 87; iSPNs: 130 . 31 ± 7 . 87 ms ACSF , 134 . 43 ± 8 . 89 ms IEM-1460 , N = 52 ) . Upon closer inspection , the decrease in calcium transient amplitude seen at the population level appeared to be dominated by the subset of SPNs with larger baseline responses ( for example , see brightest red cells before wash-in in Figure 1C ) . To determine whether there was selectivity for IEM-1460’s effects on SPNs with large basal responses , calcium transient amplitude was used as a feature to classify SPNs as having large or small evoked calcium transients prior to drug wash-in . Rather than specifying an arbitrary cutoff value for the transient amplitude , we used an unsupervised clustering algorithm known as a Gaussian mixture model ( GMM ) to separate SPNs into two clusters . Based on calibration data in this preparation demonstrating the relationship between calcium transient amplitude and number of action potentials ( O'Hare et al . , 2016 ) , the GMM separated SPNs into clusters corresponding to multi-action potential ( larger transients; ‘high-firing’ ) and single-action potential ( smaller transients; ‘low-firing’ ) responses ( Figure 1D ) . Compared to the use of a physiologically-based 0 . 05 ΔF/F0 cutoff value , the unbiased GMM classification was in 90 . 5% agreement . According to this pre-IEM-1460 categorization , low-firing SPNs were unaffected whereas calcium transient amplitudes of high-firing SPNs were significantly reduced by IEM-1460 ( Figure 1D ) . This selective relationship was also borne out by examining the relationship between basal calcium transient amplitude and the magnitude of IEM-1460 effect . Consistent with a selective inhibition of multi-action potential responses , basal calcium transient amplitudes linearly predicted the inhibitory effect of IEM-1460 in both SPN subtypes ( Figure 1E ) . Moreover , IEM-1460 did not affect spike probability in either SPN subtype ( Figure 1—figure supplement 1 ) . These pharmacological experiments in acute brain slices indicate that IEM-1460 promotes an indirect pathway timing advantage and selectively diminishes multi-action potential evoked SPN responses . Because the within-cell experimental design of measuring effects before and after IEM-1460 application did not exclude the possibility that changes in calcium signals occurred during the 20 min wash-in period independently of IEM-1460 , we performed a separate across-group study . Brain slices were incubated with either IEM-1460 or vehicle prior to and during imaging . Group mean calcium transient amplitudes were lower in IEM-1460 relative to vehicle in both dSPNs ( vehicle: 0 . 043 ± 0 . 0011 ΔF/F0 , N = 202 cells; IEM-1460: 0 . 037 ± 0 . 0021 ΔF/F0 , N = 72 cells; t ( 272 ) = 2 . 62 , p=0 . 0093 ) and iSPNs ( vehicle: 0 . 040 ± 0 . 0014 ΔF/F0 , N = 143 cells; IEM-1460: 0 . 033 ± 0 . 0011 ΔF/F0 , N = 56 cells; t ( 197 ) = 2 . 93 , p=0 . 0038 ) and IEM-1460-treated slices showed a preference for faster indirect pathway activation relative to vehicle-treated slices ( t ( 197 ) = 3 . 83 , p=1 . 41×10−7 , N = 143 and 56 independent dSPN/iSPN pairs ) . These results are generally consistent with findings from within-cell pre-post measurements . To further test whether IEM-1460 selectively inhibited multi-spike SPN responses using methodology that did not involve inferring action potentials through calcium imaging , we used conventional electrophysiological methods to record cortically-evoked SPN firing in cell-attached mode . Brief single-pulse electrical stimuli ( 300–600 μs ) were calibrated to elicit a stable multi-action potential response in SPNs prior to taking a baseline measurement . Responses to the same stimulus were then recorded after wash-in of IEM-1460 or vehicle . Consistent with the calcium imaging results , IEM-1460 decreased evoked SPN firing ( t ( 7 ) = 2 . 37 , p=0 . 029 , n = 8 ) while vehicle had no significant effect ( p=0 . 76 , n = 8 ) . Moreover , the same selectivity for modulating multi-action potential responses was observed in that the magnitude of IEM-1460’s effect correlated with the size of baseline responses and there was no effect on single-action potential responses ( Figure 1—figure supplement 2 ) . This result confirms that IEM-1460 , which inhibits FSI firing ( Figure 1—figure supplement 1 ) , selectively reduces multi-action potential SPN responses to afferent stimulation as suggested by calcium imaging experiments ( Figure 1D , E ) . Altogether , this series of experiments identifies a pharmacological agent that potently inhibits FSI activity and modulates all of the habit-predictive SPN firing properties . These results were surprising for two reasons . First , rather than a blockade of FSI activity causing disinhibition of SPNs as we had hypothesized , we found that when FSI activity was reduced , SPN activity was also reduced . This result suggests that FSI activity is capable of promoting , rather than inhibiting , SPN activity at least in the acute brain slice preparation . Secondly , although IEM-1460 strikingly affected the same features of DLS output that predict the expression of habitual behavior ( calcium transient amplitude in both pathways and relative pathway timing ) ( O'Hare et al . , 2016 ) , the directionality of these effects was opposite in all measures . Therefore , these results revise the overall hypothesis to involve a gain , rather than loss , of FSI activity as a candidate mechanism for habitual behavior . While IEM-1460 has been shown to have selective effects on the firing of FSIs in striatum , its effect of inhibiting AMPAR-mediated excitatory postsynaptic currents ( EPSCs ) in cholinergic interneurons ( CINs ) ( Gittis et al . , 2011b ) leaves open the possibility that CINs might contribute to our observed IEM-1460 effects . To isolate the effects of FSIs , the light-activated hyperpolarizing proton pump Archaerhodopsin-3 fused to green fluorescent protein ( Arch-GFP ) was Cre-dependently expressed in parvalbumin ( PV ) -expressing cells . Pvalb-Cre mice were crossed to a line which Cre-dependently expressed Arch-GFP ( See Materials and methods ) . Control experiments showed that , as predicted , 532 nm light drove outward currents in FSIs but not SPNs ( Figure 2—figure supplement 1 ) . Additionally , Arch expressed in PV+ cells ( PV-Arch ) abolished high-frequency firing of FSIs in response to somatic current injection ( Figure 2—figure supplement 1 ) and had no effect on SPN firing in the same recording configuration ( Figure 2—figure supplement 1 ) . To examine the contribution of FSI activity to SPN firing , cortically-evoked SPN action potentials were recorded in cell-attached mode , as in the cell-attached IEM-1460 experiments , while nearby PV+ interneurons ( ~0 . 5 mm radius from recorded SPN ) were silenced in alternating trials with 532 nm light exposure ( Figure 2A ) . In this configuration , PV-Arch effectively blocked evoked FSI firing ( Figure 2B ) . We found that optical inhibition of PV+ interneurons reliably decreased evoked SPN firing ( Figure 2C , left and middle panels ) . Given that IEM-1460 selectively reduced the probability of multi-action potential SPN responses , we examined whether optical inhibition of PV+ neurons had a similar selectivity . Analysis of SPN responses by trial ( paired consecutive laser OFF/ON sweeps ) , rather than by cell , indicated that single-action potential events and failures were unaffected when FSIs were silenced ( Figure 2C , right panel ) . Moreover , a single-exponential fit of all trial-by-trial data showed a selective contribution of FSIs to multi-spike SPN responses ( Figure 2C , right panel ) . Consistent with the IEM-1460 results in 2PLSM calcium imaging ( Figure 1D–E ) and cell-attached recording ( Figure 1—figure supplement 2 ) experiments , this optogenetic result indicates that FSIs promote multi-action potential SPN responses to cortical excitation in the brain slice and that the effects of IEM-1460 on striatal output occur primarily through a reduction of striatal FSI activity . While results thus far show that FSIs appear capable of specifically modulating habit-predictive properties of striatal output , we next examined whether FSI activity was different as a result of experience . We measured FSI synaptic and cellular electrophysiological properties in DLS brain slices prepared from habitual and goal-directed mice . Pvalb-Cre mice were bilaterally injected with AAV5-Ef1a-DIO-eYFP in the DLS to label PV+ interneurons and subsequently trained on an operant task in which they learned to press a lever for sucrose pellet rewards . Lever presses were reinforced on a random interval ( RI ) schedule to induce habit formation ( Dickinson et al . , 1983; Hilário et al . , 2007 ) or on an abbreviated random ratio ( RRshort ) schedule to produce goal-directed behavior ( O'Hare et al . , 2016 ) ( Figure 3—figure supplement 1 ) . Habit was measured by evaluating the sensitivity of the learned lever press behavior to devaluation of the sucrose pellet reward . Goal-directed performance is known to be highly sensitive to outcome devaluation whereas habitual performance is less sensitive ( Dickinson et al . , 1983; Hilário et al . , 2007; Dickinson , 1985 ) . The sucrose pellet reward was devalued by inducing sensory-specific satiety . Specifically , mice were pre-fed with the reward pellets or , as a control for general satiety-related behavioral changes , identically-sized normal grain pellets . On separate but consecutive days , mice were alternately pre-fed 1 . 3 g of either the sucrose pellet reward ( devalued condition ) or the grain-only pellet ( non-devalued condition ) , counterbalancing which pre-feed condition was tested first . Lever press rates were then measured during brief 3 min probe tests without reinforcement . Habitual behavior was quantified in individual mice as the log2 ratio of the devalued versus non-devalued lever press rates ( normalized devalued lever press rate; NDLPr ) . RI-trained mice with an NDLPr ≥ 0 , that is , insensitive to outcome devaluation , were considered to be habitual . RRshort-trained mice with an NDLPr < 0 were considered to be goal-directed ( Figure 3—figure supplement 1 , shaded regions ) . Mice not meeting either inclusion criterion were not used for the electrophysiological studies . We first examined whether excitatory synaptic transmission onto FSIs was altered with habit formation . Spontaneous EPSCs ( sEPSCs ) were recorded in the presence of the GABAA receptor antagonist picrotoxin ( 50 μM ) . No difference was detected in sEPSC frequency or amplitude between goal-directed and habitual FSIs ( Figure 3A ) . Additionally , paired-pulse ratios of evoked EPSCs measured at a 50 ms inter-stimulus interval were similar between groups ( Figure 3B ) . During these recordings , we also did not observe any group differences in a number of passive membrane properties ( Figure 3—figure supplement 1 ) . Rather than changes in synaptic strength , we instead found robust differences in FSI firing responses to somatic current injection . FSIs from habitual mice displayed higher firing rates compared to FSIs from goal-directed mice ( Figure 3C ) . Action potential kinetics did not appear to explain these group differences in firing rates as action potential waveforms were not appreciably different between groups ( Figure 3—figure supplement 1 ) . However , the duration over which firing could be sustained markedly differed between the two behavioral groups ( Figure 3D ) . The majority of FSIs from goal-directed mice were unable to maintain high-frequency firing for the entire duration of the 500 ms current injection ( <250 ms of firing in 10/15 cells ) whereas nearly all FSIs from habitual mice maintained such activity ( >450 ms firing in 7/9 cells ) . Interestingly , the distribution of goal-directed FSI response durations was strongly bimodal whereas that of habitual FSI response durations was not ( Figure 3D ) . The group difference in response durations explained the difference in firing rates between FSIs of habitual and goal-directed mice since , when firing rates were normalized to the duration of firing instead of duration of the current step , there was no longer a group difference in firing rate ( Figure 3E ) . Habitual behavior was associated with increased FSI firing in response to somatic current injection . However , it was afferent activation that initially revealed habit-predictive striatal output properties ( O'Hare et al . , 2016 ) . Therefore , in order for FSI plasticity to alter striatal output , it must be sufficient to differentially drive FSI firing in response to similar coincident synaptic excitation . FSI firing was monitored in cell-attached mode in response to electrical stimulation of excitatory afferents . We found that FSIs of habitual mice fired more readily than those from mice with goal-directed behavior ( Figure 3F ) . This habit-related difference in FSI excitability was not readily explained by other aspects of lever pressing performance including the total number of lever presses or rewards delivered over the course of training ( Figure 3—figure supplement 1 ) . We noted the apparent bimodal distribution of total rewards delivered for goal-directed subjects ( p=0 . 013 , Hartigans’ dip test; Figure 3—figure supplement 1 ) and wondered if the number of rewards received by an animal was related to the similarly-distributed FSI response durations to current injection ( Figure 3D ) . Instead , we found that response durations from both modes of the distribution were commonly found in FSIs from the same goal-directed mouse ( for example , 494 . 7 and 180 . 9 ms ) . Together , these experiments show that FSIs undergo long-lasting , experience-dependent plasticity with habit formation and that this plasticity is sufficient to increase FSI firing . Since photo-inhibiting FSIs produces striatal output properties that directly oppose those seen in habit ( Figure 1 ) , we inhibited FSIs after habit training to determine the necessity of FSI activity for expression of habitual behavior . Mice underwent habit-training protocols in the operant lever press task and then , prior to testing the degree of habitual responding , FSIs were inhibited chemogenetically . We selected a chemogenetic approach to allow for continuous modulation of activity during the 3 min probe tests which measure habitual behavior . Drd1a-tdTomato::Pvalb-Cre mice were bilaterally injected in DLS with AAV vectors Cre-dependently encoding either the inhibitory hM4D chemogenetic receptor ( Armbruster et al . , 2007 ) ( PV-hM4D ) or eYFP ( PV-eYFP ) ( Figure 4A , B ) . Both groups underwent the same habit-promoting RI reinforcement protocol and learned similarly ( Figure 4C ) . For both the devalued and non-devalued conditions , after each pre-feeding period and thirty minutes prior to the outcome devaluation probe tests , the hM4D agonist clozapine N-oxide ( CNO , 5 mg/kg ) was delivered intraperitoneally ( Figure 4D ) . Chemogenetic inhibition of PV+ interneurons did not affect operant behavior in general , as evidenced by indistinguishable lever press rates between groups in the non-devalued ( grain pellets ) condition ( Figure 4—figure supplement 1 ) . In contrast , a comparison of sensitivity to outcome devaluation between groups revealed that habit expression was suppressed in PV-hM4D mice relative to PV-eYFP controls ( Figure 4E ) . Mean NDLPr for RI-trained PV-eYFP control mice measured at 0 . 46 ± 0 . 27 , indicating that control mice were insensitive to outcome devaluation , i . e . habitual . By contrast , PV-hM4D mice , which received the same RI training schedule and showed comparable rates of lever pressing ( Figure 4C ) , displayed a mean NDLPr of −0 . 60 ± 0 . 30 . A negative NLDPr indicates sensitivity to outcome devaluation , i . e . goal-directed responding . These findings show that acute suppression of FSI activity in DLS causes habit-trained subjects to behave as though they were goal-directed . To understand how chemogenetic suppression of FSI firing affects striatal activity in vivo , single unit recordings were performed in a cohort of Drd1a-tdTomato::Pvalb-Cre mice implanted in DLS with multi-electrode arrays and injected with the Cre-dependent hM4D inhibitory chemogenetic virus . Single units corresponding to both FSIs and SPNs were recorded in freely-moving mice ( Figure 5A–D ) for 30 min before intraperitoneal ( i . p . ) injection of CNO ( 5 mg/kg ) or vehicle and during the period of 30–60 min after injection . As expected for the inhibitory hM4D receptor , CNO significantly decreased FSI firing rates compared to vehicle-injected controls ( CNO: 59 . 61 ± 8 . 08% baseline; vehicle: 86 . 89 ± 11 . 66% baseline ) ( Figure 5E ) . In line with previous ex vivo ( Koós and Tepper , 1999; Koos et al . , 2004 ) and in vivo ( Mallet , 2005; Gittis et al . , 2011b ) studies , we further found that suppressing FSI activity caused an overall increase in SPN firing ( i . e . disinhibitory effect ) relative to vehicle ( CNO: 472 . 00 ± 149 . 12%; vehicle: 188 . 02 ± 45 . 94%; Figure 5F ) . In contrast to the straightforward effect of CNO on FSI activity , the effect of CNO injection on SPNs was far more variable . Post-CNO SPN firing rates ranged from 32 . 5% to 2511 . 1% of baseline ( CV = 147% ) with 26% of SPNs displaying negative modulation . In acute slice experiments , FSIs had displayed an unexpected and selective effect of promoting multi-action potential responses ( Figures 1D , E and 2C ) but not otherwise affecting spike probability ( Figure 1—figure supplement 1 ) . To assess whether FSIs also promoted activity in identifiable subsets of SPNs in vivo , we analyzed the baseline firing patterns in single SPNs prior to CNO injection . SPN spiking was categorized into discrete frequency bands by deriving instantaneous firing rate from interspike intervals ( ISIs ) and was then normalized to total number of ISIs for each single unit . This analysis defined the fraction of ISIs corresponding to each frequency band for each SPN and was independent of local field potentials . We found that the baseline ( pre-CNO ) fraction of ISIs falling within the highest rate frequency band , gamma-frequency ( 30–100 Hz ) , linearly predicted how firing rates in individual SPNs changed when FSI activity was suppressed ( Figure 5G , left; see Figure 5—figure supplement 1 for example units ) . That is , the higher the fraction of gamma-frequency spikes an SPN fired , the more likely it was to fire less when FSIs were chemogenetically inhibited . No such relationship was observed in response to vehicle ( Figure 5G , right ) . Since neurons with higher firing rates would be expected to have shorter ISIs in general , we examined the possibility that the fraction of gamma ISIs in SPNs might simply relate to mean firing rate . However , we found that the proportion of gamma-frequency ISIs was unrelated to mean firing rate in baseline single unit SPN recordings before either CNO or vehicle administration ( pre-CNO: p=0 . 25 , n = 23; pre-vehicle: p=0 . 28 , n = 20 ) . Additionally , we found that SPNs fire significantly more gamma-frequency spikes than expected by Poisson processes matched to firing rate ( pre-CNO: t ( 44 ) = 5 . 76 , p=7 . 67×10−7 , n = 23 SPNs and rate-matched simulations; pre-vehicle: t ( 38 ) = 8 . 24 , p=5 . 59×10−10 , n = 20 SPNs and rate-matched simulations ) . Whereas baseline firing rates non-specifically predict fold change in firing rate after both CNO and vehicle injection ( CNO: r ( 22 ) = −0 . 61 , p=0 . 0022 , n = 23; vehicle: r ( 19 ) = −0 . 45 , p=0 . 045 , n = 20 ) , the excess probability of gamma-frequency ISIs ( observed – expected ) specifically predicts rate modulation after CNO ( r ( 22 ) = −0 . 52 , p=0 . 011 , n = 23 ) but not vehicle ( r ( 19 ) = 0 . 045 , p=0 . 85 , n = 20 ) . Directly comparing correlation coefficients using Fisher r-to-z transformations showed that baseline firing rates did not predict SPN rate modulation by CNO better than by vehicle ( z = −0 . 68 , p=0 . 25 ) whereas baseline excess gamma specifically predicted modulation by CNO ( z = −1 . 88 , p=0 . 030 ) . Therefore , gamma-frequency spiking represents a feature of interest in SPNs that predicts whether these output neurons will fire more or less as a consequence of reducing FSI activity . These results demonstrate that FSIs modulate SPN activity in a more complicated manner than previously appreciated . While FSIs can have an overall strongly inhibitory effect in vivo on SPN firing as traditionally assumed , we also found evidence that they potentiate activity in a select population of SPNs that displays higher fractions of gamma-frequency spiking . This selective potentiation may be akin to a winner-take-all ‘focusing’ mechanism that increases the signal-to-noise ratio in corticostriatal transmission . According to such a mechanism , the subset of recruited SPNs would be facilitated while the less-relevant SPNs with low fractions of gamma spiking would be suppressed . With the recent availability of tools to study specific , genetically-defined types of neurons , critical roles for interneurons in facilitating behavioral adaptations to experience are becoming increasingly apparent . In brain regions other than the striatum , interneuron activity appears to most commonly serve as a gate for the induction of long-lasting plasticity elsewhere in the local circuitry ( Kuhlman et al . , 2013; Kvitsiani et al . , 2013; Wolff et al . , 2014; Yazaki-Sugiyama et al . , 2009 ) . Although the potential for FSIs themselves to exhibit long-lasting activity-dependent plasticity is well-documented in acute brain slice experiments ( Mathur et al . , 2013; Orduz et al . , 2013; Hainmüller et al . , 2014; Sarihi et al . , 2012; Dehorter et al . , 2015 ) , we are aware of only one report in which these interneurons were found to undergo experience-dependent plasticity and contribute to the expression of an adaptive behavior or memory ( Donato et al . , 2013 ) . Here we provide the first such example for striatal interneurons . We find that FSIs are a site of adaptive plasticity that drives circuit and behavioral hallmarks of habit . The habit-associated changes in FSI excitability appear distinct ( Figure 3 , Figure 3—figure supplement 1 ) from previously reported plasticity processes which included activity-induced changes in FSI-SPN synapses selectively at direct pathway SPNs ( Mathur et al . , 2013 ) and changes in firing rate related to the modulation of afterhyperolarization currents by parvalbumin expression levels ( Orduz et al . , 2013 ) . Further characterizing the plasticity mechanisms we find in habit represents an important area for future research as it may reveal a useful target for pharmacological modulation of FSI activity . The approach we took to reveal the microcircuit mechanisms for habit was to identify a potential source for the broad local DLS circuit reorganizations of SPN firing properties that strongly correlate with habit ( Figure 1A , B ) . To do this , we first examined how FSIs influenced striatal output using a pharmacological approach that inhibits excitatory synapses on striatal FSIs ( and also CINs ) . In brain slices from untrained mice , IEM-1460 treatment showed striking specificity in that it modulated all of the previously described ( O'Hare et al . , 2016 ) habit-predictive properties of evoked SPN firing ex vivo: gain of dSPN and iSPN responses ( Figure 1D , E ) , and the relative timing of firing between dSPNs and iSPNs ( Figure 1F ) . IEM-1460 also showed specificity in that it did not affect properties such as spike probability ( Figure 1—figure supplement 1 ) that are not predictive of habit . Unexpectedly , we found that the directionality by which FSIs modulated these properties was opposite to our original hypothesis: instead of the expected disinhibition of SPNs , silencing FSIs reduced SPN output ( Figure 1B–E ) . FSI inhibition also altered the timing of direct and indirect pathway neuron firing in a direction that opposed the habit circuit signature ( Figure 1B , F ) and closely resembled previous observations in lever-press trained , goal-directed mice ( O'Hare et al . , 2016 ) . This suggests that , in DLS , relative pathway timing is altered with habit formation but not with requisite goal-directed learning . Thus , the modest nature of the timing shift after pharmacological FSI blockade in untrained mice is likely due to a floor effect . Altogether , the observed effects of FSIs on SPNs lead to the prediction that an increase in FSI activity with habit formation would generate the evoked SPN properties that correlate with habit behavior ( Figure 1B ) ( O'Hare et al . , 2016 ) . Accordingly , in habitual mice , we found that FSI firing was increased , and under the same cortical afferent stimulation conditions that evoke habit-predictive SPN firing properties ( Figure 3F ) . This series of observations leads to a model of the striatal circuit basis for habitual behavior whereby habit formation is accompanied by a long-lasting increase in FSI excitability . In this setting , incoming cortical activity would be predicted to recruit more FSI activity that would in turn drive more firing of SPNs and shift their latencies such that direct pathway SPNs would tend to fire relatively sooner . While anatomical and electrophysiological studies have long supported that striatal FSIs are critical for striatal circuit function ( Gittis et al . , 2010; Koós and Tepper , 1999; Koos et al . , 2004; Mallet , 2005; Taverna et al . , 2007; Straub et al . , 2016; Szydlowski et al . , 2013 ) , an understanding of their specific behavioral contributions is much less developed . Prior in vivo studies have identified correlations of FSI activity with behaviors involving choice and reward-related actions ( Gage et al . , 2010; Schmitzer-Torbert and Redish , 2008 ) , while more recent correlations of FSI activity with head movement velocity suggest another mechanism ( Kim et al . , 2014 ) . In the present study , by chemogenetically inhibiting PV+ interneurons in vivo , we found that FSI activity in DLS is required for the expression of a learned habit ( Figure 4E ) ; an automated , reward-insensitive behavior quite different from behaviors previously studied . Previous pharmacological inactivation studies have demonstrated a role for DLS in habit expression ( Packard and McGaugh , 1996; Zapata et al . , 2010 ) , indicating that general disruption of DLS activity also impairs established habitual behavior . Interestingly , in the present study , chemogenetic inhibition of FSI activity drove an overall increase in projection neuron activity ( Figure 5F ) which suggests that reducing FSI activity specifically may impair habit expression differently than a general inactivation of the circuitry . While the disruption of habit by chemogenetically inhibiting FSIs supports a critical role for FSIs in this behavior , this experiment does not identify FSI plasticity as a mechanism for the expression of habit since artificially manipulating the activity of any cell that plays an otherwise critical role in the function of an implicated brain region might similarly disrupt behavior . Rather , in this study , a specific role for FSI plasticity as a mechanism for habit expression is indicated by the observations that these interneurons modulate those specific striatal output properties that correlate with habit ( Figure 1 and 2 ) and show long-lasting changes in excitability after habit learning ( Figure 3D , F ) . Using opto- and chemo-genetic manipulations , we further found that FSIs , which are GABAergic , enhance activity in subsets of SPNs both in the acute slice and in vivo . Although it is unclear what if any relationship exists between the SPN subpopulations identified ex vivo versus in vivo , there exist multiple intriguing parallels . In the acute slice , only activity of those SPNs which displayed burst-like , multi-action potential responses to single-pulse stimuli ( ‘high-firing’ SPNs ) was suppressed when FSIs were silenced ( Figures 1D , E and 2C ) . In vivo , the activity of SPNs showing the highest fractions of gamma-frequency spiking was suppressed , instead of disinhibited , when FSI activity was chemogenetically reduced ( Figure 5G ) . In both cases , the SPNs were distinguished by a higher propensity for burst-like firing patterns . It was further notable that the fraction of SPNs negatively modulated by reduced FSI activity was similar in both preparations ( 29% ex vivo compared to 26% in vivo ) . Conversely , we also found that less-active SPNs were not significantly modulated in the slice ( Figures 1D , E and 2C ) and SPNs with less gamma-frequency spiking were disinhibited in vivo when FSI activity was reduced ( Figure 5G ) . This finding is reminiscent of a previous in vivo report that SPNs with weaker responses to cortical microstimulation displayed the most marked disinhibition upon GABAA receptor blockade ( Mallet , 2005 ) . An important future direction will be to determine whether there are unique biological properties that distinguish the subset of SPNs whose activity is promoted , as opposed to inhibited , by FSIs . Although an activity-promoting effect of GABAergic FSIs may appear counterintuitive , previous computational ( Humphries et al . , 2009 ) and biological ( Bracci and Panzeri , 2006 ) studies describe such a phenomenon based in part on the ‘up’ and ‘down’ resting membrane potential states of SPNs that straddle the chloride reversal potential ( ECl- ) . While a voltage-dependent excitatory effect of GABA would not necessarily affect spike probability due to a concurrent decrease in membrane resistance and the disparity between ECl- and spike threshold , such an effect could boost the glutamate-driven depolarization of an SPN in its down state ( Humphries et al . , 2009; Bracci and Panzeri , 2006 ) . Although disynaptic interneuron microcircuitry is a more common mechanism for disinhibitory effects of interneurons in other brain regions ( Wolff et al . , 2014; Lovett-Barron et al . , 2012 ) , some of our observations such as the influence of FSIs on SPN initial latency to fire ( Figure 1F ) are not consistent with the time delay necessitated by a disynaptic microcircuitry . For this reason , we instead favor a monosynaptic mechanism whereby properties of SPN resting membrane potential and firing patterns interact to yield activity-promoting effects of FSIs on SPN subsets . Based on the previous observation that habit-predictive striatal output properties are relatively uniformly distributed when elicited by strong bulk stimulation of cortical afferents ( O'Hare et al . , 2016 ) , it became apparent that habit-related adaptations of DLS broadly augment the propagation of cortical excitation into the basal ganglia . To confer specificity for certain actions , additional circuit dynamics would ostensibly be required . We hypothesized that such specificity could arise from the activation of subsets of task-specific cortical neuron projections that would in turn activate task-specific SPNs ( Rothwell et al . , 2015; Carelli and West , 1991; Gremel et al . , 2016 ) . Indeed , recent evidence suggests that spatially-clustered SPN activity encodes information relevant to locomotor behavior ( Barbera et al . , 2016 ) . In habits , one possible mechanism then is that task-specific cortical commands drive ( Smith et al . , 2012 ) , or at least initiate ( Berke et al . , 2004 ) , high-frequency firing in a cluster/subset of SPNs that would then be preferentially excited by FSIs . Additionally , in such a mechanism , feed-forward inhibition of less-active SPNs ( Mallet , 2005 ) by FSIs might then serve as a selective filter to further enhance signal-to-noise ratio in corticostriatal transmission . One testable prediction of this model is that different behaviors would reveal different subsets of gamma-rich SPNs whose activity is promoted by FSIs . Lastly , it is notable that FSIs are also implicated in some pathological settings associated with compulsive behavior . For example , fewer striatal FSIs , as determined by parvalbumin-immunopositivity , have been observed in human brains from individuals with Tourette’s syndrome ( Kalanithi et al . , 2005 ) and mouse brains in a model of OCD-like behavior ( Burguière et al . , 2013 ) . OCD is highly comorbid in Tourette’s syndrome ( Sheppard et al . , 1999 ) and disrupted habit learning has been implicated in pathological compulsivity in a variety of settings ( Graybiel , 2008; Everitt and Robbins , 2005; Gerdeman et al . , 2003 ) . Interestingly , since both of the above studies defined FSIs by parvalbumin immunoreactivity , an intriguing alternative view of those results is that parvalbumin levels are below detection threshold but cell number is not necessarily reduced . Lower parvalbumin levels are associated with a hyperexcitable FSI phenotype ( Orduz et al . , 2013 ) , which is akin to the direction of FSI plasticity we associate with habit in the present study . Thus , the finding of increased FSI excitability as a plasticity mechanism driving habitual responding also yields new insights to the potential mechanistic relatedness of habit and compulsion . All experiments were carried out under approved animal protocols in accordance with Duke University Institutional Animal Care and Use Committee standards . Mice were 2–4 months of age , in C57Bl/6 genetic background , and were hemi-/heterozygous for all transgenes . Drd1a-tdTomato line 6 BAC transgenic mice were generated in our laboratory ( RRID: IMSR_JAX:016204 ) ( Ade et al . , 2011 ) . To optically inhibit PV+ interneurons , a mouse line expressing Cre under control of the Parvalbumin ( Pvalb ) promoter ( RRID:IMSR_JAX:012358 ) was crossed to the Ai35D line from Jackson Laboratory which Cre-dependently expressed Arch3 . 0-GFP ( RRID:IMSR_JAX:012735 ) . To target PV+ interneurons with Cre-dependent viral vectors , the Drd1a-tdTomato mouse line was crossed to the Pvalb-Cre line to produce experimental progeny hemizygous for Drd1a-tdTomato and heterozygous for Pvalb-Cre . For identification of PV+ neurons in 2PLSM calcium imaging experiments , the Pvalb-Cre mouse line was crossed to the Ai9 line ( RRID:IMSR_JAX:007909 ) which Cre-dependently expressed tdTomato . The CAG-FLEX-rev-hM4D:2a:GFP plasmid was provided by the Sternson laboratory at Janelia Farm ( Addgene #52536 ) . UNC Viral Vector Core packaged this plasmid into AAV 2/5 and also provided AAV2/5-EF1a-DIO-eYFP . All viral aliquots had titers above 1 × 1012 particles/mL . Stereotaxic injections were carried out on 2–3 month old Drd1a-tdTomato::Pvalb-Cre mice under isoflurane anesthesia ( 4% induction , 0 . 5–1 . 0% maintenance ) . Meloxicam ( 2 mg/kg ) was administered subcutaneously after anesthesia induction and prior to surgical procedures for postoperative pain relief . Small craniotomies were made over the injection sites and 1 . 0 μL virus was delivered bilaterally to dorsolateral striatum via a Nanoject II ( Drummond Scientific ) at a rate of 0 . 1 μL/min . The injection pipette was held in place for 5 min following injection and then slowly removed . Coordinates for all injections relative to bregma were as follows: A/P: +0 . 8 mm , M/L: ±2 . 7–2 . 8 mm , D/V: 3 . 2 mm . Mice were allowed a minimum of 14 days recovery before behavioral training . For experiments involving chemogenetic inhibition of FSIs specifically in DLS , mice showing no expression or poor targeting ( misses were medial to DLS ) were excluded from the study prior to behavioral analysis and data unblinding . Two AAV2/5-CAG-FLEX-rev-hM4D:2a:GFP-injected mice showed expression in only one hemisphere of DLS . These mice were included for behavioral analysis and behaved no differently from bilaterally-infected mice . We note that exclusion of these two subjects does not affect the statistical significance of the result . Prior to training , animals were restricted to 85–90% baseline weight to motivate learning . Lever presses were rewarded with sucrose-containing pellets ( Bio-serv , F05684 ) and grain-only pellets ( Bio-serv , F05934 ) were used as a sensory-specific control for satiety . Mice were trained in Med Associates operant chambers housed within light-resistant , sound-attenuating cabinets ( ENV-022MD ) . Lever presses and food cup entries were recorded by Med-PC-IV software . During RR reinforcement , pellets were delivered every X times on average for an RR-X schedule . RI reinforcement gave a 10% probability of reward every X seconds for an RI-X schedule . Following random reinforcement training , subjects underwent devaluation testing to measure habitual behavior as previously described ( O'Hare et al . , 2016 ) . When training schedule was a variable , experiments were performed with experimenter blind to training schedule . For electrophysiological assessment of FSI properties , acute brain slices were prepared 0–24 hr after the final training session . Mice were excluded from analysis if they did not display the behavior that was expected based on training schedule . Specifically , mice that were trained to be habitual ( random interval reinforcement ) yet showed sensitivity to outcome devaluation ( NDLPr <0 ) were excluded . Animals were anesthetized using 2 , 2 , 2-tribromoethanol and transcardially perfused with ice-cold N-Methyl-D-glucamine ( NMDG ) solution ( Ting et al . , 2014 ) . Brains were quickly removed and 300 μm thick parasaggital sections were cut in NMDG solution using a Leica VT1200S . For electrophysiological experiments , slices recovered at 32°C in NMDG solution for 10–12 min and were then transferred to room temperature HEPES-containing holding solution ( Ting et al . , 2014 ) where they remained for the rest of the experiment . Slices remained undisturbed in the HEPES holding solution for at least one hour prior to recording . For 2PLSM calcium imaging experiments , slices were allowed to recover for approximately 45 min in NMDG solution at room temperature . Slices were then transferred to room temperature HEPES holding solution ( Ting et al . , 2014 ) shortly before bulk-loading with fura-2 , AM . Cutting and holding solutions were calibrated to 305 ± 1 mOsm/L . ACSF was calibrated to 305 ± 1 mOsm/L for 2PLSM calcium imaging and 315 ± 2 mOsm/L for electrophysiological recordings with internal solutions at 295 mOsm/L . Solutions were pH 7 . 3–7 . 4 and were carbogenated to saturation at all times . For electrophysiological recordings , IEM-1460 was dissolved in deionized , distilled water at 100 mM and added to carbogenated ACSF for a final concentration of 50 μM . Picrotoxin was dissolved at 200 mM in DMSO and added to ACSF at 50 μM . For behavioral experiments , CNO was dissolved to 10 mg/mL in DMSO and diluted in sterile 0 . 9% saline solution to administer 5 mg/kg per subject with a maximum injection volume of 0 . 5 mL . For in vivo electrophysiological recordings , CNO and vehicle were administered on different days and in counterbalanced order . Data were acquired using an Axopatch 200B amplifier ( Molecular Devices ) and a Digidata 1440A digitizer ( Axon Instruments ) . Data were digitized at 10–20 kHz and low-pass filtered at 2 kHz . Borosilicate glass pipettes were pulled to 2–5 MΩ resistance . Slices were continuously perfused with carbogenated ACSF ( 124 mM NaCl , 4 . 5 mM KCl , 1 mM MgCl2·6 H2O , 26 mM NaHCO3 , 1 . 2 mM NaH2PO4 , 10 mM glucose , 4 mM CaCl2 ) at a temperature of 29–31°C . Synaptically-evoked action potential firing was monitored in dozens of direct and indirect pathway SPNs simultaneously as previously described ( O'Hare et al . , 2016 ) in acute brain slices prepared from untrained Drd1a-tdTomato hemizygous mice ( Ade et al . , 2011 ) aged 2–4 months . Detailed methods are included below . All experiments and data analyses were performed with experimenter blind to the experimental variable ( e . g . viral construct , training schedule ) . A priori sample sizes were established based on power analyses . Data exclusion criteria and decisions were made prior to data unblinding . F statistics were calculated using repeated measures analysis of variance . For within-cell comparisons , t statistics were calculated by paired , two-sided t-tests . Otherwise , unpaired , two-sided t-tests were used . For non-normal data sets , Mann-Whitney U tests were used . All r values were obtained using Pearson correlation analyses . Normality was measured using the Kolmogorov-Smirnoff test of the data against a hypothetical normal cumulative distribution function . Unless otherwise indicated ( e . g . Figure 2C , right panel ) , N values denote number of replicates considered biologically distinct for statistical measures ( Blainey et al . , 2014 ) ( see Table 1 for further detail ) . Technical replicates within a single biological sample were averaged to obtain a single value . For all statistical tests , confidence interval was set to α = 0 . 05 .	From biting fingernails to the daily commute , habits are sets of actions that can be completed almost without thinking and that are difficult to change or stop . Behavioral neuroscientists refer to habits as “stimulus-response” behaviors , and know that forming a new habit requires a region deep within the brain called the dorsolateral striatum . Indeed , in this region , the outgoing neurons – which make up 95% of the cells - respond differently to incoming signals in mice that have learned habits compared to non-habitual mice . However a question remained: what exactly was producing these differences ? O’Hare et al . have now found , unexpectedly , that the answer resides not in the 95% of outgoing neurons , but rather in a rare type of cell known as the fast-spiking interneuron . This cell is connected to many others and it appears to act like a conductor , orchestrating the previously identified changes in the output neurons . These findings were made using mice that had been trained to press a lever for a sugar pellet reward . Habit was measured by how long mice kept pressing even if they had just been allowed to eat their fill of pellets and the test lever was no longer dispensing pellets . Habitual mice continue to press the lever in this circumstance , while other mice do not . O’Hare et al . found that inactivating the “conductor” cell made the output neurons respond in the opposite way to how they normally respond in habitual mice . Further experiments showed that fast-spiking interneurons were also more easily activated in habitual mice . To test whether this putative “conductor” cell was necessary for habitual behaviors , a technique known as chemogenetics was used to turn down its activity in habitual mice . Indeed , reducing activity in the conductor cell blocked the habitual behavior . While some habits are a helpful and economical way to get through daily life , habits are also thought to be corrupted in a number of diseases such as neurodegenerative diseases , addictions and compulsions . Identifying this specific , yet rare , cell as a critical part of maintaining habits points out a new target to consider for therapies . Further work is needed before such treatments might become available to treat habit-related disorders; though O'Hare et al . are now taking steps in this direction by trying to work out how the fast-spiking interneuron changes its own activity when a habit is formed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Striatal fast-spiking interneurons selectively modulate circuit output and are required for habitual behavior
The formation of imprinted memories during a critical period is crucial for vital behaviors , including filial attachment . Yet , little is known about the underlying molecular mechanisms . Using a combination of behavior , pharmacology , in vivo surface sensing of translation ( SUnSET ) and DiOlistic labeling we found that , translational control by the eukaryotic translation initiation factor 2 alpha ( eIF2α ) bidirectionally regulates auditory but not visual imprinting and related changes in structural plasticity in chickens . Increasing phosphorylation of eIF2α ( p-eIF2α ) reduces translation rates and spine plasticity , and selectively impairs auditory imprinting . By contrast , inhibition of an eIF2α kinase or blocking the translational program controlled by p-eIF2α enhances auditory imprinting . Importantly , these manipulations are able to reopen the critical period . Thus , we have identified a translational control mechanism that selectively underlies auditory imprinting . Restoring translational control of eIF2α holds the promise to rejuvenate adult brain plasticity and restore learning and memory in a variety of cognitive disorders . Imprinting is a form of early learning where exposure to a stimulus becomes the triggering signal of a vital behavior ( Jin et al . , 2016; Horn , 2004 ) . A particular feature of imprinting is that it occurs exclusively within a short critical period ( CP ) ( Jin et al . , 2016; Bolhuis , 1991; Nevitt et al . , 1994 ) , when structural and functional changes take place ( Hensch , 2004 ) . Imprinting drives a vigorous following behavior in chickens , key for filial attachment ( Horn , 2004; Insel and Young , 2001 ) . This rather unique and precocious behavior is advantageous for investigating experience-driven activation of molecular pathways around birth ( Bredenkötter and Braun , 1997; Bock and Braun , 1999; McCabe et al . , 1982 , 1981 ) . Unders tanding the biological basis of imprinting can shed light on the mechanisms of learning in newborns and create new avenues to rejuvenate adult brain plasticity by reopening CPs . The formation of imprinted memories has been described across sensory modalities ( Nevitt et al . , 1994; McCabe et al . , 1982; Remy and Hobert , 2005; Bock et al . , 1997 ) . Interestingly , in chickens , auditory and visual imprinting relies on different brain structures . Imprinted sounds activate the mediorostral nidopallium/mesopallium ( MNM ) ( Bock and Braun , 1999; Wallhäusser and Scheich , 1987 ) , where neural responsiveness increases after training ( Bredenkötter and Braun , 1997 ) . In contrast , the intermediate medial mesopallium ( IMM , former IMHV ) ( Horn , 2004; McCabe et al . , 1982 ) is required for visual imprinting , where neural responses shift to favor the imprinted object ( Horn et al . , 2001 ) . While the brain circuits and neurophysiological changes have been uncovered ( Horn , 2004; Scheich , 1987 ) , much less is known about the molecular machinery linking experience and the formation of imprinted memories in each sensory modality . While imprinting requires protein synthesis ( Gibbs and Lecanuet , 1981 ) , little is known about the underlying translational control mechanisms . The translation of mRNA into protein occurs in three steps: initiation , elongation and termination and can be regulated through several signaling pathways ( Sonenberg and Hinnebusch , 2009 ) . Translation initiation is believed to be the rate-limiting step and a key target for translational control ( Sonenberg and Hinnebusch , 2009; Buffington et al . , 2014 ) . A major way in which translation initiation is regulated is by modulating the formation of the ternary complex via phosphorylation of the translation-initiation factor eIF2α . In rodents , protein synthesis controlled by phosphorylation of eIF2α is critically required for long-lasting forms of synaptic plasticity ( Costa-Mattioli et al . , 2007; Di Prisco et al . , 2014 ) as well as long-term memory storage in several systems ( Costa-Mattioli et al . , 2007 , 2005; Zhu et al . , 2011; Stern et al . , 2013; Ounallah-Saad et al . , 2014; Ma et al . , 2013 ) . Here we asked whether this central translational control mechanism plays a role in imprinting in newborn chickens and can be used to restore imprinting outside of the CP . Dark reared chickens were placed in a running wheel in front of an LCD screen and a speaker for training . Visual and auditory imprinting were tested separately 24 hr after training ( Figure 1a ) . Stimuli consisted of animated movies showing a virtual object , and artificial sounds synchronized to movements of the object in the screen ( Figure 1b , see supplementary materials ) . The imprinting was assessed by the preferential approach to the imprinted stimuli , either visual or auditory , compared to the approach to novel stimuli . The preference for imprinted stimuli is commonly used as an index of long-term memory storage ( Horn , 2004 ) . Individuals' preference was measured by calculating an index , where positive and negative values indicate preference for the imprinted or novel stimulus , respectively ( Figure 1c and d ) . This index accounts for fluctuations in baseline locomotion across trials , as described in the Method section . Consistent with previous studies ( Yamaguchi et al . , 2012 ) , chickens showed imprinting to either visual or auditory cues one day after hatching ( P1 ) but not after four days ( P4 ) ( Figure 1c and d ) , indicating that the CP for imprinting ends before P4 . 10 . 7554/eLife . 17197 . 003Figure 1 . Behavioral paradigm and the critical period for imprinting . ( a ) Schematic sequence of behavioral experiments . Dark-reared chicks were trained in a running wheel and tested the day after for visual and auditory imprinting . ( b ) During imprinting training , the chickens were presented with audiovisual stimulation . An animated object moved across the screen while a sound was presented every 3 s , coupled with pulsating movements of the object . ( c ) Auditory imprinting ( left ) was assessed by comparing the approaching behavior on the wheel to the imprinted sound or a novel sound . This procedure generated robust auditory imprinting when training was performed the day after hatching ( gray , n = 13 ) but was ineffective four days after hatching ( black , n = 12 ) ( right ) . ( d ) Visual imprinting ( left ) was assessed independently , by comparing the approaching behavior to the imprinted or a novel image . Similarly to auditory imprinting , visual imprinting was strong in P1 ( gray , n = 13 ) but absent in P4 ( black , n = 12 ) ( right ) . Plots show mean and SEM , * indicates p<0 . 05 from two-sample t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 00310 . 7554/eLife . 17197 . 004Figure 1—source data 1 . Preference indexes of trained chickens during ( P1 ) or after the critical period ( P4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 004 To assess whether a protein synthesis is enhanced after imprinting in both MNM and IMM we optimized an in vivo surface sensing of translation ( SUnSET ) protocol ( Schmidt et al . , 2009 ) for monitoring protein synthesis in vivo in these areas . Briefly , the antibiotic puromycin ( PMY ) incorporated into newly synthesized proteins can be detected through immunolabeling and used to monitor translation . Because brain tissue incorporates PMY more slowly compared to other tissues ( Flexner et al . , 1962 ) pilot experiments were conducted , showing that IP-injected PMY accessed the chicken’s brain within 3–4 hr . Thus PMY was injected 1 hr before a 2 hr training and samples were collected 4 hr after injection to capture training-induced translation . We found that imprinting training increased translation both in MNM and IMM ( Figure 2b , c ) after a 2 hr training session , compared to controls , which were running on the wheel but presented with an empty screen . To further estimate the time-window during which auditory and visual imprinting are sensitive to protein synthesis inhibition , we trained chickens for 1 or 2 hr on P1 . Two-hour ( Figure 3a , right panel ) training triggered robust auditory imprinting , which was blocked by the protein synthesis inhibitor clycloheximide ( CHX ) injected immediately after training ( Figure 3a , left panel ) . In contrast , one-hour training did not elicit significant auditory imprinting ( Figure 3a , right panel ) . Interestingly , the temporal dynamics of protein synthesis dependency of visual imprinting was different . While one-hour training triggered visual imprinting that was suppressed by CHX ( Figure 3b , left panel ) , visual imprinting after two-hour training was not blocked by post-training administration of CHX ( Figure 3b right panel ) . Consistent with the effect on behavior , CHX effectively blocked imprinting-triggered protein synthesis in both MNM and IMM areas ( Figure 3c , d ) . Taken together , our results show that both auditory and visual imprinting trigger new protein synthesis , which is required for both auditory and visual imprinting . 10 . 7554/eLife . 17197 . 005Figure 2 . Experience-dependent increase in translation assessed with SUnSET . ( a ) Temporally optimized SUnSET protocol to detect changes in translation in vivo ( left ) induced by the imprinting training . Schematic sagittal view of the chicken forebrain , showing the position of MNM and IMM ( right ) . ( b ) The auditory imprinting area MNM ( left ) exhibits increased puromycin incorporation ( green ) after imprinting training , compared with MNM samples of chickens running on the wheel but not presented with the imprinting object . S6 ( red ) marker was used to identify cells somas . ( c ) In IMM ( left ) translation rates were also increased in trained animals . Sample sizes: MNM untrained ( six chickens , 48 images at 10X , zoom 3X ) ; MNM trained ( six chickens , 48 images at 10X , zoom 3X ) ; IMM untrained ( six chickens , 48 images at 10X , zoom 3X ) ; IMM trained ( six chickens , 48 images at 10X , zoom 3X ) . Bar plots show mean and SEM; * indicates p<0 . 05 from unpaired Mann-Whitney test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 00510 . 7554/eLife . 17197 . 006Figure 2—source data 1 . SUnSET results from trained and untrained chickens . Puromycin signal was measured in MNM and IMM . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 00610 . 7554/eLife . 17197 . 007Figure 3 . Protein synthesis requirement in auditory and visual imprinting . ( a ) The protein-synthesis inhibitor cycloheximide ( CHX , n = 12 ) injected immediately after 1 hr training ( left ) had no effect on the auditory preference index compared to controls ( n = 15 ) and vehicle-injected ( n = 14 ) groups . In contrast , 2 hr training , which induced stronger preference to the imprinted sound , was blocked by CHX-treatment ( n = 11 ) compared to controls ( n = 13 ) and vehicle-injected group ( n = 9 ) . ( b ) Visual imprinting was already robust after 1 hr training ( left ) in controls ( n = 15 ) and chickens injected with vehicle ( n = 14 ) , and blocked by CHX-administration ( n = 12 ) . On the other hand , 2 hr training ( right ) also induced robust preference to the imprinted visual object in controls ( n = 13 ) and vehicle-injected chickens ( n = 9 ) but was not blocked by CHX administration ( n = 12 ) . Plots show mean and SEM , * indicates p<0 . 05 from two-ways ANOVA test , Bonferroni Post hoc test . ( c ) SUnSET protocol used to detect experience-dependent translation changes in MNM and IMM in the presence or absence of CHX . ( d , e ) Puromycin ( green ) incorporation is decreased in trained animals treated with CHX . S6 ( red ) was used to identify cell somas . Sample sizes: MNM trained ( five chickens , 40 images at 10X , zoom 3X ) ; MNM trained and CHX administration ( six chickens , 48 images at 10X , zoom 3X ) ; IMM trained ( five chickens , 39 images at 10X , zoom 3X ) ; IMM trained and CHX administration ( six chickens , 47 images at 10X , zoom 3X ) . Bar plots show mean and SEM; * indicates p<0 . 05 from unpaired Mann-Whitney test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 00710 . 7554/eLife . 17197 . 008Figure 3—source data 1 . Preference indexes and SUnSET results from control chickens and injected with cycloheximide . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 008 To investigate whether the translational program controlled by eIF2α is involved in imprinting , we first measured levels of phosphorylated eIF2α ( p-eIF2α ) in MNM and IMM after training P1 chickens . Intriguingly , training significantly decreased p-eIF2α in the auditory area MNM ( Figure 4a , left panel ) , but not in the visual area IMM ( Figure 4a , right panel ) . To examine whether a reduction in eIF2α phosphorylation is required for auditory imprinting we treated chickens before training with Sal003 , an inhibitor of the eIF2α phosphatase complexes ( McCamphill et al . , 2015 ) , which increases p-eIF2α levels ( Figure 4b and Figure 4—figure supplement 1 ) and decreases translation ( Figure 4—figure supplement 2 ) . Interestingly , increasing p-eIF2α with Sal003 prevented auditory imprinting , but had no effect on visual imprinting ( Figure 4c ) . These results indicate that decreasing p-eIF2α is only required for auditory imprinting . 10 . 7554/eLife . 17197 . 009Figure 4 . Translational control of auditory imprinting by eIF2α . ( a ) After 2 hr imprinting training , IMM and MNM were punched out for western blot analysis . The ratio of phosphorylated eIF2α ( p-eIF2α ) and non-phosphorylated eIF2α was measured in controls and after training in MNM ( left ) and IMM ( right ) brain tissue . Trained chicks ( n = 7 ) exhibited decreased eIF2α phosphorylation compared to the untrained ( n = 6 ) in MNM but not in IMM . Representative western blots are shown below each panel . * indicates p<0 . 05 from unpaired Mann-Whitney test . ( b ) Left , drugs injected for targeting the eIF2α pathway . Right , schematic effect of pharmacological manipulations on the eIF2α pathway . ( c ) Auditory ( left ) but not visual ( right ) imprinting is blocked by Sal003 injection ( n = 12 ) compared to controls injected with vehicle ( n = 9 ) . ( d ) Auditory imprinting ( left ) was enhanced in chickens injected with the PKR inhibitor PKRi ( n = 26 ) , compared to controls injected with saline vehicle ( n = 14 ) . On the other hand , PKRi ( n = 26 ) had no effect on visual imprinting ( right ) , compared to saline injection ( n = 14 ) . ( e ) Auditory imprinting ( left ) but not visual imprinting ( right ) was enhanced by ISRIB administration ( n = 11 ) compared to controls injected with vehicle ( n = 13 ) . Bar plots represent mean and SEM , * indicates p<0 . 05 from unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 00910 . 7554/eLife . 17197 . 010Figure 4—source data 1 . Western blots of p-eIF2α/ total eIF2α ratio and behavioral pharmacology after targeting the eIF2α pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 01010 . 7554/eLife . 17197 . 011Figure 4—figure supplement 1 . Sal003 increases eIF2α phosphorylation . Western blots for p-eIF2α and total eIF2α of brain samples obtained from chickens 2 hr after injecting Sal003 ( S ) or vehicle ( V ) . Sal003 treatment increased eIF2α phosphorylation . * indicates p<0 . 05 from Mann-Whitney U Test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 01110 . 7554/eLife . 17197 . 012Figure 4—figure supplement 2 . ISRIB and Sal003 injection modulate translation in vivo . Sal003 and ISRIB can bi-directionally regulate protein synthesis . Both in MNM and IMM Sal003 reduces puromycin ( green ) incorporation while ISRIB enhances it . S6 ( red ) was used to localize cell somas . Bar plots show mean and SEM; different letters inside bars indicate statistically significant differences ( p<0 . 05 ) between groups from Kruskal-Wallis test , Dunn’s multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 012 We next asked whether decreasing p-eIF2α would selectively enhance auditory imprinting . To this end , we first blocked the activity of the eIF2α kinase PKR , with a specific PKR inhibitor ( Zhu et al . , 2011 ) ( PKRi ) . PKRi-injected chickens showed significantly stronger auditory imprinting compared to controls ( Figure 4d ) . However , PKRi failed to affect visual imprinting ( Figure 4d ) . Given that average locomotion towards the computer screen in both treated and control conditions was similar , the changes induced by altering eIF2α phosphorylation cannot be attributed to changes in overall motor activity . To further demonstrate that auditory imprinting could be enhanced by reducing eIF2α-mediated translational control , chickens were injected with ISRIB , a compound that blocks the translational effects induced by p-eIF2α ( Sidrauski et al . , 2013 ) and increases translation ( Figure 4—figure supplement 2 ) . Consistent with the PKRi-experiments , injection of ISRIB immediately after training enhanced auditory imprinting ( Figure 4e ) but not visual imprinting . Hence , a reduction in p-eIF2α-mediated translational control enhances auditory but not visual imprinting . Plasticity in dendritic spines , the major site of excitatory inputs in neurons , is thought to be crucial during CPs ( Roberts et al . , 2010 ) and part of the cellular substrate of memory ( Lamprecht and LeDoux , 2004; Bourne and Harris , 2007; Nishiyama and Yasuda , 2015 ) . Given that ( a ) long-term remodeling of spines requires protein synthesis ( Nishiyama and Yasuda , 2015 ) and ( b ) translational control by p-eIF2α selectively regulates auditory imprinting , we next examined the role of this translational control mechanism in structural plasticity in imprinting-relevant brain regions . To measure changes in dendritic spine number and morphology after training ( Figure 5b ) , we used the sparse Diolistic labeling technique ( Figure 5a ) . The spines were classified ( by observers blind to treatment ) in stubby , filopodia , thin and mushroom ( Figure 5c ) , a method that is informative about the functionality and maturity of spines ( Bourne and Harris , 2007 ) and has been used in studies of learning-related structural plasticity ( Sanders et al . , 2012 ) . 10 . 7554/eLife . 17197 . 013Figure 5 . Translational control of experience-dependent structural plasticity . ( a ) Example diolistic labeling of a type I IMM neuron ( 63X ) , used to analyze the number and the shape of dendritic spines in MNM and IMM after training . ( b ) Representative confocal images of dendritic segments of IMM cells from untrained animals ( 63X , zoom 3X ) . ( c ) Schematic length ( L ) and shape criteria used for spine classification . ( d ) Trained chickens showed an increased number of mushroom spines ( red ) and a decrease in thin spines ( blue ) in MNM . The increase in mushroom spines induced by training was blocked by Sal003 . Samples size: untrained ( four chickens , 12 cells , 45 dendrites ) ; imprinted ( five chickens , 15 cells , 50 dendrites ) ; Sal003 ( five chickens , 15 cells , 55 dendrites ) . ( e ) Trained chickens showed an increase in mushroom spines ( red ) and a decrease in thin spines ( blue ) in IMM . In contrast to the changes in MNM , the increase in mushroom spines was not blocked by Sal003 . Sample sizes: untrained ( four chickens , 11 cells , 35 dendrites ) ; imprinted ( four chickens , 10 cells , 33 dendrites ) ; Sal003 ( five chickens , 16 cells , 48 dendrites ) . Total number of spines did not show significant differences across groups in either region . Bar plots show mean and SEM; different letters inside bars indicate statistically significant differences ( p<0 . 05 ) between groups from Kruskal-Wallis test , Dunn’s multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 01310 . 7554/eLife . 17197 . 014Figure 5—source data 1 . Dendritic spines numbers in MNM and IMM of untrained , trained and Sal003-treated chickens . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 014 While training failed to affect the total number of spines ( Figure 5d–e ) , it significantly increased the number of mushroom spines and decreased the number of thin spines in MNM ( Figure 5d ) and IMM ( Figure 5e ) , compared to control animals with experience on the running wheel but not subject to audiovisual training . We next examined whether blocking eIF2α dephosphorization with Sal003 prevents training-induced changes in structural plasticity . Remarkably , Sal003 administration blocked the training-induced increase in the number of mushroom spines only in MNM ( Figure 5d , e ) . These results indicate that eIF2α phosphorylation not only controls the imprinting behavior during the CP but also structural plasticity , a potential cellular substrate of memory storage ( Lamprecht and LeDoux , 2004; Nishiyama and Yasuda , 2015 ) in a key forebrain area involved in auditory imprinting . Identifying the mechanisms that open the CPs could lead to novel therapeutic opportunities for a variety of cognitive disorders ( Hensch , 2004 ) . Given that ( a ) behavioral training decreases p-eIF2α ( Figure 4a ) , ( b ) blocking p-eIF2α-mediated translation enhances auditory imprinting elicited by weak-training protocol ( Figure 4d–e ) and ( c ) Sal003-mediated increase in p-eIF2α blocks auditory imprinting , we wondered whether the drugs enhancing imprinting during the CP ( PKRi and ISRIB ) would restore imprinting outside the CP ( Figure 6a ) . Remarkably , treatment with either PKRi or ISRIB ( Figure 6a ) on P4 selectively re-opened the CP for auditory imprinting ( Figure 6b ) , again without affecting visual imprinting ( Figure 6c ) . Hence , by promoting brain plasticity , the reduction of p-eIF2α-mediated translational control enhances auditory imprinting . 10 . 7554/eLife . 17197 . 015Figure 6 . Reopening the critical period for visual and auditory imprinting through eIF2α . ( a ) Chickens were trained 4 days after hatching ( P4 ) and tested 24 hr after training . To target translational control by eIF2α , chickens were injected with PKRi or ISRIB . ( b ) Controls injected with vehicle ( n = 12 ) did not show auditory imprinting at P4 but the critical period in animals treated with PKRi ( n = 13 ) or ISRIB ( n = 13 ) was reopened . ( c ) Visual imprinting was not restored in chickens injected with PKRi ( n = 13 ) and ISRIB ( n = 13 ) or injected with vehicle ( n = 12 ) . * indicates p<0 . 05 from two-ways ANOVA test , Bonferroni Post hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 01510 . 7554/eLife . 17197 . 016Figure 6—source data 1 . Preference indexes of animals trained in P4 and injected with PKRi , ISRIB or vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 016 Imprinting allows newborns to adjust behavior in response to relevant sensory experience , immediately after birth ( Horn , 2004; Bolhuis , 1991 ) . Despite having been studied for decades , the mechanism mediating the formation of imprinted memories remains elusive . Here we showed that , although visual and auditory imprinting require newly synthesized proteins , eIF2α-mediated translational control bidirectionally regulates auditory but not visual imprinting and related changes in structural plasticity . Remarkably , targeting this translational control mechanism pharmacologically recovers auditory imprinting after the closing of the critical period . Critical periods in the auditory system have been widely studied across species ( Scheich , 1987; Riebel et al . , 2002; Yang et al . , 2012; Insanally et al . , 2009 ) . Yet the mechanisms engaged during the CP for auditory imprinting have not been elucidated . One major limitation has been the design of stringent experimental approaches that control for social experience and innate biases , while achieving robust auditory imprinted memories ( Van and Bolhuis , 1991 ) . We aimed to address these concerns by: ( 1 ) raising chickens in darkness and constraining social interaction , ( 2 ) imprinting chickens to more than one type of object and sound , and ( 3 ) increasing the length of training compared to other studies ( Wallhäusser and Scheich , 1987; Van and Bolhuis , 1991 ) to achieve significant memory retention longer than 24 hr after training . These improvements , in addition to the novel custom-made audiovisual animation used for training , provided a stronger experimental design for assessing auditory and visual imprinting . Several lines of evidence support the modality-specific role of eIF2α . First , training decreased phosphorylation of eIF2α in the auditory-imprinting relevant area MNM , but not in IMM ( Figure 4a ) . Second , pharmacologically increasing eIF2α phosphorylation with Sal003 selectively disrupted auditory imprinting ( Figure 4c ) . Third , inhibiting eIF2α phosphorylation with PKRi or directly blocking p-eIF2α-mediated translational control with ISRIB , enhanced auditory imprinting after weak training ( Figure 4d–e ) . Interestingly , although Sal003 and ISRIB altered protein synthesis in IMM , these manipulations had no detectable effect on the formation of visual memories . The reason why eIF2α is not involved in visual imprinting is not yet understood . It is possible that expression of eIF2α kinase or phosphatase complexes differs between visual and auditory areas , or that upstream signaling pathways fail to engage eIF2α dephosphorylation . It would be interesting to test whether other tasks involving memory formation , such as one-trial avoidance learning ( Atkinson et al . , 2008 ) , also require translational control by eIF2α . An appealing idea is that other translational control pathways such those controlled by the mechanistic target of rapamycin mTORC1 ( Sonenberg and Hinnebusch , 2009 ) mediate the formation of visual memories . While previous studies in adult rodents suggest that eIF2α-mediated translation regulates the two major forms of synaptic plasticity ( Costa-Mattioli et al . , 2007; Zhu et al . , 2011; Stern et al . , 2013 ) , here we report for the first time that experience-dependent structural plasticity of dendritic spines requires eIF2a dephosphorylation . Furthermore , and consistent with our behavioral results , eIF2α-mediated translation exclusively regulated spine remodeling in the auditory but not in the visual area . This result is particularly important since structural plasticity is crucial during CPs ( Roberts et al . , 2010; Mataga et al . , 2004 ) but the underlying molecular mechanisms were unknown . Different forms of structural plasticity have been linked to memory , including spine turnover and morphological changes of preexisting spines ( Lamprecht and LeDoux , 2004 ) . In this case , the structural plasticity found in IMM and MNM could be consistent with potentiation and enlargement of specific dendritic spines , favoring the detection of imprinted stimuli . While we did not observe changes in spine density , the increase in mushroom spines and decrease in thin spines may suggest coordinated structural plasticity as previously reported in hippocampal slices ( Bourne and Harris , 2011 ) . Thus , our results shed light on the biological basis of experience-dependent spine remodeling and uncovered eIF2α as a major player in spine remodeling . Another interesting question is whether translational control by eIF2α in glial cells affects imprinting . Glutamate application induces a transient increase of eIF2α phosphorylation in glial cells in vitro ( Flores-Méndez et al . , 2013 ) . This effect has been linked to glutamate removal from the synaptic cleft by glial cells ( Flores-Méndez et al . , 2013 ) . However , its contribution to memory formation in vivo remains untested . In future studies , it will be important to dissect the role of p-eIF2a in memory formation at the cellular level . Behavior is shaped during sensitive periods in early postnatal life , characterized by epochs of heightened brain plasticity ( Hensch , 2004; Nabel and Morishita , 2013 ) . Reactivating such plasticity in the adult brain has the potential to rehabilitate brain function after CPs are closed ( Hensch , 2004; Nabel and Morishita , 2013; Hübener and Bonhoeffer , 2014 ) . This has been successfully achieved in the visual cortex of rodents through direct manipulation of inhibitory synaptic transmission , either pharmacologically ( Hensch et al . , 1998 ) or through transplantation of embryonic inhibitory neurons ( Davis et al . , 2015 ) . Moreover , in mice and humans , releasing ‘epigenetic brakes' , could reopen auditory CPs ( Yang et al . , 2012; Gervain et al . , 2013 ) . Our results uncover a translational control mechanism as a novel target for reopening CPs . Indeed , two different strategies , either blocking p-eIF2α-mediated translation or inhibiting the upstream kinase PKR , enabled chickens to imprint to sounds after the end of the CP , suggesting that blocking p-eIF2α-mediated translation control enhances CP-mediated plasticity . A recent report shows that reducing p-eIF2α-mediated translational control in the VTA can convert adult into adolescent mice with respect to their vulnerability to cocaine-induced changes in synaptic strength and behavior ( Huang et al . , 2016 ) . Based on these results and the evolutionarily conserved nature of this process , we speculate that reopening CPs through blockade of eIF2α-mediated translational control could be used to recover plasticity in the mature brain and treat cognitive dysfunctions . We used newly hatched chicks of both sexes from the White Leghorn strain Gallus gallus domesticus ( Charles River supplier ) . Fertilized eggs ( embryonic ages E14-17 ) were obtained and subsequently incubated in darkness at 37–38°C under controlled humidity ( Grumbach , compact S84 ) . Upon hatching , chickens were transferred to individual compartments of a brooder maintained at 37–38°C ( Brinsea , TLC-5 ) , where they remained in darkness until each experiment . Water and food was provided . It has been shown that chickens are able to eat and drink water in the dark and that this housing does not impact visual acuity or locomotion , compared to chickens reared under light conditions ( Yamaguchi et al . , 2012 ) . These experiments were approved by the institutional animal care committee ( IACUC ) at Albert Einstein College of Medicine ( protocol 20140910 ) . Training sessions and tests were performed in a sound proof chamber ( IAC acoustics ) at 37°C in the dark , except for the light coming from the monitor . All experiments and drug manipulations were performed blind to treatment . On the training day , each chicken was placed under white light for 30 min . This priming procedure has been extensively used in visual imprinting ( Bolhuis et al . , 2000; Nakamori et al . , 2010 ) . After priming , chickens were placed in a running wheel ( internal diameter = 18 cm ) in front of a computer monitor ( ACER LCD , 17'' ) . Magnets mounted on the wheel ( Gibbs and Lecanuet , 1981 ) allowed the precise measurement of the approaching behavior by a counter ( Med Associates , DIG-700G , DIG-726 ) . Each magnet count generated a TTL signal , whose timing was stored in a computer for offline analysis . Visual stimuli consisted of custom-made animations ( Blender , http://www . blender . org/ ) of either a blue or red rectangular prism coupled to a sound . Both figures had exactly the same volume and followed the same rotation and movement across a virtual room ( Video 1 and Video 2 ) . This method made it possible to synthesize arbitrary movement patterns while controlling luminosity , color and shape . Objects changed shape ( expansion and contraction ) synchronously with sound . Two different sounds were synthesized using Audacity software ( Audacity 2 . 1 . 0 ) . The frequency range for both sounds was 0–3 KHz . Sound one consisted in frequency steps and sound two was composed of frequency sweeps ( see supplementary material ) . Each sound was played 12 times during a minute , every 3 s . The start of each animation was commanded by software written in Matlab , which was interfaced to Med Associates equipment through a USB DAQ card ( National instruments USB-6008 ) . 10 . 7554/eLife . 17197 . 017Video 1 . Stimulus A presented to chickens . This animation was played on a screen during training . For auditory and visual imprinting tests only the auditory or the visual component was presented . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 01710 . 7554/eLife . 17197 . 018Video 2 . Stimulus B presented to chickens . This animation was played on a screen during training . For auditory and visual imprinting tests only the auditory or the visual component was presented . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 018 Audiovisual training stimuli were presented in 4 min bouts followed by 1 min of silence and darkness . If the chicken did not move the wheel during the first half hour of exposure , the experiment was interrupted and not included in the sample . Training length varied from 0 to 120 min , depending on the protocol . To investigate long-lasting effects on imprinting we tested chickens the day after training . Visual and auditory imprinting were tested independently in a sequential test , where the novel and imprinted stimuli are presented in alternation . While other studies have used a simultaneous choice test ( Yamaguchi et al . , 2012 ) , the sequential test allowed us to randomize stimulus presentation , measure baseline locomotion and assess the response to novel and imprinted ( Video 3 ) stimuli independently . Each test included 5 presentations of the imprinted stimulus and 5 presentations of the novel stimulus . The duration of each presentation was 1 min . Baseline locomotion was measured during 30 s between trials . Imprinted and novel stimuli were alternated over five consecutive blocks . The first stimulus that started the sequence was picked randomly . Although this method differs from previous reports where fixed sequences were used ( Bolhuis et al . , 2000; Town and McCabe , 2011; McCabe and Horn , 1988 ) , randomization prevents biases and motivation changes over time emerging from fixed sequences . 10 . 7554/eLife . 17197 . 019Video 3 . Chicken imprinted to stimulus B approaching the screen . This approach behavior was quantified during the presentation of imprinted or novel stimuli to compute a preference index . DOI: http://dx . doi . org/10 . 7554/eLife . 17197 . 019 Previous studies have used different criteria and indexes to quantify the strength of imprinting . Such quantifications have included differences in time spent in the proximity of the imprinted object ( Yamaguchi et al . , 2012 ) , differences in locomotion toward the imprinted and novel stimulus ( Bolhuis et al . , 2000 ) , differences in locomotion during the presentation of imprinted and novel objects and the absence of a stimulus ( Maekawa et al . , 2006 ) , and number of chickens within a group selecting the imprinted stimulus over several trials ( Wallhäusser and Scheich , 1987 ) . In this study , we normalized differences between locomotion to novel and imprinted stimuli by the average baseline locomotion in the wheel when no stimulus was presented . Therefore , to assess imprinting , we calculated a preference index ( PI ) , PI = ∑ ( ImprintedSTL - NovelSTL ) / BaselineA where STL indicates stimulus-triggered locomotion either during the presentation of the imprinted stimulus ( ImprintedSTL ) or presentation of the novel stimulus ( NovelSTL ) , and baselineA refers to the average baseline locomotion across the experiment . An advantage of this quantification over previous methods is that: ( 1 ) it takes into account the fluctuations in basal locomotion before each stimulus presentation , and ( 2 ) it weights differences in approaching behavior by average locomotion . The sensitive period for filial imprinting has been reported to close within 3–4 days after hatching ( Yamaguchi et al . , 2012 ) . To ensure the training and preference tests captured this sensitivity , the ability of chickens to develop a preference to visual and auditory stimuli within the first 4 days after hatching was measured immediately and 24 hr after training . We optimized previously reported in vivo SUnSET protocols in muscle fibers ( Goodman and Hornberger , 2013; Goodman et al . , 2011 ) for monitoring protein synthesis in the chick brain . It has been shown that PMY injected intravenously takes 2–4 hr to be incorporated into the brain ( Flexner et al . , 1962 ) . This contrasts with the fast incorporation ( approximately 30 min ) into muscle ( Goodman and Hornberger , 2013; Goodman et al . , 2011 ) and other organs ( Flexner et al . , 1962 ) . In pilot experiments , we determined that 3–4 hr after injecting a low dose of PMY ( MP Biomedicals , 0 . 04 mg/g , diluted in distilled H20 , IP ) was the optimal time period for detecting the incorporation of PMY in newly synthesized proteins . This information was used to adjust the timing of PMY injection in our behavioral pharmacology experiments . To simultaneously assess experience-dependent translation across sensory modalities and brain regions , in the same animal , we identified a training schedule that reliably triggered auditory and visual imprinting . Since 2 hr but not 1 hr training ( Figure 3a , b ) triggered both auditory and visual imprinting , we used the former schedule . Four hours after PMY injection chicks were decapitated and brains were rapidly ( 2–3 min ) placed in cold PFA ( 4% ) overnight at 4°C . A vibratome ( Leica VT 1000S ) was used for making 100 µ sagittal cross sections . After three 10 min washing with PBS , samples were incubated overnight at 4°C in a solution containing antibodies against PMY ( EMD Millipore , cat# MABE343 , RRID:AB_2566826 ) and S6 ( Cell signaling , cat# 2217 , RRID:AB_331355 ) to identify cell somas . Samples were washed in PBS ( three 10 min wash ) and placed for 1 . 5 hr in a solution containing Alexa-488 ( Invitrogen , cat# A21202 , RRID:AB_2535788 ) and Alexa-568 ( Invitrogen , cat# A10042 , RRID:AB_2534017 ) against the primary antibody host species . After washing again 3 times for 10 min in PBS , samples were covered with Prolong Gold mounting media ( Molecular probes , cat# P36935 ) . A confocal microscope ( Zeiss LSM 510 Meta Duo V2 ) was used to collect images from IMM and MNM ( 10X , zoom 3 ) . All images were taken blind to the experimental groups . IMM is located 2 . 5 mm from the dorsal surface of the brain and 0 . 5–1 mm from the caudal edge of the forebrain , limited below and laterally by the lateral ventricle . MNM is located 0 . 5–1 mm lateral from the midline , 3 mm from the dorsal surface of the brain and 5 mm from the caudal edge of the forebrain , below the lateral pallial lamina that separates the hyperpallium and mesopallium ( Puelles et al . , 2007 ) . All compared samples were processed the same day , using the same protocol , and images were taken with equal microscope settings . Control animals were housed in the same conditions as trained animals but presented with an empty screen . Images were analyzed using ImageJ software ( NIH , 1 . 50i ) . Threshold was adjusted by the S6 signal to select cell somas . PMY signal was detected using the selection created for the S6 channel . To compare across groups all measures were normalized to the average intensity of the control group . To investigate the involvement of protein synthesis in long-term memory formation during imprinting we injected cycloheximide ( Tocris , CHX , 1 mg/kg , IP ) , diluted in 0 . 1% DMSO and saline , immediately after training . Since 1 hr training was enough to generate visual ( Figure 3a ) but not auditory imprinting ( Figure 3b ) , we injected CHX immediately after 1 hr and 2 hr training , and tested the effect on imprinting 24 hr later for each sensory modality , independently . We used the specific blocker of eIF2α phosphatases Sal003 ( Sigma Aldrich , 0 . 2 mg/Kg , diluted in 0 . 1% DMSO and 0 . 9% Saline , IP ) to test whether a reduction in eIF2α phosphorylation is required for imprinting . We used 2 hr training for this experiment , which reliably triggered strong visual and auditory imprinting , and injected Sal003 before training to ensure translation was inhibited during and immediately after training . To specifically enhance the formation of imprinted memories by reducing eIF2α–mediated translational control , we conducted two independent manipulations: animals were injected immediately after training with either the specific inhibitor of the eIF2a kinase PKR ( PKRi; EMD Millipore , 0 . 1 mg/Kg , diluted in 0 . 1% DMSO and 0 . 9% saline ) or ISRIB ( Sigma Aldrich , 2 . 5 mg/Kg , diluted in 50% DMSO and 50% saline , IP ) , which blocks the translational effect induced by p-eIF2α . To avoid a ceiling effect masking the enhancement of imprinting , we used 1 hr training ( weak training ) and tested preference 24 hr after training . Lysates of IMM and MNM ( anatomical boundaries described above ) were obtained from brain tissue , punched out from 0 . 75- to 1mm-thick sagittal brain slices collected from imprinted and control animals . We used antibodies against eIF2α ( Cell Signaling Cat #9722 , RRID:AB_2230924 ) , p-eIF2α ( Ser51 ) ( Cell Signaling Cat #9721 , RRID:AB_330951 ) , following standard protocols described before ( Costa-Mattioli et al . , 2007 ) . Control tissue samples were obtained from chickens that ran on the wheel towards a screen displaying only a static image of an empty room , as shown in Figure 1b ( left panel ) . We tested whether reducing p-eIF2α by PKRi and ISRIB administration could reopen the CP for each sensory modality using 2 hr training on P4 . Since injecting PKRi and ISRIB immediately after 1 hr training did not have an effect on visual imprinting , we injected PKRi ( Stern et al . , 2013; Ingrand et al . , 2007 ) ( 0 . 1 mg/Kg , IP ) and ISRIB ( 2 . 5 mg/kg ) before training to control whether the lack of effect on visual imprinting was due to the time of the injection . Imprinting was assessed 24 hr after training as described above . Brains were rapidly dissected ( in 2–3 min ) and placed in paraformaldehyde ( 4% ) for 1 hr , then transferred to the phosphate buffer solution . A vibratome ( Leica VT 1000S ) was used for making 200 uM slices . Tungsten beads coated with lipophilic dye ( DiI ) were delivered to each slice using a modified gene gun ( Gan et al . , 2000 ) . The dye was allowed to spread overnight . The next day , each slice was mounted using ProLong Gold mounting media . A confocal microscope ( Zeiss LSM 510 Meta Duo V2 ) was used to collect Z-stacks ( 63X , zoom 3 ) from areas of interest containing labeled dendritic branches . Images of secondary branches , within 50–75 µm from the soma , were used for spine analysis . Dendritic spines were counted blind to experimental groups using Image J software ( Version 1 . 50a ) . A multicolored lookup table ( Fire ) was used to reliably visualize individual spines . Two 10 μm segments were marked randomly along each secondary dendritic branch . Spines along each of the two segments were counted by a blind experimenter . The spines ‘head width , presence of neck and overall length were used for classifying them in filopodia , stubby , thin , or mushroom , using published criteria ( Bourne and Harris , 2007; Sanders et al . , 2012; Chakravarthy et al . , 2006 ) . Briefly , spines without clear head and neck , and shorter than 1 μm , were categorized as stubby . Spines longer than 1 μm were classified as mushroom or thin , depending on whether a head and neck were observed . Protrusions longer than 2 μm were categorized as filopodia . To investigate if eIF2α was required for structural plasticity , we injected chickens with Sal003 ( i . p . , 0 . 2 mg/kg ) and trained them for 2 hr . The day after the training , we labeled dendritic arbors and assessed dendritic spines , as described above . Statistical analyses were performed using SigmaPlot ( Systat Software ) . Data distribution normality was assessed using the Shapiro-Wilk and F-test to evaluate the differences of variances . When variances were significantly different the Welch’s correction was used . Statistics were based on the two-sided Student’s t test , or the two-way ANOVA and Bonferroni post-hoc test for multiple comparisons of normally distributed samples . Otherwise the Mann-Whitney or the Kruskal-Wallis and Dunn’s multiple comparisons tests were used . Within-group variation is indicated by standard errors of the mean of each distribution , which are depicted in the graphs as error bars . p<0 . 05 was considered significant .	Shortly after hatching , a chick recognizes the sight and sound of its mother and follows her around . This requires a type of learning called imprinting , which only occurs during a short period of time in young life known as the “critical period” . This process has been reported in a variety of birds and other animals where long-term memory formed during a critical period guides vital behaviors . In order to form imprinted memories , neurons must produce new proteins . However , it is not clear how new experiences trigger the production of these proteins during imprinting . Unraveling such mechanisms may help us to develop drugs that can recover plasticity in the adult brain , which could help individuals with brain injuries relearn skills after critical periods are closed . It is possible to imprint newly hatched chicks to arbitrary sounds and visual stimuli by placing the chicks in running wheels and exposing them to repeated noises and videos . Later on , the chicks respond to these stimuli by running towards the screen , mimicking how they would naturally follow their mother . This system allows researchers to measure imprinting in a carefully controlled laboratory setting . A protein called elF2α plays a major role in regulating the production of new proteins and has been shown to be required for the formation of long-term memories in adult rodents . Batista et al . found that elF2α is required to imprint newly hatched chicks to sound . During the critical period , this factor mediates an increase in “memory-spines” , which are small bumps on neurons that are thought to be involved in memory storage . On the other hand , elF2α was not required to imprint newly hatched chicks to visual stimuli , suggesting that there are different pathways involved in regulating imprinting to different senses . Batista et al . also demonstrate that using drugs to increase the activity of eIF2α in older chicks could allow these chicks to be imprinted to new sounds . The next steps following on from this work are to identify proteins that eIF2α regulates to form memories , and to find out why eIF2α is only required to imprint sounds . Future research will investigate the mechanisms that control visual imprinting and how it differs from imprinting to sounds .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Translational control of auditory imprinting and structural plasticity by eIF2α
The RIG-I-like receptors RIG-I , LGP2 , and MDA5 initiate an antiviral response that includes production of type I interferons ( IFNs ) . The nature of the RNAs that trigger MDA5 activation in infected cells remains unclear . Here , we purify and characterise LGP2/RNA complexes from cells infected with encephalomyocarditis virus ( EMCV ) , a picornavirus detected by MDA5 and LGP2 but not RIG-I . We show that those complexes contain RNA that is highly enriched for MDA5-stimulatory activity and for a specific sequence corresponding to the L region of the EMCV antisense RNA . Synthesis of this sequence by in vitro transcription is sufficient to generate an MDA5 stimulatory RNA . Conversely , genomic deletion of the L region in EMCV generates viruses that are less potent at stimulating MDA5-dependent IFN production . Thus , the L region antisense RNA of EMCV is a key determinant of innate immunity to the virus and represents an RNA that activates MDA5 in virally-infected cells . Viral infection in mammals triggers a rapid innate immune response involving the production of antiviral proteins and proinflammatory mediators , prominent among which are the type I interferons ( IFN-α/β; hereafter , IFN ) ( Stetson and Medzhitov , 2006; Takaoka and Yanai , 2006 ) . IFNs are secreted cytokines that act on all nucleated cells to induce the transcription of more than 300 IFN-stimulated genes , whose products collectively limit virus replication and spread ( Haller et al . , 2007; Schoggins and Rice , 2011 ) . IFN gene transcription is triggered by the activation of pattern recognition receptors that detect viral invasion . These receptors include members of the RIG-I-like receptor ( RLR ) family ( Goubau et al . , 2013 ) , a sub-group of DExD/H-box helicases that surveys the cytosol for the presence of atypical RNAs associated with viral infection . The RLR family comprises three members: RIG-I ( retinoic acid-induced gene I ) , MDA5 ( melanoma differentiation-associated gene 5 ) , and LGP2 ( laboratory of genetics and physiology 2 ) . Binding of agonistic RNA by RIG-I or MDA5 triggers a signalling cascade that leads to the activation of transcription factors , including several of the IFN regulatory factors ( IRFs ) , which translocate into the nucleus to induce expression of IFN and other genes . LGP2 lacks a signalling domain and cannot act in the same manner . Rather , LGP2 is thought to potentiate MDA5-dependent IFN induction although the exact mechanism by which this occurs remains unclear ( Satoh et al . , 2010; Bruns et al . , 2012; Childs et al . , 2013 ) . MDA5 and RIG-I are activated by different RNA viruses ( Kato et al . , 2006 ) . RIG-I is non-redundant for detection of influenza or Sendai virus but dispensable for responses to picornaviruses whereas the opposite is the case for MDA5 ( Kato et al . , 2006; 2008 ) . The basis for these differences stems from the ability of RIG-I and MDA5 to recognize different RNA patterns . RIG-I binds to and is activated by base-paired RNA containing a 5′ triphosphate extremity ( Hornung et al . , 2006; Pichlmair et al . , 2006; Schlee et al . , 2009; Schmidt et al . , 2009 ) as found in influenza and Sendai virus genomes ( Baum et al . , 2010; Rehwinkel et al . , 2010; Weber et al . , 2013 ) . In contrast , MDA5 recognizes RNA independently of 5′ phosphates and can be activated in cells by transfection of synthetic poly I:C or the double-stranded ( ds ) RNA genome segments of reovirus ( Kato et al . , 2006 , 2008 ) . Based on such observations , it has been concluded that MDA5 detects long dsRNA generated during virus infection . However , natural MDA5 agonists derived from infected cells remain poorly characterised . Some studies have suggested that they might correspond to high molecular weight RNA complexes containing double and single-stranded ( ss ) regions ( Pichlmair et al . , 2009 ) . Other reports have proposed that relevant MDA5 agonists are specific viral replication intermediates ( Feng et al . , 2012; Triantafilou et al . , 2012 ) . Finally , it has been suggested that MDA5 might also be activated by RNA products of RNAseL cleavage ( Malathi et al . , 2010; Luthra et al . , 2011 ) or by mRNA bearing unmethylated cap structures ( Züst et al . , 2011 ) . Altogether , these studies provide a glimpse into the possible nature of MDA5 agonists in virally-infected cells but fall short of identifying them with precision . Immunoprecipitation of RIG-I from cells infected with influenza A virus ( IAV ) or Sendai virus has allowed the identification of physiological RIG-I agonists ( Baum et al . , 2010; Rehwinkel et al . , 2010 ) . In this study , we use an analogous approach to investigate the nature of MDA5 agonists in cells infected with encephalomyocarditis virus ( EMCV ) , a member of the Cardiovirus genus of picornaviruses . Picornaviruses are single-stranded positive-strand ( sense ) RNA viruses that replicate in infected cells via a negative-strand ( antisense ) intermediate . We purify RNA directly from complexes obtained by immunoprecipitation of LGP2 and show that this method enriches for MDA5 stimulatory RNA corresponding to a portion of the EMCV antisense RNA . Deletion of the region encoding this antisense RNA generates viruses that produce less stimulatory RNA and are less potent at inducing IFN in infected cells or mice . Conversely , in vitro synthesis of the same sequence generates an MDA5 agonistic RNA . Thus , a discrete region of the EMCV negative-strand RNA acts as a physiologically-relevant MDA5 agonist in infected cells . To confirm that both MDA5 and LGP2 are required for IFN responses to EMCV ( Kato et al . , 2006; Satoh et al . , 2010 ) , we used mouse embryonic fibroblasts ( MEFs ) carrying null mutant alleles of the genes Ifih1 and Dhx58 encoding MDA5 and LGP2 , respectively . We infected Ifih1−/− ( MDA5-deficient ) , Ifih1−/+ ( MDA5-sufficient ) , Dhx58−/− ( LGP2-deficient ) or Dhx58+/+ ( LGP2-sufficient ) MEFs and assessed the induction of IFN-β and the interferon-stimulated protein IFIT-1 . The upregulation of Ifit1 or Ifnb1 mRNA was greatly impaired in MDA5- or LGP2-deficient MEFs infected with EMCV ( Figure1—figure supplement 1A , B ) . The same cells responded normally to RIG-I-dependent viruses such as IAV and to known RIG-I agonists such as in vitro transcribed ( IVT ) RNA ( Figure1—figure supplement 1A , B ) . To begin to define the MDA5/LGP2 agonist , we isolated the EMCV genome from purified EMCV particles and transfected it into reporter cells together with a plasmid encoding a luciferase gene under the control of the IFN-β promoter . As reporter cells , we used an easily transfectable subclone of HEK293 cells that expresses all RLRs ( Figure 1—figure supplement 2A ) and can respond , albeit weakly , to MDA5 agonists ( data not shown; Figure 1 ) . Because transfection of positive-stranded viral RNA can lead to viral replication ( even though EMCV replicates in HEK293 cells only poorly ) , we performed the IFN reporter assay in the presence of ribavirin , an inhibitor of viral RNA synthesis . As seen in Figure 1A , EMCV genomes did not stimulate the IFN-β reporter , in contrast to the genomes of IAV , which directly activate RIG-I ( Baum et al . , 2010; Rehwinkel et al . , 2010; Weber et al . , 2013 ) . To determine whether viral replication generates stimulatory RNA , we extracted total RNA from HeLa cells that had been infected with EMCV in the presence or absence of ribavirin . RNA isolated from cells in which EMCV viral replication had been permitted to take its course ( DMSO control ) potently induced the IFN-β reporter upon transfection into HEK293 cells ( Figure 1B ) . In contrast , RNA extracted from HeLa cells treated with ribavirin was non-stimulatory ( Figure 1B ) . Treatment of the reporter HEK293 cells themselves with ribavirin did not affect the response ( Figure 1—figure supplement 2B , C ) , which indicates that the stimulatory RNA is preformed in EMCV-infected HeLa cells . Furthermore , the response in the HEK293 reporter cells was dependent on MDA5 as demonstrated using RNA interference-mediated MDA5 knockdown ( Figure 1—figure supplement 2D ) . Altogether these data indicate that MDA5 and LGP2 activation results exclusively from RNA generated during active EMCV replication , as recently suggested ( Feng et al . , 2012; Triantafilou et al . , 2012 ) . 10 . 7554/eLife . 01535 . 003Figure 1 . IFN-α/β induction requires EMCV replication . ( A ) EMCV and IAV RNA genomes were extracted from purified viral particles and tested at the indicated doses in an IFN-β promoter luciferase reporter assay in HEK293 cells in the presence of ribavirin . IVT-RNA was included as a positive control . ( B ) HeLa cells were infected with EMCV ( MOI 1 ) for 16 hr in the presence of DMSO or ribavirin . RNA was extracted ( HeLa EMCV RNA ) and tested at the indicated doses in the IFN-β promoter luciferase reporter assay in HEK293 cells . RNA extracted from uninfected HeLa cells ( NI RNA ) and IVT-RNA were included as negative and positive controls , respectively . ( C and D ) HeLa cells were either not infected ( NI ) or infected with EMCV or IAV at MOI 1 for 16 hr . RNA was extracted from cell lysates and separated into double-stranded ( ds; panel C ) or single-stranded ( ss; panel D ) fractions . The fractions were analysed on a 1% agarose gel and the indicated amounts of RNA were then tested at the indicated doses in the IFN-β promoter luciferase reporter assay in HEK293 cells . dsRNA bands are indicated on the gel picture . ssRNA runs as a smear and only the ribosomal RNA bands are identifiable , as indicated . ssRNA was used as a ladder . Although the experiment depicted was carried out in the absence of ribavirin , identical results were obtained in the presence of the drug indicating that the stimulatory capacity of ssRNA fractions is not due to subsequent replication and formation of dsRNA ( data not shown ) . ( E ) Total RNA from HeLa cells infected with EMCV at MOI 1 for 16 hr ( HeLa EMCV RNA ) or control IVT RNA was either left untreated ( NT ) or digested with calf intestinal phosphatase ( CIP ) , base-paired specific RNase ( RNaseIII ) , 5′ RNA polyphosphatase ( PP ) , or single-strand-specific RNAse ( RNaseA ) . Digested RNA samples were analysed on a 1% agarose gel and the indicated amounts of RNA were tested in the IFN-β promoter luciferase reporter assay in HEK293 cells . One representative of three ( A , C , D ) or two ( B and E ) experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 00310 . 7554/eLife . 01535 . 004Figure 1—figure supplement 1 . LGP2 and MDA5 are required for IFN-α/β production in response to EMCV . ( A and B ) MEFs of the indicated genotypes were either not infected ( NI ) , infected with EMCV or IAV at an MOI of 1 , or transfected with 1 μg of in vitro transcribed RNA ( IVT RNA ) . Ifit1 ( A ) and Ifnb1 ( B ) mRNA was measured 16 hr posttreatment by quantitative RT-PCR and normalised to gapdh . Error bars represent standard deviation of three biological replicates . Data are from one representative of two experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 00410 . 7554/eLife . 01535 . 005Figure 1—figure supplement 2 . Ribavirin abrogates EMCV RNA infectivity but does not decrease IFN-β reporter activity in 293 cells . ( A ) HEK 293 cells were either left untreated or treated with 100 units/ml of IFN for 16 hr . RNA was then extracted and the expression of MDA5 , LGP2 , and RIG-I were analysed by RT-PCR and normalised to GAPDH . Error bars represent standard deviation of three biological replicates . Data are from one representative of two experiments . ( B and C ) The indicated amount of RNA isolated from HeLa cells infected with EMCV at MOI 1 for 16 hr ( HeLa EMCV RNA ) was transfected into HEK293 cells expressing the IFN-β promoter luciferase reporter in the presence of DMSO or ribavirin . Transfection with water ( mock ) or IVT RNA at the indicated doses was used as negative and positive controls , respectively . After a 16-hr incubation , the accumulation of infectious virus in culture supernatants of reporter cells transfected with HeLa EMCV RNA was determined by plaque assay ( pfu/ml = plaque forming units/ml ) ( B ) , while luciferase activity was measured in parallel ( C ) . Standard deviation in ( A ) represents the variation of three biological replicates . ( D ) Indicated amounts of RNA isolated from Vero cells infected with EMCV ( Vero EMCV RNA ) or IAV ( Vero IAV RNA ) at MOI 1 for 16 hr were transfected into HEK293 cells pretreated for 48 hr with 25 nM of siRNA control ( ctrl ) or targeting MDA5 and expressing the IFN-β promoter luciferase reporter . Luciferase activity was measured 16 hr posttransfection . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 005 One feature of the replication cycle of positive-strand RNA viruses is the generation of a negative-strand RNA that , together with the annealed positive strand , forms a long dsRNA structure . To characterise the ‘strandedness’ of the IFN stimulatory RNA generated upon EMCV replication , we extracted total RNA from non-infected or either IAV or EMCV-infected HeLa cells and separated it into ds and ssRNA fractions ( Feng et al . , 2012 ) . As expected ( Pichlmair et al . , 2006 ) , the ssRNA but not the long dsRNA fraction extracted from cells infected with IAV , stimulated the IFN reporter ( Figure 1C , D ) . In contrast , both ds and ssRNA fractions from cells infected with EMCV triggered the IFN-β reporter ( Figure 1C , D ) . Similarly , treatment of total RNA from EMCV-infected cells with RNase A or RNase III , which digest unpaired or base-paired RNA , respectively , strongly reduced stimulatory activity ( Figure 1E ) . In contrast , removal of 5′ phosphates by digestion with calf intestinal phosphatase ( CIP ) or 5′ polyphophatase ( PP ) did not impact stimulatory activity although it abrogated that of IVT control RNA ( Figure 1E ) . Altogether these data indicate that MDA5/LGP2 stimulatory RNA accumulates in EMCV-infected cells during virus replication and contains both ssRNA and dsRNA features , as previously suggested ( Pichlmair et al . , 2009 ) . To further purify the RNA responsible for activation , we established a method to isolate LGP2-associated RNA from infected cells . We immunoprecipitated ( IP ) LGP2 from EMCV-infected HeLa cells transiently expressing a FLAG-tagged LGP2 protein ( Figure 2A ) , extracted RNA from the precipitates and analysed its stimulatory activity in reporter cells ( Figure 2B ) . Notably , RNA associated with the LGP2 precipitates , but not with control ( ctrl ) precipitates , stimulated the IFN-β luciferase reporter in HEK293 cells and induced expression of the IFNB1 gene in HeLa cells ( Figure 2B , C ) . Moreover , when compared to input material , stimulatory activity was significantly enriched in the LGP2-immunoprecipitate and was selectively depleted from the unbound fraction ( Figure 2B , C ) . We additionally transfected the LGP2-associated RNA into MEFs and confirmed its ability to stimulate IFN production by ELISA ( Figure 2D ) . Importantly , activity was lost in LGP2- or MDA5-deficient MEFs ( Figure 2D ) , indicating that the LGP2-associated RNA isolated from immunoprecipitates is a pure MDA5/LGP2 agonist . Consistent with this notion , its IFN stimulatory capacity was unaffected by treatment with CIP , which inactivates most RIG-I agonists , or Terminator ( Term ) , which digests RNA with 5′ monophosphates ( Figure 2E ) . Furthermore , digestion with RNaseT1 and RNaseIII completely abolished stimulatory potential ( Figure 2E ) , again suggesting the presence of unpaired and base-paired RNA regions as previously observed with total RNA from EMCV-infected cells ( Figure 1E ) . Finally , we incubated purified RNA from EMCV-infected HeLa cells together with recombinant FLAG-tagged LGP2 ( Figure 2F ) . FLAG immunoprecipitation allowed enrichment for stimulatory activity when compared to a control IP , demonstrating that stimulatory RNA can associate with LGP2 in vitro in the absence of additional proteins ( Figure 2F ) . Altogether these results indicate that LGP2 selectively associates with a pool of MDA5 agonists in EMCV-infected cells and that LGP2 IP is a suitable approach to enrich for such agonists . 10 . 7554/eLife . 01535 . 006Figure 2 . LGP2 pulldown captures MDA5 agonistic RNA from EMCV-infected cells . ( A ) Schematic representation of the experimental setup for LGP2 immunoprecipitation ( IP ) . Precipitation efficiency was routinely verified by immunoblotting with an anti-FLAG antibody; an example is shown in the lower panel . ( B and C ) The indicated amounts of RNA from EMCV-infected FLAG-LGP2-expressing HeLa cells ( input ) , RNA associated with LGP2 or control ( ctrl ) immunoprecipitations ( IP ) , or RNA remaining after LGP2 or control precipitations ( unbound ) were tested for the ability to stimulate the IFN-β promoter reporter assay in HEK293 cells ( B ) or induce IFNB1 mRNA ( normalised to GAPDH ) in HeLa cells ( C ) . ( D ) The indicated amounts of RNA from samples processed as in ( B ) and ( C ) were transfected into WT , LGP2-deficient , or MDA5-deficient MEFs . Supernatants were harvested 16 hr later and mIFN-α levels measured by ELISA ( left panel ) . IVT RNA transfections were used as positive controls ( right panel ) . Error bars represent the standard deviation of three replicate transfections . ( E ) LGP2-associated RNA was not treated ( NT ) or digested with RNaseT1 ( specific for ssRNA ) , RNaseIII ( specific for base-paired RNA ) , CIP or Terminator ( Term , an RNase specific for 5′monophosphate RNA ) and subsequently tested for the ability to stimulate the IFN-β reporter in HEK293 cells . The activity of all enzymes was validated in control samples ( not shown ) . ( F ) RNA from HeLa cells infected with EMCV ( input ) was incubated with recombinant FLAG-tagged LGP2 protein and anti-FLAG or control IP was performed as in ( A ) . IP-associated RNA was isolated and tested , in parallel with input RNA , at the indicated doses in the IFN-β promoter luciferase reporter assay on HEK293 cells . Schematic representation of the experiment is shown on the left , results are presented on the right . One representative of the three ( A–C ) or two ( D–F ) experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 006 Having established a method to purify IFN stimulatory RNA from LGP2/RNA complexes isolated from EMCV-infected HeLa cells , we subjected it to deep sequencing analysis . We pooled RNA extracted from multiple independent control or LGP2 IPs and carried out Illumina sequencing in duplicate ( ‘Material and methods’ ) . Approximately , 30 million reads of 60 nts in length were obtained from each sequencing sample and were mapped to the human and the EMCV genomes . The total number of reads mapping to EMCV represented around 30% in LGP2 IP samples vs only 4% and 6% in ctrl IP and input , respectively ( Tables 1 and 2; Figure 3A ) , which indicated that the RNA in the LGP2 IP is specifically enriched for EMCV-derived sequences . To allow better comparison across samples , the number of reads from LGP2 IP , ctrl IP and input samples was first normalised to the total number of reads ( displayed as reads/million ) and then aligned to the EMCV genome ( Figure 3 , Figure 3—figure supplement 1 ) . Surprisingly , the distribution of sequences in the LGP2 IP sample was not uniform but displayed a number of discrete peaks concentrated in the 5′ region of the EMCV genome . In particular , one peak from position 735 nts to 905 nts on the antisense RNA was strongly enriched ( 25 , 000 reads/million ) over input and ctrl IP samples ( non detectable and 60 reads/million , respectively ) ( Figure 3B ) . A smaller peak ( ±5 , 000 reads/million ) in the corresponding part of the sense strand was also enriched in LGP2 IP samples ( Figure 3B ) . This region encodes the leader ( L ) protein of EMCV and is henceforth referred to as the L region . 10 . 7554/eLife . 01535 . 007Table 1 . Total number of reads aligning to the EMCV genome in LGP2 IP , ctrl IP , or input samplesDOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 007LGP2 IPctrl IPinputNumber of reads31 , 662 , 25537 , 235 , 33235 , 725 , 972EMCV2 , 747 , 427243 , 4721 , 380 , 128EMCV ( + ) 1 , 305 , 129239 , 6251 , 380 , 058EMCV ( − ) 1 , 442 , 2983 , 84770Total numbers of reads , reads matching both strands ( EMCV ) , sense strand ( EMCV ( + ) ) or antisense strand ( EMCV ( − ) ) of EMCV RNA sequences . 10 . 7554/eLife . 01535 . 008Table 2 . Percentage of reads mapping either the sense ( + ) or the antisense ( − ) strand in the L region compared to the full length EMCV genomeDOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 008L regionTotal EMCV ( + ) ( − ) ( + ) ( − ) LGP2 IP19 . 4980 . 5147 . 5052 . 50ctrl IP78 . 5521 . 4598 . 171 . 83input100 . 000 . 00100 . 000 . 0010 . 7554/eLife . 01535 . 009Figure 3 . The L antisense RNA region is enriched in LGP2 pulldowns from EMCV-infected cells . ( A ) RNA from LGP2 IP , control ( ctrl ) IP , or total RNA ( input ) from EMCV-infected cells ( Figure 2A , B ) was sequenced . Reads corresponding to human or EMCV sequences are shown as a percentage of the total number of reads that could be aligned to any sequence in bioinformatic databases . ( B ) All reads obtained from LGP2 IP or ctrl IP from EMCV infected cells were normalised to the number of reads per million to allow for comparison across samples . Results were mapped to the EMCV genome ( depicted above; the red arrows indicate the respective position of the primer used for the reverse transcription prior to quantitative RT-PCR in [C] ) . The total numbers of reads are shown in Table 1 . The vertical axis shows the number of normalised reads mapping to a particular position on the EMCV genome ( horizontal axis ) . The positive and negative numbers represent reads that align to the sense or the antisense strand , respectively . Numbering along the x-axis indicates nucleotide position on the sense ( + ) strand . One experiment of two is shown . ( C ) The amount of L antisense ( AS ) ( primer 1 in B ) or VP1 AS ( primer 2 in B ) RNA was analysed by strand-specific RT-PCR in LGP2 IP , ctrl IP , and input samples from an independent experiment . Vertical axis represents RNA level , calculated relative to the data obtained from a standard curve of cDNA from EMCV-infected cells . Error bars represent the standard deviation of four independent samples . ns = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 00910 . 7554/eLife . 01535 . 010Figure 3—figure supplement 1 . Comparison of read-distribution along the EMCV genome for LGP2-associated RNA and input material . Strand-specific analysis of sequencing results from LGP2 IP vs input RNA from EMCV-infected cells ( Figure 3 ) . Number of reads per million obtained from Illumina sequencing were mapped to their starting position on the EMCV genome . The vertical axis shows the number of normalised reads mapping to a particular position . The positive and negative numbers represent reads that align to the sense or the antisense strand , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 01010 . 7554/eLife . 01535 . 011Figure 3—figure supplement 2 . LGP2 directly binds L region antisense RNA . ( A ) Schematic representation of the experimental setup . Total RNA purified from HeLa cells infected with EMCV ( input ) was mixed with recombinant FLAG-tagged LGP2 before immunoprecipitation with an anti-FLAG ( LGP2 IP ) or control ( ctrl IP ) antibody . ( B ) Input RNA and RNA from precipitated material was then extracted and subjected to strand-specific RT-PCR using primers specific for the L antisense region . Vertical axis represents RNA level , calculated by comparison with a standard dilution curve of cDNA from EMCV-infected cells . The arrow indicates the position of the antisense primer ( primer 1 ) used for the reverse transcription of the L region on the EMCV genome . Error bars represent standard deviation of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 011 To validate these findings , we used strand-specific RT-PCR with primers for the L antisense or , as a control , the VP1 antisense regions ( primer localisation indicated by red arrows on the schematic representation of the EMCV genome in Figure 3B ) . This analysis confirmed that the L antisense but not the VP1 antisense region , was enriched in LGP2 IP samples compared to ctrl IPs and input material ( Figure 3C ) . These findings were further validated in the LGP2 in vitro reconstitution assay , which confirmed that RNA that binds to LGP2 is enriched for the L antisense region ( Figure 3—figure supplement 2 ) . In sum , LGP2 association with EMCV RNA is strongly biased towards a discrete area of the negative ( antisense ) strand of the L protein-encoding region . We next examined if the L antisense ( AS ) RNA sequence was sufficient to trigger an MDA5-dependent IFN response . We generated , by in vitro transcription , RNAs corresponding either to the AS or the sense strand of the EMCV L region . After CIP treatment to remove any RIG-I-stimulatory activity linked to the presence of the 5′ triphosphates , we tested the stimulatory potential of these RNAs in MEFs deficient for the Ddx58 gene encoding RIG-I or in MDA5-deficient MEFs by IFN-β promoter luciferase assay ( Figure 4A , B ) . The RNA containing the L AS derived sequence was clearly stimulatory in contrast to the one derived from the L sense sequence . Moreover , the activity of the CIP-treated L AS was greatly reduced in the MDA5-deficient but not RIG-I-deficient MEFs ( Figure 4A ) . As a control , we used non-CIP treated IVT RNA corresponding to the sense sequence of the neomycin gene ( Rehwinkel et al . , 2010 ) , which showed the expected RIG-I dependence ( IVT RNA , Figure 4C ) . We conclude that an IVT RNA corresponding to the EMCV L AS RNA sequence found in LGP2 immunoprecipitates ( Figure 3 ) can trigger an MDA5-dependent IFN response . 10 . 7554/eLife . 01535 . 012Figure 4 . In vitro transcribed L AS RNA triggers an MDA5-dependent IFN response . ( A–C ) L antisense ( AS ) ( A ) , L sense ( B ) or IVT RNA ( C ) sequences were in vitro transcribed and all RNA except from the control IVT RNA ( C ) were CIP treated to remove any 5′ phosphates . The indicated amount of RNA were then transfected into RIG-I- , MDA5-deficient or sufficient MEFs expressing the IFN-β reporter . Reporter activity was measured 16 hr later . One of two experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 012 To ask whether the L region of the EMCV genome is important for the generation of IFN stimulatory RNA , we used mutant EMCV viruses with partial ( EMCV ΔLac and EMCV ΔLzn ) or complete ( EMCVΔL ) deletions of that region ( Dvorak et al . , 2001; Figure 5A ) . We infected HeLa cells , extracted RNA and subjected it to RT-PCR analysis . Using specific primers ( shown as arrows in Figure 5A ) , we confirmed the loss of the L region in the mutant viruses ( Figure 5B , left panel ) . Because all three viruses are attenuated due to the absence of the L protein , a known antagonist of IFN induction ( Hato et al . , 2007 ) , we also amplified the Vp1 region to verify that similar levels of EMCV RNA were present in all samples ( Figure 5B , right panel ) . We then assessed the IFN stimulatory potential of these samples using the IFN-β reporter assay . In all cases , RNA extracted from HeLa cells infected with EMCV ΔLac , ΔLzn or ΔL was slightly less stimulatory than RNA extracted from cells infected with wild-type ( WT ) EMCV ( Figure 5C ) . More importantly , only a low amount of stimulatory RNA was recovered from LGP2 IPs following infection with the mutant viruses in contrast to infection with WT EMCV ( Figure 5D ) . These results are consistent with the earlier indications that the major species of stimulatory RNA associating with LGP2 in EMCV-infected cells derives from the L antisense region . 10 . 7554/eLife . 01535 . 013Figure 5 . The L region of EMCV is required for the generation of LGP2-associated stimulatory RNA . ( A ) Schematic representation of the L region of EMCV genome and L region mutant viruses used in this study . The crosses indicate the position of the two point mutations in EMCV ZnC19AC22A . The marked positions indicate the limits of the deletions present in EMCV ΔL , ΔLac , and ΔLzn , respectively . ( B–F ) FLAG-LGP2-expressing HeLa cells were infected with EMCV WT , ZnC19AC22A , ΔL , ΔLac , or ΔLzn viruses at MOI 1 for 16 hr before lysis and total RNA extraction ( input; B , C , E ) or lysis followed by LGP2 or control ( ctrl ) immunoprecipitation and RNA extraction from precipitates ( IP; D and F ) . ( B ) Quantitative RT-PCR analysis of viral sequences in input samples using primers specific for the L region ( left panel ) or the Vp1 region ( right panel ) and normalised to GAPDH . ND = non-detected . The position of the primers used for amplifying the L region is depicted in ( A ) . ( C–F ) The stimulatory potential of the indicated doses of input ( C and E ) , LGP2 IP or ctrl IP RNA ( D and F ) was assessed by IFN-β promoter luciferase reporter assay in HEK293 cells . ( B–F ) One experiment of two is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 01310 . 7554/eLife . 01535 . 014Figure 5—figure supplement 1 . 200 ng of RNA isolated from HeLa cells infected with EMCV WT , ZnC19AC22A or ΔL at MOI 1 for 16 hr ( HeLa EMCV RNA ) was transfected into MDA5-sufficient or MDA5-deficient bone marrow-derived DCs in presence of ribavirin . Transfection with ribavirin only and IVT RNA was used as negative and positive controls , respectively . Supernatants were harvested 16 hr later and mIFN-α levels measured by ELISA . ND = non detected . One of two experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 014 Deletion of the L region results in a virus that is both depleted of L RNA and L protein . To verify that the lack of LGP2-associated stimulatory RNA upon infection with ΔLac , ΔLzn or ΔL viruses was due to the absence of L region RNA rather than absence of L protein function , we additionally used an EMCV ZnC19AC22A strain . This virus carries two mutations in the zinc domain of the L protein , which inactivate function ( Hato et al . , 2007 ) but should not impact on the production of L RNA ( even if that RNA now carries two nucleotide substitutions ) . Reassuringly , the amount of stimulatory RNA in total cell extracts ( Figure 5E ) or associated with LGP2 precipitates ( Figure 5F ) was comparable upon infection with EMCV ZnC19AC22A and EMCV WT even though , as before , it was markedly reduced after EMCVΔL infection ( Figure 5E , F ) . We conclude that the lack of stimulatory LGP2-associated RNA observed after infection with EMCVΔL is specifically due to the loss of L region RNA rather than the loss of L protein function . We subsequently assessed the importance of the L RNA sequence for IFN responses in primary cells by comparing infection with EMCVΔL and EMCV ZnC19AC22A mutant viruses . We used dendritic cells grown from mouse bone marrow ( BM-DCs ) , which produce vast amounts of IFNs in response to virus infection ( Diebold et al . , 2003 ) . To allow efficient replication of the L protein-deficient viral strains , BM-DCs were derived from IFN-α/β receptor-deficient mice ( IFNAR1 knockout [KO] ) . To monitor EMCV detection and the activation of the downstream RLRs signalling pathway , we assessed the induction of both Ifit1 and Ifnb1 , which are direct transcriptional targets of IRF-3 ( Grandvaux et al . , 2002 ) . Upon infection with WT EMCV , induction of Ifit1 and Ifnb1 was limited ( Figure 6A , B ) . It was markedly greater in response to infection with EMCV ZnC19AC22A , which encodes the non-functional mutant L protein ( Figure 6A , B ) , consistent with the fact that the L protein inhibits IFN induction ( Hato et al . , 2007 ) . In contrast , infection with EMCVΔL lacking both L protein and EMCV L region RNA induced lower levels of Ifnb1 or Ifit1 than infection with EMCV ZnC19AC22A at two different multiplicities of infection ( MOI ) ( Figure 6A , B ) . EMCV ΔL and EMCV ZnC19AC22A replicated to similar levels , indicating that any differences in stimulation by these viruses are not caused by variations in viral RNA levels ( Figure 6C ) . Similar results were obtained when the mutations were introduced into the EMCV mengo strain background ( data not shown ) . These data show that the L region RNA of EMCV is important for RLR stimulation and viral restriction in infected cells and that this is independent on its ability to encode a functional L protein . 10 . 7554/eLife . 01535 . 015Figure 6 . L region RNA is required for IFN-α/β production in response to EMCV . ( A–C ) IFNAR1-deficient GM-CSF bone marrow-derived DCs were either not infected ( NI ) or infected with the indicated viruses at an MOI of 1 or 10 . Levels of Ifit1 ( A ) , Ifnb1 ( B ) or EMCV Vp1 ( C ) RNA were analysed 16 hr later by quantitative RT-PCR and normalised to gapdh . ( D and E ) 293T cells were transfected with the influenza vRNP reconstitution system using the influenza NS wt segment or NS segment carrying the L AS RNA sequence and incubated for 24 hr . RNA was then extracted and the IFN stimulatory activity was tested on MEF wt , MDA5 or RIG-I deficient by luciferase assay ( D ) and the expression of the L AS sequence was verified by strand-specific RT-qPCR with primer specific for the L antisense RNA ( E ) . ( A–E ) One experiment of two is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 015 In parallel , we also examined the IFN response to infection with an EMCV mengo strain mutant virus carrying the L region of the foot-and-mouth disease virus ( FMDV ) . FMDV belongs to a different picornavirus genus and the FMDV L region sequence and FMDV leader protein share little similarity with those of EMCV . Notably , the induction of both Ifit1 and Ifnb1 in response to EMCV LFMDV was comparable to that induced by EMCVΔL ( Figure 6A , B ) and replication of the two strains was indistinguishable . These data suggest that the EMCV L AS RNA sequence rather than the position of the L region in the EMCV genome is the key determinant in innate stimulation . To further address this issue , we introduced the EMCV L antisense RNA sequence into the influenza virus NS segment , a known RIG-I agonist ( Rehwinkel et al . , 2010 ) . However , as shown in Figure 6D , E , the resulting chimeric NS RNA remained dependent on RIG-I and not MDA5 for its stimulatory activity . This suggests that the L region antisense sequence needs to be somehow processed or released in the context of EMCV infection to trigger an MDA5-dependent response and this does not happen in the context of the influenza virus NS segment . Finally , we assessed whether L region RNA is also important for responses to EMCV in vivo . Mice deficient in IFNAR1 ( to allow replication of the attenuated viruses ) were infected with EMCV WT , EMCVΔL and EMCV ZnC19AC22A strains and the outputs of innate immune stimulation were measured . Higher levels of IFN-α were found in the serum of mice 24 hr after infection with EMCV ZnC19AC22A when compared to EMCVΔL and WT ( Figure 7A ) . In addition , the expression of Ifit1 and Ifnb1 was also much higher in hearts from mice infected with EMCV ZnC19AC22A compared to mice infected with EMCVΔL and WT ( Figure 7B , C ) . Measurement of VP1 mRNA in hearts confirmed broadly similar levels of infection by all viruses ( Figure 7D ) . We conclude that the presence of L region RNA independently of the function of the L protein is important for innate immune responses to EMCV infection in vivo . 10 . 7554/eLife . 01535 . 016Figure 7 . The L region is required for IFN-α/β responses to EMCV in vivo . ( A and B ) IFNAR1 KO mice were injected intraperitoneally with PBS ( not infected; NI ) or with 10E6 pfu of the indicated viruses . ( A ) mIFN-α levels in the serum were measured by ELISA after 24 hr . ( B–D ) RNA was extracted from hearts 24 hr postinfection and analysed by quantitative RT-PCR for levels of Ifit1 ( B ) , Ifnb1 ( C ) , or EMCV Vp1 ( D ) RNA normalised to gapdh expression . Combined results of two independent experiments are shown . Each symbol represents an individual mouse . ns = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01535 . 016 Several types of RNA can trigger MDA5-dependent responses in experimental settings ( Malathi et al . , 2007 , 2010; Pichlmair et al . , 2009; Luthra et al . , 2011; Züst et al . , 2011; Feng et al . , 2012; Triantafilou et al . , 2012 ) but it remains unclear which RNAs serve as natural MDA5 agonists during virus infection . For example , some of the RNA species extracted from infected cells may not have access to MDA5 during infection or may be produced in quantities or at times that are irrelevant for innate immune recognition . One way to identify those RNAs most likely to be relevant agonists in infected cells is to isolate them directly from RLR complexes present in those cells ( Baum and García-Sastre , 2010; Rehwinkel et al . , 2010 ) . In this study , we describe a method to isolate relevant MDA5 agonists from cells infected with EMCV by immunoprecipitation of LGP2/RNA complexes . We show that this method simultaneously enriches for MDA5 stimulatory activity and for a discrete species of antisense RNA mapping to the EMCV L region . Notably , we establish a causal connection between the two events by showing that deletion of the L region from the virus genome abrogates the association of stimulatory RNA with LGP2 in infected cells and reduces the innate stimulatory capacity of the virus , both in vitro and in vivo . Finally , we show that in vitro synthesis of L region antisense RNA is sufficient to generate an MDA5 agonist . Our results demonstrate that the antisense L region of EMCV associates with LGP2 and is a key determinant of MDA5 stimulation . Curiously , this determinant is derived from the same region that encodes a protein product that suppresses IFN induction . The fact that deletion of this region in EMCV reduces viral stimulatory activity is therefore counterintuitive and strengthens the notion that it is a physiologically-relevant major determinant of innate immunity to the virus . The finding that pulldown of LGP2 allows isolation of MDA5 stimulatory RNA is in agreement with the fact that both RLRs are required for IFN production upon EMCV infection ( Figure 1—figure supplement 1 ) and with previous reports describing LGP2 as a positive regulator of MDA5 signalling ( Venkataraman et al . , 2007; Satoh et al . , 2010; Bruns et al . , 2012; Childs et al . , 2013 ) . Interestingly , despite numerous attempts and experimental permutations , we have never been able to isolate stimulatory RNA directly from MDA5 pulldowns ( data not shown ) . One explanation may be that LGP2 possesses higher avidity for RNA than MDA5 ( Bruns et al . , 2012; Childs et al . , 2013 ) and is therefore more prone to retain bound agonist during cell lysis and precipitation . Another possibility relates to the fact that MDA5 forms filaments along RNA upon activation ( Peisley et al . , 2011 , 2012; Berke and Modis , 2012; Berke et al . , 2012; Wu et al . , 2012 ) and that such structures—when formed in infected cells—may be difficult to isolate . These considerations raise the question of whether LGP2 precipitation selects relevant MDA5 agonists . For example , given that LGP2 can bind to stimulatory RNAs in vitro ( Figure 2F ) , is it possible that the complexes that we precipitate result from post-lysis ‘mopping-up’ of stimulatory RNA that would never in the intact cell have come into contact with MDA5 ? We believe this is unlikely based on the fact that the functional relevance of the L region identified in the LGP2 pulldowns could be validated in the context of infection using mutant viruses . Furthermore , we find that LGP2 pulldown both enriches for MDA5 agonists and depletes them from input material . Therefore , it appears that the universe of LGP2-bound RNAs largely overlaps with that of MDA5 agonists in EMCV-infected cells . Although we do not exclude the possibility that there are MDA5 agonists that are not captured by LGP2 pulldown ( see below ) , our findings suggest a model where LGP2 pre-selects RNAs for MDA5 activation prior to the formation of MDA5 filaments . Whether the two proteins form a physical complex remains to be established—it is possible that LGP2 acts upstream of MDA5 to ‘handover’ agonistic RNA and that interaction with MDA5 , if any , may be indirect or transient . Deep sequencing of LGP2-associated stimulatory RNA was revealing . We were not able to identify specific enrichment for any human sequences ( data not shown ) , a finding that does not support a dominant role for RNAseL cleaved self RNA in MDA5 activation ( Malathi et al . , 2007 ) . This is in agreement with another recent publication that also failed to detect a significant role for RNaseL activity in EMCV-dependent IFN induction ( Feng et al . , 2012 ) . Instead , the sequencing data , confirmed by strand-specific RT-PCR analysis , identified a different species of small RNA that was highly enriched in LGP2 pulldowns . This is a fragment of 171 nucleotides that corresponds to the antisense RNA in the L region of the genome . Its short length and failure to be accompanied by the complementary strand are at odds with the notion that MDA5 is activated by long dsRNA ( Kato et al . , 2006 , 2008 ) . However , recent data indicate that MDA5 can form filaments along short dsRNA of around 100 nucleotides in length ( Peisley et al . , 2011 ) . Moreover , the fact that LGP2-associated RNA appears to be single-stranded is in agreement with the observation that the stimulatory activity of the LGP2-associated RNA is sensitive to RNases that degrade ssRNA ( Figure 2E ) . Interestingly , this LGP2-associated RNA is also sensitive to dsRNase treatment , which suggests the presence of base-paired regions important for MDA5/LGP2 recognition . Consistent with that notion , heat denaturation destroyed its activity ( data not shown ) . In silico analysis reveals the potential of the L antisense RNA to form hairpins ( data not shown ) , and such secondary and tertiary structure features may help to efficiently trigger MDA5 activation . Interestingly , by in vitro transcription of a sequence corresponding to the L antisense region we have succeeded in creating de novo an RNA that triggers MDA5 activation ( Figure 4 ) . What endows this RNA with MDA5 agonistic activity and whether the in vitro transcribed version fully corresponds to that naturally formed in EMCV-infected cells will require further investigation . Another intriguing question is how a fragment of L AS RNA is generated upon viral infection . We have not been able to find a report describing the generation of small antisense RNA fragments during EMCV replication . However , Northern blots of RNA from EMCV-infected cells screened with an L AS sequence probe reveal many fragments smaller than the EMCV genome ranging between 200 and 2000 nucleotides ( data not shown ) . Moreover , a previous study has reported the generation of a short subgenomic RNA fragment derived from the degradation of the non-polyadenylated genome of flaviviruses by the exonuclease XRN1 ( Funk et al . , 2010 ) . It is possible that a similar mechanism could generate an L antisense RNA fragment . Alternatively , stress granule formation , known to interfere with picornavirus replication ( Borghese and Michiels , 2011 ) , could potentially lead to degradation of viral RNA and generate the L region antisense RNA agonist . Interestingly , MDA5 and LGP2 have been recently shown to associate with stress granules ( Onomoto et al . , 2012; Langereis et al . , 2013 ) . Picornavirus replication involves a stage where the positive-stranded genome anneals with a newly-formed negative strand in a full-length double-stranded RNA structure known as the replicative form ( RF ) . Negative-strand RNA from the RF then serves as the template to produce new progeny viral genomes through a replicative intermediate ( RI ) consisting of many positively-stranded progeny RNA genomes in the process of being synthesised and still partially hybridised to the full-length antisense RNA template . A recent study showed that negative-strand RNA synthesis is absolutely required for IFN induction during infection of mengovirus , a strain of EMCV ( Feng et al . , 2012 ) . Our results that a small RNA derived from the negative-strand RNA of EMCV associates with LGP2 and activates MDA5 is in line with that report . However , it has also been shown that both the RF and RI forms can activate MDA5 upon transfection ( Feng et al . , 2012; Triantafilou et al . , 2012 ) . It is conceivable that RF and RI forms could be processed by RNases within the cell following transfection or infection , releasing the L antisense fragment that associates with LGP2 . Alternatively , it is possible that the L region antisense fragment enriched in our IP is only one of several agonists generated upon EMCV replication and that the RF and RI forms constitute a different set of agonists that do not strongly associate with LGP2 and have therefore been missed in our approach . This hypothesis is consistent with our data , which show that EMCVΔL viruses still possess the ability to induce low levels of IFN and IFIT-1 ( Figures 6 and 7 ) and generate stimulatory RNA ( Figure 5 ) . Such data suggest the existence of at least one additional stimulatory RNA species beside the L region antisense fragment and help explain why EMCVΔL is an attenuated strain: the additional stimulatory RNA means that it cannot fully evade innate immune detection while the lack of the L protein means that it cannot suppress the consequences of detection . Although it remains to be established whether this putative additional stimulatory RNA species corresponds to the RF/RI ( or even acts via MDA5—Figure 5—figure supplement 1 ) , these observations suggest the need to determine to what extent different innate immune stimuli dominate at different stages of infection . Our investigation has resulted in the identification of a specific region of the EMCV negative-strand RNA as a determinant of innate immunity to the virus . To our knowledge , this is the first study to identify an MDA5 agonist bound to LGP2 in infected cells . Picornaviruses constitute a large family of viruses that encompasses many human and animal pathogens including some of economical or medical importance including Poliovirus , Coxsackievirus , Rhinovirus , and EMCV ( Tuthill et al . , 2010 ) . The L regions of picornaviruses can show wide variation and it will be interesting to determine to what extent L region RNA or other small RNAs can act as MDA5 agonists across picornavirus genera . Elucidating the molecular basis of picornavirus detection will help understand the general rules underlying innate virus detection and may suggest new strategies to control viral infection . The human LGP2 sequences were amplified from cDNA derived from IFN-A/D ( PBL Assay Science , Piscataway , NJ ) -treated HEK293 cells using a forward primer containing the 3xFLAG epitope sequence with the following oligos ( FW LGP2: 5′gccgccatggactacaaagaccatgacggtgattataaagatcatgacatcgattacaaggatgacgatgacaaggagcttcggtcctaccaatggga-3′ , RV LGP2: 5′-tcagtccagggagaggtccga-3′ ) . PCR products were then introduced into the pcDNA3 . 1 TOPO plasmid ( Life Technologies , Foster City , CA ) to generate the 3xFLAG LGP2 expression constructs . For production of the recombinant LGP2 protein , the 3xFLAG-human LGP2 sequence was amplified using the forward primer 5′-actcgagttatggactacaaagaccatgacgg-3′ and the reverse primer 5′-ttgcggccgctcagtccagggagaggtccga-3′ and cloned into the XhoI and NotI sites of the pBacPAK-His3-GST plasmid . Recombinant 3xFLAG LGP2 was expressed as a GST-tagged protein in SF9 insect cells using a baculovirus expression system and purified on a single step by affinity chromatography using Glutathione Sepharose matrix ( GE Healthcare Life Sciences , Buckinghamshire , UK ) . The 3xFLAG LGP2 was eluted by GST tag cleavage using in-house 3C enzymatic digestion . A final polishing step was then performed on a superdex 200 10/300 GL column ( GE Healthcare ) . Protein purity was then verified on an acrylamide gel and the protein yield was quantified using a Nanodrop apparatus ( ThermoScientific , Wilmington , DE ) . The M2 ( Sigma ) and IgG1 isotype match control ( BD Pharmingen , San Diego , CA ) antibodies were used for immunoprecipitation . M2 antibody was used at 1/5000 dilution for Western blot . Anti-mouse HRP antibody ( Southern Biotech , Birmingham , AL ) was used at 1/10 , 000 dilution for Western blot . IFN A/D ( PBL Assay Science ) was used at 100 units/ml for 24 hr to pre-treat cells . Ribavirin powder ( Sigma ) was reconstituted in DMSO and used at 4 mM final concentration . HeLa , BHK21 , and Vero cells were from Cancer Research UK ( CRUK ) Cell services . HEK293 selected for responsiveness to RLR agonists were previously described ( Pichlmair et al . , 2009 ) . WT , Ifih1−/− , Ifih1+/− , Ddx58−/− , Ddx58+/+ , Ddx58−/+ , Dhx58−/− and Dhx58+/+ , MEFs were generated as previously described ( Kato et al . , 2006 ) , immortalised with simian virus 40 large T antigen and selected on puromycin ( final concentration , 2 mg/ml ) for 2 weeks . HeLa and BHK21 cells were grown in 10% FCS-containing minimum essential medium ( CRUK ) or Glasgow media ( Life Technologies ) , respectively . All other cell lines were grown in Dulbecco’s modified Eagle’s medium containing 10% FCS and 2 mM glutamine . Mouse BM-DCs were generated using GM-CSF as described previously ( Inaba et al . , 1992 ) . Briefly , femur and tibia were collected from both hindquarters . Bones were flushed with RPMI 1640 ( Life Technologies ) media containing 10% FCS , 100U/ml Penicillin/Streptomycin , 5 μM β-mercaptoethanol and 200 units/ml of GM-CSF ( CRUK ) and passed through a 70 μM cell strainer . Cells were then cultured for 5 days with medium renewal every 2 days . Influenza A virus ( IAV ) ( PR8 strain ) and EMCV were as previously described ( Pichlmair et al . 2009 ) . EMCV Znc19AC22A , EMCV ΔL and EMCV LFMDV ( mengovirus strain ) , were produced as described ( Feng et al . , 2012 ) . Other EMCV mutants were generated as follows: pEC9 EMCV WT , pEC9 EMCV ΔL , pEC9 EMCV ΔLac and pEC9 EMCV ΔLzn plasmids ( kind gift from Ann C Palmenberg ) were in vitro transcribed to generate full length EMCV RNA using T7 Megascript kit ( Ambion ) as per manufacturer’s instructions . Products were then digested with DNase I and purified with phenol:chloroform:isoamylalcohol ( 25:24:1 ) , followed by chloroform extraction and ethanol precipitation . RNAs were transfected into HeLa cells to generate viruses , which were subsequently amplified on BHK21 cell monolayers until cytopathic effects were observed . Cell lysates were then freeze-thawed three times , cleared and centrifuged for 2 hr at 22 , 000 rpm ( SW 32 Ti Rotor , Beckman ultracentrifuge ) at 4°C on 20% sucrose cushion to purify viral particles . The pelleted viral particles were resuspended in 10 mM Tris pH7 , 2 mM MgCl2 containing buffer and viral content quantified by plaque assay on Vero cells . IFNAR1 KO mice were obtained from Michel Aguet ( University of Lausanne ) and backcrossed 14 times to C57BL/6J . The mice were bred at CRUK or at St Mary’s Hospital ( kind gift from Cecilia Johansson ) in specific-pathogen free conditions . For infection , 10- to 12-week-old C57BL/6-IFNAR1 KO mice were injected intraperitoneally with 106 pfu of the indicated virus in 200 μl of PBS . Control mice were injected with 200 μl of PBS . Serum and organs were collected from culled animals 24 hr after injection . All animal experiments were performed in accordance with national and institutional guidelines for animal care and were approved by the London Research Institute Animal Ethics Committee and by the UK Home Office ( project licence PPL 80/2309 ) . The IVT RNA used as a positive control for IFN induction was the IVT neomycin sequence previously described ( Rehwinkel et al . , 2010 ) . The in vitro transcribed RNA derived from the L AS and sense RNA sequence were generated by in vitro transcription using the T3 and Sp6 Megascript kit ( Ambion Life Technologies ) using the PCR with primer fw 5′-gcgcactctctcacttttga-3′ and rv 5′-aaatttaggtgacactatagaagcgctcgaaaacgacttccatgt-3′ or fw 5′-aaaattaaccctcactaaagggagaacttgcgcgcactctctcac-3′ and rv 5′-tcgaaaacgacttccatgtct-3′ for the production of the L AS and sense templates . Viral genomic RNAs were extracted from purified viral suspension using Trizol LS reagent ( Life Technologies ) as per manufacturer’s recommendations . For RNA extracted from infected HeLa cells , HeLa cells were infected with EMCV or IAV at an MOI of 1 for 16 hr before total RNA was extracted using Trizol . For separation of RNA into ss and dsRNA fractions , ssRNAs were first precipitated in presence of 2M LiCl . Double-stranded RNAs were then ethanol precipitated from the supernatant in presence of 0 . 7M LiCl . RNA pellets were washed with 70% ethanol , dried , and resuspended in RNase free water . Calf intestinal phosphatase ( New England Biolabs , Ipswich , MA ) , Terminator ( Epicenter Biotechnologies , Madison , WI ) , RNaseT1 ( Ambion Life Technologies ) , Polyphosphatase ( Epicenter Biotechnologies ) , RNase A ( Sigma , St . Louis , MO ) and RNase III ( Ambion Life Technologies ) were used as per manufacturer’s recommendations . A control reaction omitting the enzyme was carried out in parallel . RNA was recovered by extraction with phenol:chloroform:isoamylalcohol ( 25:24:1 ) , followed by chloroform extraction and precipitation with ethanol and sodium acetate in the presence of glycogen ( Ambion ) . All RNAs were quantified using a Nanodrop apparatus . For quantitative RT-PCR analysis of Ifnb1 , Ifit1 , IFNB1 , gapdh , GAPDH , VP1 and L genes , RNA was extracted from 1 × 105 MEFs , BM-DCs or HeLa cells using the RNeasy kit ( Qiagen , Valencia , CA ) or from infected mouse organs using Trizol . To generate the NS L AS Influenza segment , we amplified the L EMCV fragment amplified using the forward primer 5′-aaaccatggatggccacaaccatggaac-3′ and the reverse primer 5′-aaaccatggctgtaactcgaaaacgactt-3′ and introduced it into NcoI site of the pPolI_NS plasmid encoding the NS segment sequence of Influenza WSN strain ( gift from Ervin Fodor ) . vRNP were then reconstituted in HEK 293T cells as previously described in Rehwinkel et al . ( 2010 ) . RNA was treated with DNase I ( Qiagen ) prior to reverse transcription using superscript II reverse transcription reagents ( Life Technologies ) according to the manufacturer’s instructions . PCR was performed with TaqMan Universal PCR master Mix ( Applied Biosystem Life Technologies ) and the following taqman reagent assays: mouse Ifit1 ID Mm00515153_m1 , mouse Ifnb1 ID Mm00439546_s1 , mouse gapdh ID 4308313 , human IFNB1 ID Hs02621180_s1 , human gapdh ID 402869 . To detect viral RNA , Express Syber GreenER reagent ( Life Technologies ) was used in combination with the following primers: fw 5′-gcgcactctctcacttttga-3′ and rv 5′-tcgaaaacgacttccatgtct-3′ for detection of the L region or fw 5′-cctcttctccccctttgtgt-3′ and rv 5′-caggtccggcactataaacc-3′ for detection of the VP1 region . Data were normalized to levels of gapdh . For strand-specific detection of viral RNA , RNA was first reverse transcribed using Superscript II ( Life Technologies ) in the presence of primers specific for L antisense ( 5′- ggccgtcatggtggcgaataagcgcactctctcacttttga-3′ ) or VP1 antisense ( 5′- ggccgtcatggtggcgaataacaggtccggcactataaacc-3′ ) . cDNAs were then subjected to Exonuclease I ( New England Biolabs ) treatment and purified using Qiaquick PCR purification kit ( Qiagen ) . In the next step , quantitative RT-PCR was performed in presence of Express Syber GreenER ( Invitrogen ) using the following primers: common forward primer 5′-aataaatcataaggccgtcatggtggcgaataa-3′ in combination with either L reverse primer 5′-aataaatcataatcgaaaacgacttccatgtct-3′ or 1D reverse primer 5′- aataaatcataacctcttctccccctttgtgt-3′ to detect L or 1D region respectively . RNA levels were calculated by comparison with a dilution curve of cDNA from EMCV-infected cells . IFN-β luciferase reporter assay: 2 . 5 × 105 of HEK293 cells or IFN-A/D treated MEFs were transfected with 200 ng of p125Luc ( gift from T Fujita , Kyoto university , Japan ) and 50 ng pRL-TK ( Promega , Madison , WI ) using Lipofectamine 2000 ( Life Technologies ) as per manufacturer’s instructions . Cells were incubated for 6–8 hr and were then transfected with water ( mock control ) or with test or control RNAs using lipofectamine . Luciferase activity was analysed in cell lysates 12 to 16 hr later using the Dual luciferase reporter Assay system ( Promega ) . In all cases , firefly luciferase values were divided by Renilla luciferase values to normalise for transfection efficiency . All data are shown as fold increase relative to reporter cells transfected with water alone . MEF assay: MEFs were pre-treated with 100 units/ml of IFN-A/D ( PBL interferon ) for 24 hr and plated into 24-well plates at 0 . 5 × 105 cell/well . The cells were then transfected with test or control RNAs using lipofectamine . After overnight culture , mIFN-α was measured in cell supernatants by ELISA as described previously ( Rehwinkel et al . , 2010 ) or MEF RNA was extracted with an RNeasy kit ( Qiagen ) following manufacturer’s instructions for RT-PCR analysis of Ifit1 and Ifnb1 induction . HeLa assay: HeLa cells were pre-treated with 100 units/ml of IFN-A/D ( PBL Assay Science ) for 24 hr and plated into 24-well plates at 0 . 5 × 105 cell/well . The cells were then transfected with the indicated amounts of test or control RNAs using lipofectamine . After overnight culture , HeLa RNA was extracted for RT-PCR analysis of IFNB1 induction . Around 10–15 million HeLa cells were transfected with 30 µg of 3xFLAG LGP2 IP plasmids using lipofectamine reagent following the manufacturer instructions . The cells were incubated for 16 hr , lipofectamine was washed away and the cells were incubated in fresh medium for 6–8 hr . HeLa cells expressing the 3xFLAG constructs were then infected with EMCV at an MOI of 1 for 16 hr . The cells were subsequently washed and lysed in lysis buffer ( 10 mM Tris pH 7 . 4 , 2 . 5 mM MgCl2 , 200 mM NaCl , 0 . 5% NP40 , 1X protease inhibitor cocktail [Roche , Mannheim , Germany] , 0 . 5 U/ml RNasin [Promega] ) . Lysate was incubated for 30 min on ice , cleared by centrifugation for 10 min and the supernatant was collected . At this stage a small fraction of the input was collected for protein and RNA extraction . 5 μg of M2-anti-FLAG ( Sigma ) or IgG1 isotype control antibody was added to 500 μl of lysate and incubated on a rotating shaker for 1 hr at 4°C . 300 μl of washed Gamma Bind Plus Sepharose Beads ( GE Healthcare Bioscience AB ) were then added to the mixture for another 1 hr . The beads were then precipitated by centrifugation and washed four times with 1 ml of lysis buffer . The beads were then split into two samples for protein or RNA extraction . Proteins were extracted from Protein/RNA complexes by boiling the beads for 5 min in SDS sample buffer for Western blot analysis . RNAs were purified from the beads by Phenol/chlorophorm extraction followed by ethanol precipitation . RNA was then quantified using a Nanodrop apparatus and the same amount of RNA from each sample was tested for IFN stimulatory activity as indicated . For the IP using recombinant LGP2 , 5 μg of the purified protein was incubated with 20 μg of total EMCV-infected RNA ( MOI 1; 16 hr ) in lysis buffer for 1 hr at 4°C on a rotating shaker . The rest of the IP was performed the same way as for the LGP2 IP from infected cells . Input material pooled from five replicate cultures was subjected to immunoprecipitation with anti-FLAG or control IgG1 antibody ( see above ) . RNA extracted from 10 independent IPs was pooled and ribosomal RNA was removed using Ribo-Zero rRNA Removal Kit ( Human . Mouse . Rat ) ( Epicenter RZH1046 , Madison , WI ) according to the manufacturer’s protocol . After ribosomal RNA removal , the RNA was tested for its ability to activate the IFN-β promoter luciferase reporter . The RNA was then prepared for Illumina sequencing using an optimised version of the original Directional mRNA-Seq Library Prep protocol ( Pre-Release Protocol Rev . A ) . The original protocol was optimised from 1 μg of total RNA input to a total input of up to 173 ng . Reagents supplied by Illumina included 10× v1 . 5 sRNA 3′ Adaptor; SRA 5′ Adaptor; SRA RT primer; PCR Primer GX1/GX2 . The remaining reagents recommended on the protocol were outsourced from alternative vendors . To analyse both coding and non-coding RNA regions , the poly ( A ) purification step was omitted . After fragmentation of the total RNA ( optimised incubation of 2 min at 80°C ) , the RNA was checked for efficient fragmentation sizing , ribosomal RNA contamination and RIN using the Agilent 2100 Bioanalyzer QC Pico RNA chip . After library preparation , the PCR reaction was optimised by replacing the recommended PCR polymerase with Kapa Hifi DNA Polymerase ( KAPA Biosystems , Wilmington , MA ) in addition to optimising the number of PCR cycles from 12 to 18 cycles using the recommended PCR cycling conditions . The final library preparation was then size selected at 150 bp–450 bp using a 2% Agarose E-gel ( Invitrogen ) to remove unwanted regions in excess of 500 bp which was detected using the Agilent 2100 Bioanalyzer QC DNA 1000 chip . Next generation sequencing and library preparation was performed in the Advanced Sequencing Facility ( ASF ) at the London Research Institute on the Genome Analyzer IIx ( GAIIx ) with a single-end 72 bp sequencing run alongside a PhiX control of 1–5% in every lane of the flowcell . Sequencing typically generated ∼30 million 60 bp single-end reads . Alignment to the EMCV strain pEC9 genome ( genbank accession ID DQ288856 ) was performed using Bowtie ( Langmead et al . , 2009; version 0 . 12 . 7 ) permitting a maximum of three mismatches per read . Genome-wide coverage was calculated using the genomeCoverageBed function in BEDTools ( Quinlan and Hall , 2010; version 2 . 16 . 2 ) and all subsequent plots were generated using the statistical programming language , R ( R core team 2012 , R: A language and environment for statistical computing . R Foundation for Statistical Computing , Vienna , Austria; version 2 . 15 ) . To generate more comparable data across samples the number of reads were normalised to the number of reads per million . An unpaired two-tailed Student’s t test was used to determine statistically significant differences . p values of less than 0 . 05 were considered statistically significant . GraphPad Prism version 6 for Macintosh ( GraphPad Software ) was used for statistical analysis of data .	A virus is basically molecules of DNA or RNA inside a protein shell , and in order to reproduce , it must infect a living cell and take control of it . However , the attacked cell will fight back and try to eliminate the invader . Activation of this so-called innate immune response requires the host cells to recognize that they have been infected , which they do by detecting the tell-tale molecules that indicate the presence of the virus . When an RNA virus infects a cell , the tell-tale molecules are often atypical RNA molecules carried by the virus or produced as the virus replicates . Recognition of this ‘foreign’ material by receptor proteins inside the cell triggers the production of molecules called interferons , which activate the innate defence systems that eliminate the virus . Different receptor proteins recognize different RNA viruses . For example , a receptor called MDA5 is known to respond to the picornaviruses , some of which can cause inflammation of the brain and heart muscle . However , the identities of the specific RNA molecules that are recognized by the MDA5 receptor have not been known . Deddouche et al . have now identified one such RNA molecule with the help of a second receptor protein , called LGP2 . The LGP2 receptor is not able to give the signal to produce interferons , so it is thought to work by binding to the RNA molecule to form a complex that is then relayed to MDA5 to give this signal . By isolating the complexes of LGP2 receptor from picornavirus-infected cells and sequencing the associated RNA , it was possible to identify the mystery RNA trigger . Deddouche et al . then tested picornaviruses in which this piece of RNA had been deleted from the genome , and found that the mutant viruses triggered a much weaker interferon response . By providing insight into the ways that some viruses are detected by the innate immune system , this research may inform future work on the development of new treatments to control viral infection .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2014	Identification of an LGP2-associated MDA5 agonist in picornavirus-infected cells
Bradykinin ( Bk ) is a potent inflammatory mediator that causes hyperalgesia . The action of Bk on the sensory system is well documented but its effects on motoneurons , the final pathway of the motor system , are unknown . By a combination of patch-clamp recordings and two-photon calcium imaging , we found that Bk strongly sensitizes spinal motoneurons . Sensitization was characterized by an increased ability to generate self-sustained spiking in response to excitatory inputs . Our pharmacological study described a dual ionic mechanism to sensitize motoneurons , including inhibition of a barium-sensitive resting K+ conductance and activation of a nonselective cationic conductance primarily mediated by Na+ . Examination of the upstream signaling pathways provided evidence for postsynaptic activation of B2 receptors , G protein activation of phospholipase C , InsP3 synthesis , and calmodulin activation . This study questions the influence of motoneurons in the assessment of hyperalgesia since the withdrawal motor reflex is commonly used as a surrogate pain model . The nanopeptide bradykinin ( Bk ) is an important mediator of pain and inflammation ( Dray and Perkins , 1993; Calixto et al . , 2000 ) . It causes hyperalgesia ( Manning et al . , 1991; Dalmolin et al . , 2007 ) by exciting and/or sensitizing components of the pain pathway , including primary afferent terminals , sensory ganglia , and dorsal horn neurons ( Dray and Perkins , 1988; Dray et al . , 1988; Thayer et al . , 1988; McGuirk and Dolphin , 1992; Rueff and Dray , 1993; Jeftinija , 1994; Cesare et al . , 1999; Wang et al . , 2005 ) . Consistent with this , systemic blockade of Bk receptors produces analgesia ( Steranka et al . , 1988; Correa and Calixto , 1993; Perkins et al . , 1993; Levy and Zochodne , 2000; Ferreira et al . , 2002 ) . Although much has been reported on the action of Bk on the sensory system , its effects on motoneurons remain under-explored . Earlier and indirect evidence suggested that Bk may transynaptically activate motoneurons through activation of transmitter release from primary nociceptive afferents ( Dunn and Rang , 1990 ) . However , kinin , cleaved to Bk by kallikrein in inflammatory responses , is extensively distributed throughout the motor areas of the CNS ( Walker et al . , 1995; Raidoo and Bhoola , 1998 ) and in particular in the ventral horn of the spinal cord ( Lopes and Couture , 1997; Li et al . , 1999 ) . Furthermore , Bk receptors have been detected in the membrane of spinal motoneurons ( Lopes et al . , 1995 ) suggesting that pain-related behaviors such as the withdrawal reflex may also arise from a direct Bk-evoked sensitization of motoneurons . The purpose of our study was to determine whether direct activation of Bk receptors sensitizes lumbar motoneurons of neonatal rats and , if so , to characterize the ionic mechanisms involved . From an in vitro hemicord preparation , the direct application of Bk ( 4–8 µM ) above the ventral horn column ( L3–L5 , Figure 1A ) evoked sustained ventral root activity without inducing antidromic discharge from the dorsal roots ( Figure 1B , left ) . This motor output was not exclusively caused by local reverberating circuits , because it persisted in the presence of the glutamate receptor antagonist kynurenate at a concentration that blocked the monosynaptic reflex ( 1 . 5 mM , Figure 1B , right and Figure 1C ) . We tested whether a change in the motoneuronal processing of sensory inputs by Bk might contribute to an increase in the motor output . To avoid sustained ventral root discharge , subthreshold concentrations of Bk ( 0 . 5–1 µM ) were used . With these low concentrations of Bk , the motoneuron spiking probability in response to supramaximal dorsal root stimulation was increased ( Figure 1D–F ) . Specifically , Bk had no effect on the transient short latency reflexes ( number of events: 144 ± 16 for control vs 143 ± 14 during Bk; p = 0 . 68 , Wilcoxon paired test , n = 7 ) but recruited a long-lasting reflex ( number of events: 537 ± 104 for control vs 2493 ± 1207 during Bk; p = 0 . 015 , Wilcoxon paired test , n = 7 ) such that the distribution of peristimulus time histograms ( PSTHs ) shifted from unimodal to bimodal ( Figure 1E , F ) . This long-lasting reflex has been previously shown to result from sustained firing of motoneurons involved in muscle spasms ( Bennett et al . , 2004; Li et al . , 2004 ) . 10 . 7554/eLife . 06195 . 003Figure 1 . Bradykinin potentiates the gain of sensory inputs into the motor system . ( A ) Drawing of a midsagittally hemisected rat spinal cord illustrating localized Bk application to the lumbar motor column , and dorsal ( DR ) and ventral ( VR ) roots used for reflex testing . ( B ) Responses to ventral application of bradykinin ( Bk , 4 µM pipette concentration ) recorded via the lumbar L5 dorsal ( DR L5 ) and ventral ( VR L5 ) roots before ( left ) and after ( right ) the fast glutamatergic synaptic transmission was blocked by kynurenate ( 1 . 5 mM ) . ( C ) Ventral root response to ipsilateral dorsal root stimulation before ( left ) and after ( right ) application of kynurenate ( 1 . 5 mM ) . Single arrows indicate the monosynaptic reflex ( mono ) and the stimulus artifact ( stim ) . ( D ) Five superimposed responses recorded from an L5 ventral root induced by stimulations of the ipsilateral dorsal root before ( control , black trace ) and after local application of low concentrations of Bk ( 1 µM pipette concentration , red trace ) . ( E ) Average peristimulus time histogram ( PSTH , bin width: 20 ms ) of L5 ventral root recordings collected before ( black ) and after ( red ) the local application of Bk ( 1 µM pipette concentration ) . ( F ) Group means quantification of the PSTH for the transient short latency and long-lasting reflexes computed over a window 10–40 ms and 500–4000 ms post stimulus , respectively , before ( black ) and after ( red ) local application of Bk . Error bars indicate SEM . p < 0 . 05 ( Wilcoxon paired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 00310 . 7554/eLife . 06195 . 004Figure 1—figure supplement 1 . At the right , five superimposed responses recorded under spantide ( 2 µM ) from the L5 ventral root of an hemichord preparation and induced by stimulations of the ipsilateral dorsal root before ( control , black trace ) and after local application of low concentrations of Bk ( 1 µM pipette concentration , red trace ) . At the left , group means quantification of the PSTH for the transient short latency and long-lasting reflexes computed over a window 10–40 ms and 500–4000 ms post stimulus , respectively , before ( black ) and after ( red ) local application of Bk . Error bars indicate SEM . ns , not significant , p < 0 . 05 ( Wilcoxon paired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 004 To explore a potential sensitization of the firing properties of lumbar motoneurons by Bk , we recorded motoneurons in whole cell configuration from in vitro slice preparations ( Table 1 and Figure 2 ) . From a resting potential adjusted to −70 mV with bias current , an incrementing series of hyperpolarizing and depolarizing pulses was delivered before and after bath application of Bk ( 4–8 µM ) . A few minutes ( 2–3 min ) after the application of Bk , the major effect was an increase in motoneuron excitability , as reflected by a lower rheobase ( 528 ± 118 pA for control vs 299 ± 156 pA during Bk; p = 0 . 028 , Wilcoxon paired test , n = 10 cells; Table 1 and Figure 2A , B ) . As a consequence , Bk evoked a leftward shift of the f-I curve to lower current values and a slight increase in the slope of the f-I curve ( 0 . 05 ± 0 . 004 Hz/pA for control vs 0 . 06 ± 0 . 006 Hz/pA during Bk; p = 0 . 027 , Wilcoxon paired test , n = 10 cells; Table 1 and Figure 2B ) . Thus , current steps that elicited only a single spike prior to Bk typically induced repetitive spiking during Bk application ( Figure 2A ) . The higher motoneuron excitability was associated with an apparent increase in input resistance ( 55 . 6 ± 5 . 4 MΩ for control vs 62 . 9 ± 7 . 2 MΩ during Bk; p = 0 . 004 , Wilcoxon paired test , n = 10 cells; Table 1 ) as seen by the increased slope of the voltage responses to hyperpolarizing pulses ( Figure 2A , C ) . Bk lowered the threshold for action potential generation ( −50 . 0 ± 2 . 1 mV for the control vs −52 . 8 ± 2 . 5 mV during Bk; p = 0 . 009 , Wilcoxon paired test , n = 10 cells; Table 1 ) which , although not wider ( 0 . 51 ± 0 . 02 ms for control vs 0 . 51 ± 0 . 02 ms during Bk; p = 0 . 77 , Wilcoxon paired test , n = 10 cells; Table 1 ) , was smaller in amplitude ( 67 . 0 ± 2 . 3 mV for control vs 63 . 6 ± 2 . 2 mV during Bk; p = 0 . 049 , Wilcoxon paired test , n = 10 cells; Table 1 and Figure 2D ) . Hyperpolarizing current steps evoked a marked depolarizing sag ( inward rectification ) which is often associated with activation of the hyperpolarization-activated inward current Ih ( Figure 2A ) ; the amplitude of the sag was not affected by Bk ( 12 ± 2 . 6% for control vs 12 . 3 ± 2 . 5% during Bk; p = 0 . 95 , Wilcoxon paired test , n = 10 cells; Table 1 ) . However , the medium afterhyperpolarization ( mAHP ) was enhanced both in amplitude and duration ( amplitude: −9 . 1 ± 0 . 9 mV for the control vs −11 . 2 ± 1 . 5 mV for Bk , p = 0 . 02; duration: 41 . 1 ± 2 . 5 ms for control vs 48 . 6 ± 3 . 4 ms for Bk , p = 0 . 014; Wilcoxon paired test , n = 10 cells; Table 1; Figure 2E ) . In a recent publication , we demonstrated that at temperatures above 30°C , neonatal rat lumbar motoneurons show marked bistability , characterized by their ability to generate a slow afterdepolarization ( sADP ) that outlasted a brief high amplitude depolarizing pulse ( Bouhadfane et al . , 2013 ) . Bk increased the amplitude of the post-stimulus sADP ( 11 . 2 ± 1 . 4 mV for control vs 18 ± 2 . 6 mV for Bk; p = 0 . 008 , Wilcoxon paired test , n = 10 cells; Table 1 ) and triggered self-sustained spiking in all motoneurons tested ( Figure 2F ) . These results show that Bk sensitizes motoneurons by increasing their excitability and promoting bistability . 10 . 7554/eLife . 06195 . 005Table 1 . Effects of bradykinin on passive and active membrane properties of lumbar motoneuronsDOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 005ControlBradykininN1010Rm ( MΩ ) 55 . 6 ± 5 . 462 . 9 ± 7 . 2AP amp ( mV ) 67 . 0 ± 2 . 363 . 7 ± 2 . 2AP dur ( ms ) 0 . 51 ± 0 . 020 . 51 ± 0 . 02AP threshold ( mV ) −50 . 0 ± 2 . 1†−52 . 8 ± 2 . 5f-I slope ( Hz/pA ) 0 . 05 ± 0 . 0040 . 06 ± 0 . 006Rheobase ( pA ) 528 ± 118299 ± 156sADP ( mV ) 11 . 2 ± 1 . 4†18 ± 2 . 6AHP amp ( mV ) −9 . 1 ± 0 . 9−11 . 2 ± 1 . 5AHP dur ( ms ) 41 . 1 ± 2 . 548 . 6 ± 3 . 4Sag ( % ) 12 ± 2 . 612 . 3 ± 2 . 5Statistical significance was assessed by a Wilcoxon paired test . p < 0 . 05 , †p < 0 . 01 , n = number of cells . Mean firing frequency was measured at two times the rheobase . AHP = afterhyperpolarization . 10 . 7554/eLife . 06195 . 006Figure 2 . Bradykinin enhances repetitive firing and promotes self-sustained spiking . ( A–C ) Typical responses of a motoneuron to incrementing 1-s current injections ( A ) with its respective frequency–current ( B ) and voltage–current ( C ) relationships before ( black traces ) and after ( red traces ) bath applying bradykinin ( Bk , 8 µM ) . Large hyperpolarizing currents revealed the presence of an inward rectification causing a depolarizing ‘sag’ ( arrowheads ) . Initial potential was held at −70 mV . Note that a negative current injection was employed to counter the Bk-induced depolarization so that Bk measurements were also made at −70 mV . ( D ) Action potentials aligned in time on their peaks before and during Bk . ( E ) Afterhyperpolarization following a single action potential ( truncated ) . Initial holding potential , −60 mV . ( F ) Superimposed voltage traces recorded in response to a 2-s depolarizing pulse from a holding potential of −60 mV before ( black trace ) and after ( red trace ) bath applying Bk ( 8 µM ) . Bottom traces in A , E , and F are the injected current protocol . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 006 To test whether Bk directly excites motoneurons , we recorded activity from 200 lumbar motoneurons in response to a brief ( ≤2 min ) bath application of Bk ( 8 µM ) . Among them , 154 ( 77% ) displayed a reversible membrane depolarization ( Figure 3A ) while the remaining 46 had no responses . The onset of the depolarization was slow , taking close to 1 min to peak , and subsided slowly over 3–5 min during washout . During the peak of the depolarizing response , spikelets , indicative of electrical coupling to other neurons , were sometimes seen ( Figure 3A , inset ) . The amplitude of the depolarization was not changed after blockade of glutamatergic input with kynurenate ( 1 . 5 mM ) or by CNQX ( 10 µM ) and AP-5 ( 20 µM ) ( 92 . 9 ± 17 . 8% of the control; p = 0 . 62 , Wilcoxon paired test , n = 4 cells; Figure 2A , right ) . These effects were thus not exclusively caused by a feed-forward network of excitatory synaptic input to motoneurons , but rather probably involved a direct postsynaptic action . Supporting this idea , the Bk-induced depolarization persisted in the presence of 0 . 5–1 μM TTX ( 12 . 9 ± 0 . 6 mV; n = 154; Figure 3B ) . The effects of Bk were fully reversible without tachyphylaxis; when Bk was repetitively applied at intervals of 20 min , the amplitude of the second response was not significantly reduced from the first response ( 95 . 6 ± 13 . 9% of the first response; p = 0 . 22 , Wilcoxon paired test , n = 7 cells; Figure 3B ) . Previous indirect evidence suggested that Bk's actions on lumbar motoneurons were secondary to the activation of NK1 receptors by substance P released from primary afferent C-fibers ( Dunn and Rang , 1990 ) . To test this hypothesis , we added Bk in the presence of spantide ( 2–5 µM ) , a selective NK1 receptor antagonist ( IC50 = 1 . 5 µM; Merali et al . , 1988 ) , that has been shown to depress the responses of neonatal rat motoneurons to substance P ( Yanagisawa and Otsuka , 1990 ) . Spantide did not attenuate Bk-induced responses ( response in spantide 92 . 1 ± 18% of control response; p = 0 . 62 , Wilcoxon paired test , n = 4 cells; see Figure 3—figure supplement 1 ) . Note that the long-lasting reflex induced by Bk from an in vitro hemicord preparation was not occluded by spantide ( 2 µM ) ( number of events: 580 ± 58 for control vs 1859 ± 549 during Bk; p = 0 . 03 , Wilcoxon paired test , n = 6; see Figure 1—figure supplement 1 ) . Together , these results argue for a likely direct effect of Bk on lumbar motoneurons . 10 . 7554/eLife . 06195 . 007Figure 3 . Bradykinin depolarizes lumbar motoneurons by a direct postsynaptic action of B2 receptors . ( A ) Voltage trace in response to bradykinin ( Bk ) collected under kynurenate ( 1 . 5 mM ) . The asterisk indicates the point shown by the trace at right with higher temporal resolution , where spikelets occurred . ( B ) Sequential depolarizations recorded under TTX ( 1 µM ) induced by two successive applications of Bk ( Bk , 8 µM ) with intervals of 20 min ( C ) Voltage traces collected under TTX during application of either Lys-[Des-Arg9]-Bk , a selective B1 receptor agonist ( red trace ) , or [Hyp3]-Bk , a selective B2 receptor agonist ( black trace ) . ( D ) Voltage traces collected under TTX in response to bradykinin ( Bk ) , before ( black ) and after ( red ) pretreatment with the selective B2 receptor antagonist HOE-140 . At the right of each panel , graphs show the mean peak amplitude of membrane depolarizations induced by Bk or by one of its agonists . Drug application periods are indicated by lines above the records . Error bars indicate SEM . ns , not significant , p < 0 . 05 . ( A , B , D: Wilcoxon paired test; C: Mann–Whitney test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 00710 . 7554/eLife . 06195 . 008Figure 3—figure supplement 1 . At the left , superimposed voltage traces recorded from a L5 motoneuron in response to bradykinin ( Bk ) collected under TTX ( 0 . 5 µM ) before and after the application of spantide ( 2 µM ) . At the right , histogram plotting the peak amplitude of membrane depolarizations induced by Bk before ( black ) and after ( red ) the superfusion of the medium . ns , not significant ( Wilcoxon paired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 008 Bk exerts its biological effects through the activation of two receptors , called B1 and B2 ( Hall , 1992 ) . To determine which receptor subtype ( s ) mediate Bk's effects on motoneurons , we first examined the ability of selective agonists to reproduce the depolarization . Under TTX , the selective B1 agonist , Lys-[Des-Arg9]-Bk ( Ki = 1 . 7 nM; Marceau et al . , 1998 ) , at concentrations as high as 3 µM , failed to induce a significant depolarization ( 2 . 2 ± 0 . 6 mV , n = 10 cells; Figure 3C ) . In contrast , the selective B2 receptor agonist [Hyp]3-Bk ( Ki = 314 pM; Windischhofer and Leis , 1997 ) , used at 2 µM evoked a membrane depolarization similar to responses evoked by Bk ( 11 . 2 ± 1 . 5 mV n = 10 cells for [Hyp]3-Bk vs 12 . 9 ± 0 . 6 mV , n = 154 for Bk; p = 0 . 48 , Mann–Whitney test; Figure 3C ) . Consistent with an involvement of B2 receptors , the response to Bk was almost abolished by 2 µM HOE 140 ( 11 . 7 ± 2 . 7% of the control depolarization; p = 0 . 03 , Wilcoxon paired test , n = 6 cells; Figure 3D ) , a potent and selective B2 receptor antagonist ( IC50 = 420 pM; Eggerickx et al . , 1992 ) . Together , these results strongly suggest that Bk depolarizes motoneurons mainly via activation of B2 receptors . To determine the ionic mechanisms involved , motoneurons were held by voltage-clamp at −70 mV , a potential close to the normal resting membrane potential ( Tazerart et al . , 2007 ) . To investigate the current in relative isolation , TTX ( 0 . 5 µM ) was added in the aCSF . Under these conditions , Bk reversibly generated an inward current whose time course was similar to that seen in current-clamp recordings ( Figure 4A ) . Part of the Bk response seems thus independent of voltage-gated channels . The inward current amplitude was steeply dependent on Bk concentration , with an EC50 of 680 nM and maximal amplitude ( 322 ± 26 pA , n = 14 ) at 4–8 μM ( Figure 4B ) . The peak current density was 1 . 5 ± 0 . 15 pA/pF ( n = 14 ) . In the following experiments , the Bk-sensitive current was investigated by means of slow voltage ramp to minimize inactivating voltage-gated currents . To ensure a maximum response and to avoid potential fluctuation of the effective peptide concentration , Bk was applied at 8 µM . A significant Bk-induced decrease in the slope conductance was observed in I/V curves ( 18 . 7 ± 2 . 4 pA/mV before Bk vs 15 . 6 ± 2 . 2 pA/mV during Bk; p = 0 . 016 , Wilcoxon paired test , n = 7 cells; Figure 4C ) , an effect often interpreted as a consequence of the inhibition of a K+ conductance . Consistent with this interpretation , the linear I/V relationship of the Bk-sensitive-current ( obtained by subtracting the control from the Bk currents: Figure 4C , orange line with open circles ) reversed at an extrapolated value close to the predicted equilibrium potential for K+ ( Ek: −100 mV ) . The combination of traditional voltage-gated K+ channel blockers [TEA ( 10 mM ) , 4-aminopyridine ( 2 mM ) , the HCN-channel blocker ZD7288 ( 20 µM ) ] and the Ca2+-activated K+ channel blocker apamin ( 100 nM ) , did not change either the Bk-induced decrease in slope conductance ( 33 . 2 ± 3 . 2 pA/mV before Bk vs 21 . 1 ± 2 . 3 pA/mV during Bk; p = 0 . 02 , Wilcoxon paired test , n = 15 cells; Figure 4D1 ) or the peak depolarization magnitude measured in current clamp from −70 mV ( 94 . 5 ± 15 . 8% of the control; p = 0 . 54 , Wilcoxon paired test , n = 7 cells; Figure 4D2 ) . The contribution of K+ ions was further examined by raising the extracellular K+ ( [K+]o ) to 9 mM , a manipulation that almost eliminates the driving force for K+ at resting membrane potential by setting Ek at ∼ −70 mV according the Nernst equation . Under high [K+]o , Bk still induced a depolarization from −70 mV without a measurable change in input resistance ( 45 . 3 ± 13 . 9 MΩ before Bk vs 46 . 5 ± 13 . 2 MΩ during Bk; p = 0 . 81 , Wilcoxon paired test , n = 7 cells ) . However , the depolarization was significantly smaller than that seen in normal aCSF ( 64 . 2 ± 17 . 7% of the control value measured in normal aCSF; p = 0 . 047 , Wilcoxon paired test , n = 7 cells; Figure 4E ) . These data suggest that a second ionic component contributes to the Bk depolarization in addition to a reduction in K+ conductances . The residual Bk-induced depolarization observed in high [K+]o appears to be sodium-mediated because it disappeared after reducing [Na+]o from 152 to 26 mM with equimolar choline-chloride ( 4 . 1 ± 2 . 8% of the control value measured in high [K+]o; p = 0 . 016 , Wilcoxon paired test , n = 7 cells; Figure 4F ) . In sum , our data indicate that there are at least two ionic components in the response to Bk: a K+ component that produces an input resistance increase and likely involves a reduction of a resting K+ conductance , and a Na+ component that produces depolarization without measurable change in input resistance . 10 . 7554/eLife . 06195 . 009Figure 4 . Bradykinin inhibits a leak K+ current and activates a Na+-dependent nonselective cationic current . ( A ) Representative inward current induced by bradykinin ( Bk , 8 µM ) in lumbar motoneuron . Voltage clamp , Holding potential , −70 mV . ( B ) Dose-response curve of Bk-induced changes in peak holding current . Numbers in the parenthesis are number of cells recorded . ( C ) At the left , I–V relationships reconstructed from voltage ramp data under TTX ( 0 . 5 µM ) before ( black trace ) and at the peak of the response to bath-applied Bk ( red trace ) . The orange trace with open circles illustrates the I–V relation of the difference current representing the Bk-sensitive current . At the right , current traces from a representative cell and histogram plotting changes in the slope conductance induced by Bk . ( D1 ) At the left , I–V relationships reconstructed from voltage ramp data in a medium containing TTX ( 0 . 5 µM ) , TEA ( 10 mM ) , 4-aminopyridine ( 2 mM ) , the HCN-channel blocker ZD7288 ( 20 µM ) , and apamin ( 100 nM ) before ( black trace ) and after ( red trace ) a bath application of Bk ( 8 µM ) . The orange trace with open circles illustrates the mean I–V relation of the difference current representing the Bk-sensitive current . At the right , current traces from a representative cell and histogram plotting changes in the slope conductance induced by Bk . ( D2 ) Superimposed voltage traces under TTX ( 0 . 5 µM ) and TEA ( 10 mM ) in response to Bk before ( black ) and after ( red ) the superfusion of a medium containing TEA ( 10 mM ) , 4-aminopyridine ( 2 mM ) , the HCN-channel blocker ZD7288 ( 20 µM ) and apamin ( 100 nM ) . ( E ) Superimposed voltage traces under TTX ( 0 . 5 µM ) and TEA ( 10 mM ) in response to Bk before ( black ) and after ( red ) the superfusion of a medium with high [K+]o ( 9 mM ) . ( F ) Superimposed voltage traces under TTX ( 0 . 5 µM ) , TEA ( 10 mM ) , and high [K+]o ( 9 mM ) in response to Bk before ( black ) and after ( red ) the superfusion of a medium with low [Na+]o . At the right of each panel , histogram plotting the peak amplitude of membrane depolarizations induced by Bk before ( black ) and after ( red ) the superfusion of the medium . Drug application periods are indicated by horizontal lines . Error bars indicate SEM , ns , not significant , p < 0 . 05 ( Wilcoxon paired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 009 The foregoing results suggest that reduction of a resting K+ conductance associated with an increase of the input resistance might help Bk to generate the slow depolarization . To test the contribution of a resting K+ conductance , we used quinidine ( 200–400 µM ) , a broad-spectrum blocker of neuronal background K+ channels ( IC50 ≈ 100 µM; Lesage and Lazdunski , 2000 ) . Quinidine induced alone a large depolarization ( +6 . 8 ± 3 . 9 mV , n = 6 cells , not shown ) and reduced the slope conductance itself ( 22 ± 3 . 8 pA/mV before quinidine vs 14 . 2 ± 3 . 4 pA/mV during quinidine; p < 0 . 01 , One-way ANOVA test , n = 6 cells; Figure 5A1 ) . Further , quinidine prevented the slope conductance decrease induced by Bk ( 14 . 2 ± 3 . 4 pA/mV for control vs 13 ± 3 pA/mV during Bk; p > 0 . 05 , One-way ANOVA test , n = 6; Figure 5A1 ) and partially occluded the Bk-induced membrane depolarization ( 57 . 7 ± 21 . 5% of control; p = 0 . 03 , Wilcoxon paired test , n = 6; Figure 5A2 ) . These results show that the blockade of a resting K+ conductance can both mimic and partially occlude the effects of Bk . 10 . 7554/eLife . 06195 . 010Figure 5 . Pharmacological profile of the bradykinin-induced current . ( A1–C1 ) At the left , superimposed I–V relationships reconstructed from voltage ramp data ( representative data in inserts ) before ( black trace ) and after ( red trace with filled circles ) quinidine ( 200 µM , A1 ) , extracellular acidosis ( B1 ) , or barium ( 300 µM , C1 ) . In each panel the orange trace with open circles illustrates the mean I–V relation obtained during bradykinin application ( Bk , 8 µM ) under quinidine ( A1 ) , extracellular acidosis ( B1 ) , or barium ( C1 ) . At the right of each panel , histogram plotting changes in the slope conductance . ( A2–C2 ) At the left , superimposed voltage traces in response to Bk before ( black ) and after ( red ) quinidine ( 200 µM , A2 ) , extracellular acidosis ( B2 ) , or barium ( 300 µM , C2 ) . At the right of each panel , histogram plotting the peak amplitude of membrane depolarizations induced by Bk . ( D–E ) At the left , superimposed voltage traces under high [K+]o in response to Bk before ( black ) and after ( red ) applying ruthenium red ( 200 µM , D ) or lowering temperature from 34 to 24°C ( E ) . At the right of each panel , histogram plotting the peak amplitude of membrane depolarizations induced by Bk . All recordings were performed under TTX ( 1 µM ) and TEA ( 10 mM ) . Error bars indicate SEM , p < 0 . 05 , p < 0 . 01 ( A1 , B1 , C1: One-Way ANOVA; A2 , B2 , C2 , D , E: Wilcoxon paired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 010 We further characterized the nature of the resting K+ conductance modulated by Bk . Because a fraction of background K+ channels is downregulated by extracellular acidosis ( Talley et al . , 2000; Patel and Honore , 2001 ) , motoneurons were exposed to Bk before and after lowering the pH from 7 . 4 to 6 . 4 . As expected , acidification induced a membrane depolarization ( +4 . 3 ± 0 . 6 mV , n = 5 cells , not shown ) but did not significantly decrease the slope conductance itself ( 26 . 4 ± 6 . 9 pA/mV at pH 7 . 4 vs 21 . 2 ± 5 . 2 pA/mV at pH 6 . 4; p > 0 . 05 , One-Way ANOVA test , n = 5 cells; Figure 5B1 ) . Furthermore , the response to Bk was not affected by acidification to pH 6 . 4 , that is , the slope conductance decrease ( 21 . 2 ± 5 . 2 pA/mV for control vs 11 . 2 ± 3 . 6 pA/mV during Bk; p < 0 . 05 , One-Way ANOVA test , n = 5 cells; Figure 5B1 ) and the membrane depolarization from −70 mV were not affected ( 92 . 6 ± 16 . 0% of values at pH 7 . 4; p = 0 . 69 , Wilcoxon paired test , n = 6 cells; Figure 5B2 ) . We next tested Ba2+ , which is known to block a subset of resting K+ currents . The Bath application of Ba2+ ( 300 µM ) caused a significant depolarization ( +6 . 6 ± 0 . 8 mV , n = 6 cells , not shown ) and also by itself decreased the slope conductance ( 29 . 3 ± 3 . 2 pA/mV for control vs 22 ± 2 . 6 pA/mV during barium; p < 0 . 01 , One-Way ANOVA test , n = 6 cells; Figure 5C1 ) . Ba2+ reduced but did not prevent the slope conductance decrease induced by Bk ( 22 ± 2 . 6 pA/mV for control vs 18 . 1 ± 2 . 4 pA/mV during Bk; p < 0 . 05 , One-Way ANOVA test , n = 6; Figure 5C1 ) , and it occluded the Bk-induced membrane depolarization from −70 mV ( 58 . 6 ± 17 . 7% of control; p = 0 . 03 , Wilcoxon paired test , n = 6 cells; Figure 5C2 ) . In sum , it appears that a Ba2+-sensitive , pH-insensitive resting K+ conductance accounts for a substantial fraction of the K+ component modulated by Bk . In addition to Bk's inhibition of a resting K+ current , we also characterized the activation of the conductance carried by Na+ . We recently characterized a thermosensitive nonselective cationic current ( ICaN ) in lumbar motoneurons of neonatal rats ( Bouhadfane et al . , 2013 ) . To determine the contribution of ICaN to the Bk-evoked depolarization , we examined the effect of ruthenium red , routinely used to block channels that engender ICaN ( Wu et al . , 2010 ) . In high [K+]o , the Bk-induced depolarization from −70 mV ( where the effect of the resting K+ current is minimized ) was nearly abolished by 10 µM ruthenium red ( 16 . 6 ± 6 . 5% of control response; p = 0 . 03 , Wilcoxon paired test , n = 6 cells; Figure 5D ) . This apparent ICaN was also thermosensitive because lowering the temperature from 34 to 24°C induced a significant decrease in the Bk-induced depolarization at −70 mV in high [K+]o ( 26 . 2 ± 9 . 8% of control values; p = 0 . 03 , Wilcoxon paired test; Figure 5E ) . In sum , it appears that a current with properties of a thermosensitive ICaN might account for part of the Bk-induced depolarization . Increases in the concentration of free cytosolic Ca2+ ( [Ca2+]i ) are fundamental in recruiting ICaN from neonatal rat lumbar motoneurons ( Bouhadfane et al . , 2013 ) . We therefore examined whether an increase in intracellular calcium ( [Ca2+]i ) was required for the Bk-induced depolarization . To demonstrate a Bk-induced rise of [Ca2+]i in motoneurons , we first loaded the ventral horn with the calcium indicator Oregon Green-BAPTA 1 AM ( OGB1-AM ) along with the glial cell marker Sulforhodamine-101 ( SR-101 ) using the extracellular bolus loading technique ( see ‘Materials and methods’ ) . Since the astroglial cells were labeled by both dyes , they appeared as yellow cells , while neurons were labeled by green OGB1-AM only . This allowed the clear separation of neurons from astroglial cells . After a brief equilibration period , we imaged calcium changes from somata of lumbar motoneurons using a three-dimensional random-access two-photon microscope ( Figure 6A ) . Bk induced a rise in [Ca2+]i ( 13 . 4 ± 0 . 7% ΔF/F ) in 65% of motoneuron somata ( n = 177 out of 272 cells; n = 5 slices ) , with very slow decay times ( Figure 6B , C ) ; there was no change in [Ca2+]i in the remaining motoneurons . In the next step , we explored the relationship between spatio-temporal changes in [Ca2+]i in motoneuron somata and dendrites along with the Bk-induced membrane depolarization . Here , the patch pipette was filled with the cell-impermeable calcium indicator OGB1 ( 120 µM ) and a morphometric dye , Alexa 594 ( 20 µM ) to allow simultaneous two-photon imaging and anatomical tracing of the cellular processes ( Figure 6D ) . After Bk application , the proximal dendrites up to 50 µm away from the soma showed a clear rise in [Ca2+]i ( 49 . 8 ± 8 . 7% ΔF/F; n = 5 cells; Figure 6E , F; red trace ) , which was followed with a delay by the somatic response ( 44 . 5e ± 8 . 9% ΔF/F; n = 5 cells; Figure 6E , F; blue trace ) . All of the above results were obtained with blocked voltage-gated sodium and potassium channels ( 0 . 5 µM TTX and 10 mM TEA , respectively ) , showing the Bk-induced response alone , without spiking activity of the motoneurons . To obtain a physiologically relevant comparison to the Bk-induced responses , we imaged [Ca2+]i changes induced by action potentials evoked by depolarizing pulses in the absence of TTX and TEA ( Figure 6G ) . Our results show that a single action potential induced a [Ca2+]i rise significantly higher in the proximal dendrites than in the soma ( Casoma = 6 . 1 ± 1 . 5% ΔF/F; Cadendrite = 16 . 9 ± 3 . 6% ΔF/F; n = 5 cells; p = 0 . 02 , Student's paired t-test ) . The Bk-induced rise in [Ca2+]i was significantly higher than that evoked by a single action potential in both the soma and the dendrites ( p = 0 . 003 and p = 0 . 008 , Student's unpaired t-test ) . In conclusion , the data obtained with calcium imaging show that Bk induces a significant increase in [Ca2+]i . 10 . 7554/eLife . 06195 . 011Figure 6 . The Bk-induced depolarization of lumbar motoneurons is associated with [Ca2+]i rise in dendrites . ( A ) z-projection of two-photon image stack showing cells loaded by bolus injection of OGB1-AM ( 1 mM ) and SR-101 ( 300 µM ) . ( B ) Bradykinin ( Bk , 8 µM ) increases calcium levels in the motoneuron somata . Each line is a different motoneuron , aligned to the time of Bk application , see color code ( right ) for quantification of increase in calcium fluorescence ( C ) Time course of the mean ( ±SEM ) population calcium response to Bk application; n = 177 neurons . ( D ) z-projection of a two-photon image stack showing a representative motoneuron loaded with OGB1 ( 120 µM ) and Alexa 594 ( 20 µM ) . ( E ) Simultaneously recorded dendritic ( red trace ) and somatic ( blue trace ) calcium transients of the motoneuron shown on ( D ) . ( F ) Average dendritic ( red trace ) and somatic ( blue trace ) calcium responses along with the simultaneously recorded average voltage trace ( black trace ) acquired from the patched cells . n = 5 neurons . ( G ) Average somatic ( blue trace ) and dendritic ( red trace ) calcium traces of 5 sweeps in response to action potentials evoked by somatic current injection . Inset , points of interest on soma and dendrites ( black and white diamonds , respectively ) . ( H–I ) At the left , superimposed voltage traces under TTX ( 0 . 5 µM ) and TEA ( 10 mM ) in response to Bk ( Bk , 8 µM ) before ( black ) and after ( red ) the superfusion of ( H ) cadmium ( 200 µM , H ) or intracellular diffusion of BAPTA ( 10 mM , I ) . At the right of each panel , histogram plotting the peak amplitude of membrane depolarizations induced by Bk before ( black ) and after ( red ) the drug application . Error bars indicate SEM , p < 0 . 01 ( Wilcoxon paired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 011 We dissected the sources of the Bk-induced increase in [Ca2+]i . Calcium can enter the cytoplasm via several routes including the NMDA-receptor channels , voltage-gated calcium channels and release from internal calcium stores . Blockade of NMDA-receptor channels with kynurenate ( 1 . 5 mM ) or CNQX ( 10 µM ) and AP-5 ( 20 µM ) did not prevent the depolarization induced by Bk ( Figure 3B ) . Similarly , the effects of Bk on motoneurons were not significantly reduced by cadmium blockade of voltage-gated calcium channels ( 100 µM; 83 ± 9 . 7% , of control depolarization; p = 0 . 46 , Wilcoxon paired test , n = 8 cells; Figure 6H ) . The dependence of the depolarization on [Ca2+]i was then studied by recording motoneurons after intracellular perfusion of the Ca2+ chelating agent BAPTA ( 10 mM ) from the pipette solution . In these experiments , we applied Bk as soon as possible after establishing whole-cell access , so that we could record a control response before full diffusion of BAPTA into the cell . The first application caused a robust depolarization similar to that in the control cells ( 12 . 7 ± 2 . 4 mV , n = 8 cells vs 12 . 9 ± 0 . 6 mV for control , n = 154 cells , p = 0 . 94 , Mann–Whitney test ) . 20 minutes after establishing the whole cell recording , the depolarization induced by Bk declined to 28 . 8 ± 6 . 4% of its original value ( p = 0 . 008 , Wilcoxon paired test , n = 8 cells; Figure 6I ) . Combined with our calcium imaging results , these experiments suggest that the Bk-induced increase in [Ca2+]i observed in our experiments was likely caused by Ca2+ released from intracellular stores rather than by Ca2+ influx through Ca2+-permeable membrane channels . To test whether a G-protein-mediated pathway is involved , we diffused the nonhydrolyzable GDP analog GDPβS ( 2 mM ) , a broad spectrum G-protein inhibitor , from the pipette . Immediately after breakthrough , before the passive loading of GDPβS into the cell , the first application of Bk induced the expected transient slow depolarization ( 15 . 7 ± 2 . 4 mV , n = 6 cells ) . After diffusion of GDPβS into the cell , the subsequent application of Bk failed to induce a depolarization ( 9 . 7 ± 5 . 7% of control , p = 0 . 03 , Wilcoxon paired test , n = 6 cells; Figure 7A ) . Bk receptors are generally described as signaling through Gq α-subunit ( Gutowski et al . , 1991; LaMorte et al . , 1993 ) , but they can also interact with other G proteins such as Gs ( Liebmann et al . , 1996 ) and Gi α-subunits ( Ewald et al . , 1989; Yanaga et al . , 1991 ) . We showed that Bk signaling is independent of Gi or Gs protein by showing that after 7 hr of preincubation with either the Gi inhibitor pertussis toxin ( PTX , 2 µg . ml−1 ) or the Gs inhibitor cholera toxin ( CTX , 2 µg . ml−1 ) the magnitude of Bk responses was not disturbed ( CTX: 14 . 7 ± 2 . 7 mV , n = 5 cells , p = 0 . 6; PTX: 16 . 7 ± 1 . 8 mV , n = 4 cells p = 0 . 3; control: 12 . 9 ± 0 . 6 mV , n = 54 cells , Mann–Whitney test; data not shown ) . These data suggest the contribution of Gq to the transduction process . 10 . 7554/eLife . 06195 . 012Figure 7 . Signal transduction mechanism underlying the Bk effects . ( A–F ) At the left , superimposed voltage traces under TTX ( 0 . 5 µM ) and TEA ( 10 mM ) in response to bradykinin ( Bk , 8 µM ) before ( black ) and after ( red ) GDPßS ( 2 mM , A ) , U73122 ( 10 µM , B ) , Xestospongin C ( 2 . 5 mM , C ) , Chelerythrine ( 10 µM , D ) , H89 ( 10 µM , E ) or W-7 ( 100 µM , F ) . At the right of each panel , histogram plotting the peak amplitude of membrane depolarizations induced by Bk before ( black ) and after ( red ) the drug application . Error bars indicate SEM , *p < 0 . 05 ( Wilcoxon paired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 012 Classically , Gq proteins are coupled to a phospholipase C ( PLC ) signaling pathway ( Selbie and Hill , 1998 ) . To test whether Bk acts by this pathway , we used the PLC inhibitor U73122 ( IC50 = 5 µM; Bleasdale et al . , 1990 ) . U73122 ( 10 µM ) greatly reduced Bk-induced depolarization ( 24 . 6 ± 4% of control; p = 0 . 03 , Wilcoxon paired test , n = 6 cells; Figure 7B ) . The PLC pathway bifurcates to generate inositol 1 , 4 , 5-trisphosphate ( InsP3 ) and calcium release from InsP3-sensitive intracellular stores on the one hand , and production of diacylglycerol ( DAG ) which activates protein kinase C ( PKC ) on the other . We tested the role of InsP3 by adding xestospongin C , an alkaloid InsP3 receptor antagonist ( IC50 = 0 . 4 µM; Gafni et al . , 1997 ) . Intrapipette loading with xestospongin ( 2 . 5 mM ) induced a significant reduction of the Bk-induced depolarization ( 38 . 8 ± 17 . 4% of control , p = 0 . 015 , Wilcoxon paired test , n = 8 cells , Figure 7C ) . In contrast , bath application of chelerythrine ( 10 µM ) or intracellular diffusion of PKC 19–36 ( 10 µM ) , two potent inhibitors of PKC ( IC50 = 0 . 66 µM and 0 . 3 µM respectively; Herbert et al . , 1990; Smith et al . , 1990 ) , did not reduce the Bk response ( pooled data:10 . 5 ± 17 . 8% of control , p = 1 , Wilcoxon paired test , n = 5 cells 3 with chelerythrine and 2 with PKC 19–36 , Figure 7D ) indicating that the DAG pathway is not involved . Similar results were collected with either 10 µM bath application of H89 or 10 µM intracellular administration of PKA 6–22 ( 95 . 9 ± 12 . 9% of control , p = 0 . 31 , Wilcoxon paired test , n = 6 cells 3 with H89 and 3 with PKA 6–22 , Figure 7E ) , two potent blockers of PKA ( IC50 = 0 . 14 µM and 2 nM respectively; ( Glass et al . , 1989; Davies et al . , 2000 ) . Multiple Ca2+-sensing proteins detect the elevation of free cytosolic Ca2+ and become downstream elements in intracellular calcium signaling pathways . Among them , calmodulin ( CaM ) , a small cytoplasmic Ca2+-sensing protein , regulates many ion channels in a Ca2+-dependent manner . Supporting a role for CaM in the Bk response , application of W-7 ( 100 µM ) , a membrane permeable CaM inhibitor ( IC50 ≈ 30 µM; Hidaka et al . , 1981 ) , abolished the Bk-induced depolarization ( 3 . 1 ± 5%; p = 0 . 02 , Wilcoxon paired test , n = 6 cells; Figure 7F ) . This result , and the requirement for intracellular calcium build-up , suggested that the Bk-induced depolarization may be mediated by Ca2+/CaM-dependent protein kinase II ( CaMKII ) . To test this hypothesis , we inhibited CaMKII by preincubating the neurons ( ≥90 min ) with KN-93 ( 10 μM ) , a potent CaMKII inhibitor ( IC50 = 0 . 4 µM; Sumi et al . , 1991 ) . However , the Bk-evoked motoneuron depolarization was not significantly affected by KN-93 ( KN-93: 10 . 9 ± 2 . 4 mV , n = 6 cells; Control: 12 . 9 ± 0 . 6 mV , n = 154; p = 0 . 52 , Mann–Whitney test , data not shown ) . In sum , Bk appears to inhibit a leak K+ conductance and activate ICaN via a transduction system activating PLC-/InsP3-mediated release of intracellular Ca2+ and CaM . The excitatory effects of Bk on motoneurons are reminiscent of its action on in vitro spinal cord preparations from neonatal rats where it slowly depolarizes ventral roots ( Dray et al . , 1988; Dunn and Rang , 1990; Dray et al . , 1992; Rueff and Dray , 1993 ) ; see also ( Figure 1B ) . The actions of Bk in generating motor outputs have long been viewed as secondary to release of substance P from capsaicin-sensitive sensory neurons ( Bynke et al . , 1983; Geppetti , 1993; Jeftinija , 1994 ) . This indirect mechanism for exciting motoneurons is not used since the substance P receptor antagonist spantide did not antagonize responses to Bk in our study . Our results suggest a direct postsynaptic action of Bk on motoneurons that is still seen when sensory inputs are blocked by TTX or cadmium . Further supporting this , Bk receptors have been mapped to the membranes of motoneurons ( Lopes et al . , 1995 ) , Bk induces direct elevation of free calcium in motoneurons , and the depolarization in motoneurons is greatly reduced by calcium chelators infused into the motoneurons themselves . On the other hand , microglial cells express functional Bk receptors ( Noda et al . , 2003; Ifuku et al . , 2007 ) and Bk can act as a neuron-glia signaling molecule ( Parpura et al . , 1994; Heblich et al . , 2001 ) . Therefore , we cannot be rule out the possibility that bath-applied Bk activates glial cells resulting in the release of an unidentified chemical substance which may mediate some of Bk's effects on motoneurons . Bk has long been shown to depolarize sensory neurons by inhibiting K+ channels ( Weinreich , 1986; Brown and Higashida , 1988; Jones et al . , 1995; Villarroel , 1996; Cruzblanca et al . , 1998; Liu et al . , 2010 ) . We have found that inhibition of K+ channels seems also to occur in motoneurons , as seen by a decrease of the slope conductance which reverses near Ek and is influenced by changes in [K+]o and addition of the K+ blocker Ba2+ . The Bk-sensitive K+ current might be derived from a member ( s ) of the two-pore domain K+ ( K2P ) channels that are substrates for resting K+ currents in neurons ( Enyedi and Czirjak , 2010 ) . This structurally distinct group of K+ channels includes over 15 members classified into six functional subgroups: ( 1 ) TWIK-1 and -2; ( 2 ) TASK-1 , -3 and -5; ( 3 ) TREK-1 , and -2 and TRAAK; ( 4 ) TASK-2 , TALK-1 and -2; ( 5 ) THIK-1; ( 6 ) TRESK . Functional distinction between K2P channel subtypes is still a delicate issue because of a lack of selective pharmacological reagents ( Lesage , 2003; Lotshaw , 2007 ) . However , several features help to separate them ( Patel and Honore , 2001 ) . In addition to quinidine sensitivity , we show that the response to Bk is reduced by a low concentration of Ba2+ but not by acidosis . The relatively weak pH sensitivity of the response is not consistent with the properties of TASK , TALK , and TRESK channels that are downregulated by acidosis ( Duprat et al . , 1997; Sano et al . , 2003; Kang and Kim , 2004; Dobler et al . , 2007; Mathie et al . , 2010; Callejo et al . , 2013 ) . Likewise , the contribution of TREK/TRAAK channels is unlikely because they are weakly inhibited by or resistant to Ba2+ ( Lesage et al . , 2000; Patel and Honore , 2001 ) . Thus the most likely candidates for the Ba2+-sensitive K+ current are TWIK and THIK channels ( Chavez et al . , 1999; Patel et al . , 2000; Rajan et al . , 2001; Lloyd et al . , 2009 ) though TWIK channels are much more sensitive to Ba2+ ( IC50: 0 . 1 mM ) than THIK ( IC50: ∼1 mM ) ( Patel and Honore , 2001; Rajan et al . , 2001 ) . The contribution of these channels in generating the Bk effects remains to be established . While it is clear that Bk inhibits a Ba2+-sensitive K+ current , it is also clear that it activates a Ba2+-insensitive Na+ conductance . This conductance shares features of ICaN ( Partridge and Swandulla , 1988 ) , as it appears to be: ( 1 ) voltage-independent , ( 2 ) calcium-activated , ( 3 ) blocked by the cationic channel blocker ruthenium red , and ( 4 ) eliminated by lowering extracellular Na+ . The fact that the Na+-dependent conductance mediates a significant current without affecting the input resistance suggests that it may have a dendritic location and help regulate the flow of distal synaptic inputs to the soma . The initial rise of [Ca2+]i at the dendritic level supports this hypothesis and is consistent with the location of B2 receptors in the peripheral processes of motoneurons ( Lopes et al . , 1995 ) . Channels capable of mediating a Na+ response similar to that reported in this study are mammalian homologues of Drosophilia transient receptor potential ( TRP ) channels ( Wu et al . , 2010 ) . Consistent with this , Bk has been shown to activate different classes of TRP channels ( Premkumar and Ahern , 2000; Delmas et al . , 2002; Sugiura et al . , 2002; Bandell et al . , 2004; Zhang et al . , 2008 ) and to induce a Na+-mediated ICaN in cultured spinal sensory neurons ( Burgess et al . , 1989; Dray et al . , 1992; McGehee et al . , 1992; Seabrook et al . , 1997 ) . The cationic current could be produced by gating of more than a single class of channels . Given the sensitivity of Bk-induced responses to temperature , thermo-TRP channels ( including TRPV and TRPM ) might be activated by Bk in motoneurons . Still , it should be stressed that additional studies will be necessary to precisely define the channel type ( s ) that mediate Bk responses . Although Bk appears to act by at least two ionic mechanisms in motoneurons , activation of a single B2 receptor class appears to occur: B2 ( but not B1 ) receptors are expressed in motoneurons ( Lopes et al . , 1995 ) , and B2 agonists and antagonists mimic or block all the effects of Bk . Several groups have provided data in different systems suggesting that Bk depolarizes neurons by triggering a PLC signaling pathway giving rise to the InsP3-dependent release of Ca2+ and DAG ( Yano et al . , 1984 , 1985; Derian and Moskowitz , 1986; Francel and Dawson , 1986; Francel et al . , 1987; Jackson et al . , 1987; Thayer et al . , 1988; Perney and Miller , 1989; Gutowski et al . , 1991; Hall , 1992; Premkumar and Ahern , 2000; Ferreira et al . , 2004 ) . Our pharmacological data support this pathway , as the Bk response is: ( 1 ) blocked by GDPβS; ( 2 ) blocked by the PLC antagonist U73122; ( 3 ) blocked by the InsP3 receptor antagonist xestospongin C . The DAG arm of the bifurcating PLC pathway appears not to be involved , as the Bk-induced depolarization was not inhibited by PKC antagonists . Consistent with a PLC/InsP3 pathway , we found that a rise of intracellular Ca2+ is essential for Bk responses . Since blockade of membrane Ca2+ channels does not affect Bk-induced depolarizations , and ICaN is mainly mediated by Na+ , the [Ca2+]i increase must arise from the release of Ca2+ from intracellular stores . Most directly , we showed that the introduction of the Ca2+ buffer BAPTA into the cytoplasm greatly reduced the Bk response , and blockade of the Ca2+ sensor CaM fully abolished Bk responses . As summarized in Figure 8 , these data suggest that the two ionic components are mediated by B2 receptors acting via the single PLC/InsP3/calcium signaling pathway . 10 . 7554/eLife . 06195 . 013Figure 8 . Overview of the signal transduction cascade for excitatory actions of Bk on lumbar motoneurons . InsP3 , inositol 1 , 4 , 5-trisphosphate; PIP2 , phosphatidylinositol-4 , 5-diphosphate; PLC , Phospholipase C; ER , endoplasmic reticulum . DOI: http://dx . doi . org/10 . 7554/eLife . 06195 . 013 Bk not only depolarizes spinal motoneurons but also sensitizes them , so they are more activated by small inputs than under control conditions . The leftward shift in the f–I relation can be explained as an increase in responsiveness due to an increase in input resistance . The depolarizing effect of the Na+ current would also be enhanced by the simultaneous increase in input resistance due to reduction of a resting K+ conductance . This synergistic mechanism might promote the burst-evoked sADP , which helps to generate self-sustained spiking ( Bouhadfane et al . , 2013 ) . Following traumatic injury , increases in the concentration of kinins in the spinal cord ( Xu et al . , 1991 ) might thus predispose motoneurons to express self-sustained plateau potentials , thereby contributing to muscle spasms ( Nickolls et al . , 2004 ) . In keeping with this hypothesis , chronic spinal rats show the emergence of long-lasting reflexes in vivo ( Murray et al . , 2010 ) , similar to those we recorded in response to Bk in vitro ( see Figure 1D ) . The spinal cord also represents a relevant site of action for kinins to produce hypernociception ( Steranka et al . , 1988; Perkins et al . , 1993; Ferreira et al . , 2002 ) . The flexor withdrawal motor reflex has commonly been used as a surrogate pain model ( Barrot , 2012 ) ; however , the contribution of motoneuron hyperexcitability has never been questioned . A role for Bk in hyperalgesic assessment methods at the level of motoneurons remains to be substantiated . It would be expected that excitation of sensory interneurons by Bk works in concert with direct sensitization of motoneurons to augment the withdrawal flexor reflex . It is also important to emphasize that the neonatal response to nociception differs from that found in adults ( Fitzgerald , 2005 ) . It seems therefore rational that Bk actions might undergo marked postnatal changes . Details of the in vitro preparations have been previously described ( Bouhadfane et al . , 2013; Brocard et al . , 2013 ) and are only summarized here . For the isolated spinal hemicord preparation , the spinal cord was transected at T10 and longitudinally cut along the midline . The hemispinal cord with the medial side up was transferred to the recording chamber and perfused with oxygenated normal Krebs solution composed of the following ( in mM ) : 120 NaCl , 3 KCl , 1 . 25 NaH2PO4 , 1 . 3 MgSO4 , 1 . 2 CaCl2 , 25 NaHCO3 , 20 D-glucose , pH 7 . 4 , ∼32°C pH 7 . 4 . For the slice preparation , the lumbar spinal cord was isolated in oxygenated ( 95% O2/5% CO2 ) ice-cold ( <4°C ) low-sodium medium , composed of the following ( in mM ) : 232 sucrose , 3 KCl , 1 . 25 KH2PO4 , 4 MgSO4 , 0 . 2 CaCl2 , 26 NaHCO3 , 25 D-glucose , pH 7 . 4 . The lumbar spinal cord was then introduced into a 4% agar solution , quickly cooled , mounted in a vibrating microtome ( VT1000S; Leica , Germany ) and sliced ( 350 µm ) through the L3-5 lumbar segments . Slices were immediately transferred into the holding chamber filled with oxygenated aCSF solution ( composed of [in mM]: 120 NaCl , 3 KCl , 1 . 25 NaH2PO4 , 1 . 3 MgSO4 , 1 . 2 CaCl2 , 25 NaHCO3 , 20 D-glucose , pH 7 . 4 , ∼30°C ) . Following a 1-hr resting period , individual slices were transferred to a recording chamber that was continuously perfused ( ∼4 ml/min ) with the same medium heated to ∼34°C . All solutions were oxygenated with 95% O2/5% CO2 . Electrophysiological data were acquired through a Digidata 1440a interface using the Clampex 10 software ( Molecular Devices ) . For the spinal hemicord preparation , motor outputs were recorded using extracellular stainless steel electrodes placed in contact with lumbar ventral roots ( L3–L5 ) and insulated with Vaseline . The ventral root recordings were amplified ( ×2000 ) , high-pass filtered at 70 Hz , low-pass filtered at 3 kHz , and sampled at 10 kHz . Monopolar stainless steel electrodes were also placed in contact with the dorsal roots and insulated with Vaseline to deliver a brief supramaximal stimulation ( 0 . 2 ms duration ) . Dorsal root stimulation was repeated five times with an interstimulus interval of 60 s for each trial . For the slice preparation , whole-cell patch-clamp recordings were made from motoneurons located in the lateral ventral horn using a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) in current- or voltage-clamp mode . Motoneurons were visually identified with video microscopy ( E600FN; Nikon Eclipse , Japan ) coupled to infrared differential interference contrast , as the largest cells located in layer IX . The image was enhanced with a KP-200/201 infrared-sensitive CCD camera ( Hitachi , Japan ) and displayed on a video monitor . Only one motoneuron was recorded from each slice . Patch electrodes ( 2–3 MΩ ) were pulled from borosilicate glass capillaries ( 1 . 5 mm OD , 1 . 12 mm ID; World Precision Instruments , Sarasota , FL ) on a Sutter P-97 puller ( Sutter Instruments Company , Novato , CA ) and filled with intracellular solution containing ( in mM ) : 140 K+-gluconate , 5 NaCl , 2 MgCl2 , 10 HEPES , 0 . 5 EGTA , 2 ATP , 0 . 4 GTP , pH 7 . 3 ( 280–290 mOsm ) . In some recordings , BAPTA ( 10 mM ) , GDPßS ( 2 mM ) , or xestospongin ( 2 . 5 mM ) were added in the pipette solution to chelate intracellular Ca2+ , to inhibit G-proteins or to block InsP3 receptors , respectively . Recordings were digitized on-line and filtered at 10 kHz ( Digidata 1322A , Molecular Devices ) . Access resistance was monitored periodically throughout the experiments . For calcium-indicator loading of cell populations ( Stosiek et al . , 2003 ) , we used multicell bolus loading with Oregon Green 488 BAPTA-1 acetoxymethyl ester ( 1 mM OGB1-AM; Life Technologies , France ) , which resulted in the staining of virtually all cells near the ejection site , including neurons and astrocytes . Astrocytes were identified through additional staining with the astrocytic marker Sulforhodamine-101 ( 300 µM SR-101 ) , which permitted a clear separation of the astroglial ( yellow ) and neuronal ( green ) networks . According to the criteria stated above , motoneuron somata were manually selected in a three-dimensional space after acquiring a reference z-stack of the volume on a custom-built three dimensional fast acousto-optical trajectory scanning microscope ( Femtonics Ltd , Budapest , Hungary ) ( Katona et al . , 2012 ) with a femtosecond pulsed laser tuned to 830 nm ( Mai Tai HP , Spectra Physics , Mountain View , California ) . For each motoneuron selected , acquisition points were placed in the soma in the cytoplasmic area avoiding the nucleus . The selected neurons were imaged at 33 kHz/point . Imaging was controlled using the MES software package ( Femtonics Ltd ) based on Matlab ( Mathworks ) . Fluorescence was measured as the average pixel value over selected somatic and proximal dendritic regions . Calcium changes were calculated as the relative change in fluorescence [ΔF/F = ( F − F0 ) / ( F0 − Fb ) ] , where F0 is baseline fluorescence and Fb is background fluorescence measured from a region lacking labeled cells away from the recorded area . In some experiments , whole-cell recording and simultaneous calcium imaging were performed on motoneurons identified using 900-nm oblique illumination . In these experiments , the recording electrodes ( 2–5 MΩ ) contained 120 µM OGB-1 and 20 µM Alexa-594 ( Life Technologies ) dissolved in the same intracellular solution as stated above . To ensure equilibration between the pipette solution and cytosol , the acquisition was started at least 15 min after establishment of the whole-cell configuration . For each subcellular region of interest ( soma and dendrites ) five to ten acquisition points were manually defined , outside the region of the nucleus , and averaged during analysis . For simultaneous calcium imaging and whole cell recordings , detection of calcium transients was triggered by the onset of spikes or the onset of the Bk response measured in the electrophysiological recording . All imaging sessions were conducted at 34°C . Electrophysiological data were analyzed off-line with Clampfit 10 software ( Molecular Devices ) or the MES sofware package ( Femtonics Ltd ) . Only cells exhibiting a stable holding membrane potential , access resistance ( no more than 20% variation ) and an action potential amplitude larger than 45 mV were considered . All reported membrane potentials were corrected for liquid junction potentials . Passive membranes properties of cells were measured by determining from the holding potential the largest voltage deflections induced by small current pulses that avoided activation of voltage-sensitive currents . The input resistance and input conductance were measured by the slope of the linear portion of the I/V relationship . There was evidence of inward rectification ( ‘sag’ ) during strong hyperpolarization . The size of the sag was measured from the voltage response of the cell to the hyperpolarizing current pulse adjusted to elicit peak voltage deflections to −120 mV , and was defined as 100 × ( 1 − Vss/Vpeak ) , where Vpeak was the peak voltage deflection from the holding potential ( −70 mV ) and Vss was the steady-state voltage deflection from the holding potential . Firing properties were investigated with 1-s-long depolarizing current pulses of varying amplitudes . Firing frequency was calculated as the average action potential frequency over the entire duration of the pulse . The rheobase was defined as the minimum current intensity necessary to induce an action potential . To investigate the action potentials and afterhyperpolarizations ( AHPs ) , single spikes were evoked by brief intracellular pulses from a holding potential of −60 mV . Peak spike amplitude was measured from the threshold potential , and spike duration was defined as the width of the action potential at 50% of the peak . The peak amplitude and duration ( to half of the peak height ) of AHPs were measured from the holding potential . To investigate the sADP , a train of spikes was evoked by a 2-s current pulse at holding potential of −60 mV . The peak amplitude of the sADP was measured from the holding potential . The peak amplitude of the depolarization induced by bath application of Bk was measured from the holding potential . When the Bk-induced depolarization reached a steady-state level , the membrane potential was manually moved back to its original holding level for direct comparison of membrane properties before and during the application of the nanopeptide . A standard sigmoidal curve was fit to the relation between log of agonist dose and the Bk-induced depolarization . The dose that produced 50% of the maximal effect ( EC50 ) was measured from the curve . To characterize the ionic mechanism ( s ) underlying the Bk-evoked depolarization , the membrane potential was initially held in voltage clamp at −70 mV , which approximates the resting membrane potential of motoneurons ( Tazerart et al . , 2007 ) , and then ramped to 0 mV over 3 s . To show the reversal potential , the I–V relationship of the current was constructed by subtracting the I–V relationship obtained before Bk application ( control ) from that performed in the presence of the nanopeptide . The current generated was fitted by a linear regression which was used to extrapolate the reversal potential at which the currents intersect . Motor outputs recorded on the ventral roots in response to a brief dorsal root stimulation were quantified by cumulative counts of spikes generated in PSTHs ( bins width: 20 ms ) over a time window of 4000 ms post-stimulation . PSTHs were constructed from 5 consecutive rectified responses . As defined previously ( Murray et al . , 2010 ) , we computed the transient short latency and long-lasting reflexes over a window of 10–40 ms and 500–4000 ms post stimulus , respectively . Counts were corrected for spontaneous activity by subtracting the number of spontaneous events arising prior to the stimulus . Calcium fluorescence changes were analyzed using the MES software package ( Femtonics Ltd ) based on Matlab ( Mathworks ) . Data are presented as means ± SEM . The statistical tests used are given in the text and figure legends . Values of p < 0 . 05 were considered significant ( GraphPad Software , San Diego California USA ) . Normal aCSF was used in most cases for electrophysiological recordings . In whole-cell current-clamp and voltage-clamp recordings , TTX ( 0 . 5–1 µM ) was used to eliminate the effect of presynaptic input , except when firing properties of motoneurons were studied . Low Na+ extracellular solution was prepared by replacing NaCl with equivalent concentration of choline chloride . High-K+ aCSF was prepared by adding 1 M KCl . The effect of extracellular pH on membrane current was examined by lowering pH with 1 . 0 N HCl solution . Drugs were purchased from the following sources: apamin ( 100 nM ) , 4-aminopyridine ( 2 mM ) , BAPTA ( 10 mM ) , barium ( 100–300 µM ) , cadmium chloride ( 100 µM ) , CTX ( 2 µg . ml−1 ) , chelerythrine ( 10 µM ) , CNQX ( 10 µM ) , GDPßS ( 2 mM ) , [Hyp3]-Bk ( 2 µM ) , kynurenate ( 1 . 5 mM ) , pertussis toxin ( PTX: 2 µg . ml−1 ) , PKA 6–22 ( 10 µM ) , PKC 19–36 ( 10 µM ) , R 715 TFA ( 5 µM ) , tetraethylammonium chloride ( TEA , 10 mM ) from Sigma–Aldrich; AP-5 ( 20 µM ) , bradykinin ( Bk , 0 . 01–16 µM ) , H89 ( 10 µM ) , Hoe 140 ( 2 µM ) , KN-93 ( 2 µM ) , Lys-[Des-Arg9]-Bk ( 2 µM ) , quinidine ( 200–400 µM ) , ruthenium red ( RR , 10 µM ) , spantide ( 2–5 µM ) , tetrodotoxin citrate ( TTX , 20 nM–1 µM ) , U73122 ( 10 µM ) , W-7 ( 100 µM ) , ZD7288 ( 20 µM ) from Tocris ( Bristol , UK ) . In most experiments , drugs were delivered via superfusion using a syringe pump . In some experiments , cells were loaded with BAPTA , GDPßS , or xestospongin C by passive diffusion from the patch pipette . In some experiments performed using the in vitro hemicord preparation , a pipette with a wide drip tip containing Bk was placed ∼25–50 µM upstream of the ventral horn . Leakage of the solution from the pipette was monitored by including neutral red in the intrapipette solution . This configuration restricts the application of drug to the ventral horn ( see Figure 1A ) and avoids mechanical artifacts .	When we accidentally place our hand on a hot stove , we normally experience a painful sensation that starts with the sensory nerves under our skin . These nerves respond by transmitting electrical impulses to our brain , where the painful sensation is then processed . At the same time , these impulses are also transmitted to the motor nerves that control the muscles in our hand to trigger an immediate reflex to withdraw the hand from the hot stove . Pain therefore has a useful role as it can reduce how bad an injury is . People with a condition called hyperalgesia have an increased sensitivity to pain . This condition can result from a chemical called bradykinin ‘sensitizing’ the sensory nerves , causing them to transmit more electrical impulses in response to pain than normal . This makes the injury feel much more painful , and can make the pain last for longer than is beneficial . It was less clear whether bradykinin also affects motor nerves and so triggers a withdrawal reflex . By recording the electrical activity of motor nerve cells taken from the spinal cords of newborn rats , Bouhadfane et al . now show that these motor nerves become more active when exposed to bradykinin . Nerve cells generate electrical signals when ions—such as potassium , sodium , and calcium ions—move through channels in the membranes of the cell . Therefore , to investigate how bradykinin influences the electrical activity of motor nerves , Bouhadfane et al . exposed the cells to drugs that inhibit particular ion channels . This revealed that bradykinin sensitizes the motor nerves by blocking a type of potassium ion channel and activating another ion channel that mainly transports sodium ions . Furthermore , Bouhadfane et al . were able to identify the signaling pathways that allow bradykinin to affect the motor nerve cells . The study implies that the neuronal circuitry for pain does not rely exclusively on sensory nerve cells but should also integrate motor nerve cells . A future challenge remains in developing a protocol to resolve the contribution of motor nerve cells to hyperalgesia assessed by reflex withdrawal .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Sensitization of neonatal rat lumbar motoneuron by the inflammatory pain mediator bradykinin
Understanding chromatin function requires knowing the precise location of nucleosomes . MNase-seq methods have been widely applied to characterize nucleosome organization in vivo , but generally lack the accuracy to determine the precise nucleosome positions . Here we develop a computational approach leveraging digestion variability to determine nucleosome positions at a base-pair resolution from MNase-seq data . We generate a variability template as a simple error model for how MNase digestion affects the mapping of individual nucleosomes . Applied to both yeast and human cells , this analysis reveals that alternatively positioned nucleosomes are prevalent and create significant heterogeneity in a cell population . We show that the periodic occurrences of dinucleotide sequences relative to nucleosome dyads can be directly determined from genome-wide nucleosome positions from MNase-seq . Alternatively positioned nucleosomes near transcription start sites likely represent different states of promoter nucleosomes during transcription initiation . Our method can be applied to map nucleosome positions in diverse organisms at base-pair resolution . The eukaryotic genome is compacted into chromatin ( Kornberg , 1974 ) which is comprised of nucleosomes , each consisting of approximately 147 base pairs ( bp ) of DNA wound around a histone protein octamer ( Kornberg and Lorch , 1999 ) . The helical DNA makes direct contact with the histones every 10 base pairs , with the major groove of DNA alternating between facing towards and away from the histone core ( Luger et al . , 1997 ) . Shifting the histones relative to the DNA sequence by a few base pairs can change the accessibility of sequence elements to DNA binding proteins if they are located in the linker sequences between nucleosomes , or may switch these elements between facing towards and away from nucleosomes if they are located within nucleosomal DNA ( Jiang and Pugh , 2009b; Segal and Widom , 2009b; Zhang and Pugh , 2011 ) . The location of nucleosomes with respect to DNA sequences influences many biological processes . Nucleosomes restrict the accessibility of DNA sequences to protein factors , such as transcriptional regulators and the transcription machinery ( John et al . , 2011; Li et al . , 2007; Liu et al . , 2006; Zhou and O'Shea , 2011 ) . The positions and occupancy of nucleosomes can influence the interplay between transcription factors ( Mirny , 2010 ) and the level ( Carey et al . , 2013; Kim and O'Shea , 2008 ) , dynamics ( Lam et al . , 2008 ) , and differences in gene expression between cells ( Dadiani et al . , 2013; Raser and O'Shea , 2004; Tirosh and Barkai , 2008 ) . Recently , nucleosome organization has also been suggested to affect how gene promoters interpret dynamic signaling information at the single cell level ( Hansen and O'Shea , 2013; Hao and O'Shea , 2012 ) , and heterogeneity in promoter nucleosome positions has been linked to differences in gene expression ( Small et al . , 2014 ) . Knowing the precise location that nucleosomes occupy with respect to DNA sequence is crucial for understanding how these biological processes are influenced by eukaryotic chromatin . Genome-wide nucleosome positions are commonly mapped with micrococcal nuclease digestion based high-throughput sequencing ( MNase-seq ) ( Hughes and Rando , 2014 ) . In this method , histone-DNA interactions protect DNA from MNase digestion and the protected DNA fragments are sequenced and aligned to genome sequences to infer the location of nucleosomes ( Clark , 2010; Kaplan et al . , 2009; Rando , 2010; Zhang and Pugh , 2011 ) . Although MNase exhibits sequence preference when digesting DNA devoid of histones ( Horz and Altenburger , 1981 ) , genome-wide analyses of nucleosomes with MNase-based methods are generally consistent with studies using MNase-independent methods ( Hughes and Rando , 2014 ) , such as DNase I chromatin digestion ( Hesselberth et al . , 2009 ) and chemical cleavage ( Brogaard et al . , 2012 ) . Studies that apply MNase-based methods typically report the position of a nucleosome as the average of the bulk nucleosome population ( referred to as the 'consensus center' , Figure 1A ) ( Struhl and Segal , 2013; Zhang and Pugh , 2011 ) . However , if nucleosomes have overlapping positions in a significant portion of the population , the effect of averaging over heterogeneous nucleosome positions can lead to discrepancy between the consensus center of nucleosomes and the most representative nucleosome positions ( Figure 1A ) . A variety of methods have been developed to improve the precision of nucleosome mapping from MNase-seq data , such as peak finding of nucleosome occupancy ( Zhang and Pugh , 2011 ) and filtering of single-end digestion patterns ( Weiner et al . , 2010 ) , but determining the precise locations of individual nucleosomes within a cell population remains a challenge due to substantial variability in the mapped locations of digested nucleosomes – the midpoints of paired-end sequenced nucleosomes or the endpoints of single-end sequenced nucleosomes ( Figure 1B ) . This variability may arise from a cluster of overlapping and stably positioned nucleosomes , as well as from effects causing different degrees of digestion of the same nucleosome by MNase , such as nucleosome breathing and nuclease trimming – all of which influence the distribution of the aligned reads and are difficult to disentangle ( Clark , 2010 ) . Recently , a chemical cleavage approach that uses a genetically engineered histone H4 to chemically cleave DNA sequences in contact with the nucleosome dyad allowed direct measurement of nucleosome positions with unprecedented resolution ( Brogaard et al . , 2012; Moyle-Heyrman et al . , 2013 ) . However , the requirement for genetic engineering of essential histones limits its current application to genetically tractable organisms . Therefore , novel experimental or analytical approaches that are generally applicable in eukaryotes are still needed to determine the accurate positions of nucleosomes in vivo . Here , we report a computational approach to determine in vivo nucleosome positions from paired-end MNase-sequencing data . 10 . 7554/eLife . 16970 . 003Figure 1 . Illustration of the Template-Based Bayesian ( TBB ) approach for determining nucleosome positions . ( A ) Diagram illustrating the heterogeneous nucleosome positions and the consensus centers of nucleosomes along a genomic region in a population of cells . Blue ovals illustrate individual nucleosomes and dotted lines mark all nucleosome positions . ( B ) Example of digested nucleosome reads , their nucleosome positions and the overall occupancy . ( C ) Illustration of the computational pipeline of the TBB approach . Occupancy of sequencing read midpoints indicates the number of midpoints at every base pair for yeast Chr 8 , 204 , 500–206 , 500 bp . Blue ovals illustrate overlapping TBB nucleosome positions and are colored according to the magnitude of their coefficients β . Two common presentations of nucleosome sequencing data are shown for comparison: the light gray area represents the nucleosome occupancy generated by smoothing sequencing read midpoints with a Parzen window approach ( band size of 20 bp ) ( Albert et al . , 2007; Tsankov et al . , 2010 ) ; the dark gray area ( Fragment extension ) represents the nucleosome occupancy generated by extending 73 bp on both ends from the sequencing read midpoints . ( D ) Histogram showing the distance between adjacent TBB nucleosome positions in a combination of the T1 and T2 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 00310 . 7554/eLife . 16970 . 004Figure 1—figure supplement 1 . Diagrams of nucleosome digestion variability template estimation . Diagrams illustrating the estimation of the nucleosome digestion variability template from the length distribution of paired-end sequencing reads . The length of a sequenced DNA fragment is decomposed into the sum of the length of the nucleosome core ( 147 bp ) and the digestion errors ( ε1 , ε2 ) on both ends . Assuming that ε1 and ε2 are sampled randomly from a distribution of the digestion error ( ε ) , the digestion error ε can be then estimated from the length distribution extracted from paired-end sequencing data , and is used to infer the nucleosome digestion variability template around a true nucleosome center . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 00410 . 7554/eLife . 16970 . 005Figure 1—figure supplement 2 . Length distribution of nucleosome reads . Plots showing the length distribution of paired-end sequencing reads within gene coding regions and non-coding regions ( left ) , and in the promoters of genes and non-promoter regions ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 005 Applying template-based deconvolution to experimental data has many applications in biology . For example , in super-resolution fluorescence microscopy , the locations of individual fluorophores within the diffraction limit of light can be identified by deconvoluting light intensity information with a function describing the distribution of light intensity from individual light spots ( Betzig et al . , 2006; Huang et al . , 2009; Rust et al . , 2006 ) . Inspired by this , we use the size distribution of MNase digested nucleosome fragments to infer a digestion variability template for nucleosomes , and report a Bayesian method that makes use of these templates to identify the individual positions of nucleosomes at a base-pair resolution , hereafter referred to as the template-based Bayesian ( TBB ) approach . This approach can be applied to data generated through both paired-end and single-end sequencing to map chromatin structure in diverse organisms . Here we demonstrate the template-based Bayesian approach with paired-end sequencing of MNase digested nucleosomes in yeast and human cells . We show that the periodic occurrences of dinucleotide sequence motifs relative to the nucleosome dyad can be directly determined from MNase based nucleosome positions and are conserved in vivo in both yeast and human cells . Leveraging this method , we find that alternative configurations of nucleosomes are a common feature in both yeast and human chromatin . The alternatively positioned nucleosomes around gene transcription start sites represent configurations that differ in their compatibility with the assembly of the pre-initiation complex . A 3-step model for transcription initiation can reconcile the competition between nucleosomes and the transcriptional machinery observed from genomic analysis . We reasoned that it might be possible to resolve the positions of multiple overlapping nucleosomes if we could estimate the degree to which MNase digestion contributes to the deviation of the midpoints from a true nucleosome center . Since the overall digestion of nucleosomes is reflected in the length of nucleosomal DNA fragments , we tested the idea that we might be able to estimate the variation in the midpoints from the length of digested nucleosomes ( Figure 1—figure supplement 1 ) , and use this information to infer the positions of individual nucleosomes through deconvolution . The variability of digested nucleosomes could come from two sources: the technical variation that is associated with nuclease cleavage , such as variable trimming at nucleosome ends , and biological variation that directly influences the length of DNA wound around histones , such as nucleosome breathing and remodeling ( Polach and Widom , 1995 ) . While the technical effects are likely to affect both ends of nucleosomal DNA equally , the biological effects may create bias for specific nucleosomes and/or a specific side of the nucleosomes . However , the biological effects are generally believed to be transient and rare at a given genomic location within a population of cells ( Andrews and Luger , 2011 ) ; we thus assumed that the digestion variation was equivalent at both ends of the nucleosome when averaged over the genome and population . The biological variation at individual nucleosomes could generate large shifts in read midpoints due to length differences in nucleosomes ( likely by multiples of 10 bp due to the unwrapping of each helical turn of DNA ) , and could be identified as alternative nucleosome positions if they were present in a significant fraction of the bulk nucleosome population . Nucleosomes with substantially smaller size , such as sub-nucleosomes ( Rhee et al . , 2014 ) , can be identified based on the sequenced fragment size ( <100 bp ) and were excluded from our analysis . The template-based Bayesian approach includes four major steps ( Figure 1C , see Materials and methods ) . First , we estimated the distribution of MNase digestion variability from the length distribution of DNA sequencing reads , assuming that MNase digestion is not biased towards either end of the nucleosome on average . This produced a digestion variability template , describing the overall deviations in read midpoints relative to the true nucleosome center ( Figure 1—figure supplement 1 ) . Second , each chromosome was segmented into regions with similar sequencing coverage to increase detection sensitivity , with a requirement to maintain sufficient segment length for estimating statistical properties of nucleosome positions . Third , within each segment , the occupancy of read midpoints ( the number of paired-end read midpoints at each base pair , Figure 1C , gray trace ) was deconvoluted with the digestion variability template to estimate the expected read occupancy at every base pair if digestion errors had not occurred ( coefficient β , Figure 1C , black trace ) . Finally , the locations of nucleosomes were summarized based on this expected read occupancy and statistical thresholds . To control for falsely identified nucleosome positions arising from the sequence preference of MNase , we set the statistical thresholds using a false discovery rate ( FDR≤0 . 05 ) estimated based on a simulated MNase digestion dataset that matches the experimentally observed digestion sequence distribution . In our study , the length distribution of paired-end sequencing reads from the gene coding and promoter regions was indistinguishable from the rest of the genome ( Figure 1—figure supplement 1–2 ) . Therefore , a single digestion variability template was applied across the yeast genome . We represented the locations of nucleosomes using two metrics: the consensus centers of nucleosomes and the TBB positions of nucleosomes ( Supplementary file 1 ) . The consensus centers of nucleosomes were defined to represent non-overlapping nucleosomes with a minimum size of 147 bp ( Figure 1C , blue bars ) , by applying a Gaussian window smoothing method to the expected read occupancy ( coefficient β ) from our model – they are equivalent to the nucleosome positions commonly described in the literature ( Albert et al . , 2008; Jiang and Pugh , 2009b; Struhl and Segal , 2013 ) ( hereafter termed ‘consensus centers’ ) . The TBB nucleosome positions describe the center locations of individual nucleosomes across the genome without limitation on minimum spacing ( Figure 1C , red dots and blue ovals ) ( hereafter termed ‘TBB nucleosome positions’ ) , which represent individual stably positioned nucleosomes over a statistical threshold . We titrated MNase to digest yeast chromatin and selected for analysis two nucleosome samples that were digested to different sizes ( ‘T1’ and ‘T2’ , Figure 2—figure supplement 1 ) . Both samples were digested to primarily mono-nucleosomes without generating over-represented sub-nucleosome fragments . We generated paired-end data sets of 'T1' and 'T2' , each with a coverage of ~2 . 8 read midpoints per base pair; as an average between these two data sets , we identified ~61 , 800 consensus centers of nucleosomes and ~125 , 000 TBB nucleosome positions , covering at least 75 . 5% of the S . cerevisiae genome . The spacing between adjacent TBB nucleosome positions peaked at 11 and 21 bp ( Figure 1D ) , consistent with the rotation of DNA sequence by one or two helical turns around histone cores ( Albert et al . , 2007; Brogaard et al . , 2012 ) , and suggesting that the same nucleosome commonly occupies overlapping configurations in different cells . The consensus centers of nucleosomes identified by the template-based Bayesian approach were consistent with the published methods employing Gaussian smoothing algorithms ( often referred to as 'Parzen window' ) ( Albert et al . , 2007 , 2008; Clark , 2010; Tsankov et al . , 2010 ) ( Figure 2A ) , with more than 90% of the nucleosomes matching within a single base pair ( Figure 2—figure supplement 2 ) . These consensus centers also showed general agreement with the published nucleosome reference maps ( Jiang and Pugh , 2009a ) ( Figure 2B and Figure 2—figure supplement 3 ) , validating the overall biological relevance of our data sets . We then evaluated the accuracy of our identified TBB nucleosome positions by comparing them to individual nucleosome positions mapped through a recently reported chemical approach ( Brogaard et al . , 2012 ) , which provides a reference that is independent of MNase cleavage specificity . The nucleosome positions identified by these two approaches were consistent ( Figure 2C and Figure 2—figure supplement 4 ) , demonstrating that the TBB approach can precisely map the positions of individual nucleosomes . Further , we assessed the absolute precision of the TBB approach by using it to map nucleosome positions from in silico generated MNase-seq data sets where the simulated nucleosome positions are known . We systematically tested the precision of the TBB approach with varying sequencing coverages , relative nucleosome population abundance and spacing among overlapping nucleosomes ( Figure 3 and Material and Methods ) . The majority of the TBB nucleosome positions identified in these simulated data sets were within 2–3 bp of the originally simulated nucleosome positions , even for the cases where nucleosome positions were separated by only 10–15 bp or differ by more than 3-fold in their relative occupancy ( Supplementary file 2 ) . 10 . 7554/eLife . 16970 . 006Figure 2 . Genome-wide evaluation of the TBB approach and alternative nucleosome positions . Histogram of the nearest distance: ( A ) between the consensus centers of nucleosomes determined by the TBB approach and by the Parzen window approach; ( B ) between the consensus centers determined by the TBB approach and the reference MNase nucleosome positions ( Jiang and Pugh , 2009a ) ; ( C ) between the TBB nucleosome positions and the nucleosome positions mapped by the chemical approach ( Brogaard et al . , 2012 ) ; ( D , E ) between the consensus centers of nucleosomes ( D ) and the TBB nucleosome positions ( E ) mapped in two independent experiments . The median of the distance between matched consensus centers or TBB nucleosome positions is reported for each comparison . ( F ) Example of stable TBB positions that are tolerant to MNase digestion , chr 8 , 341651 – 341771 . Randomly selected reads fragments are shown to represent the locations of sequenced tags . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 00610 . 7554/eLife . 16970 . 007Figure 2—figure supplement 1 . Titration of MNase digestion . ( A–C ) Titration of MNase digestion for obtaining mononucleosomes . T0 , T1 , T2 and T3 correspond to MNase digestion with 0 . 5U , 1U , 2U and 4U MNase . ( A ) Bioanalyzer analysis of purified nucleosomal DNA after MNase digestion . The mononucleosome fractions of the second and third samples were isolated for paired-end sequencing , termed 'T1' and 'T2' , respectively . ( B ) Quantification of the molar fraction of mono- , di- and tri-nucleosomal DNA for the samples in ( A ) . ( C ) Read length distribution of all digested samples with fragment length over 100 bp . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 00710 . 7554/eLife . 16970 . 008Figure 2—figure supplement 2 . Cumulative distribution of the nearest distance analysis in Figure 2A . Cumulative distribution of the nearest distance between the consensus centers of nucleosomes determined by the TBB approach and by the Parzen window approach . Circles mark the distance that matches 50% and 75% in the cumulative probability distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 00810 . 7554/eLife . 16970 . 009Figure 2—figure supplement 3 . Cumulative distribution of the nearest distance analysis in Figure 2B . Red traces and gray traces show the cumulative distribution of the nearest distance between the consensus centers determined by the TBB approach ( red ) or randomly generated consensus centers ( gray ) and the reference MNase nucleosome positions ( Jiang and Pugh , 2009a ) . Circles mark the distance that matches 50% and 75% in the cumulative probability distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 00910 . 7554/eLife . 16970 . 010Figure 2—figure supplement 4 . Cumulative distribution of the nearest distance analysis in Figure 2C . Magenta traces and gray traces show the cumulative distribution of the nearest distance between the TBB nucleosome positions ( magenta ) or randomly generated nucleosome positions ( gray ) and the nucleosome positions mapped by the chemical approach ( Brogaard et al . , 2012 ) . Circles mark the distance that matches 50% and 75% in the cumulative probability distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01010 . 7554/eLife . 16970 . 011Figure 2—figure supplement 5 . Cumulative distribution of the nearest distance analysis in Figure 2D . Green traces and gray traces show the cumulative distribution of the nearest distance between between the TBB consensus centers ( green ) or randomly generated consensus centers ( gray ) mapped in two independent experiments . Circles mark the distance that matches 50% and 75% in the cumulative probability distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01110 . 7554/eLife . 16970 . 012Figure 2—figure supplement 6 . Cumulative distribution of the nearest distance analysis in Figure 2E . Blue traces and gray traces show the cumulative distribution of the nearest distance between between the TBB nucleosome positions ( Blue ) or randomly generated nucleosome positions ( gray ) mapped in two independent experiments . Circles mark the distance that matches 50% and 75% in the cumulative probability distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01210 . 7554/eLife . 16970 . 013Figure 3 . Nucleosome detection from in silico MNase-seq datasets . ( A ) Plots summarizing the distance between the detected TBB nucleosome positions in the in silico datasets and the nearest simulated primary and alternative nucleosome positions . ( 'C' , total sequencing coverage of all overlapping nucleosomes; 'E' , the effective magnitude ( relative occupancy of neighboring nucleosomes ) ; and 'O' , the offset ( spacing between nearby nucleosome positions ) . ( B ) Examples of nucleosome detection in the simulation at different coverage , effective magnitude and offset ( different values are highlighted in red ) . Sequencing read midpoints ( gray ) were distributed randomly around the simulated nucleosome positions ( blue dots ) according to the digestion variability template . The coefficients ( blue trace ) and nucleosome positions ( red dots ) determined by the TBB approach are shown for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 013 The accuracy and reproducibility of MNase-based mapping of nucleosomes may be limited by MNase digestion variability . The degree to which chromatin is digested by MNase depends on the DNA sequence and the preparation of nucleosomal DNA samples ( Dingwall et al . , 1981; Horz and Altenburger , 1981 ) , and may differ significantly between experiments even when experimental protocols have been carefully followed . To evaluate the tolerance of the TBB nucleosome positions to different levels of MNase digestion , we compared the consensus centers and TBB nucleosome positions in the ‘T1’ and ‘T2’ data sets that represent digestion to different fragment sizes ( Figure 2—figure supplement 1 ) . Both the consensus centers and the TBB nucleosome positions were in agreement between these two experiments ( Figure 2D–E , Figure 2—figure supplements 5–6 ) , with a median difference of 2 bp across the genome; in comparison , randomly generated consensus centers and nucleosome positions gave rise to a median difference of 49 bp and 36 bp , respectively . Individual nucleosome positions can be reliably detected regardless of the differences in MNase digestion ( Figure 2F ) , demonstrating the robustness of the determined nucleosome positions to digestion variability . Overall , these results demonstrate that the positions of nucleosomes can be reliably and accurately detected at base-pair resolution with paired-end MNase-seq coupled with the template-based Bayesian approach . 10 . 7554/eLife . 16970 . 014Figure 4 . Dinucleotides frequency of nucleosome positions . ( A–C ) Normalized frequency of AA/AT/TA/TT and CC/CG/GC/GG dinucleotides of DNA sequences aligned at the centers of nucleosomes , for all TBB nucleosome positions in yeast , both before ( A ) and after ( C ) correction for MNase digestion bias , and for all TBB nucleosome positions on human chromosome 12 , position 38 , 000 , 000 bp to 48 , 000 , 000 bp , a 10 Mbp region randomly chosen in the human genome ( B ) . ( D ) The frequency of AA/AT/TA/TT for 62 , 035 randomly selected TBB nucleosome positions ( blue trace ) , and for genome locations with an average distance of either 2 bp ( orange trace ) or 5 bp ( black trace ) from these selected TBB nucleosome positions . The distance was randomly perturbed by between 0–4 bp or 0–10 bp for each nucleosome positions , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01410 . 7554/eLife . 16970 . 015Figure 4—figure supplement 1 . Dinucleotides frequency of nucleosome consensus centers . ( left ) Dinucleotide frequency for all 62 , 035 consensus centers of nucleosomes identified in experiment T1 . ( right ) Dinucleotide frequency for all consensus centers of nucleosomes identified on human chromosome 12 , position 38 , 000 , 000 bp to 48 , 000 , 000 bp . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01510 . 7554/eLife . 16970 . 016Figure 4—figure supplement 2 . Dinucleotides frequency of selected 147 bp nucleosome reads . ( left ) Dinucleotide frequency for 62 , 035 randomly selected sequence fragments that are exactly 147 bp in length in experiment T1 . ( right ) Dinucleotide frequency for all human nucleosome reads that are exactly 147 bp in length . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01610 . 7554/eLife . 16970 . 017Figure 4—figure supplement 3 . Dinucleotides frequency of TBB positions . Dinucleotide frequency for 62 , 035 ( the same number as consensus centers ) randomly selected TBB nucleosome positions from experiment T1 . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01710 . 7554/eLife . 16970 . 018Figure 4—figure supplement 4 . MNase-digestion correction for dinucleotides frequency of TBB positions . Dinucleotide frequency for TBB positions from experiment T1 ( blue , pink ) and the nucleosome positions determined from the simulated MNase digestion data set ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01810 . 7554/eLife . 16970 . 019Figure 5 . Alternatively positioned nucleosomes at transcription start sites . ( A , B ) Examples of a uniquely positioned nucleosome ( A ) , and alternatively positioned nucleosomes at the TSS ( B ) . Blue and magenta traces show the sequencing read midpoint occupancy and fitted coefficient β from experiments ‘T1’ and ‘T2’ , respectively . Ovals indicate the TBB nucleosome positions and are colored based on coefficients β . ( C ) Heat map showing the TBB nucleosome positions ( 'Positions' ) , the occupancy of read midpoints from two experiments ( T1 and T2 ) , and the occupancy of read midpoints from the two MNase bias simulations ( 'T1 DG control' and 'T2 DG control' ) . All data are aligned by the centers of selected unique or alternative nucleosome positions ( essentially the consensus center of +1 nucleosomes ) that overlap with the transcription start site ( TSS ) ( 4672 open reading frame in total ) . The order of transcripts is ranked by the maximum space between TBB nucleosome positions within each group , as illustrated by the diagrams on the left . The positive direction of the x-axis indicates 5’ to 3’ for all transcripts . ( D ) Correlation coefficient of the sequencing read midpoint occupancy ( un-smoothed ) in experiment T1 and T2 for each gene in panel C . ( E ) Graph showing the end location of paired-end sequencing reads of the upstream and downstream nucleosomes in panel C ( later defined as the proximal and distal nucleosomes , respectively ) . The ends are aligned at the position of upstream nucleosomes ( proximal nucleosomes ) . ( F ) Length distribution of paired-end sequencing reads in the unique , proximal and distal nucleosomes at gene promoters . In both ( E ) and ( F ) , if the midpoint of a sequencing read is within 5 bp of the position of a nucleosome , it is counted as a read of this nucleosome . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 01910 . 7554/eLife . 16970 . 020Figure 5—source data 1 . Source data for Figure 5C , with arrays of positions , midpoints from T1 and T2 used for generating the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 02010 . 7554/eLife . 16970 . 021Figure 5—figure supplement 1 . Reads occupancy between alternatively positioned nucleosomes . Bar graphs showing the ratios of the reads corresponding to alternatively positioned nucleosomes . The genes are the same as the panels in Figure 5C from the top to the bottom . Numbers on top of each panel indicate their ranks in Figure 5C . The ratios of the reads were calculated by dividing the total number of sequencing read midpoints 60 bp to the left of the group center by the total number of sequencing read midpoints 60 bp to the right of the group center , and were binned on a log2 scale . The standard deviation of the read ratios on a log2 scale ( std ) is labeled in the top right corner for each graph . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 02110 . 7554/eLife . 16970 . 022Figure 5—figure supplement 2 . Heat map showing the TBB nucleosome positions and the midpoint read occupancy . ( A ) Heatmaps are the same as Figure 5C , except that the position clusters with more than 2 positions are excluded . ( B ) Plots show the average midpoint occupancy of genes in groups of 1000 . ( C ) Bar graphs are the same as Figure 5—figure supplement 1 , except that the clusters with more than 2 positions are excluded . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 022 Several methods have been developed to determine the genome-wide locations of nucleosomes , each of which addresses different experimental and computational issues ( Albert et al . , 2007a; Brogaard et al . , 2012; Polishko et al . , 2012; Schep et al . , 2015; Tirosh , 2012; Tsankov et al . , 2010; Valouev et al . , 2011; Weiner et al . , 2010; Zhong et al . , 2016 ) . We briefly summarized the approaches , performance and limitations of these methods for reference ( Supplementary file 3 ) . Nucleosomes in vivo are thought to be enriched for genome sequences that show a pattern of periodic AA/AT/TA/TT or CC/CG/GC/GG motifs that oscillate every 10 bp with opposite phases ( Albert et al . , 2007a; Brogaard et al . , 2012; Gaffney et al . , 2012; Kaplan et al . , 2009; Mavrich et al . , 2008; Satchwell et al . , 1986; Segal et al . , 2006 ) . This dinucleotide sequence pattern matches the bending property of preferred nucleosomal DNA sequences , as the double stranded DNA is twisted and bent to maintain contact with the histone cores ( Drew and Travers , 1985 ) . However , this periodic pattern was not observed around the consensus centers of nucleosomes determined using MNase based mapping methods ( Valouev et al . , 2011 ) ( Figure 4—figure supplement 1 ) and only emerged from nucleosome fragments selected to be exactly 147 bp ( Figure 4—figure supplement 2 ) ( Albert et al . , 2007a; Gaffney et al . , 2012; Kaplan et al . , 2009 ) , perhaps due to the imprecision of the determined nucleosome centers . In contrast , when we aligned DNA sequences at the TBB nucleosome positions in yeast , the dinucleotide AA/AT/TA/TT and CC/CG/GC/GG motifs displayed 10 bp periodicity across almost the full length of the nucleosome core ( Figure 4A and Figure 4—figure supplement 3 ) , illustrating the power of the TBB approach to precisely map nucleosome positions from MNase-based methods . To control for the influence of MNase digestion bias , we simulated MNase digestion based on its observed sequence preference ( Gaffney et al . , 2012 ) , and identified nucleosome positions from the in silico MNase digestion data set . We used these in silico nucleosome positions to normalize the bias that was merely a result of MNase digestion ( Figure 4—figure supplement 4 ) . After these corrections , we observed dinucleotide AA/AT/TA/TT and CC/CG/GC/GG motifs displaying 10 bp periodicity ( Figure 4C ) , consistent with the dinucleotide frequency around the chemically determined nucleosome positions in yeast ( Brogaard et al . , 2012 ) . Randomly shifting the determined TBB nucleosome positions by an average of 2 bp greatly reduced the observed dinucleotide periodicity and shifting the positions by an average of 5 bp completely eliminated the periodic pattern ( Figure 4D ) , demonstrating the importance of precisely mapping nucleosome positions in vivo . Paired-end nucleosome sequencing has been used to map chromatin structure in higher eukaryotes ( Gaffney et al . , 2012 ) , but a base-pair resolution nucleosome map has been lacking for organisms other than yeast . We applied the TBB approach to determine the positions of human nucleosomes from a recent paired-end MNase-seq study ( Gaffney et al . , 2012 ) . The frequency of AA/AT/TA/TT and CC/CG/GC/GG motifs displayed clear 10 bp periodicity relative to human TBB nucleosome positions , which was comparable to the motif periodicity observed in yeast ( Figure 4A and B , Figure 4—figure supplement 2 ) . In contrast , when DNA sequences were aligned at the consensus centers of human nucleosomes , the frequency of these dinucleotide motifs did not exhibit a clear periodicity with respect to the consensus centers ( Figure 4—figure supplement 1 ) . With the template-based Bayesian approach , we demonstrated that nucleosome positions determined from MNase-based methods indeed present periodic dinucleotide frequency globally , and that this feature may be a conserved signal for nucleosome positioning in vivo . Positioning of nucleosomes over yeast transcription start sites ( TSSs ) is influenced by promoter sequence signals and trans acting factors ( Jiang and Pugh , 2009b; Radman-Livaja and Rando , 2010; Segal and Widom , 2009b ) . This observation , together with our finding that overlapping positions are commonly observed in yeast and human chromatin , motivated us to examine the relationship between TSSs and overlapping nucleosome positions ( resulting from alternative positions ) . We identified all TBB nucleosome positions that were closely associated with TSSs ( Jiang and Pugh , 2009a ) and organized them into groups consisting of either unique or overlapping TBB nucleosome positions for each gene ( Figure 5A and B ) . If a group contained several overlapping TBB nucleosome positions , the center of this group was defined as the numeric average of these positions . All groups were aligned at their centers and ranked by the spacing between TBB nucleosome positions within the groups ( Figure 5C ) . The TBB nucleosome positions exhibited a bimodal shape for at least three quarters of the genes with nucleosome-covered TSSs ( Figure 5C , ‘positions’ ) , suggesting that alternative positioning is a common feature of nucleosomes near TSSs . The number of sequenced read midpoints showed the same bi-furcating shape and was highly consistent between two experiments ( Figure 5C , ‘T1 midpoints’ and ‘T2 midpoints’ ) . The occupancy of nucleosome midpoints at each gene correlated well between two experiments ( T1 and T2 ) , indicating alternative nucleosomes can be reliably observed ( Figure 5D ) . The distribution of read midpoints was symmetric along the direction of gene transcription ( Figure 5—figure supplement 1 ) , and could not be explained by the sequence bias of MNase digestion ( Figure 5C ) . When we examined the midpoints attributed to each nucleosome position within each group ( Figure 5—figure supplements 1–2 ) , we found that most groups showed less than 2-fold variation in the read occupancy between alternative TBB nucleosome positions , suggesting that the different positions are significantly represented in the cell population and therefore are likely of biological relevance . These alternative positions were not a consequence of differential digestion at nucleosome edges ( Weiner et al . , 2010 ) , as both ends of paired-end sequencing reads showed two peaks that match the edges of alternatively positioned nucleosomes ( Figure 5E ) . Differential digestion at nucleosome edges would produce two peaks of read ends on one side and just one peak on the other read end ( Weiner et al . , 2010 ) . Of the genes with their TSSs covered by nucleosomes , approximately one quarter contain more than two TBB nucleosome positions at TSSs , suggesting that chromatin structure at the transcription initiation site is complex and heterogeneous . Alternative positions of nucleosomes around the TSS may allow the transcription start site of a gene to be exchanged between accessible and inaccessible states . To determine if this is the case , we analyzed the location of transcription start sites with respect to the alternative TBB nucleosome positions . We defined the nucleosome with its center position closer to the TSS as the 'proximal' nucleosome , and the nucleosome with its center position further away from the TSS as the 'distal' nucleosome ( Figure 6A ) . The unique , proximal and distal nucleosomes were digested by MNase to similar sizes , further suggesting that the alternative nucleosome positions are not a result of MNase digestion bias ( Figure 5F ) . 10 . 7554/eLife . 16970 . 023Figure 6 . Alternatively positioned nucleosomes and transcription pre-initiation complex . ( A ) Plots ( left ) showing the locations of TSSs relative to the average centers of all overlapping TBB nucleosome positions shown in Figure 3C ( 'All' , black ) to uniquely positioned nucleosomes ( magenta ) and to the distally ( red ) and proximally ( cyan ) positioned nucleosomes . The gray area marks the region covered by the nucleosome core . The cartoon diagrams on the right illustrate the location of the TSS relative to the nucleosome dyad . ( B ) Area showing the average occupancy of subunits of the pre-initiation complex ( TBP , TFIIA , TFIIB , TFIID , TFIIE , TFIIF , TFIIH , TFIIK and RNA polymerase II ( Pol II ) ; determined by ChIP-exo ) ( Rhee and Pugh , 2012 ) aligned at the center of the proximal ( cyan ) and the distal nucleosomes ( red ) , determined by the TBB approach . ( C ) Bar graph showing the distribution of TATA box and TATA-like sequences of all genes ( Rhee and Pugh , 2012 ) aligned at the dyad of the proximal nucleosomes . ( D ) Illustration of the 3-step model for transcription initiation mediated by alternatively positioned nucleosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 02310 . 7554/eLife . 16970 . 024Figure 6—source data 1 . Source data for Figure 6 , with genes and coordinates of TSSs , unique , proximal and distal nucleosome positions . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 024 Among uniquely positioned nucleosomes , TSSs weree enriched ~10 bp inside the boundary of the nucleosome core ( Figure 6A , magenta curve ) . Among the alternatively positioned nucleosomes , the proximal nucleosomes were located much closer to the TSS where the nucleosome core restricts the accessibility of these factors and therefore competes for active transcription ( Figure 6A , cyan curve ) . For the distal nucleosomes , TSSs were aligned near the edge of the nucleosome core ( Figure 6A , red curve ) , making these sites and the upstream regulatory elements accessible to other DNA binding factors . Alternatively positioned nucleosomes at TSSs gave rise to different accessibilities of the promoter elements within a population of cells , which may reflect the different activities in gene transcription . If one focused only on the average centers of these nucleosomes ( essentially the consensus centers of the +1 nucleosomes ) , one would draw the conclusion that nucleosomes occlude the TSSs of most genes ( Figure 6A , black line ) . To explore how alternative TBB nucleosome positions might influence transcription , we focused on the initial steps of this process: the assembly of the transcription pre-initiation complex ( PIC ) , which generally requires binding of TBP ( TATA-binding protein ) ; and the subsequent recruitment of general transcription factors and RNA polymerase II ( Pol II ) ( Buratowski et al . , 1989; Green , 2000; Orphanides et al . , 1996; Roeder , 1996 ) . Recently , it was suggested that the +1 nucleosome downstream of the TSS assists in the assembly of the PIC ( Rhee and Pugh , 2012 ) . However , the binding of the PIC extends into the boundary of the +1 nucleosomes ( Rhee and Pugh , 2012 ) , presenting a physical conflict between PIC binding and nucleosomes . To test if the resolution of alternatively positioned nucleosomes would provide a better understanding of the interplay between nucleosomes and the assembly of the PIC , we analyzed the occupancy of the PIC ( determined with the ChIP-exo method ( chromatin immunoprecipitation with lambda exonuclease digestion ) by Rhee and Pugh [Rhee and Pugh , 2012] ) around the identified distal and proximal nucleosomes , respectively . Most subunits of the PIC , including TBP , TFIIA , TFIIB , TFIIE , TFIIF , TFIIH and TFIIK , showed peak enrichment at the edge of the distally positioned nucleosome ( Figure 6B , cyan ) , suggesting that the +1 nucleosome in the distal position marks a boundary for the assembly of the PIC . This boundary may be provided by the positioned nucleosomes or by binding of one or more components of the transcription complex . In contrast , the proximally positioned nucleosome covered most of the regions enriched for PIC binding ( Figure 6B , red ) , suggesting that the +1 nucleosome in the proximal position is not compatible with the binding of transcription machinery . Intriguingly , TATA-box and TATA-like sequences were enriched in front of the proximal nucleosomes ( Figure 6C ) , suggesting that proximal nucleosomes do not block recognition of most TATA elements . Thus , the proximal nucleosome may represent a promoter state that is not associated with active transcription and is populated prior to the assembly of the PIC ( Figure 6D , Proximal state ) . The distal nucleosome may represent a promoter state in the process of transcription initiation ( Figure 6D , Distal state ) . Once transcription begins , the barrier of nucleosomes no longer hinders Pol II as its occupancy extends through the region covered by the +1 nucleosome ( Figure 6B ) . The location of uniquely positioned nucleosomes relative to TSSs was similar to that of the distal nucleosomes ( Figure 6A , magenta and red curves ) , suggesting that the binding and assembly of PIC may encounter less competition from chromatin at these genes . We found that genes with uniquely positioned nucleosomes at their TSSs were enriched in highly expressed genes ( Newman et al . , 2006 ) ( top 500 genes , p = 0 . 001 , Fisher’s exact test ) . More strikingly , we found that 73% of the genes encoding ribosomal proteins had TSSs with uniquely positioned nucleosomes ( p<0 . 00001 , Fisher’s exact test ) , highlighting a connection between the configuration of nucleosomes ( unique vs alternative ) and gene expression . DNA sequence features , such as periodic occurrence of dinucleotides ( AA/AT/TA/TT or CC/CG/GC/GG ) and consecutive dA or dT sequences ( poly ( dA:dT ) ) , can influence nucleosome positions ( Struhl and Segal , 2013 ) . We therefore tested the possibility that DNA sequence determinants may underlie the difference between uniquely , proximally , and distally positioned nucleosomes around TSSs . The frequency of dinucleotides occurred periodically ~10 bp from the positions of either proximal or distal nucleosomes and resembled the periodic feature of nucleosomes in yeast genome ( R2 = 0 . 75 , Figure 7A ) . This periodicity was less pronounced in the uniquely positioned nucleosomes ( R2 = 0 . 40 between dinucleotide frequency in uniquely positioned nucleosomes and all nucleosomes in yeast genome; Figure 7A ) . Autocorrelation analysis of the dinucleotide frequency revealed a lack of periodicity in DNA sequences around the uniquely positioned nucleosomes , further highlighting the difference between uniquely and alternatively positioned nucleosomes ( Figure 7B ) . 10 . 7554/eLife . 16970 . 025Figure 7 . Sequence features of uniquely and alternatively positioned nucleosomes . ( A ) Plots showing the frequency of AA/AT/TA/TT and CC/CG/GC/GG dinucleotides of DNA sequences aligned at the TBB nucleosome positions of either unique , proximal or distal nucleosomes at gene promoters ( illustrated by the diagrams ) . ( B ) Plots showing the normalized dinucleotide frequency ( smoothed with a 3 bp window ) of DNA sequences aligned at unique nucleosomes , alternative nucleosomes ( proximal + distal ) at gene promoters or all TBB nucleosome positions , and the autocorrelation analysis ( performed in MATLAB ) of the dinucleotide frequency within nucleosome core ( −73–73 bp ) . ( C , D ) Plots showing the frequency of poly ( dA:dT ) 6 sequences aligned at unique nucleosomes or the average centers of alternatively positioned nucleosomes . The positive direction of the x-axis indicates 5’ to 3’ for all transcripts . Black arrows mark the enriched poly ( dA:dT ) 6 signals . Overall , 1172 genes contain unique nucleosomes and 3469 genes contain proximal and distal nucleosomes at their promoters . Gray traces present the analysis for a random permutation control of selected promoters and the location of nucleosome positions ( A , C , D ) . The number of these random locations matches the number of nucleosomes in each plot . DOI: http://dx . doi . org/10 . 7554/eLife . 16970 . 025 Poly ( dA:dT ) tracks are important for nucleosome depletion and correlate with transcriptional activity ( Hughes et al . , 2012; Iyer and Struhl , 1995; Raveh-Sadka et al . , 2012; Segal and Widom , 2009a; Struhl , 1985 ) . When we examined the occurrences of poly ( dA:dT ) sequences at least 6 nucleotides in length , we found that the uniquely positioned nucleosomes were flanked by a strong enrichment of poly ( dA:dT ) 6 sequences ( Figure 7C ) . A similar but weaker enrichment of poly ( dA:dT ) 6 sequences was also observed flanking both the proximally and the distally positioned nucleosomes ( Figure 7D ) . In comparison , no enrichment of poly ( dA:dT ) 6 sequences was observed in randomly selected DNA sequences ( Figure 7C and D , gray curve ) . Overall , uniquely positioned nucleosomes and alternatively positioned nucleosomes presented different DNA sequence features , suggesting that these sequence determinants may contribute to the different configurations of chromatin states at gene promoters . Knowing the precise location of nucleosomes in vivo is key to understanding how diverse biological processes are regulated ( Jiang and Pugh , 2009b ) . Although MNase-seq methods have been widely used for mapping nucleosomes in many organisms , the commonly reported consensus centers of nucleosomes do not precisely reflect the location of nucleosomes with respect to DNA sequence . We developed a new computational method to accurately determine the positions of nucleosomes at base pair resolution based on paired-end sequencing data , providing a general and accessible approach to accurately map the basic chromatin structure in eukaryotes . The periodicity in the frequency of dinucleotide AA/AT/TA/TT and CC/CG/GC/GG motifs has been observed in vivo in several organisms , including unicellular fungi ( Albert et al . , 2007a; Brogaard et al . , 2012 ) , insects ( Mavrich et al . , 2008 ) , chicken ( Satchwell et al . , 1986 ) and human ( Gaffney et al . , 2012 ) , and was thought to strongly influence the positioning of nucleosomes ( Kaplan et al . , 2009; Struhl and Segal , 2013 ) . However , this feature was only obvious when the sequences of nucleosomal DNA fragments exactly 147 bp in length were selected for analysis ( Albert et al . , 2007a; Gaffney et al . , 2012; Mavrich et al . , 2008; Segal et al . , 2006 ) , and was much weaker or even disappeared around the consensus centers of nucleosomes that were determined with an entire data set ( Brogaard et al . , 2012; Valouev et al . , 2011 ) . Thus , differences in the selection and analysis of MNase-seq data may result in different conclusions about nucleosome positioning in vivo ( Gaffney et al . , 2012; Valouev et al . , 2011 ) . With the high-precision nucleosome position maps generated by TBB , we directly demonstrated that the DNA sequences around the mapped nucleosome positions exhibited 10 bp periodicity in the frequency of dinucleotide motifs in both yeast and human cells , arguing that this sequence feature is a universal signal for positioning nucleosomes in vivo at the whole genome level . Positioning of nucleosomes is dynamic and linked to many biological processes . Identifying overlapping nucleosome positions that cannot co-exist on the same DNA molecule at the same time provides a snapshot of possible chromatin states in dynamic processes . We observed at least two states of chromatin where gene transcription is initiated: a state where a nucleosome covers the transcription start site and competes with transcription machinery , but leaves the core promoter sequences mostly accessible ( 'proximal state' ) ; and another state where a nucleosome barely covers the transcription start site and seems poised for the assembly of the pre-initiation complex ( 'distal state' ) . We propose a 3-step model ( Figure 6D ) to describe the transcription initiation steps influenced by nucleosomes , including a state where RNA polymerase transcribes through the nucleosome covered sequence ( 'open state' ) . The transition from the 'proximal state' to the 'distal state' and to the 'open state' could be actively driven by the assembly of the PIC; in this case , different nucleosome states may reflect rate-limiting steps during transcription initiation . Alternatively , this transition could be controlled by transcription factors , chromatin modulators and even the preference of DNA sequences as a means to regulate and assist gene transcription . Interestingly , we found that TSSs of highly expressed genes are enriched with uniquely positioned nucleosomes , which resemble nucleosomes in their 'distal state' . It is possible that the +1 nucleosomes of these most highly expressed genes are oriented in a way that allows constitutive assembly of the PIC and therefore frequent transcription activity . Further genomic analysis and mechanistic studies will shed light on the generalization and the underlying principles of this model . DNA sequences may play a role in determining whether nucleosomes have unique or alternative positions . Unique nucleosome positions may result from the combination of the strong nucleosome boundary signal from poly ( dA:dT ) sequences and the lack of a rotational positioning signal from periodic dinucleotide motifs . Conversely , alternatively positioned nucleosomes may be a consequence of a weak boundary signal and a strong rotational positioning signal that allows nucleosomes to interconvert between multiple equivalent positions . The exact positions of these nucleosomes and how alternative nucleosomes are positioned relative to each other may be further controlled by other factors associated with promoters , such as chromatin modifiers , remodelers and transcription factors . The configurations of nucleosomes ( unique vs . alternative ) may influence the transcription of a gene; if promoter DNA sequences contain information that helps to specify the nucleosome configurations , eukaryotic genome could use nucleosome configurations as a means to encode transcriptional programs . It is worth noting that the template-based Bayesian approach does not capture all possible positions of nucleosomes within the cell population – it identifies the nucleosome positions that are represented with strong statistical confidence ( Materials and methods ) , reflecting the different configurations of nucleosomes that are likely to be relevant in biological processes . Revealing these heterogeneous configurations of nucleosomes in cells will provide new insights into how chromatin regulates or is influenced by biological processes . Furthermore , mutually exclusive positions of protein-DNA complexes can imply dynamic information . As an example , the alternatively positioned nucleosomes at TSSs suggest a change in chromatin structure during transcription initiation . Similar analytic approaches can be applied to other systems to reconstruct dynamic processes . Detailed conditions of cell culture , chromatin isolation , MNase digestion and DNA extraction have been previously described ( Zhou and O'Shea , 2011 ) . Briefly , 100 OD600 units of yeast cells ( EY57; genotype K699 , ade2-1 trp1-1 can1-100 leu2-3 , 112 his3-11 , 15 ura3 GAL+ ) growing in synthetic complete medium at 30°C with shaking were harvested at OD600 ~0 . 5 for each experiment . Cells were crosslinked with 1% formaldehyde for 15 min at room temperature and quenched with 125 mM glycine for 5 min . Collected cells were lysed mechanically with glass beads at 4°C with Mini-Beadbeater-24 ( Biospec ) in lysis buffer ( 50 mM HEPES , pH 7 . 5 , 150 mM NaCl , 2 mM EDTA , 1% Triton X-100 , 0 . 1% Sodium-Deoxycholate ) and the chromatin pellet fraction was digested with either 0 . 5 , 1 , 2 or 4 U MNase ( Worthington Biochemical ) at 37°C for 30 min . DNA fragments corresponding to mono-nucleosomes were extracted from an agarose gel after electrophoretic separation . For each experiment , 0 . 5 μg of mono-nucleosomal DNA ( ~5 pmol ) was used for sequencing library construction according to the Illumina Truseq paired-end protocol . ~1 pmol of paired-end ligated library was recovered from the agarose gel for each experiment before PCR amplification . Since the amount of ligated library was sufficient for sequencing , we only performed 4 cycles of PCR amplification with the Illumina primer cocktail to obtain a double stranded sequencing library . The number of sequencing clusters generated on the flow-cell represented at most 1/10 , 000 of the amplified library , such that the chance was very small to have more than one copy of duplicated library molecules clustered on the flow-cell ( p<1 . 8 × 10–5 , binomial distribution ) . Therefore , we treated each paired-end read as an individual nucleosome from a single cell . Libraries were sequenced with an Illumina Hiseq 2000 and the paired-end sequencing reads were aligned to the S . cerevisiae genome Scer02 with bowtie 2 . 0 ( Langmead et al . , 2009 ) . We mapped reads with two or fewer mismatches , and insert lengths were restricted to sizes between 100 bp and 300 bp . Reads with multiple reportable alignments were randomly assigned to one such alignment . The midpoints of all sequenced DNA fragments were randomly rounded to an integer genomic coordinate and used to generate sequencing read midpoint occupancy maps . We obtained 33 . 7 million and 34 . 3 million mappable paired-end reads for the T1 and T2 experiments , respectively . All analyses presented in this work were done with the T1 dataset unless specified . The consensus centers of nucleosomes were determined based on the fitted coefficient β with a standard Parzen window peak calling method ( Tsankov et al . , 2010 ) . TBB nucleosome positions were determined at FDR ≤ 0 . 05 , where the null distributions of nucleosome positions were estimated from randomly sampled nucleosome reads distributed based on the local read density . We developed two different null distributions: a random null distribution where simulated sequencing reads were assumed to be uniformly distributed within each region , and an MNase digestion-aware null where the simulated sequencing reads were sampled to match the observed dinucleotide end frequency of the paired-end reads . We applied the digestion-aware null to identify TBB nucleosome positions with a threshold of FDR ≤ 0 . 05 , resulting in 62 , 035 consensus centers and 120 , 347 TBB nucleosome positions in the T1 experiment , and 61 , 532 consensus centers and 129 , 505 TBB nucleosome positions in the T2 experiment . When we attempted to identify possible nucleosome positions that were entirely due to the sequence bias of MNase digestion , we analyzed the simulated MNase digestion data though our TBB pipeline and applied the random null distribution with a threshold of FDR ≤ 0 . 05 . This allowed us to identify 49 , 739 and 50 , 076 positions to analyze and correct for influence of MNase digestion bias . Analyses on human nucleosomes were performed using paired-end sequencing reads with length between 126 bp and 184 bp ( Gaffney et al . , 2012 ) . We randomly selected a 10 Mbp region on human chromosome 12 , position 38 , 000 , 000 bp to 48 , 000 , 000 bp , to test the TBB approach . In total , we identified 43 , 991 consensus centers of nucleosomes and 251 , 948 TBB nucleosome positions in this selected region . All the identified positions were used in our analysis . All TBB nucleosome positions determined in T1 were used for the dinucleotide frequency analysis unless specified . DNA sequences from 100 bp upstream to 100 bp downstream of each TBB nucleosome position were aligned and the frequency of AA/AT/TA/TT and CC/CG/GC/GG dinucleotides was calculated for every base pair . This calculated di-nucleotide frequency was then normalized to the average dinucleotide frequency and presented in the plots . For the permutation of TBB nucleosome positions , we first randomly selected 62 , 035 TBB nucleosome positions ( the same number as consensus centers of nucleosomes ) . For each of these positions , we randomly shifted the genome coordinate of the position by 0–4 bp with an overall average of 2 bp , or by 0–10 bp with an overall average of 5 bp . The frequency of AA/AT/TA/TT dinucleotides with respect to these permutated positions is presented in the plot Figure 4D , in comparison to that of the original TBB nucleosome positions . To determine if the observed overlapping TBB nucleosome positions are biologically meaningful , we excluded TBB nucleosome positions whose occupancy is less than 4 . 8 reads ( about half of the median sequencing read occupancy of all TBB nucleosome positions ) , which account for ~5% of the detected TBB nucleosome positions . The position occupancy was calculated as the average read count within a 3 bp window centered at TBB nucleosome positions . We then combined individual TBB nucleosome positions that are separated by less than 80 bp into groups of overlapping nucleosomes ( overlap by ~ half of a nucleosome ) . In total , 57 , 643 groups containing either unique or overlapping TBB nucleosome positions were identified in the yeast genome . Groups that contain more than 9 nucleosome positions ( 139 groups , 0 . 24% ) were further excluded from this analysis because they likely reside in regions with delocalized nucleosomes . The remaining groups were compared to the positions of TSSs of all ORFs in yeast genome ( Jiang and Pugh , 2009a ) , and 4641 were identified to overlap with TSSs ( TSSs within 73 bp distance of any TBB nucleosome position in the group ) . All 4641 were analyzed and presented in Figure 5C . 3307 of these groups contain no more than 2 TBB nucleosome positions and their analyses are shown in Figure 5—figure supplement 2 . The occupancy data for the subunits of the PIC was generated from the supplementary data of the work of Rhee and Pugh ( Rhee and Pugh , 2012 ) . The genome coordinates for PIC occupancy were calculated based on the positions of the +1 nucleosomes kindly provided by Ho Sung Rhee . Both TATA-box sequences and TATA-like sequences identified by Rhee and Pugh ( Rhee and Pugh , 2012 ) were analyzed in this study and presented in Figure 6C . However , a substantial fraction of the TATA-like sequences are located next to or even after the annotated transcription start site – TATA-like sequences of ~600 genes are within 5 bp of TSSs . If we exclude these genes from our analysis , the majority of the TATA-box/TATA-like sequences are located outside the boundary of the proximal nucleosomes . We estimated the digestion variability template from the distribution of sequencing read lengths . This digestion variability reflects a combination of factors that influences the size of protected nucleosomal DNA , such as over- or under- digestion of intact nucleosomes . This estimation was based on a simple , symmetric model of MNase digestion for each sequenced fragment . Using this model , we obtained a nonparametric maximum likelihood estimator of the distribution of digestion variability ( e ) at each end of a sequenced fragment ( e1j , e2j ) . Therefore , we can infer the digestion variability template as the distribution of the variability in paired-end read midpoints due to digestion . We denoted the length of each paired-end read j as ℓj and assumedℓj=ℓ0+e1j+e2j , e1j , e2j∼IID . ℓ0 is defined as the length of DNA covering the histone core octamer ( 147 bp ) ( Kornberg and Lorch , 1999 ) and the e1j and e2j terms are the digestion errors at each end of the read j . We assumed MNase has the same propensity towards over-or under-digestion at each end on average; therefore , we modeled these errors as bounded and symmetric between the ends of sequencing reads . Along the genome , the distributions of digestion errors at the ends of each read are mirror images of each other , but the two errors of the same read were independent . So positive values of e1j or e2j imply that DNA sequence not covered by the histone core octamer was left at the ends after MNase digestion , and negative values imply that MNase over-digested the nucleosomes . Under this model , each sequencing read’s midpoint varies around its nucleosome’s center according to the distribution of dj≡12 ( e1j−e2j ) . The template t→ specifies this distribution , expressed in vector form and transformed to account for the random rounding of fragment centers to integer positions . Hence , tk=P ( dj=k ) +12 ( P ( dj=k−12 ) +P ( dj=k+12 ) ) For k=−w , … , w , yielding a vector t→ of length 2w+1 . To setup the estimation task , we defined two probability distributions , p ( i ) =Pr ( lj=i ) , q ( i ) =Pr ( e1 , j=1 ) . It is known that lj≥0 , which implies e1 , j , e2 , j≥−⌊l02⌋ . If the longest observed fragment length is lmax , we also have Pr ( lj >lmax ) =0 . The non-parametric Maximum Likelihood Estimation ( MLE ) also enforces lj≤lmax , which implies e1 , j , e2 , j≤lmax−l0+⌊l02⌋ . We could then writep ( i ) =∑k=−⌊l02⌋lmax−l0+⌊l02⌋q ( k ) q ( i−l0−k ) . The resulting log-likelihood for the observed fragment lengths isℓ ( q ) =∑j=1Mlog p ( lj ) . We maximized l ( q ) numerically , using a multivariate logit transformation on the values q ( k ) to avoid bounded optimization . Using the L-BFGS algorithm ( Zhu et al . , 1997 ) on a laptop with a Core i5 processor and 8GB of RAM , this maximization requires ~40 s for a typical experiment . The complexity of this computation depends only on the number of distinct read lengths observed ( i . e . , 151 bp , 152 bp , etc ) . We obtained the template distribution t→ from q via a convolution sum and linear transformation . We first obtained the distribution of e1−e2 viau ( i ) =P ( e1−e2=i ) =∑k=−⌊l02⌋lmax−l0+⌊l02⌋q ( k ) q ( k−i ) . We finally transformed the distribution u ( i ) to the desired template t ( i ) by accounting for random rounding , ast ( k ) =12u ( 2k−1 ) +u ( 2k ) +12u ( 2k+1 ) . The resulting digestion variability template accurately reflects both variation from enzymatic digestion and the details of the preprocessing . Each chromosome was segmented into disjoint , contiguous regions with similar statistical properties ( sequencing coverage ) . This was done to control for local variation in sequencing coverage and nucleosome occupancy , allowing for efficient estimation of local structure with minimal sensitivity to distant observations . To accomplish this , we started from ORFs and refined these natural demarcations into a statistically useful segmentation . We first enumerated all open reading frames ( ORFs ) and intergenic regions on each given chromosome . Merging overlapping ORFs into single segments yielded a starting set of contiguous , non-overlapping segments . We then iteratively merged the most similar short segments until all segments exceed a minimal length ( 800 bp ) . The similarity was defined based on sequencing read midpoints per base pair within each segment . This yielded a segmentation for which each segment has sufficient length to estimate local properties of the nucleosome position distribution . For the S . cerevisiae genome , we obtained a total of 5135 segments for our T1 experiment with a mean segment length of 2351 bp ( minimum length of 800 bp and maximum length of 14 , 499 bp ) . The results on the T2 experiment were extremely similar with 5118 segments and a mean length of 2359 bp . This segmentation was fixed and used in all subsequent computation . The region of human chromosome 12 , position 38 , 000 , 000 bp to 48 , 000 , 000 bp was segmented based on the annotated introns and exons of human genome hg18 ( Roche , NimbleGen ) . Other segmentation techniques based on the biological structure of each chromosome can be used and may be more appropriate for mammalian genomes . Our probabilistic model for the observed read midpoints captures the variation due to nuclease digestion as well as the variation from biological sources ( for example , biological heterogeneity among cells , stochastic fluctuation in the selection of fragments for sequencing , and other factors in sample preparation and analysis ) . To accomplish this , we extended discrete convolution models from signal processing to account for the particular types of structure and variation found in MNase-seq data . For each chromosome , we defined yk as the number of sequencing read midpoints aligned to position k , where k ranges from 1 to N ( the length of the chromosome ) . Given the segment s:{1 . . . N}⟶{1 . . . S} , which maps the N base pair locations to S segments on this chromosome , we positedyk∣λk ~ Poisson ( λk ) λ→ ( N×1 ) ≡X ( N× ( N−2⌊ℓ0/2⌋ ) ) β→ ( ( N−2⌊ℓ0/2⌋ ) ×1 ) βk>0 for k=⌊ℓ0/2⌋+1…N−⌊ℓ0/2⌋logβk∼Normal ( μsk , σsk2 ) where X specifies the contribution of a nucleosome positioned at kto the expected number of reads at position m due to digestion variability , and sk is the region corresponding to position k . X was defined as the matrix generated by the convolution of t→ with a sequence of length N . ℓ0 stands for 147 bp as mentioned before . To complete the model specifications , we placed priors on μs and σs2 . We use independent conjugate priors for σs2 , assuming 1σs2∼Gamma ( α0 , γ0 ) . Our priors for μs are fully conjugate and independent across segments; we assumed p ( μs∣σs2 ) ∼N ( μ0 , σs2nsτ0 ) where ns is the length of segment s . To estimate β and detect individual nucleosome positions , we sampled from the posterior distribution of β→ using a Markov chain Monte Carlo ( MCMC ) sampler . By iterating through a carefully engineered sequence of random draws , this sampler obtains approximate samples from the given posterior . These draws of β→ form the basis for all of our subsequent summaries , particularly the detection of nucleosomes positions ( using selected posterior probabilities ) and the estimation of consensus positions ( using selected posterior expectations ) . This sampler consisted of two alternating updates . At each iteration r , our algorithm The first is a standard conjugate normal update , given the log-normal model for β→ , and operates independently across segments . Full details can be found in Blocker and Airoldi ( 2016 ) . The conditional posterior of β→ ( t ) ∣ ( μ→ ( t ) , σ→2 ( t ) ) is not part of any standard family , so we turned to Hamiltonian Monte Carlo ( HMC ) . The dimensionality of β→ makes a single HMC update for the entire vector both computationally infeasible and numerically unstable . For efficient computation , we leveraged the conditional independence structure of this conditional posterior . Subvectors of β→ separated by at least 2w entries are conditionally independent given ( μ→ ( t ) , σ→2 ( t ) ) and the entries of β→ between them . Consider the subvectors β→[j1:j2] and β→[k1:k2] , with j1<j2<k1<k2 . The elements of β→[j1:j2] only affect j1<j2<k1<k2 , and the elements of β→[k1:k2] only affect λ→[j1−w:j2+w] . Hence , if β→[k1:k2] , then β→[j1:j2] and β→[k1:k2] are conditionally independent given μ→ and σ→2 . We first fixed the length of each subvector that will be updated via a single HMC step to B>4w . Next , considered two partitions of β→ into subvectors:β→[1:B] , β→[B+2w+1:2B+2w] , … , β→[nb ( B+2w ) +1:N] , β→[B/2+1:3B/2] , β→[3B/2+2w+1:5B/2+2w] , … , β→[nb ( B+2w ) B/2+1:N] . Within each partition , the subvectors are conditionally independent , and the union of these partitions includes all entries of β→ . For each iteration of our sampler , we cycled through each of these partitions , updating each subvector of β→ with an HMC step . As each subvector within each partition is conditionally independent , we can execute all HMC steps in parallel for each partition . This allows us to distribute the computational burden over hundreds of CPUs . Each of these distributed HMC steps is computationally efficient , as the log-conditional posterior’s value and gradient can be computed via a convolution , lowering the computational cost to o ( BlogB ) with the fast Fourier transform . A Python implementation of the sampler is available on GitHub , http://www . github . com/airoldilab/cplate . We defined consensus centers of nucleosomes as the average centers of possible overlapping nucleosomes within a small region . Each such position represents a cluster of nearby nucleosome positions and is similar to a 'nucleosome' obtained by standard methods ( Albert et al . , 2007; Shivaswamy et al . , 2008; Tirosh , 2012; Tsankov et al . , 2010 ) . Based on the nucleosome position distribution β→ , we determined the consensus centers of nucleosomes as the centers obtained from a standard Parzen window peak calling method ( Tirosh , 2012 ) . We smoothed each draw of β→ obtained from our sampler with a Gaussian window with a standard deviation of 20 bp . We then averaged these smoothed maps across draws and performed a greedy search for local maxima , enforcing a minimal spacing constraint of 147 bp . The comparisons presented in the main article are based on the standard practice of applying the same smoothing and search procedure directly to the midpoint counts ( Albert et al . , 2007; Shivaswamy et al . , 2008; Tsankov et al . , 2010 ) . We defined nucleosome positions as a position along the genome with a greater chance observing a nucleosome than expected under a locally uniform distribution of sequencing read midpoints . With this definition in hand , we used carefully selected posterior probabilities ( estimated using the above algorithm ) to find those positions with strong support from the observed data . Formally , to detect individual nucleosome positions , we first definedCp , l ( k ) =∑i=−ppβk+i∑i=−llβl+i for each position k . We then estimated Cp , l ( k ) =∑i=−ppβk+i∑i=−llβl+i for each k using the MCMC sampler described previously . We then calibrated these Bayesian summaries against those implied by a null distribution of nucleosome positions within each region , constraining the number of reads within each segment to match the number observed and using information on the digestion variation in the form of the estimated template . We approximated this null distribution by repeatedly randomly permuting the observed reads within each segment . In order to generate the Bayesian summaries implied by the null distribution described above , we then ran the proposed MCMC sampler on paired end sequencing reads drawn from the null , using the template and segmentation estimated from the observed data . From the sampler’s output , we obtained an estimate of the distribution of P ( Cp , l ( k ) > ( 2p+1 ) / ( 2l+1 ) ∣y→ ) over positions k under the null . We compared this to the distribution of posterior probabilities for the observed data and set a detection threshold to control the FDR using the method of Storey and Tibshirani ( Storey and Tibshirani , 2003 ) . For the datasets analyzed in this work , a threshold of 0 . 8 on P ( Cp , l ( k ) > ( 2p+1 ) / ( 2l+1 ) ∣y→ ) typically yields an FDR of 5% or less for the experimental data . We have considered two null distributions in this work , both of which preserve the sequencing coverage within the identified regions as the experimental dataset . The first is a random null distribution , where simulated sequencing reads are assumed to be uniformly distributed within each region . The second is a MNase digestion-aware null , where simulated sequencing reads are assumed to be uniformly distributed subject to the observed distribution of the dinucleotide ends . These two null distributions are applied to identify TBB positions in two scenarios . The MNase digestion-aware null is used to identify experimental TBB positions that are statistically significant over the sequence bias of MNase digestion . The random null is used to set the threshold for determining possible nucleosome positions that result from MNase digestion bias over a uniform background . In all cases , the false discovery rate is controlled to be less than 5% , allowing a fair comparison between different data sets . For each region of the genome , we tabulated the full table of dinucleotide counts from all aligned paired-end reads . For each pair of cut dinucleotides , we enumerated all potential paired-end reads with matching cut dinucleotides with centers falling in the given region . We then sampled uniformly from this set of potential reads with replacement to match the observed number of reads with the given cut dinucleotides . This yields a sample of reads exactly matching the observed cut dinucleotide distribution with fragment centers random within each region conditional on cut dinucleotides . These simulated controls were then passed through the same pipeline as the observed reads and used to set thresholds based on the stated FDR-controlling procedure . All comparisons are based on matched distances , as in ( Brogaard et al . , 2012 ) . For example , when we compared our identified consensus centers of nucleosomes to those identified from a previous study ( Jiang and Pugh , 2009a ) , we calculated the distance between every center in our dataset to its nearest center in the published study and summarized them into distance probability ( Figure 2B–E ) and cumulative ( Figure 2—figure supplement 3–6 ) plots . Similarly , when we evaluated the performance of the TBB method on an in silico MNase-seq dataset , we computed the distance between every identified TBB position from the in silico dataset and its nearest simulated nucleosome position . When we evaluated reproducibility between replicates , we considered the set of all best-match distances obtained by matching each replicate against the other to ensure symmetry . We used two methods to generate random nucleosome positions and consensus centers as controls to estimate the detection accuracy of the TBB approach . In the first method , we randomly generated the genomic coordinates of nucleosome positions on each chromosome to match the number of experimentally detected TBB nucleosome positions or consensus centers . In the second method , we took into account the spacing features between TBB nucleosome positions or consensus centers . In this way , the randomly generated genomic coordinates and the experimental determined data have the same distribution of spacing between adjacent positions . The random nucleosome positions or consensus centers maps generated by these two methods yielded similar results and only the result of the second method is shown in the plots ( Gray trace , Figure 2—figure supplements 3–5 ) . The median distance between the random nucleosome positions and the chemical positions is 18 bp , and is 36 bp for the spacing between the random nucleosome positions and the TBB positions determined here . The median distance between the random consensus centers and either the reference centers or the consensus centers determined here is 49 bp in both cases . The median distance between the random nucleosome positions and the chemical positions is much smaller than the rest of the comparisons because the number of chemical positions was three times larger than the number of TBB nucleosome positions and 5 times larger than the number of consensus centers determined in this study . The true positions of nucleosomes in vivo are generally not available with the current experimental and computational approaches . We thus performed a set of in silico experiments to evaluate the performance of the TBB method in a setting where ground truth of nucleosome positions is available . We first simulated the true positions of nucleosomes: we generated the primary nucleosome positions to represent the most frequent ( strongest ) positions among a set of overlapping nucleosomes , and then added alternative positions around the primary positions to account for the other overlapping nucleosomes ( Figure 3A ) . In each set of in silico experiments , we systematically varied the occupancy ( coverage ) , spacing ( offset ) , and relative strength of primary and alternative nucleosome positions ( effective magnitude ) according to a factorial design that spans the 5th to 95th percentiles of the corresponding properties observed in our yeast experiments ( Figure 3A; Supplementary file 2 ) . At each of the simulated nucleosome positions , we randomly generated sequencing reads based on the digestion variability template estimated from T1 , and constructed 10 artificial chromosomes to represent the in silico MNase-seq data sets . The occupancy of sequencing read midpoints in these simulated data sets resembles that determined from our biological samples . We then applied the TBB approach to identify nucleosome positions in these in silico data sets and compared them with the simulated nucleosome positions ( both simulated primary and alternative positions ) . We found that the TBB approach can reliably identify primary nucleosome positions ( 50% and 85% of the primary positions within 2 bp and 4 bp , respectively ) across all settings . Detection of the alternative positions is similarly reliable ( 50% and 75% within ~3 bp and ~7 bp , respectively ) if the alternative positions are populated at least 1/3 as frequently as the nearest primary positions ( effective magnitude smaller than 0 . 6 ) ( Supplementary file 1 ) . Detailed methods and discussion about the in silico validation can be found in Extended Experimental Procedures . To estimate the precision of the TBB approach in identifying nucleosome positions , we simulated nucleosome positions on a set of artificial chromosomes , generated in silico MNase-seq sequencing read midpoint datasets based on the experimental digestion variation , estimated the in silico TBB nucleosome positions with these in silico datasets , and compared them to the original simulated nucleosome positions . The differences between these identified TBB in silico positions and the original simulated positions reflect the precision of the TBB approach . To mimic the organization of nucleosomes in the genome , we simulated nucleosome positions based on observed in vivo organization around genes and constructed simulated artificial chromosomes with units of genes . Each artificial chromosome contains 1100 genes , and each gene was 3501 bp in length , consisting of a 1000 bp promoter region before its transcription start site ( TSS ) and 2500 bp following the TSS . The in vivo organization of nucleosomes around genes was determined from the identified consensus centers from the T1 experiment and averaged across all ORFs . As traditionally annotated , the nucleosomes after the TSS were numbered incrementally from +1 , and the nucleosomes before TSS were numbered decrementally from −1 . The average positions of these nucleosomes relative to TSSs were used in the construction the simulated nucleosome positions . Meanwhile , the number of sequencing reads within each consensus position was used to simulate the in silico MNase-seq datasets . To test the ability of the proposed model to identify overlapping nucleosomes , we built overlapping nucleosome positions into our simulation . We first generated nucleosome positions downstream of the TSS ( corresponding to the positions of the +1 , +2 , +3 , … nucleosomes ) and upstream of the TSS ( corresponding to the positions of the −1 , −2 , −3 , … nucleosomes ) to represent the most frequent ( strongest ) positions among a set of overlapping nucleosomes ( termed 'primary positions' ) . Then we added positions around the primary positions ( termed 'alternative positions' ) to represent overlapping nucleosomes . In the simulation , we varied the relationships between the primary positions and the alternative positions to explore the performance of the TBB model . For simplicity , we assumed the alternative positions are symmetric to the primary positions . We designed a simulation with three factors , varied at the gene level: coverage ( the expected number of reads per gene ) , the spacing between primary nucleosome positions and alternative positions ( which we refer to as offset ) , and the relative magnitudes of primary and alternative positions ( which we refer to as effective magnitude and is defined as the percentage of reads attributed to the primary positions ) ( Figure 3A ) . Coverage had 10 levels , spanning the 5th to 95th percentile observed gene-level coverages in increments of 10% . Alternative position spacing had 10 levels , spanning from 0 bp ( no alternative positions ) to 45 bp in increments of 5 bp . We tested 11 levels for the relative magnitude between alternative positions and primary positions , spanning from 0 ( no alternative positions ) to 1 ( alternative positions of the same magnitude as primary positions ) in increments of 0 . 1 , where the effective magnitude ranged from 1 to 1/3 . We used a full factorial design on these three factors , yielding 1100 distinct relationships between the primary and alternative positions for each of 10 simulated chromosomes . To generate our in silico MNase-seq dataset , we followed a modified version of the generative process described above . For each gene , we first drew coefficients for its subset of β→ from an upper-truncated log-normal distributed with parameters estimated from those regions in T1 with similar coverage . These corresponded to 'background' positions and introduce a realistic level of variation into the simulations; biologically , such background could originate from a combination of low-occupancy nucleosome positions and naked DNA obtained during the MNase-seq process . Then , we set the entries of β→ corresponding to the gene’s primary and alternative positions deterministically . The sum of the coefficients for these positions was fixed to the total occupancy of the gene minus the sum of the background positions . The relative magnitudes were determined by the design described above , with two alternative positions placed symmetrically around each primary position at the designated spacings . Thus , for a given level of coverage , the expected number of reads within each cluster was fixed , but its distribution across primary and alternative positions varies . We convolved these β→ vectors with the template estimated from the experimental data to obtain vectors of expected read counts λ→ . Finally , we generated y→∼iid Poisson ( λ→ ) to obtain simulated read counts . This entire procedure was repeated for each replicate , yielding 10 artificial chromosomes of length 3 , 851 , 100 bp each . The simulated read midpoint occupancy was similar to the midpoint occupancy observed in vivo around ( Figure 3B ) . Based on our in silico results , the TBB method appears extremely accurate for calling primary nucleosome positions . It can estimate 50% of such positions within 2–3 bp and 95% within 4–5 bp for all simulated conditions , as shown in Supplementary file 1 Its performance remains strong for the estimation of alternative positions . As the data in Supplementary file 1 hows , more than 50% of the simulated alternative positions were mapped within 2 bp when the effective magnitude is less than 0 . 71 ( alternative positions populated at least as much as 20% of their corresponding primary positions ) . When the effective magnitude reaches 0 . 56 or less ( populated as much as 40% of their corresponding primary positions ) , we mapped over 50% of alternative positions within a single base pair . With the spacing between the alternative and primary positions ranging from 5–45 bp , the median error for estimating alternative positions is no more than 2 bp . We observed stronger dependence of the TBB method’s performance on the spacing between alternative and primary positions: we generally attained higher reliability for the larger offsets , with over 85% of alternative positions estimated within 8 bp when the offset is 10 bp , and within 6 bp when the offset is 40 bp ( http://www . github . com/airoldilab/cplate ) . All sequencing data are deposited in the NCBI SRA database under accession number SRP023122 . All software for the template-based Bayesian model and in silico MNase-seq experiments used in this paper are available at http://www . github . com/airoldilab/cplate .	Plants , animals and other eukaryotes wrap their DNA around complexes of proteins called histones to form repeating units known as nucleosomes . The interaction between histones and DNA is strong , and so the DNA region inside a nucleosome has limited access to other proteins , including those that drive the expression of genes . Moving a nucleosome slightly can change the access to its DNA and significantly impact how the genes in the region are regulated . Nevertheless , determining the position of nucleosomes accurately or testing how nucleosomes are different between individual cells are challenging tasks . Most methods for identifying nucleosomes use an enzyme called micrococcal nuclease ( or MNase for short ) to break down the DNA that isn’t protected in nucleosomes , followed by high-throughput DNA sequencing to identify the DNA fragments that remain . However , this technique , known as MNase-seq , is limited because it only measures an average location of the nucleosomes across millions of cells . Now , Zhou , Blocker et al . have developed a new computational approach to identify nucleosome positions more accurately using MNase-seq data obtained from both yeast and human cells . This approach revealed that in more than half of the yeast genome , a given nucleosome is found at slightly different positions in different cells . Nucleosomes positioned near the beginning of a gene mark it open or closed for binding by the cell’s gene expression machinery . Zhou , Blocker et al . suggest that the nucleosomes’ positions influence how gene expression starts via a multi-step process . Following on from this work , the next step is to use the newly developed method to study how nucleosome positions change when other regulators of gene activity bind and when genes are activated or repressed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "computational", "and", "systems", "biology", "genetics", "and", "genomics" ]	2016	A computational approach to map nucleosome positions and alternative chromatin states with base pair resolution
Calcium ( Ca2+ ) dysregulation is a hallmark of heart failure and is characterized by impaired Ca2+ sequestration into the sarcoplasmic reticulum ( SR ) by the SR-Ca2+-ATPase ( SERCA ) . We recently discovered a micropeptide named DWORF ( DWarf Open Reading Frame ) that enhances SERCA activity by displacing phospholamban ( PLN ) , a potent SERCA inhibitor . Here we show that DWORF has a higher apparent binding affinity for SERCA than PLN and that DWORF overexpression mitigates the contractile dysfunction associated with PLN overexpression , substantiating its role as a potent activator of SERCA . Additionally , using a well-characterized mouse model of dilated cardiomyopathy ( DCM ) due to genetic deletion of the muscle-specific LIM domain protein ( MLP ) , we show that DWORF overexpression restores cardiac function and prevents the pathological remodeling and Ca2+ dysregulation classically exhibited by MLP knockout mice . Our results establish DWORF as a potent activator of SERCA within the heart and as an attractive candidate for a heart failure therapeutic . Cardiovascular disease is the leading cause of death and disability in industrialized nations and its prevalence is rising rapidly . The molecular mechanisms that drive the progression of heart failure are poorly understood due to the complex and multifactorial nature of the disease . Among the many pathological features of heart failure , the most prominent and widespread is aberrant Ca2+ cycling , which reduces myocardial contractility and initiates pathological remodeling ( Piacentino et al . , 2003 ) . Dilated cardiomyopathy ( DCM ) , which is characterized by ventricular chamber enlargement and systolic dysfunction , is the third most common cause of heart failure and the most frequent reason for heart transplantation ( Maron et al . , 2006 ) . While many cases of DCM are idiopathic in nature , direct links have been established between the development of DCM as a consequence of inflammatory , metabolic , or toxic insults or by genetic mutations in Ca2+ regulatory proteins , contractile proteins or cytoskeletal proteins that reside at the sarcomeric Z-disc ( Arber et al . , 1997; Cahill et al . , 2013; McNally et al . , 2013 ) . Regardless of the cause of DCM , progressive chamber dilation and heart failure are driven by Ca2+ dysregulation including alterations in Ca2+ cycling and homeostasis ( Luo and Anderson , 2013; Minamisawa et al . , 1999 ) . Ca2+ is a ubiquitous intracellular second messenger involved in the regulation of a broad range of cellular processes including muscle contraction , energy metabolism , proliferation and apoptosis . The involvement of Ca2+ in so many fundamental events demands its precise control , which predominantly occurs at the level of the sarco ( endo ) plasmic reticulum ( SR ) , the major intracellular Ca2+ storage site . In the heart , Ca2+ plays a crucial role in connecting membrane excitability with contraction , a process known as excitation contraction-coupling . During each cycle of contraction and relaxation , Ca2+ is released from the SR via ryanodine receptors ( RyRs ) into the cytoplasm where it binds to myofilament proteins to induce sarcomere shortening ( Bers , 2002 ) . Relaxation is initiated by Ca2+ re-sequestration into the SR , a process that is mediated by a SR Ca2+-ATPase ( SERCA ) , which uses the energy generated from ATP hydrolysis to pump Ca2+ against its concentration gradient back into the lumen of the SR . A universal cause of the decreased contractile performance of the failing heart is impaired Ca2+ sequestration into the SR and a reduction in SERCA activity and protein ( Luo and Anderson , 2013 ) . Hence , augmenting SERCA activity has been suggested as an attractive clinical approach for treating heart failure by preserving cardiac contractile function ( Gwathmey et al . , 2013; Kranias and Hajjar , 2012; Pleger et al . , 2013 ) . Consistent with this hypothesis , overexpression of SERCA2a , the predominant cardiac isoform of SERCA , has been shown to improve cardiac function and ameliorate the progression of cardiovascular disease in several rodent and large animal models of heart failure ( Kawase et al . , 2008; Miyamoto et al . , 2000; Prunier et al . , 2008 ) . In the heart , it has been shown that the activity of SERCA is inhibited by the binding of two small transmembrane peptides , phospholamban ( PLN ) and sarcolipin ( SLN ) , which lower the affinity of SERCA for Ca2+ and decrease the rate of Ca2+ re-uptake into the SR ( MacLennan and Kranias , 2003; Nelson et al . , 2014; Vangheluwe et al . , 2006 ) . Our lab recently discovered and characterized a novel micropeptide named DWarf Open Reading Frame ( DWORF ) , which binds directly to SERCA and enhances its activity by displacing the SERCA inhibitory peptides PLN and SLN ( Nelson et al . , 2016 ) . The discovery of DWORF as a potent stimulator of SERCA activity and cardiac contractility provides a novel therapeutic target through which to preserve cardiac contractile function and restore Ca2+ homeostasis in the context of heart failure . In this study , we investigate the molecular determinants of the DWORF-SERCA regulatory complex and explore the therapeutic potential of DWORF overexpression as a means to increase SERCA activity and cardiac contractility in the context of heart failure . We use a combination of techniques to examine the interaction of SERCA with DWORF and PLN to precisely demonstrate that SERCA has a higher apparent affinity for DWORF than for PLN . We also examine the stoichiometric parameters of the DWORF-SERCA complex and analyze the ability of DWORF and PLN to homo- or hetero-oligomerize into higher order structures . Additionally , we show that in vivo overexpression of DWORF relieves the inhibitory effects of PLN on SERCA , even in the context of super-inhibition of SERCA via cardiac-specific PLN overexpression . Lastly , we use a well-characterized mouse model of DCM , muscle-specific LIM protein ( MLP ) knockout mice , to show that DWORF overexpression enhances cardiac function and prevents adverse cardiac remodeling to abrogate the heart failure phenotype observed in these mice ( Arber et al . , 1997 ) . The MLP KO mouse model of heart failure strongly reproduces the morphological and clinical characteristics of DCM and heart failure in human patients , which highlights the potential of DWORF overexpression as a clinically relevant therapy ( Hoshijima et al . , 2006 ) . The interaction of PLN with SERCA has been extensively studied ( Kranias and Hajjar , 2012; MacLennan and Kranias , 2003; Hou et al . , 2008; Hou and Robia , 2010; Kelly et al . , 2008; Kimura et al . , 1998; Robia et al . , 2007 ) . In contrast , due to the very recent discovery of DWORF , very little is known about the SERCA/DWORF regulatory complex . To examine the apparent binding affinity and stoichiometry of SERCA in complex with DWORF in live cell membranes , we performed fluorescence resonance energy transfer ( FRET ) experiments using transfected AAV-293 cells . We sampled large populations of cells ( ~1000 cells per experiment ) coexpressing mCerulean ( Cer ) -SERCA2a and either yellow fluorescent protein ( YFP ) -DWORF or –PLN and compared each cell’s FRET efficiency ( Cer excitation , YFP emission ) with its YFP-DWORF or –PLN fluorescence intensity , which is an index of protein expression ( Hou et al . , 2008; Kelly et al . , 2008 ) . For both DWORF and PLN , FRET efficiency increased with increasing protein expression , a relationship that can be approximated by a hyperbolic fit of the form y= ( FRETmax ) x/ ( Kd +x ) ( Hou et al . , 2008 ) ( Figure 1—figure supplement 1A ) . FRETmax is defined as the maximal FRET and represents the intrinsic FRET of the bound complex , while Kd represents the protein concentration at which half-maximal FRET is achieved [apparent dissociation constant; in arbitrary units ( AU ) ] . Multiple independent experiments were performed and representative data are shown in Figure 1—figure supplement 1A . The mean PLN-SERCA2a FRETmax value was 29 . 9 ± 2 . 1% , which is similar to previous results ( Hou et al . , 2008; Hou and Robia , 2010; Kelly et al . , 2008 ) , while the FRETmax value for DWORF-SERCA2a was 16 . 3 ± 1 . 7% ( Figure 1—figure supplement 1B ) . This difference in FRETmax values is consistent with an increased FRET distance for the DWORF-SERCA2a complex compared to PLN-SERCA2a due to the shorter cytoplasmic domain of DWORF ( Bidwell , 2012 ) . Importantly , SERCA2a exhibited a higher apparent affinity for DWORF than for PLN as evidenced by a reduction in Kd ( Figure 1A ) . Additionally , we performed progressive acceptor photobleaching experiments to determine the stoichiometry of the SERCA regulatory complexes with PLN or DWORF . We observed a linear increase in donor fluorescence with decreasing acceptor fluorescence , consistent with a 1:1 stoichiometry of the PLN:SERCA and DWORF:SERCA complexes ( Figure 1B and Figure 1—figure supplement 1C , D ) . PLN has been well described to exist as both a monomer , which is a potent inhibitor of SERCA , and a less inhibitory pentamer ( Kimura et al . , 1998 ) . The modulation of the PLN monomer/pentamer ratio is an important determinant of SERCA activity and therefore cardiac contractility . We performed additional progressive acceptor photobleaching experiments to determine if DWORF is capable of homo-oligomerizing with itself or hetero-oligomerizing with PLN . These experiments did not detect DWORF-DWORF FRET ( Figure 1C ) or PLN-DWORF FRET ( Figure 1D ) , suggesting that DWORF does not form homo- or hetero-oligomers at the concentrations achieved here , while PLN-PLN FRET experiments showed the expected high-order oligomerization that has been previously described ( Figure 1E ) ( Kelly et al . , 2008; Robia et al . , 2007 ) . These results suggest that DWORF exists as a monomer that is available for interaction with SERCA at all times . We previously generated DWORF transgenic ( Tg ) mice using the α-myosin heavy chain ( αMHC ) promoter to overexpress DWORF specifically in the heart ( Nelson et al . , 2016 ) . Cardiomyocytes from DWORF Tg mice have a cellular phenotype that mimics that observed in PLN null mice , including an increase in peak Ca2+ transient amplitude , faster cytosolic Ca2+ decay rates , higher SR Ca2+ load and enhanced cardiomyocyte contractility ( Nelson et al . , 2016; Luo et al . , 1994 ) . Our previous work indicates that DWORF activates SERCA by displacing its negative regulator , PLN , and suggests that the profile of enhanced contractility in DWORF Tg animals is due to the ability of DWORF to compete PLN off of SERCA and relieve its inhibitory effects . To investigate this in vivo , we crossed our DWORF Tg mice with the well-characterized αMHC-PLN transgenic mice ( PLN Tg ) ( Kadambi et al . , 1996 ) to generate double transgenic ( PLN/DWORF Tg ) animals . Cardiomyocytes from PLN Tg animals exhibit a cellular phenotype opposite that of DWORF Tg mice , with reduced peak Ca2+ transient amplitude , slower transient decay rates , and reduced fractional shortening due to super-inhibition of SERCA ( Kadambi et al . , 1996 ) . We hypothesized that overexpression of DWORF in PLN Tg mice would lead to displacement of the excess PLN from SERCA and relieve its inhibitory effects . Baseline cardiac phenotyping of wild-type ( WT ) , PLN Tg , DWORF Tg , or PLN/DWORF Tg mice by echocardiography ( ECHO ) indicated that all genotypes had similar cardiac function as measured by ejection fraction or fractional shortening ( Figure 2—figure supplement 1A , B ) . Additionally , all genotypes analyzed had comparable cardiac dimensions as assessed by ECHO ( Figure 2—figure supplement 1C , D ) , normal heart weight to tibia length measurements ( Figure 2—figure supplement 1E ) , and similar histological appearances ( Figure 2—figure supplement 1F ) , indicating that overexpression of PLN , DWORF , or a combination of the two did not lead to adverse remodeling . To analyze the cellular phenotype of these animals , we isolated cardiomyocytes from WT , PLN Tg , DWORF Tg , and PLN/DWORF Tg mice and performed Ca2+ transient measurements while simultaneously monitoring sarcomere shortening . Consistent with previous findings ( Nelson et al . , 2016 ) , we found that DWORF Tg animals had enhanced Ca2+ cycling with increased peak Ca2+ transient amplitude and faster transient decay rates ( Figure 2A–C ) accompanied by increased fractional shortening ( Figure 2D , E ) . Measurements from PLN Tg cardiomyocytes also recapitulated previous findings and displayed the opposite phenotype characterized by diminished peak Ca2+ transient amplitude , slower decay rates and reduced fractional shortening , indicating a strong inhibition of SERCA activity translating into reduced cardiomyocyte contractility ( Figure 2A–E ) ( Kadambi et al . , 1996 ) . Remarkably , cardiomyocytes isolated from PLN/DWORF Tg animals exhibited a complete prevention of impaired Ca2+ cycling associated with PLN overexpression ( Figure 2A–E ) . PLN/DWORF Tg mice displayed a profile of enhanced Ca2+-handling almost identical to that of DWORF Tg animals , indicating that DWORF overexpression can relieve the super-inhibition of SERCA caused by overexpression of PLN . To directly assess SERCA enzymatic activity in cardiac homogenates from WT , PLN Tg , DWORF Tg and PLN/DWORF Tg mice , we performed oxalate-supported Ca2+-dependent Ca2+-uptake measurements ( Nelson et al . , 2016; Bidwell and Kranias , 2016 ) . Consistent with previously published reports , hearts over-expressing PLN showed a reduction in SERCA activity at lower concentrations of Ca2+ substrate quantified as a lower affinity of SERCA for Ca2+ ( an increase in KCa ) ( Figure 2F , G ) ( Nelson et al . , 2016; Kadambi et al . , 1996 ) , while DWORF Tg hearts exhibited the opposite phenotype with a significant increase in the affinity of SERCA for Ca2+ as indicated by a decrease in KCa ( Figure 2F , G ) . SERCA activity assays performed in homogenates from PLN/DWORF Tg mice mirrored those of DWORF Tg mice , indicating that the super-inhibition of SERCA caused by PLN overexpression can be completely nullified in the presence of excess DWORF . Importantly , western blot analysis and quantitative RT-PCR performed on cardiac tissue from WT , PLN Tg , DWORF Tg , and PLN/DWORF Tg mice showed no significant differences in protein or RNA expression levels of any of the major Ca2+-handling proteins , indicating that the results observed were not due to compensatory responses ( Figure 2—figure supplement 2A–D ) . We also analyzed the phosphorylation state of PLN to verify that our observations were not due to post-translational modifications of the protein that are known to strongly regulate its ability to inhibit SERCA and saw no significant changes amongst genotypes ( Figure 2—figure supplement 2B and D ) ( Luo et al . , 1998 ) . Taken together , these results support previous data indicating that DWORF overexpression enhances cardiac Ca2+ cycling and contractility through displacement of PLN from SERCA , thereby relieving its inhibitory effects ( Nelson et al . , 2016 ) . To further substantiate these findings , we analyzed the interaction of SERCA2a with PLN and DWORF in a heterologous expression system . HEK293 cells were co-transfected with equal amounts of Myc-tagged SERCA2a and HA-tagged PLN in the presence of increasing levels of HA-DWORF , and Myc-SERCA2a/HA-peptide interactions were assessed by Myc ( SERCA2a ) immunoprecipitation and western blot analysis . We observed a strong reduction in the interaction of HA-PLN with SERCA2a when co-expressed with HA-DWORF , and this occurred in a dose-dependent manner ( Figure 2—figure supplement 3A ) . Consistent with previous findings ( Nelson et al . , 2016 ) , using the same heterologous expression system we found that co-expression of DWORF with SERCA2a did not change the apparent affinity of SERCA for Ca2+ , but it relieved the inhibition of PLN on SERCA in a dose-dependent manner ( Figure 2—figure supplement 3B , C ) . These results substantiate the hypothesis that the overexpression of DWORF could be a powerful means of enhancing SERCA activity via the displacement of PLN and therefore may enhance cardiac contractility in the setting of heart failure and prevent the progression of the disease . To directly assess the potential of DWORF as a therapeutic for heart failure , we crossed our DWORF Tg mice with the well-characterized MLP KO mouse model of DCM . The MLP protein is expressed in cardiac and skeletal muscle and is predominantly localized adjacent to the Z-disc where it plays a structural role and also acts as a stress signaling molecule that transduces mechanical stress into biochemical signals ( Arber et al . , 1997; Arber et al . , 1994; Heineke et al . , 2005; Knöll et al . , 2002 ) . The adult-onset DCM phenotype exhibited by MLP KO mice mimics that of human DCM and is characterized by progressive dilation of all four cardiac chambers , ventricular wall thinning , a reduction in cardiac contractility and elongation of action potential duration ( Arber et al . , 1997; Hoshijima et al . , 2006 ) . Notably , defects in SR Ca2+ cycling have been shown to be important determinants of cardiac dysfunction and the transition to heart failure in MLP KO mice ( Minamisawa et al . , 1999 ) . We have previously shown that DWORF mRNA and protein levels are dramatically reduced in human ischemic heart failure and in mouse models of cardiovascular disease , indicating that a decrease in DWORF expression may contribute to the Ca2+ dysregulation that drives cardiac decompensation ( Nelson et al . , 2016 ) . We measured DWORF expression in cardiac tissue from WT and MLP KO mice and found a reduction in both protein and RNA levels in MLP KO hearts ( Figure 3A , B ) , suggesting that loss of DWORF expression may contribute to the DCM phenotype . To evaluate whether DWORF overexpression provides cardioprotection in MLP KO mice , we crossed our DWORF Tg animals with MLP KO mice to create MLP KO/DWORF Tg mice . Cardiac function was assessed in 8-week-old mice by ECHO ( Figure 3C ) . Consistent with previous reports , MLP KO mice showed a marked reduction in left ventricular ( LV ) function compared to WT animals as measured by ejection fraction ( Figure 3D ) and fractional shortening ( Figure 3E ) . Cardiac-specific overexpression of DWORF in MLP KO mice resulted in a significant improvement of LV function as evidenced by increases in ejection fraction and fractional shortening to values approaching those of WT mice ( Figure 3C–E ) . We also assessed whether DWORF loss-of-function exacerbated the MLP KO phenotype and indeed observed that MLP/DWORF double KO ( dKO ) mice showed a further decline in cardiac function as compared to MLP KO mice ( Figure 3C–E ) . Cardiac dimensions were calculated from M-mode ECHO tracings and MLP KO mice showed an increase in LV internal diameter both during diastole , or relaxation ( Figure 3F ) , and systole , or contraction ( Figure 3G ) , consistent with the clinical presentation of DCM . Concurrent loss of DWORF protein in MLP/DWORF dKO mice resulted in a slight but non-significant increase in chamber dilation , while DWORF overexpression led to a dramatic reduction in LV chamber dilation and near complete prevention of the MLP KO DCM phenotype ( Figure 3F , G ) . Diastolic dysfunction coexists in human patients with dilated cardiomyopathy , and it has previously been shown that the progression to heart failure in MLP KO mice may be anticipated by diastolic cardiac dysfunction ( Lorenzen-Schmidt et al . , 2005 ) . We evaluated LV diastolic function in our mice by pulse-wave Doppler echocardiography of transmitral valve blood flow and by mitral annular tissue Doppler ( Table 1 ) . We found that the E/A ratio ( ratio of the early [E] to late [A] ventricular filling velocities , Figure 3H ) and E/E’ ratio ( ratio of early filling [E] to early diastolic mitral annular velocity [E’] , Figure 3I ) of MLP KO and MLP/DWORF dKO were significantly greater than WT animals . In MLP KO/DWORF Tg mice , both the E/A ratio ( Figure 3H ) and E/E’ ratio ( Figure 3I ) were indistinguishable from those of WT mice , indicating that DWORF overexpression ameliorates the diastolic dysfunction observed in MLP KO mice ( Figure 3H and I and Table 1 ) . Collectively , these results indicate that the restoration of SERCA activity and enhancement of Ca2+ cycling in MLP KO mice via DWORF overexpression is sufficient to prevent the onset of DCM in MLP KO mice and their subsequent transition to heart failure . Histological analysis of 8-week-old MLP KO hearts showed characteristic morphological defects consistent with DCM including ventricular and atrial chamber dilation , wall thinning and cardiac enlargement , and these features were exacerbated in MLP/DWORF dKO mice ( Figure 4A ) . In sharp contrast , overexpression of DWORF in MLP KO mice prevented the spectrum of morphological defects observed in MLP KO hearts ( Figure 4A ) . Significant ventricular cardiomyocyte hypertrophy was observed in both MLP KO and MLP/DWORF dKO mice compared to WT animals as assessed by cross-sectional area analysis ( Figure 4—figure supplement 1A , B ) and isolated cardiomyocyte length and width measurements ( Figure 4—figure supplement 1C–E ) . DWORF overexpression in MLP KO mice significantly blunted this hypertrophic response and cell size parameters were indistinguishable from WT mice in the MLP KO/DWORF Tg group ( Figure 4—figure supplement 1A–E ) . Additionally , MLP KO/DWORF Tg hearts had heart weight to tibia length ( Figure 4B ) and lung weight to tibia length ( Figure 4C ) measurements comparable to those of WT mice , while MLP KO and MLP/DWORF dKO mice showed significant increases in these parameters , indicative of advanced heart failure . MLP/DWORF dKO mice also had a significantly higher liver weight to tibia length ratio compared to any of the other genotypes assessed , indicating that these animals were in a particularly aggravated state of congestive heart failure ( Figure 4D ) . Quantification of cardiac fibrosis by Picrosirius Red staining revealed significant myocardial fibrosis in MLP KO mice that was mildly exacerbated in MLP/DWORF dKO mice and dramatically reduced in MLP KO/DWORF Tg mice at 8 weeks of age ( Figure 5A , B ) . Quantitative RT-PCR revealed a robust induction of the cardiac fetal gene program in MLP KO mice , a molecular marker of pathological cardiac hypertrophy ( Figure 5C ) . This response was significantly inhibited in MLP KO/DWORF Tg mice , which is consistent with the preservation of ventricular function in these animals ( Figure 5C ) . Ultrastructural analysis of MLP KO mice by electron microscopy revealed a striking disruption of cardiac myofibrillar organization characteristic of the late phases of DCM in both mice and humans ( Figure 5D ) . Overexpressing DWORF in MLP KO mice resulted in complete prevention of these ultrastructural defects , indicating a preservation of cardiac function and cardiomyocyte architecture ( Figure 5D ) . To gain further insight into the mechanisms responsible for the dramatic improvement of cardiac function in MLP KO mice by DWORF overexpression , we isolated cardiomyocytes from our animals and performed intracellular Ca2+ transients and fractional shortening measurements . Compared to WT mice , MLP KO and MLP/DWORF dKO cardiomyocytes exhibited marked reductions in Ca2+ transient amplitude ( Figure 6A ) , significant prolongation of the transient decay rate ( Figure 6B ) , and decreased fractional shortening ( Figure 6C ) , collectively indicating diminished SERCA activity and Ca2+ cycling . Additionally , sarcomere relaxation kinetics were significantly slowed in MLP KO and MLP/DWORF dKO cardiomyocytes ( Figure 6D ) . Overexpression of DWORF in MLP KO mice resulted in an increase in cardiomyocyte Ca2+ transient amplitude ( Figure 6A ) , faster transient decay rates ( Figure 6B ) , enhanced fractional shortening ( Figure 6C ) and increased sarcomere relaxation kinetics ( Figure 6D ) to levels that prevented the phenotype observed in MLP KO mice and surpassed those of WT cardiomyocytes . Sarcomere shortening kinetics were similar across all genotypes analyzed ( Figure 6E ) , indicating a specific alteration in cardiomyocyte relaxation kinetics in this animal model . We directly confirmed that SERCA enzymatic activity was enhanced in MLP KO/DWORF Tg animals by performing oxalate supported Ca2+-dependent Ca2+-uptake measurements in cardiac homogenates and observed a strong leftward shift of the SERCA activity curve ( Figure 6F ) , indicating an increase in the affinity of SERCA for Ca2+ and quantified as a reduction in KCa value ( Figure 6G ) . Consistent with previous reports ( Minamisawa et al . , 1999 ) , western blotting and quantitative RT-PCR revealed that MLP gene deletion does not cause significant alterations in protein or RNA levels of any major Ca2+ handling genes in the heart , suggesting that the defects of Ca2+ cycling in MLP KO mice result from a functional impairment of excitation-contraction coupling rather than a decrease in the proteins mediating the cycling itself ( Figure 6—figure supplement 1 ) . We also analyzed the phosphorylation state and oligomerization of PLN to verify that our observations were not due to post-translational modifications of the protein that are known to strongly regulate its ability to inhibit SERCA and saw no significant changes amongst genotypes ( Figure 6—figure supplement 1A and C ) . Taken together , these findings provide evidence that the reversal of the MLP KO phenotype by DWORF overexpression mechanistically lies in the ability of DWORF to displace PLN from SERCA and enhance its activity to restore Ca2+ cycling and maintain cardiac contractility . While heart failure is a complex disease with many distinctly different causes , the functional characteristics of the failing myocardium are surprisingly consistent and include the slowing of both contraction and relaxation rates and the prolongation of the cardiac action potential ( Houser et al . , 2000 ) . Alterations in Ca2+ cycling and depressed SR Ca2+ re-uptake are universal features of heart failure that have been shown to contribute directly to the pathogenesis of cardiovascular disease ( Piacentino et al . , 2003; Luo and Anderson , 2013 ) . For this reason , significant attention has been focused on restoring Ca2+ homeostasis through enhancing SERCA activity , which has been shown to maintain cardiac contractility and prevent the progression of the disease ( Gwathmey et al . , 2013; Kranias and Hajjar , 2012; Pleger et al . , 2013; Ly et al . , 2007 ) . Here we describe a novel approach to stimulate SERCA activity through DWORF overexpression and present strong evidence demonstrating its potent ability both to enhance Ca2+ cycling and contractility and to prevent the development of cardiomyopathy in a well-characterized mouse model of DCM . In the heart , it is well established that SERCA activity is reversibly regulated by PLN , a small transmembrane protein that directly interacts with SERCA and reduces its activity by lowering its affinity for Ca2+ . Since its discovery over 40 years ago , the regulation of PLN and its interaction with SERCA have been the subject of intense research ( Kranias and Hajjar , 2012; MacLennan and Kranias , 2003; Nelson et al . , 2014; Vangheluwe et al . , 2006; Hou et al . , 2008; Hou and Robia , 2010; Kelly et al . , 2008; Kimura et al . , 1998; Robia et al . , 2007 ) . Our lab recently identified DWORF as a novel transmembrane protein that resides in the cardiac SR membrane and competes for the same binding site on SERCA as PLN and enhances SERCA activity ( Nelson et al . , 2016 ) . This work has opened up new avenues of research aimed at understanding SERCA regulation in the heart and also provides a novel mechanism to increase SERCA activity in the context of heart failure . In this study , we used FRET to investigate the stoichiometry and relative affinity of micropeptide regulatory complexes in live cell membranes and found that SERCA has a higher apparent affinity for DWORF than PLN , which makes it an attractive candidate for gene therapy . While enhancing SERCA activity by increasing its expression level has been a primary focus of gene therapy studies thus far , there is also strong experimental evidence that increasing SERCA activity through the ablation of the SERCA inhibitor PLN is beneficial ( Minamisawa et al . , 1999; Sato et al . , 2001 ) , and achieving a similar outcome through DWORF overexpression is a much more clinically relevant approach to achieve this goal . In this study , we used the MLP KO mouse model of DCM to test our hypothesis that increasing SERCA function by DWORF overexpression would prevent the development of ventricular dysfunction , fibrosis , and long-term heart failure that is characteristic of dilated cardiomyopathy . We selected this specific genetic model of DCM for several reasons . First , genetic ablation of MLP in mice has been shown to lead to adult onset DCM , which closely resembles the human disease , and MLP KO mice represent a very common and extensively used mouse model for studying the pathophysiology of dilated cardiomyopathy ( Arber et al . , 1997; Minamisawa et al . , 1999; Lorenzen-Schmidt et al . , 2005; Heineke et al . , 2010; Recchia and Lionetti , 2007; Rockman et al . , 1998 ) . Second , it has been shown that MLP KO mice exhibit the characteristic cardiomyocyte Ca2+ cycling defects that are acquired in human heart failure , including reductions in peak cardiomyocyte Ca2+ transient amplitude and prolongation of Ca2+ reuptake kinetics , which relates to decreased SERCA activity ( Arber et al . , 1997 ) , and these are the specific parameters that should be enhanced with DWORF overexpression . Lastly , it has been shown that increasing SERCA function by PLN gene deletion in MLP KO mice preserved Ca2+ homeostasis and prevented the development of DCM in this model ( Minamisawa et al . , 1999 ) . Because we believe that DWORF exerts its effect through the displacement of PLN to induce maximal cardiomyocyte contractility , we expected to see a similar degree of benefit in MLP KO mice via DWORF overexpression . Interestingly , it has been reported that in ischemic cardiomyopathy and idiopathic DCM , MLP protein levels are significantly decreased , implying that MLP deficiency may be a common pathophysiological mechanism in advanced heart failure ( Zolk et al . , 2000 ) . Our studies largely focused on Ca2+ dynamics due to the described function of DWORF as a potent stimulator of SERCA activity , but it should be noted that SERCA is not the only target that could rescue this animal model of DCM . In fact , it has previously been shown that G-protein-coupled receptor kinase 2 ( GRK2 ) inhibition via overexpression of a β-adrenergic receptor kinase peptide inhibitor ( βARKct ) prevented the development of myocardial failure in MLP KO mice ( Rockman et al . , 1998 ) . We did not directly assess β-adrenergic receptor density or responsiveness in our animals , so this could also be a contributing factor . Additionally , we showed that DWORF overexpression in MLP KO mice resulted in a dramatic attenuation of fibrosis compared to MLP KO mice , which also may contribute to its ability to prevent the disease phenotype . It remains to be seen whether DWORF overexpression can preserve cardiac function in other forms of DCM and in chronic heart failure . In the future , we will test the therapeutic potential of DWORF overexpression in additional clinically relevant models of heart failure by adeno-associated viral ( AAV ) delivery . It has previously been shown that enhancing contractility by overexpressing SERCA is protective against diabetic cardiomyopathy ( Trost et al . , 2002 ) as well as cardiac dysfunction induced by chronic pressure overload ( Miyamoto et al . , 2000; del Monte et al . , 2001; Nakayama et al . , 2003 ) and we believe that we can achieve the same results with DWORF overexpression . Despite promising results in rodent and large animal models of heart failure ( Gwathmey et al . , 2013; Kawase et al . , 2008; Miyamoto et al . , 2000; Prunier et al . , 2008; Ly et al . , 2007 ) , overexpression of SERCA by AAV gene delivery in a human clinical trial ( Calcium Up-regulation by Percutaneous administration of gene therapy In cardiac Disease , CUPID ) failed to meet its primary endpoints and was discontinued ( Greenberg et al . , 2016; Greenberg et al . , 2014 ) . A major reason this clinical trial is believed to have failed was due to lack of successful SERCA overexpression ( Greenberg et al . , 2016 ) . In spite of the failure of this clinical trial , we believe that gene therapy is still a promising approach for heart failure treatment and that DWORF overexpression may be superior to that of SERCA . First , as mentioned above , the CUPID trial likely failed primarily because SERCA , which is a large multi-pass transmembrane protein , was not efficiently delivered to the targeted cells and expressed in the cells it did infect . In this regard , DWORF may provide a more optimal protein for delivery due to its small size , which allows it to be easily packaged in AAV vectors and rapidly translated from a relatively small number of transcripts . Second , in heart failure , it is well established that SERCA levels are reduced , but there is often either no change or a slight increase in PLN expression which would dramatically increase the PLN to SERCA ratio ( Kranias and Hajjar , 2012 ) . Therefore , overexpressing SERCA alone may not be sufficient to overcome this imbalance . Since DWORF can enhance the activity of SERCA in the presence of excess PLN ( Figure 2 ) , it may prove to be more beneficial to increase the activity of the endogenous SERCA pump by expressing DWORF rather than the pump itself . Lastly , it has been shown that SERCA2a requires post-translational modification with SUMO for full activity ( Kho et al . , 2011 ) , a process that may be limited by the capacity for SUMOylation rather than SERCA abundance . Ectopic expression of DWORF could increase the activity of the available SERCA protein without the need to address SUMOylation capacity . We strongly believe that enhancing Ca2+ cycling remains a compelling pathway to target in the development of heart failure therapeutics as its disruption is a major common insult in the disease and our evidence suggests that the overexpression of DWORF represents a potent means to achieve this goal . In summary , through multiple independent assays , our results show that DWORF displays a higher apparent affinity for SERCA than PLN , partly because PLN has a high self-affinity for oligomerization . Thus , the more monomeric DWORF outcompetes PLN for binding to the pump . Overexpression of DWORF leads to the displacement of PLN from SERCA and results in enhanced SERCA activity , even in instances where PLN is overexpressed causing SERCA super-inhibition . Lastly , the prevention of the DCM phenotype of MLP KO mice by DWORF overexpression highlights the clinical potential of DWORF overexpression as a promising therapeutic for heart failure and an attractive candidate for future gene therapy studies . The objectives of the present study were to molecularly characterize the interaction of DWORF with SERCA and to directly test if DWORF overexpression could prevent the development of cardiomyopathy in a mouse model of DCM . Male mice were used for all experiments and all mice with the appropriate genotypes were used without any exclusions . With the exception of echocardiography measurements , we did not use blinding approaches . All echocardiography experiments were performed and analyzed by a single blinded operator . The sample sizes were based on previous experience and published reports . For each experiment , sample size is indicated in the figure legend and reflects the number of independent biological replicates . In general , sample size was chosen to use the least number of animals to achieve statistical significance , and no statistical methods were used to predetermine sample size . AAV-293 cells were cultured in DMEM cell culture medium supplemented with 10% fetal bovine serum ( FBS ) ( ThermoScientific , Waltham , MA ) and transiently transfected using MBS mammalian transfection kit ( Agilent Technologies , Stratagene , La Jolla , CA ) , according to manufacturer instructions . The transfected cells were trypsinized ( ThermoScientific ) and replated onto poly D lysine-coated glass-bottom chambers and allowed to adhere for 1–2 hr prior to imaging . Acceptor sensitization FRET microscopy was performed as described previously ( Hou et al . , 2008; Hou and Robia , 2010 ) . Cells were imaged with an inverted microscope ( Nikon Eclipse Ti ) equipped with an EM-CCD camera ( iXon 887 , Andor Technology , Belfast , Northern Ireland ) . Acquisition was performed with a 40 × 0 . 75 N . A . objective with 100 ms exposure for each channel: Cer , YFP , and ‘FRET’ ( Cer excitation , YFP emission ) . Fluorescence intensity was quantified from ~1000 cells per sample using automatic multiwavelength cell scoring in MetaMorph ( Molecular Devices , Sunnyvale , CA ) . FRET efficiency was calculated according to E = G/ ( G + 3 . 2 × FCer ) , where G = FFRET−a × FYFP−d × FCer ( Himes et al . , 2016 ) , where FFRET , FYFP , and FCer are the matching fluorescence intensity from FRET , YFP , and Cer images , respectively , and G represents FRET intensity corrected for the bleed-through of the channels . The parameters a and d are bleed-through constants calculated as a = FFRET/FCer for a control sample transfected with only YFP-SERCA and d = FFRET/FCer for a control sample transfected with only Cer-SERCA . These values were determined to be G = 4 . 74 a = 0 . 075 and b = 0 . 88 . Progressive acceptor photobleaching was performed as described previously ( Kelly et al . , 2008; Zak et al . , 2017 ) . Briefly , we collected images of Cer and YFP fluorescence at intervals to establish a baseline and then initiated progressive acceptor photobleaching ( Zak et al . , 2017 ) , acquiring successive images of Cer and YFP in between 10 s of exposure to illumination through a 504/12 nm bandpass filter for selective photobleaching of YFP . The images were analyzed in FIJI ( Schindelin et al . , 2012 ) , and FRET was calculated from the pre- and post-bleach donor fluorescence intensity using the equation FRET = 1- ( FDA/FD ) where FDA = the intensity of the donor before bleaching and FD = the intensity of the donor after bleaching . To distinguish between 1:1 and higher order stoichiometry , the fluorescence of the donor was plotted against the fluorescence of the acceptor at the same time point during progressive bleaching . A linear relationship was taken to indicate a 1:1 complex of Cer- and YFP-labeled proteins ( Kelly et al . , 2008; Zak et al . , 2017 ) . Animal work described in this manuscript has been approved and conducted under the oversight of the UT Southwestern Institutional Animal Care and Use Committee . Mice were housed in a barrier facility with a 12 hr light/dark cycle and maintained on standard chow ( 2916 Teklad Global , Houston , TX ) . All mouse lines used in this manuscript have been previously published ( Arber et al . , 1997; Nelson et al . , 2016; Kadambi et al . , 1996 ) . All data presented were collected from male mice . Cardiac function and heart dimensions were determined by two-dimensional echocardiography using a Visual Sonics Vevo 2100 Ultrasound ( Visual Sonics , Toronto , Canada ) on conscious mice . M-mode tracings were used to measure anterior and posterior wall thicknesses at end diastole and end systole . Left ventricular ( LV ) internal diameter ( LVID ) was measured as the largest anteroposterior diameter in either diastole ( LVID;d ) or systole ( LVID;s ) . A single observer blinded to mouse genotypes performed echocardiography and data analysis . Fractional shortening ( FS ) was calculated according to the following formula: FS ( % ) = [ ( LVID;d − LVID;s ) /LVID;d]×100 . Ejection fraction ( EF% ) was calculated by: EF ( % ) = ( [EDV − ESV]/EDV ) ×100 . EDV , end diastolic volume; ESV , end systolic volume . Diastolic function was assessed in lightly anesthetized mice ( 1 . 5%–2% isoflurane ) using pulsed wave Doppler recordings of the maximal early ( E ) and late ( A ) diastolic transmitral flow velocities and Doppler tissue imaging recordings of peak E’ velocity and peak A’ velocity in apical four-chamber view . Body temperature was maintained at 37°C throughout using a heating pad . Changes in transmitral flow pattern ( E/A ratio ) and mitral annulus velocities ( E’ , A’ ) were used to assess diastolic dysfunction . Adult mouse hearts were rapidly excised and the aorta was cannulated on a constant-flow Langendorff perfusion apparatus . Hearts were digested with perfused Tyrode’s solution ( 10 mM glucose , 5 mM HEPES , 5 . 4 mM KCl , 1 . 2 mM MgCl2 , 150 mM NaCl , 2 mM sodium pyruvate , pH 7 . 4 ) containing Liberase ( 0 . 25 mg/ml ) , and the ventricles were minced , filtered , and equilibrated with Tyrode’s solution containing 1 mM CaCl2 and bovine serum albumin at room temperature ( Nelson et al . , 2016 ) . Cardiomyocyte length and width measurements were assessed using ImageJ analysis on bright-field images taken of freshly isolated cardiomyocytes imaged with a 20X objective . Length/width measurements were taken at the longest/widest part of each cell . Adult cardiomyocytes were loaded with 0 . 5 μM Fura-2-AM ( Molecular Probes , Eugene , OR ) and placed in a heated chamber ( 37°C ) on the stage of an inverted microscope . The chamber was perfused with Tyrode’s solution containing CaCl2 ( 1 . 8 mM ) ( pH 7 . 4 ) . Cardiomyocytes were paced with an IonOptix Myocyte Calcium and Contractility System at 0 . 5 Hz using a MyoPacer field stimulator . Changes in intracellular Ca2+ levels were monitored using Fura-2 dual-excitation ( 340/380 nm ) , single emission ( 510 nm ) ratiometric imaging . Tau , the decay rate of the average Ca2+ transient trace , was determined using IonWizard 6 . 0 analysis software ( IonOptix , Westwood , MA ) . Cardiomyocyte contractility measurements were made using sarcomere length ( SarcLen ) parameters and data was processed with IonWizard 6 . 0 analysis software . Oxalate-supported Ca2+ uptake in cardiac homogenates and transfected HEK293 cells were measured as previously described in detail ( Nelson et al . , 2016; Bidwell and Kranias , 2016 ) . Briefly , mouse hearts were isolated and rapidly snap frozen in liquid nitrogen and stored at −80°C until processed . Frozen tissue samples or cultured cells were homogenized in 50 mM phosphate buffer , pH 7 . 0 containing 10 mM NaF , 1 mM EDTA , 0 . 3 M sucrose , 0 . 3 mM PMSF and 0 . 5 mM DTT . Ca2+ uptake was measured in reaction solution containing 40 mM imidazole pH 7 . 0 , 95 mM KCl , 5 mM NaN3 , 5 mM MgCl2 , 0 . 5 mM EGTA , 5 mM K+ oxalate , 1 μM ruthenium red and various concentrations of CaCl2 to yield 0 . 02 to 5 μM free Ca2+ . The reaction was initiated by the addition of ATP ( final concentration 5 mM ) . The data were analyzed by nonlinear regression with computer software ( GraphPad Software ) , and the KCa values were calculated using an equation for a general cooperative model for substrate activation . CoIPs were performed as previously described ( Nelson et al . , 2016 ) . Briefly , HEK293 cells were co-transfected with expression plasmids encoding Myc-SERCA2a and HA-PLN in the presence of increasing concentrations of HA-DWORF or control plasmid . Whole cell lysates were prepared in CoIP buffer ( 20 mM NaPO4 , 150 mM NaCl , 2 mM MgCl2 , 0 . 1% NP-40 , 10% Glycerol , 10 mM sodium fluoride , 0 . 1 mM sodium orthovanadate , 10 mM sodium pyrophosphate , 1 mM DTT and Complete protease inhibitor [Roche , Basel , Switzerland] ) . Immunoprecipitations were carried out using 1 mg of mouse monoclonal anti-Myc antibody ( Invitrogen , Carlsbad , CA ) and collected with Dynabeads ( Invitrogen , Carlsbad , CA ) . Tris/Tricine gel electrophoresis was performed using pre-cast 16 . 5% Mini-PROTEAN Tris-Tricine gels ( Bio-Rad , Hercules , CA ) . Standard western blot procedures were performed on input and IP fractions using the following antibodies: HA ( Invitrogen , Carlsbad , CA ) , Myc ( Invitrogen , Carlsbad , CA ) or GAPDH ( Invitrogen , Carlsbad , CA ) . Specific catalogue numbers for the reagents used for these CoIP studies can be found in the Key Resource Table . Hearts were isolated and fixed in 4% ( vol/vol ) paraformaldehyde in PBS for 48 hr at 4°C with gentle shaking . Hearts were dehydrated , embedded in paraffin , and sectioned . Heart sections were stained with hematoxylin and eosin ( H and E ) and Picrosirius red using standard procedures . Fibrosis was quantified using Pircorsirius red staining and ImageJ software ( NIH , Rockville , MD ) . For immunofluorescent staining , tissue sections were deparaffinized and subjected to antigen retrieval with Citra buffer ( BioGenex , Fremont , CA ) . Tissue sections were incubated with 488-conjugated Wheat Germ Agglutinin ( Invitrogen , Carlsbad , CA ) to label the cell membranes and cardiomyocytes were immunostained using a primary antibody for cardiac troponin-T ( Abcam , Cambridge , MA ) and an Alexa Fluor 555 secondary antibody ( Invitrogen , Carlsbad , CA ) . Coverslips were mounted using VECTASHIELD Antifade Mounting Media with DAPI ( Vector Laboratories , Burlingame , CA ) and confocal images were taken of the mid-LV free wall with a Zeiss LSM-800 using a 40X oil objective . Cardiomyocyte cross-sectional area was assessed using Fiji Software . Specific catalogue numbers for the reagents used for immunohistochemistry can be found in the Key Resource Table . Eight-week-old mice were perfusion fixed by transcardial perfusion using 4% paraformaldehyde and 1% glutaraldehyde in 0 . 1 M sodium cacodylate buffer ( pH 7 . 4 ) . Heart tissue was collected and samples were processed by the University of Texas Southwestern Medical Center Electron Microscopy Core facility . Briefly , fixed tissues were post‐fixed , stained , dehydrated , and embedded in EMbed‐812 resin . Tissue sections were cut and post‐stained , and images were acquired on a FEI Tecnai G2 Spirit TEM . Total RNA was extracted from adult mouse tissues using Trizol and reverse transcribed using iScript Reverse Transcription Supermix ( Bio-Rad , Hercules , CA ) with random primers . Quantitative Polymerase Chain Reaction ( qPCR ) reactions were assembled using KAPA Probe Fast qPCR Master Mix ( SIGMA , St . Louis , MO ) and the following TaqMan probes from Applied Biosystems ( Foster City , CA ) : Atp2a2 ( Mm01201431_m1 ) , Ryr2 ( Mm00465877_m1 ) , Cacna1c ( Mm01188822_m1 ) , Casq2 ( Mm00486742_m1 ) , Pln ( Mm00452263_m1 ) , Nppa ( Mm01255747_g1 ) , Nppb ( Mm01255770_g1 ) , Myh6 ( Mm00440359_m1 ) and Myh7 ( Mm01318999_g1 ) . Assays were performed using a 7900HT Fast Real-Time PCR machine ( Applied Biosystems ) . Expression was normalized to 18S mRNA using Kappa SYBR Fast qPCR Master Mix and was represented as fold change relative to wild-type . 18S and DWORF ( currently annotated as Gm34302 ) oligonucleotides were ordered from Integrated DNA Technologies: 18 s Forward: 5′- ACC GCA GCT AGG AAT AAT GGA −3′ 18 s Reverse: 5′- GCC TCA GTT CCG AAA ACC A −3′ DWORF Forward: 5′- TTC TTC TCC TGG TTG GAT GG −3′ DWORF Reverse: 5′- TCT TCT AAA TGG TGT CAG ATT GAA GT −3′ Tissues were collected and snap frozen in liquid nitrogen . Frozen samples were pulverized and homogenized in RIPA buffer ( SIGMA ) with added cOmplete , EDTA-free protease inhibitor cocktail ( Roche , Basel , Switzerland ) and PhosSTOP phosphatase inhibitors ( Roche ) on ice . Protein concentration was determined using a Pierce BCA Protein Assay Kit ( ThermoFisher Scientific , Waltham , MA ) . Samples were separated on Mini-PROTEAN TGX Precast Gels ( Bio-Rad , Hercules , CA ) or bis/acrylamide gels made by standard gel preparation . Gels were transferred to PVDF membrane ( Millipore , Immobilon-P , Burlington , MA ) , blocked in 5% milk/TBST and then incubated in primary antibodies: total PLN ( 2D12 , Invitrogen , Carlsbad , CA ) , pSer16-PLN and pThr17-PLN ( Badrilla , Leeds , UK ) ; SERCA2 ( 2A7-A1 , Invitrogen , Carlsbad , CA ) ; RyR2 ( Invitrogen , Carlsbad , CA , C3-33 ) ; LTCC ( α1C , Millipore , Burlington , MA ) ; Calsequestrin ( Invitrogen , Carlsbad , CA ) ; DWORF ( custom antibody , New England Peptide , Gardner , MA ) ( Nelson et al . , 2016 ) ; GAPDH ( Invitrogen , Carlsbad , CA ) . Western blots were washed in TBST , incubated with fluorescent or HRP-conjugated secondary antibodies ( Bio-Rad , Hercules , CA ) , and then developed using a ChemiDoc MP Imagine System ( Bio-Rad , Hercules , CA ) or autoradiograph film . Westerns were quantified using ImageJ software ( NIH ) using an internal GAPDH loading control for each western blot analyzed . Specific catalogue numbers for the antibodies used for westerns can be found in the Key Resource Table . All statistical analyses were performed using Prism 6 ( GraphPad , San Diego , CA ) . Information on the statistical analyses presented are included in each figure legend and are either mean ±SD or SEM . Two-tailed t-tests were performed to determine significance . p-Values were defined as follows: * , #p<0 . 05 , , ##p<0 . 01 , * , ###p<0 . 005 or **** , ####p<0 . 001 . All samples were included .	The heart is a muscular organ that contracts regularly to pump blood around the body , ensuring that nutrients and oxygen are carried to the cells and organs . Heart failure is a disease where the heart muscle becomes weakened , does not beat as strongly , and cannot pump blood as well as it should . Eventually , the heart can no longer deliver enough blood to meet the body’s needs . Although heart failure is a widespread disease , we still do not fully understand its underlying causes and the molecular machinery driving its progression . However , one common feature in many cases of heart failure is a problem with the supply of calcium to the heart muscle . Calcium is the molecule responsible for the process of muscle contraction; the strength of contraction depends on the amount of calcium available . Movement of calcium within heart cells is in turn controlled by an enzyme pump called SERCA . In 2016 , researchers identified a small protein , DWORF , which increased the activity of SERCA . Makarewich et al . – including many of the researchers involved in the 2016 study – therefore wanted to find out more about how DWORF and SERCA worked together . They also wanted to test if DWORF could be used to boost the heart’s ability to pump blood efficiently , and if so , whether it could treat heart failure . Genetically modified mice that produced larger than normal amounts of DWORF had more available calcium in the heart muscle , which made it contract more strongly . This was true even when the same mice were treated with an excessive amount of a specific protein ( phospholamban ) that can lower the activity of SERCA , suggesting that DWORF might have a protective effect on the heart . Experiments using mice engineered to show symptoms of heart disease confirmed that DWORF treatment did indeed help their hearts beat normally , and , crucially , prevented them from developing heart failure . This work has shown for the first time that DWORF can restore the heart’s ability to pump normally in an experimental model of heart disease . In the future , Makarewich et al . hope that DWORF could be a useful target for new , more effective drugs to treat heart failure .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2018	The DWORF micropeptide enhances contractility and prevents heart failure in a mouse model of dilated cardiomyopathy
Polyploidization , the increase in genome copies , is considered a major driving force for speciation . We have recently provided the first direct in planta evidence for polyspermy induced polyploidization . Capitalizing on a novel sco1-based polyspermy assay , we here show that polyspermy can selectively polyploidize the egg cell , while rendering the genome size of the ploidy-sensitive central cell unaffected . This unprecedented result indicates that polyspermy can bypass the triploid block , which is an established postzygotic polyploidization barrier . In fact , we here show that most polyspermy-derived seeds are insensitive to the triploid block suppressor admetos . The robustness of polyspermy-derived plants is evidenced by the first transcript profiling of triparental plants and our observation that these idiosyncratic organisms segregate tetraploid offspring within a single generation . Polyspermy-derived triparental plants are thus comparable to triploids recovered from interploidy crosses . Our results expand current polyploidization concepts and have important implications for plant breeding . The evolutionary history of flowering plants is characterized by recurrent polyploidization events ( Comai , 2005; De Bodt et al . , 2005; Otto and Whitton , 2000; Van de Peer et al . , 2017 ) . Polyploids are generally assumed to arise from unreduced gametes or somatic doubling , i . e . from defects during meiosis or mitosis ( Kreiner et al . , 2017; Mason and Pires , 2015; Ramsey and Schemske , 1998; Sattler et al . , 2016; Spoelhof et al . , 2017; Tayalé and Parisod , 2013 ) . In addition , recent in planta and in vitro assays have provided the first direct evidence that viable polyploid plants can arise from polyspermy , the fusion of one egg cell with supernumerary sperm ( Nakel et al . , 2017; Toda et al . , 2016 ) . In fact , this previous work indicates that a single Arabidopsis plant can generate several polyspermy-induced triploid seedlings ( Nakel et al . , 2017 ) . The currently favored polyploidization scenario involves the formation of unreduced male gametes and the natural occurrence of such sperm has been reported for several species ( Kreiner et al . , 2017; Mason and Pires , 2015; Ramsey , 2007 ) . Consequently , triploid plants are assumed to function as an important bridge towards polyploidization ( Comai , 2005; Felber and Bever , 1997; Ramsey and Schemske , 1998 ) and field studies have identified both auto and allopolyploid triploids ( Kyrkjeeide et al . , 2019; Lee et al . , 2001; Marques et al . , 2018; Meng et al . , 2018; Schinkel et al . , 2017; Sree Rangasamy , 1972 ) . However , the generation of triploid plants via unreduced male gametes is limited by the triploid block , which is a postzygotic hybridization barrier operating in many plants species ( Dilkes et al . , 2008; Köhler et al . , 2010; Marks , 1966; Ramsey and Schemske , 1998; Scott et al . , 1998 ) . The triploid block is explained by the unique reproductive mode of flowering plants , which involves fertilization of two female gametes , the egg and the central cell . The required sperm cell pair is typically delivered by a single pollen tube . While the fertilized egg cell gives rise to the embryo , the fertilized central cell develops into embryo-nourishing endosperm ( Russell , 1992 ) . Fertilization involving unreduced sperm consequently not only affects the ploidy status of the egg cell but also introduces additional paternal chromosome copies to the endosperm , and it is this latter tissue , which commonly mounts the triploid block that is manifested by seed abortion ( Köhler et al . , 2010 ) . In Arabidopsis thaliana , the effect of the triploid block is accession-dependent , being highly penetrant e . g . in Col-0 , but less strict in Ler and C24 ( Dilkes et al . , 2008; Scott et al . , 1998 ) . A complete triploid block has been reported in many taxa ( Ramsey and Schemske , 1998; Schinkel et al . , 2017; Sekine et al . , 2013; Stoute et al . , 2012 ) . In light of this fatal consequence , it has been suggested that there are ways to overcome this hybridization barrier ( Köhler et al . , 2010 ) . Making use of a two-component in planta assay , we here show that polyspermy can selectively polyploidize the egg cell , while rendering the genome size of the ploidy-sensitive endosperm unaffected . By introducing the triploid block suppressor admetos , we in addition show , that this unprecedented reproductive mode bypasses the triploid block . Consistent with animal nomenclature , the term polyspermy is used alone when referring to egg cell polyspermy . Central cell polyspermy is specified as such . During flowering plant fertilization , both egg and central cell fuse in a coordinated manner with a single sperm each ( Hamamura et al . , 2011; Kawashima and Berger , 2011 ) . In order to address whether during polyspermy egg cell fertilization is still coupled to the fertilization of the central cell , we aimed at analyzing endosperm in developing seeds that contain polyspermy-derived embryos . To ease the screening process , we established a novel polyspermy-detection assay termed HIPODSCO1 , which can efficiently and unambiguously detect the rare event of egg cell polyspermy already in developing seeds . HIPODSCO1 capitalizes on the pale green appearance of developing seeds defective for the gene SNOWY COTYLEDON 1 ( SCO1 ) ( Ruppel and Hangarter , 2007 ) ( Figure 1—figure supplement 1 ) and a bipartite SCO1 complementation system , provided by two different pollen donors ( Figure 1A ) . Pollen donor one contains the synthetic GAL4 transcription factor under the control of the RPS5a promoter . Pollen donor two contains a functional copy of tdTOMATO tagged SCO1 under the control of the GAL4 responsive UAS enhancer sequence . Seeds that contain a monospermy-derived embryo inherit an incomplete complementation system and will consequently be rendered pale green due to the lack of functional SCO1 . By contrast , combinations of both constructs , which can only result from polyspermy , will give rise to green seeds with a positive tdTOMATO fluorescence signal ( Figure 1B ) . It should be noted that this assay only detects polyspermy if the two sperm are derived from different pollen donors . Monopaternal polyspermy , which only delivers a single HIPODSCO1 component , does not rescue seed color . This scenario is expected to account for 50% of all polyspermy events , and escapes detection . To test the system , we compared seed color of sco1 mutant plants expressing either one of the constructs with seeds containing both , the GAL4 activator and the UAS reporter line . This experiment confirmed that only the presence of both constructs complemented the defect resulting in dark green seeds , which exhibited a tdTOMATO signal ( Figure 1C ) . The novel HIPODSCO1 assay enabled us to screen for seeds that contain polyspermy-derived embryos at an advanced seed developmental stage . We processed a total of 56 , 493 seeds seven days after pollination ( DAP ) and identified 10 normally developed seeds with a change in color ( Figure 2A , Figure 2—figure supplement 1A ) . To determine whether the candidate embryos were indeed of triparental origin , we microscopically inspected the developing seeds and found that all 10 embryos exhibited a tdTOMATO signal ( Figure 2B , Figure 2—figure supplement 1A ) . This implies that the embryo inherited two rather than one paternal copy . To identify a corresponding shift in embryo ploidy , we carried out a chromosome spread assay . Chromosome counts are technically challenging when performed on subfractions of individual seeds and some chromosomes escape detection . However , comprehensive controls and the fact that parental chromosome contributions are quantal in nature make the assay robust and reliable . Notably all embryos showed a triploid profile ( Figure 2C , Figure 2—figure supplement 1B ) . This finding is comparable to the results obtained from triploid embryos segregated from an interploidy cross between diploid and tetraploid plants and contrasts with the diploid profile detected in embryos recovered from a regular cross involving haploid gametes ( Figure 2C ) . To substantiate this result we introduced a GFP-tagged centromere-localized CENH3 reporter into PD1 ( De Storme et al . , 2016 ) . In this complementary experiment we screened 10 , 774 seeds by HIPODSCO1 and recovered three green seeds containing tdTOMATO positive embryos ( Figure 2—figure supplement 1A ) . In all seeds we detected between 11 and 15 GFP foci indicative of triploid embryos ( Figure 2E; Figure 2—source data 1 ) . Together , the analysis confirms the triparental origin of embryos in seeds with dark green color , establishing HIPODSCO1 as a powerful novel tool to identify polyspermy-derived embryos already in developing seeds . In order to determine whether egg cell polyspermy is concomitant with central cell polyspermy , we assessed the ploidy of the endosperm in developing seeds containing polyspermy-induced triparental embryos . The central cell of many flowering plants , including Arabidopsis thaliana , is homodiploid and generates a triploid nurturing tissue after sperm fusion . In fact , we detected between 12 and 15 chromosomes in the endosperm of seeds recovered from a cross involving diploid plants . By contrast , more than 15 chromosomes are detected in control interploidy crosses between diploid female and tetraploid male ( Figure 2D ) . Remarkably , in the 10 developing seeds containing triparental embryos we detected between 11 and 15 chromosomes , which is characteristic of a triploid endosperm ( Figure 2D , Figure 2—figure supplement 1C ) . The result was substantiated by a complementary experiment involving the recombinant CENH3-GFP reporter , which detected a triploid profile in the endosperm of three analyzed seeds ( Figure 2F; Figure 2—source data 1 ) . Notably , at this advanced seed developmental stage , we recovered one abnormal seed from the HIPODSCO1 assay containing an underdeveloped triploid heart-stage embryo and tetraploid endosperm ( Figure 2—figure supplement 1D ) , characteristic of triploid block-induced seed abortion ( Dilkes et al . , 2008; Kradolfer et al . , 2013; Scott et al . , 1998 ) . Together our data indicate that egg cell polyspermy can occur independent of central cell polyspermy . Such selective polyploidization of the egg cell implies that polyspermy has the potential to bypass the triploid block . To further substantiate our findings , we established a functional assay to determine the potential of polyspermy in bypassing this reproductive barrier . It was previously shown that mutations in the paternally expressed imprinted gene ADMETOS ( ADM ) suppress the triploid block in Arabidopsis thaliana ( Kradolfer et al . , 2013 ) . In fact , interploidy crosses between diploid and tetraploid plants lead to a 15 . 6 fold increase in fertile triploid seeds when the tetraploid pollen donor segregated the adm-1 allele ( Figure 3A–C , Figure 3—source data 1 ) . If polyspermy would equally trigger the triploid block , we would expect a similar increase in polyspermy frequencies when using adm-1 segregating pollen donors . In order to identify polyspermy-derived seedlings , we made use of the previously established HIPOD assay that works analogous to the HIPODSCO1 system introduced above , but positively selects triparental seedlings on the basis of herbicide resistance ( Nakel et al . , 2017 ) . A total of 116 , 279 and 113 , 777 seeds were harvested from three independent HIPOD experiments using either adm-1 or wild-type segregating pollen donors , respectively . Out of these , 47 herbicide resistant seedlings segregated from pollen donor with adm-1 background while 27 were recovered from wild type ( Figure 3D–F , Figure 3—source data 1 ) . This corresponds to an almost two fold increase in the adm-1 segregating approach ( Figure 3C ) , which is more than eight times lower than the effect observed in the interploidy cross . Previous results suggested that the egg cell block is stricter than the central cell block ( Grossniklaus , 2017; Scott et al . , 2008 ) and fertilization of the two female gametes during monospermy has been shown to occur in a coordinated fashion ( Kawashima and Berger , 2011; Hamamura et al . , 2011 ) . Our unprecedented finding that most polyspermy-derived embryos develop in the presence of a monospermy-derived endosperm show that polyspermy enables selective polyploidization of the egg cell and concomitant bypassing the triploid block . The transcript profile of triploid plants has been characterized previously and remarkably few differences were found with respect to their cognate diploid controls ( Hou et al . , 2018 ) . Polyspermy-induced plants differ from triploids derived from interploidy crosses as they inherit two sperm cytoplasms and , as shown in this work , mostly develop in a seed characterized by identical ploidies in embryo and endosperm . Given their special mode of origin , we aimed to compare the transcript profile of polyspermy-induced triparental triploids ( TT ) with that of biparental triploids ( BT ) . In addition , we compared the transcriptome profile of BT plants with that of biparental diploids ( BD ) to identify ploidy-dependent changes in the transcriptional landscape . We used ein3 mutants as pollen acceptor as they have previously been shown to attract supernumerary pollen tubes ( Völz et al . , 2013 ) . We performed RNAseq on five plants 18 days after sowing ( DAS ) each from BD , BT , and TT ( Figure 4—figure supplement 1A ) . We chose this early state as we expected potential differences to become established during seed development , which differs in the three settings . In order to assess the quality of the transcriptome data , and to identify potential transcriptome-wide differences between the mRNA profiles of the three groups of plants , we performed a hierarchical clustering , a two-dimensional principal component analysis ( PCA ) , and a two-dimensional multidimensional scaling ( MDS ) analysis of the fifteen 21 , 450-dimensional normalized expression profiles . We found that Spearman’s correlation coefficient c was greater than 0 . 94 for all of the 105 pairs of profiles ( Figure 4—figure supplement 1B ) . Notably , the sub-trees of the dendrogram obtained by hierarchical average linkage clustering did not correspond to the five biological replicates from the same genotype ( Figure 4A ) , and the 15 samples showed a high overlap in the PCA ( Figure 4B ) and MDS ( Figure 4—figure supplement 1C ) plots . These results not only reflect the high quality of the transcriptome data but also suggest a high similarity of the 15 transcriptome profiles compared . We first addressed , if there were genes with ploidy-specific expression changes by comparing transcript profiles of BD and BT . This comparison did not yield a single gene with a statistically significant differential expression . This result is in support of previous transcriptome profiling approaches that have uncovered remarkably few changes in plants with different ploidy ( Pignatta et al . , 2010; Riddle et al . , 2010; Stupar et al . , 2007; Wang et al . , 2006; Yu et al . , 2010 ) . We next compared the transcript profiles of BT and TT in order to identify specific expression changes potentially associated with polyspermy . Interestingly , also this approach did not yield genes with statistically significant differential expression . The similarity in the overall transcriptional profiles between TT and BT plants is reflected by strong similarities in various life-history traits , including flower organ size , cell size , and even fertility ( Figure 4—figure supplement 2 ) . Please note that the normalized expression data presented here does not allow any conclusions on alterations in transcriptome size , i . e . changes affecting the total number of transcripts per cell ( Coate and Doyle , 2015 ) . However , the data suggests that Arabidopsis responds in a transcriptionally balanced fashion to the inheritance of supernumerary genomes and seed homoploidy . Along these lines , also the ability to generate tetraploid offspring within a single generation , a parameter that has been described previously for interploidy cross-induced triploids was maintained: To assess whether polyspermy-derived triploids can segregate stable polyploid offspring , we harvested the seeds of polyspermy-derived triparental plants and propagated them in two successive generations . The progeny of polyspermy-derived triploids segregates a complex swarm of karyotypes , similar to what has previously been described for interploidy crosses ( Henry et al . , 2005 ) . On the basis of flow cytometric analysis , we grouped the plants into five different categories: near-diploids , 2n-3n , near-triploids , 3n-4n and near-tetraploids ( Figure 4C , Figure 4—figure supplement 3A ) . Already in the F2 generation 5 out of 109 plants were found to fall into the near-tetraploids category , while 22 plants segregated a diploid-like profile ( Figure 4C , Figure 4—figure supplement 3B ) . To determine whether any of the high-ploidy plants represented a genuine tetraploid , we collected seeds and determined the ploidy of 20 offspring per individual F2 plant . Flow cytometric analysis revealed that 3 out of 5 near-tetraploid F2 plants segregated exclusively plants that exhibit a ploidy profile characteristic to 4n plants ( Figure 4D ) . This result was confirmed in the F4 generation , which , again , revealed a homogenous tetraploid ploidy profile . Together these results show that polyspermy-derived triploid plants have the potential to generate stable tetraploid and diploid offspring within a single generation . We previously combined three distinct Arabidopsis accessions in a three parent cross ( Nakel et al . , 2017 ) and noticed that the resulting triploid hybrids initiate flowering later than their parents , an effect which was previously described also for two-accession hybrids ( Groszmann et al . , 2014; Moore and Lukens , 2011 ) ( Figure 4—figure supplement 4A ) . To address whether this phenotype has the potential to reduce gene flow , we compared flowering time between the parental line and triparental three accession hybrids , henceforth referred to as TT3 . We found that flowering is induced more than one month later in TT3 than in the parental lines , and the flowering period of TT3 is completely isolated from the parents ( Figure 4E ) . Under our plant growth conditions , Ler and Col-0 start to flower 24 . 4 ± 0 . 7 and 26 . 6 ± 1 . 0 DAS , and flowering terminates 44 . 9 ± 1 . 2 and 47 . 1 ± 0 . 9 DAS , respectively . This corresponds to a flowering period of around 20 days . The C24 flowering window is comparable but flower initiation is delayed by five days ( 31 . 8 ± 1 . 2 DAS ) . By contrast , TT3 plants initiate flowering only after 64 . 6 ± 5 . 4 days , which is around 20 , 17 and 15 days after the respective parental lines have terminated their flowering phase ( Figure 4E; Figure 4—figure supplement 4B ) . Even though these data are obtained under optimized growth condition , the results suggest that polyspermy-derived triploid three accession hybrids are reproductively isolated from the parental plants in the first generation . The triploid block is an established and widely distributed postzygotic hybridization barrier . In light of its fatal consequence , it has been suggested that there are ways to overcome this hybridization barrier ( Köhler et al . , 2010 ) . We here established a novel polyspermy detection assay that allows to identify and characterize developing embryos resulting from supernumerary sperm fusion . With this tool we were able to show that most polyspermy-derived plants develop from seeds resulting from selective egg cell polyploidization . In those seeds , supernumerary paternal copies are only transmitted to the embryo , thereby bypassing the triploid block of the endosperm . Our results expand previous polyploidization concepts , which state that the increase in genome copies is caused by infrequent meiotic or mitotic defects . In fact , the currently favored route towards polyploid plants involves unreduced male gametes; however , this scenario introduces supernumerary paternal copies also to the endosperm , which is not tolerated in many plants resulting in seed abortion ( Dilkes et al . , 2008; Ramsey and Schemske , 1998; Scott et al . , 1998; Stoute et al . , 2012 ) . In fact , this endosperm-related triploid block is considered a means of reproductive isolation ( Köhler et al . , 2010; Ramsey and Schemske , 1998 ) . Plant polyploidization via polyspermy , by contrast , often affects the embryo-derived seed fraction only and hence has the potential to bypass the triploid block . It will be a challenge for the future to determine whether and to what extent polyspermy is relevant in nature and contributed to the evolution of polyploid plants . From an evolutionary and agricultural point of view , another intriguing implication of polyspermy-induced polyploidization is the one-step combination of three genetically distinct plant genomes in a three-parent hybridization . This scenario is expected to occur rarely in nature , even when taking into account wind and animal pollination . Still , it might be relevant in an evolutionary time scale and when considering the huge seed amounts produced by many plants . In addition , our findings might have important implications for agriculture: Not only do three-parent-crosses hold the potential to instantly combine beneficial traits of three rather than two cultivars . They , in addition , might provide a means to overcome endosperm-induced hybrid incompatibilities by transmitting incompatible sperm along with compatible sperm through the egg cell only . Scrutinizing the potential of this genetic hitchhiking approach opens up new avenues for future research . Unless otherwise stated , all experiments were carried out using Arabidopsis thaliana Columbia ( Col-0 ) . The sco1 T-DNA insertion mutant ( SALK_025112 ) ( Ruppel and Hangarter , 2007 ) was obtained from European Arabidopsis Stock Center ( NASC ) ( Nottingham , UK ) . The adm-1 mutant ( Kradolfer et al . , 2013 ) was kindly provided by Claudia Köhler . For RNA-Seq , ein3-1 mutant ( Völz et al . , 2013 ) was used as female plant . Triparental triploids ( TT ) were obtained from double pollination as previously described ( Nakel et al . , 2017 ) , while biparental triploids ( BT ) were recovered from a cross of 2n female with 4n male pollen donor pRPS5a::mGAL4-VP16/-; pUAS::BAR-YFP/- . Biparental diploids ( BD ) were generated from a cross of 2n female with 2n pollen donor pRPS5a::mGAL4-VP16/-; pUAS::BAR-YFP/- . Stratified seeds ( 2 days , 4°C ) were germinated on soil in a Conviron MTPS growth chamber under long-day conditions ( 16 hr light/8 hr dark ) at 23°C . For HIPODWT , HIPODadm and experiments in Figure 4—figure supplement 2 , plants were transferred to 18°C after bolting . Tetraploid Col-0 wild-type , adm-1/- and pollen donor plants were generated by using 0 . 05% colchicine treatment following the one-drop method previously described ( Yu et al . , 2009 ) . Primer sequences are given in Supplementary file 1 . The sco1/- mutant was confirmed via PCR amplification using primer pair LBb1 . 3/YM8-RP and YM8-RP/YM7-LP for the mutant and wild-type allele , respectively . The adm-1 allele was identified as previously described ( Kradolfer et al . , 2013 ) . The presence of pRPS5a::mGAL4-VP16 and pUAS::BAR-YFP was confirmed by using TN12s/TN12as and TN26s/TN26as , respectively . The pUAS::SCO1-tdTOMATO::tNOS vector was obtained by inserting YM1-F/YM2-R amplified CDS of SCO1 from cDNA library of leaves into DR13 ( pAt5g40260::NLS-tdTOMATO::tNOS ) ( Völz et al . , 2013 ) using PacI/NotI , followed by exchanging pAt5g40260 with pUAS sequence from pUAS::BAR-YFP ( Nakel et al . , 2017 ) using AscI/PacI . The CENH3 CDS was amplified from leaf cDNA library using primer pair DT231s/DT231as and cloned into p35S::NLS-GFP using AvrII and MfeI to generate p35S::CENH3-GFP . Transgenic Arabidopsis plants were generated by following the floral dip protocol using Agrobacterium tumefaciens ( Clough and Bent , 1998 ) . Pollen donors with adm-1 background were generated by crossing adm-1 mutant with each pollen donor . For germination assay , all seeds from a single silique ( n ≥ 8 ) were sown on soil . Germination frequency was scored after 9 days . The experiment was repeated three times . Flowering time was assessed as days after sowing ( DAS ) , starting from the day of sowing till the opening of the first flower . Within three independent experiments a total of 7 plants ( TT3 ) , 36 ( Col-0 ) , 37 ( Ler ) , and 35 ( C24 ) were analyzed . Flowering period was defined as the interval between the formation of the first and the last flower . In independent experiments a total of 4 plants ( TT3 ) , 36 ( Col-0 ) , 37 ( Ler ) , and 35 ( C24 ) were analyzed . Please note that watering was stopped once the main branch stopped flowering . For the HIPODSCO1 experiment , 2–3 closed flower buds from one inflorescence of sco1/- mutant were emasculated . 3 days after emasculation , pollination was carried out using pollen grains collected individually from plants of sco1/- pRPS5a::mGAL4-VP16/+ and sco1/- pUAS::SCO1-tdTOMATO/+ . These two pollen categories were applied onto the stigmatic surface with two brushes . 7 days after pollination ( DAP ) , siliques were dissected under Leica S6E or S8apo stereomicroscope ( Leica , Germany ) for seed color screening . Seeds with a change in color were analyzed for tdTOMATO signal and embryo and endosperm chromosome number . Sample preparation and ploidy determination was performed using the same method as described previously; nuclei were marked with CyStain UV Precise P kit and assessed with CyFlow ploidy analyzer ( Nakel et al . , 2017 ) . The triploid swarm ploidies were analyzed by mixing nuclei extracted from two similar sized leaf pieces derived from a diploid control plant and a leaf sample of TT progeny . Seeds that contain TT , BT , and BD were selected on ½ MS medium containing 25 μg/ml Glufosinate-ammonium ( PPT , Sigma ) , and the seedlings were transferred to soil 9 DAS . YFP and ploidy analysis was carried out 17 DAS to confirm TT , BT and BD profile . 18 DAS , aerial tissues were frozen and grinded in pre-cooled EP tubes with metal beads and total RNA was isolated following the procedure of GeneMATRIX Universal RNA purification kit . 750 ng total RNA per sample was used to prepare sequencing libraries according to the Illumina TruSeq RNA Sample Preparation v2 Guide . Purification of the poly-A containing mRNA was performed using two rounds of poly-T Oligos attached to magnetic beads . During the second elution of the poly-A RNA , the RNA was fragmented and primed for cDNA synthesis . After cDNA synthesis the fragments were end-repaired and A-tailing was performed . Multiple indexing adapters were ligated to the ends of the cDNA fragments and the adapter ligated fragments were enriched by 10 cycles of PCR . After quali- and quantification the resulting sequencing libraries were pooled equimolar and sequenced 75 bp single-read on an Illumina NextSeq500 . The sequenced single-end reads were processed by trimmomatic version 0 . 38 ( Bolger et al . , 2014 ) . We used the parameters SE -phred33 ILLUMINACLIP: adapter_sequences . fasta:2:30:10 to remove remaining adapter sequences . For performing a quality trimming , we used the parameters SE -phred33 LEADING:5 TRAILING:5 SLIDINGWINDOW:4:15 MINLEN:36 . Next , the trimmed reads were mapped to the Arabidopsis thaliana genome from EnsemblGenomes ( Kersey et al . , 2018 ) , release 36 , by STAR version 2 . 6 . 0c ( Dobin et al . , 2013 ) with parameters --alignIntronMin 20 --alignIntronMax 50000 --outFilterMismatchNmax 2 --outFilterMultimapNmax 50 --outFilterIntronMotifs . The assignment of reads to the annotated genes of EnsemblGenomes , release 36 , and the quantification of transcript expression were performed by salmon version 0 . 11 . 3 ( Patro et al . , 2017 ) with parameters quant –libType U –gcBias –seqBias –useErrorModel . Genes not having more than zero counts in at least three out of five biological replicates in at least one genotype were removed from the transcriptome analysis , resulting in a count table of 21 , 450 genes . Hence , we still keep genes in the analysis being expressed in just one genotype . The normalization of counts , the computation of the regularized log transformation , and the identification of differentially expressed genes were performed by the R package ( R Development Core Team , 2018 ) DESeq2 ( Love et al . , 2014 ) . For each pair of samples , we computed Spearman’s correlation coefficient c based on the regularized log transformed count data , and we computed the dendrogram of the 15 samples by hierarchical average linkage clustering with the distances d = 1 – c . Principle component analysis ( PCA ) , multidimensional scaling analysis ( MDS ) , the computation of Spearman’s correlation coefficients , and hierarchical clustering were performed by functions prcomp , cmdscale , and cor of the R package stats as well as by the UPGMA algorithm implemented in the R function hclust . Alternatively to the PCA , we chose the MDS analysis as a non-linear approach for dimensional reduction that tries to preserve as closely as possible the pairwise distances of the original multidimensional data points . The prediction of differentially expressed genes was performed by the Wald test , implemented in DESeq2 , with a subsequent application of the Benjamini-Yekutieli correction procedure and a threshold of the false discovery rate of 5% . 7 DAP seeds were harvested and their embryo ( cotyledon stage ) was extracted by applying pressure with syringe needles . The separated embryo and the remaining part of the seed containing the endosperm was individually transferred into 200 μl pre-treatment solution and incubated for 4 hr at RT ( Scott et al . , 2008 ) . For aborted seeds recovered from interploidy crosses and HIPODSCO1 , the embryo was separated from the endosperm in the final fixative step on the slide . The remaining steps were carried out as described in Armstrong et al . ( 2001 ) . Images were taken using Leica DMI6000b epifluorescence inverted microscope , equipped with YFP , DAPI and DsRed filter cube . For YFP expression analysis , dissected sepals were transferred to 10% glycerol and data were collected . For chromosome spread , data from DAPI stained samples were collected with the aid of 100 × oil objective and an immersion oil Immersol 518F ( Zeiss , Germany ) or by using confocal laser scanning microscopy with Airyscan module ( Zeiss LSM 880 ) at excitation of 405 nm and emission of 421 nm . CENH3-GFP localization was observed at excitation of 488 nm and emission of 510 nm . 7 DAP silique images were captured by using Leica S8apo stereomicroscope .	Ever since Darwin published his most famous book on the theory of evolution , scientists have sought to identify the mechanisms that drive the formation of new species . This is especially true for plant biologists who have long been fascinated by the extraordinary diversity of flowering plants . Many species of flowering plant first evolved after a dramatic increase in the DNA content of an individual plant , a process termed polyploidization . Most explanations for polyploidization involve a pollen grain making sperm that mistakenly contain two sets of chromosomes rather than one . Yet , it is difficult to reconcile this explanation with an important aspect of plant reproduction – the so-called “triploid block” . Fertilization in flowering plants is more complicated than in animals . While one sperm fertilizes the egg cell to make the plant embryo , a second sperm from the same pollen grain must fertilize another cell to form the endosperm , the tissue that will nourish the embryo as it develops . This means that sperm with twice the normal number of chromosomes would affect the DNA content of both the embryo and the endosperm . Yet , an endosperm that receives extra paternal DNA typically halts the development of the seed via a process known as the triploid block , meaning it was not clear how often this process would actually result in a polyploid plant . In 2017 , researchers reported that plants can , on rare occasions , generate polyploid offspring via a different route: the fertilization of one egg with two sperm rather than one . Now , Mao et al . – who include several researchers involved in the 2017 study – show that this process , termed “polyspermy” , can introduce extra copies of DNA into just the egg cell , meaning it can bypass the triploid block of the endosperm . The experiments involved a model plant called Arabidopsis , and a screen of over 55 , 000 seeds identified about a dozen with embryos that had three parents , one mother and two fathers . Notably , most of these three-parent embryos developed in seeds that contained endosperm with the regular number of chromosomes and hence escaped the triploid block . These new results show that polyspermy provides plants with a means to essentially sneak extra copies of DNA ‘behind the back’ of the DNA-sensitive endosperm and into the next generation . They also give new insight in how polyploidization may have shaped the evolution of flowering plants and have important implications for agriculture where the breeding of new “hybrid” crops has often been limited by incompatibilities in the endosperm .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "developmental", "biology" ]	2020	Selective egg cell polyspermy bypasses the triploid block
Arthropod-borne rickettsial pathogens cause mild and severe human disease worldwide . The tick-borne pathogen Rickettsia parkeri elicits skin lesions ( eschars ) and disseminated disease in humans; however , inbred mice are generally resistant to infection . We report that intradermal infection of mice lacking both interferon receptors ( Ifnar1-/-;Ifngr1-/- ) with as few as 10 R . parkeri elicits eschar formation and disseminated , lethal disease . Similar to human infection , eschars exhibited necrosis and inflammation , with bacteria primarily found in leukocytes . Using this model , we find that the actin-based motility factor Sca2 is required for dissemination from the skin to internal organs , and the outer membrane protein OmpB contributes to eschar formation . Immunizing Ifnar1-/-;Ifngr1-/- mice with sca2 and ompB mutant R . parkeri protects against rechallenge , revealing live-attenuated vaccine candidates . Thus , Ifnar1-/-;Ifngr1-/- mice are a tractable model to investigate rickettsiosis , virulence factors , and immunity . Our results further suggest that discrepancies between mouse and human susceptibility may be due to differences in interferon signaling . Obligate cytosolic bacterial pathogens in the family Rickettsiaceae are a diverse group of arthropod-borne microbes that cause severe human disease worldwide , including spotted fever , scrub typhus , and typhus ( Bonell et al . , 2017; Fang et al . , 2017; Sahni et al . , 2019 ) . Human disease caused by the tick-borne spotted fever group ( SFG ) pathogen Rickettsia parkeri is characterized by a necrotic , dry skin lesion ( eschar ) at the infection site , as well as generalized rash , headache , fatigue , and fever ( Paddock et al . , 2008 ) . There is no approved vaccine for R . parkeri or for the more virulent rickettsial pathogens that can cause fatal or latent disease ( Osterloh , 2017 ) . Moreover , many critical aspects of disease caused by obligate cytosolic bacterial pathogens , including the mechanisms of virulence and immunity , remain unknown . R . parkeri can be handled under biosafety level 2 ( BSL2 ) conditions , and characterizing mouse models to recapitulate major features of human disease would enhance research efforts into understanding rickettsial pathogens and disease ( Osterloh , 2017; Grasperge et al . , 2012; Sunyakumthorn et al . , 2013 ) . R . parkeri is genetically similar to the more virulent human pathogens R . rickettsii and R . conorii ( Roux and Raoult , 2000; Goddard , 2009 ) , and it can be handled under BSL2 conditions . Moreover , mutants can be generated using transposon mutagenesis ( Reed et al . , 2014; Lamason et al . , 2018 ) , and small rodents including mice have been found as seropositive for R . parkeri in the wild ( Moraru et al . , 2013a; Moraru et al . , 2013b; Krawczak et al . , 2016; Barbieri et al . , 2019 ) . Thus , a mouse model for R . parkeri that recapitulates key features of human infection would greatly enhance investigations into understanding rickettsial disease . However , inbred mice including C57BL/6 and BALB/c develop no or minor skin lesions upon intradermal ( i . d . ) infection with R . parkeri ( Grasperge et al . , 2012 ) . Strategies to develop a mouse model , including infecting Toll-like receptor 4 ( TLR4 ) -deficient mice ( Grasperge et al . , 2012 ) , delivering high doses of bacteria ( Londoño et al . , 2019 ) , and delivering R . parkeri via a tick vector ( Saito et al . , 2019 ) , have been examined . However , a model that recapitulates eschar formation and dissemination via needle inoculation with low doses of R . parkeri has remained elusive . A mouse model for R . parkeri that mimics eschar formation and disseminated disease in C57BL/6 mice , the most widely used genetic background with a large variety of available mutants , would aid investigations into rickettsial virulence mechanisms , the host response to infection , and spotted fever disease . Toward better understanding the host response to R . parkeri infection , we recently investigated the relationship between R . parkeri and interferons ( IFNs ) , which are ubiquitous signaling molecules of the innate immune system that mobilize the cytosol to an antimicrobial state . Type I IFN ( IFN-I ) generally restricts viral replication , whereas IFN-γ generally restricts intracellular bacterial pathogens ( Raniga and Liang , 2018; Billiau and Matthys , 2009; Meunier and Broz , 2016 ) . We observed that mice lacking either gene encoding the receptors for IFN-I ( Ifnar1 ) or IFN-γ ( Ifngr1 ) are resistant to i . v . infection with R . parkeri , whereas double knockout ( DKO ) Ifnar1-/-;Ifngr1-/- mice succumb ( Burke et al . , 2020 ) . This demonstrates that IFNs redundantly protect against systemic R . parkeri . However , the i . v . infection route does not recapitulate eschar formation and i . d . infection may more closely mimic infection by tick bite . Further investigations into whether IFNs redundantly protect against R . parkeri in the skin may improve the mouse model for SFG Rickettsia . A robust mouse model would facilitate investigations into conserved rickettsial virulence factors , whose role in pathogenesis in vivo remain poorly understood . One virulence mechanism shared by divergent cytosolic bacterial pathogens including Rickettsia , Listeria , Burkholderia , Mycobacterium , and Shigella species is the ability to undergo actin-based motility , which facilitates cell to cell spread ( Choe and Welch , 2016; Lamason and Welch , 2017 ) . However , the pathogenic role for many actin-based motility factors in vivo remains unknown . R . parkeri actin-based motility differs from that of other pathogens in that it occurs in two phases . The first phase requires the RickA protein , which elicits actin-based motility by activating the host Arp2/3 complex ( Gouin et al . , 2004; Jeng et al . , 2004 ) ; however , RickA is dispensable for cell to cell spread in vitro ( Reed et al . , 2014 ) . The second phase requires the Sca2 protein ( Reed et al . , 2014; Kleba et al . , 2010 ) , which mimics eukaryotic formins to directly nucleate and elongate actin filaments ( Reed et al . , 2014; Haglund et al . , 2010 ) . Sca2 is required for efficient cell to cell spread , although it is not required for replication in epithelial cells or for avoiding antimicrobial autophagy ( Reed et al . , 2014; Kleba et al . , 2010; Haglund et al . , 2010; Engström et al . , 2019 ) . sca2 mutant R . rickettsii elicit reduced fever in guinea pigs when compared with wild-type ( WT ) R . rickettsii ( Kleba et al . , 2010 ) , yet the explanation for reduced fever and the pathogenic role for Sca2 in vivo remains unclear . Additionally , Sca2 is not essential for dissemination of R . parkeri within ticks ( Harris et al . , 2018 ) . A second virulence strategy employed by intracellular pathogens is the ability to avoid autophagy , which for R . parkeri requires the abundant , conserved outer membrane protein B ( OmpB ) ( Engström et al . , 2019; Engström et al . , 2021 ) . ompB mutant R . parkeri are ubiquitylated and restricted by antimicrobial autophagy in mouse macrophages , and OmpB is important for R . parkeri infection of internal organs in WT mice and for causing lethal disease in IFN receptor-deficient mice after i . v . infection ( Burke et al . , 2020; Engström et al . , 2019 ) . However , the role for R . parkeri OmpB upon i . d . infection remains unknown . An improved mouse model may improve our understanding of how conserved virulence factors including Sca2 and OmpB enhance rickettsial pathogenesis . Here , we use IFN receptor-deficient mice to examine the effects of i . d . inoculation of R . parkeri , mimicking the natural route of infection . We observe skin lesions that appear similar to human eschars , as well as disseminated lethal disease with as few as 10 bacteria . Using this model , we find that Sca2 promotes dissemination and is required for causing lethality and that OmpB contributes to eschar formation and to lethal disease . We demonstrate that immunization with sca2 or ompB mutant R . parkeri protects IFN receptor-deficient mice against subsequent challenge with WT bacteria , revealing live-attenuated vaccine candidates . Our study establishes a mouse model to investigate numerous aspects of Rickettsia pathogenesis , including eschar formation , virulence factors , and immunity . More broadly , this work also reveals that a potent , redundant IFN response protects mice from eschar-associated rickettsiosis . We sought to develop an i . d . murine infection model to better recapitulate the natural route of tick-borne R . parkeri infection . WT , single mutant Tlr4-/- , Ifnar1-/- , or Ifngr1-/- , and DKO Ifnar1-/-;Ifngr1-/- C57BL/6J mice , as well as outbred WT CD-1 mice , were infected i . d . with 107 WT R . parkeri and monitored over time . No or minor dermal lesions appeared at the site of infection in WT , single mutant Tlr4-/- , Ifnar1-/- , or Ifngr1-/- C57BL/6J mice or CD-1 mice ( Figure 1 , Figure 1—figure supplement 1a ) . In contrast , DKO Ifnar1-/-;Ifngr1-/- C57BL/6J mice developed delimited skin lesions measuring >1 cm in diameter that were necrotic , hardened , non-pruritic , and surrounded by an indurated red halo ( Figure 1b ) , similar to human eschars ( Figure 1c; Paddock et al . , 2008; Paddock et al . , 2004; Kaskas , 2014; Cragun et al . , 2010; Herrick et al . , 2016 ) . In some cases , tails of DKO Ifnar1-/-;Ifngr1-/- or single mutant Ifngr1-/- mutant mice became inflamed after i . d . or i . v . infection ( Figure 1—figure supplement 1b ) . These findings demonstrate that interferons redundantly control disease caused by R . parkeri in the skin and that i . d . infection of DKO Ifnar1-/-;Ifngr1-/- mice recapitulates the hallmark manifestation of human disease caused by R . parkeri . Our previous observations using the i . v . route revealed dose-dependent lethality in Ifnar1-/-;Ifngr1-/- mice , with 107 R . parkeri eliciting 100% lethality and 105 R . parkeri eliciting no lethality ( Burke et al . , 2020 ) . R . parkeri are present in tick saliva at a concentration of approximately 104 per 1 μl , and approximately 107 R . parkeri are found in tick salivary glands ( Suwanbongkot et al . , 2019 ) . However , the number of bacteria delivered from tick infestation likely varies depending on many factors , and we therefore sought to examine the effects of different doses of R . parkeri upon i . d . infection of Ifnar1-/-;Ifngr1-/- mice . We observed skin lesion formation at all infectious doses , from 107 to 10 bacteria ( Figure 1d ) , suggesting that i . d . infection of Ifnar1-/-;Ifngr1-/- mice elicits lesions with doses similar to what is delivered by tick infestation . We next sought to quantitatively evaluate the effects of i . d . infection by monitoring animal weight , body temperature , the degree of lesion formation , and lethality . Intradermally infected DKO Ifnar1-/-;Ifngr1-/ mice lost significant body weight ( Figure 2a , Figure 2—figure supplement 1a ) and body temperature ( Figure 2b; animals were euthanized when body temperature fell below 90°F / 32 . 2°C ) when compared with WT mice , whereas infected single mutant Tlr4-/- , Ifnar1-/- , or Ifngr1-/- mice did not . To evaluate lesion severity , we scored lesions upon infection with different doses of R . parkeri . Whereas 107 bacteria elicited similar responses as 105 , 104 , 103 , and 102 bacteria ( Figure 2c ) , lesions were less severe when mice were infected with 101 bacteria compared with 107 bacteria . If mice survived , lesions healed over the course of approximately 15–40 days post infection ( d . p . i . ) at all doses ( Figure 2—figure supplement 1b ) . To investigate whether i . d . infection by R . parkeri caused lethal disease , we monitored mouse survival over time . Upon i . d . delivery of 107 R . parkeri , 8 of 12 Ifnar1-/-;Ifngr1-/- mice exhibited lethargy , paralysis , or body temperatures below 90oF , at which point they were euthanized , whereas delivery of the same dose of bacteria to WT and single mutant mice did not elicit lesions and all survived ( Figure 2d ) . Lower doses of R . parkeri also elicited body weight loss ( Figure 2a ) , body temperature loss ( Figure 2—figure supplement 1 ) , and lethal disease ( Figure 2d ) in Ifnar1-/-;Ifngr1-/- mice . Degrees of lethality between different doses in Ifnar1-/-;Ifngr1-/- mice were not significantly different from one another , and the cause of lethality in this model remains unclear . Nevertheless , these findings reveal that i . d . infection can cause lethal disease in Ifnar1-/-;Ifngr1-/- mice with ~10 , 000-fold lower dose of bacteria than i . v . infection . It remained unclear whether i . d . infection could also be used to model dissemination from the skin to internal organs . We therefore evaluated bacterial burdens in spleens and livers of WT and Ifnar1-/-;Ifngr1-/- mice at 5 d . p . i . by measuring R . parkeri plaque-forming units ( p . f . u . ) . Bacteria were not recoverable from spleens and livers of intradermally infected WT mice , suggesting that they did not disseminate from the skin to internal organs in high numbers ( Figure 2e ) . In contrast , bacteria were recovered from spleens and livers of intradermally infected Ifnar1-/-;Ifngr1-/- mice at 5 d . p . i . ( Figure 2e ) . This demonstrates that i . d . infection of Ifnar1-/-;Ifngr1-/- mice with R . parkeri causes systemic infection and can be used as a model for dissemination from the skin to internal organs . To examine whether lesions in Ifnar1-/-;Ifngr1-/- mice had similar inflammation and necrosis as those observed in human eschars at the tissue and cellular level , we performed histologic evaluation to further characterize the tissue alterations of i . d . -infected Ifnar1-/-;Ifngr1-/- mice and used anti-Rickettsia immunohistochemistry to identify the infected cell types in the skin . WT mice infected i . d . with 103 R . parkeri exhibited no discernible inflammation in the skin at 4 , 7 , or 12 d . p . i . ( Figure 3a ) . In contrast , at 4 d . p . i . , Ifnar1-/-;Ifngr1-/- mice infected with 103 R . parkeri exhibited mild inflammation characterized by dermal inflammatory foci composed of neutrophils and macrophages , with high numbers of intralesional rickettsiae ( Figure 3b ) . By 7 d . p . i . , the inflammation throughout the skin was severe , with large coalescing foci of predominantly neutrophils and fewer macrophages within areas of fibrosis , associated with abundant rickettsiae . The epidermis had multifocal coagulative necrosis associated with fibrin thrombosis of the underlying dermal capillaries ( Figure 3c ) . The inflammation was still severe and associated with large numbers of intralesional rickettsiae at 12 d . p . i . , accompanied by extensive coagulative necrosis of the skin ( Figure 3d ) . Together , this indicates that R . parkeri causes extensive necrosis and inflammatory cell infiltration of the skin of i . d . -infected Ifnar1-/-;Ifngr1-/- mice , with primary staining within neutrophils and macrophages . We also examined internal organ tissues of mice infected i . d . by performing immunohistochemistry of brain and lung at 7 d . p . i . , as infection of these organs may result in lethality . In the lungs of Ifnar1-/-;Ifngr1-/- mice , inflammatory foci were scattered throughout the parenchyma , with several cells morphologically compatible with neutrophils containing rickettsiae ( Figure 3e ) . In one tissue section , rickettsiae were also evident in some cells lining small vessels and therefore interpreted as endothelial cells . In the brain , rickettsial immunostaining was not observed after i . d . infection with 103 R . parkeri , but infection with a higher dose ( 106 ) resulted in low number of rickettsial bacteria in the leptomeninges and the choroid plexus , in regions moderately infiltrated by neutrophils and macrophages ( Figure 3f ) . These data demonstrate that R . parkeri disseminates from the skin to internal organs , where they are again primarily found in macrophages and neutrophils . Finally , we used histology and immunohistochemistry after i . v . infection to analyze spleens and livers , as R . parkeri is abundant and easily recoverable from these organs after i . v . infection ( Burke et al . , 2020; Engström et al . , 2019 ) . In WT mice , we observed little to no inflammation or lesions . In contrast , both organs of Ifnar1-/-;Ifngr1-/- mice had fibrinoid vascular wall degeneration , endothelial hypertrophy , fibrin thrombi in medium caliber vessels , and marked inflammation that was composed predominantly of macrophages . In WT mice , rickettsiae were infrequent and were found in macrophages in the red pulp of the spleen and scarcely throughout the liver in Kupffer cells ( Figure 3g , h; Bonell et al . , 2017 ) . In contrast , in Ifnar1-/-;Ifngr1-/- mice , rickettsiae were abundant in the splenic red pulp , primarily infecting histiocytes/macrophages ( Figure 3g ) . In the liver of Ifnar1-/-;Ifngr1-/- mice , rickettsiae were abundant in macrophages within the granulomas ( Figure 3h ) . We did not observe any immunostaining for Rickettsia in endothelial cells or vessels in the spleen or liver after i . v . infection . Together , these findings reveal that macrophages are the primary cell type affected by R . parkeri in the spleen and liver after i . v . infection . We next examined whether Ifnar1-/-;Ifngr1-/- mice could be used to characterize R . parkeri virulence factors . Sca2 is a surface protein that mediates actin-based motility in rickettsial pathogens; however , its contribution to virulence in vivo remains unclear . We examined if i . v . and i . d . infections of WT and Ifnar1-/-;Ifngr1-/- mice could reveal a pathogenic role for R . parkeri Sca2 . Upon i . v . infection with 5 × 106 bacteria ( Figure 4 ) or 107 bacteria ( Figure 4b ) , we observed that sca2::Tn mutant R . parkeri caused reduced lethality compared to WT bacteria . Similarly , i . d . infection with sca2::Tn mutant bacteria elicited significantly less lethality ( Figure 4c ) and weight loss ( Figure 4d ) as compared to WT bacteria and no severe temperature loss ( Figure 4—figure supplement 1a ) . Although we sought to evaluate infection using a sca2 complement strain of R . parkeri , our attempts to generate such a strain were unsuccessful . As an alternative strategy , we examined whether the transposon insertion itself had an effect on R . parkeri survival in vivo . We evaluated infection of an R . parkeri strain that harbors a transposon insertion in MC1_RS08740 ( previously annotated as MC1_05535 ) , which has no known role in virulence ( Engström et al . , 2019 ) . I . v . infection with MC1_RS08740::Tn R . parkeri caused lethality to a similar degree as WT R . parkeri ( Figure 4a ) , demonstrating that the transposon likely does not significantly impact R . parkeri fitness in vivo . Together , these findings suggest that the actin-based motility factor Sca2 is required for causing lethal disease in Ifnar1-/-;Ifngr1-/- mice . We next examined whether Sca2 facilitates R . parkeri dissemination throughout the skin and whether Sca2 is required for lesion formation . Unexpectedly , upon i . d . inoculation , Ifnar1-/-;Ifngr1-/- mice infected with sca2::Tn mutant bacteria developed skin lesions that were of similar severity to lesions caused by WT R . parkeri; however , the lesions elicited by sca2 mutant bacteria appeared significantly earlier than lesions caused by WT bacteria ( Figure 4e ) . Further examinations will be required to better evaluate this observation; however , it may suggest that actin-based motility enables R . parkeri to avoid a rapid onset of inflammation in the skin . To evaluate R . parkeri dissemination within the skin , we used a fluorescence-based assay that measures vascular damage as a proxy for pathogen dissemination ( Glasner et al . , 2017 ) . Mice were intradermally infected with WT and sca2::Tn R . parkeri . At 5 d . p . i . , fluorescent dextran was intravenously delivered , and fluorescence was measured at the infection site ( Figure 4f , representative small black circle ) and in the surrounding area ( Figure 4f , representative large black circle ) . No significant differences were observed when comparing WT and sca2::Tn R . parkeri infections in Ifnar1-/-Ifngr1-/- mice using an infectious dose of 107 R . parkeri in the larger surrounding area ( Figure 4g ) or at the site of infection ( Figure 4—figure supplement 2a ) . Similar results were observed upon infection with 106 or 105 bacteria ( Figure 4—figure supplement 2b , c ) . However , significantly more fluorescence was observed in the skin of infected Ifnar1-/-;Ifngr1-/- mice as compared to WT mice ( Figure 4g ) , demonstrating that interferons protect against increased vascular permeability during R . parkeri infection . Fluorescence was also measured in spleens and livers; however , no differences between control or experiment groups was observed , suggesting that this assay as described is most appropriate for the skin . Together , the gross pathological analysis and fluorescence-based assay suggest that Sca2 likely does not significantly enhance R . parkeri spread throughout the skin during i . d . infection of Ifnar1-/-;Ifngr1-/- mice . Among the factors that mediate actin-based motility , the Listeria monocytogenes actin-based motility factor ActA is one of the best understood . ActA enables L . monocytogenes to spread from cell to cell ( Choe and Welch , 2016; Lamason and Welch , 2017 ) , escape antimicrobial autophagy ( Cheng et al . , 2018; Mitchell et al . , 2018; Yoshikawa et al . , 2009b; Yoshikawa et al . , 2009a ) , proliferate in mouse organs after i . v . infection ( Auerbuch et al . , 2001; Le Monnier et al . , 2007 ) , and cause lethal disease in mice ( Goossens et al . , 1992; Brundage et al . , 1993 ) . We initially hypothesized that R . parkeri Sca2 plays a similar pathogenic role in vivo to ActA , which we found is required for bacterial survival in spleens and livers upon i . v . delivery ( Figure 4h ) , in agreement with previous experiments ( Auerbuch et al . , 2001; Le Monnier et al . , 2007 ) . However , when we examined bacterial burdens upon i . v . infection of Ifnar1-/-;Ifngr1-/- mice with R . parkeri , similar amounts of WT and sca2::Tn bacteria were recovered in spleens ( Figure 4i ) . We were also surprised to find that significantly more sca2::Tn than WT R . parkeri were recovered in livers ( Figure 4i ) . The explanation for higher sca2::Tn burdens in livers remains unclear . Nevertheless , these data reveal that Sca2 is likely not essential for R . parkeri survival in blood , invasion of host cells , or intracellular survival in spleens and livers . We next evaluated the role for Sca2 in R . parkeri dissemination by measuring p . f . u . in spleens and livers following i . d . infection of Ifnar1-/-;Ifngr1-/- mice . After i . d . infection , sca2::Tn mutant bacteria were ~20-fold reduced in their abundance in spleens and ~2-fold reduced in their abundance in livers as compared to WT R . parkeri ( Figure 4j ) . Similar results were seen upon i . d . infection with lower doses of sca2::Tn and WT bacteria ( Figure 4k ) . Together , these results suggest that Sca2 is required for R . parkeri dissemination from the skin to internal organs . Sca2-mediated actin-based motility is required for efficient plaque formation and cell to cell spread by R . parkeri in vitro ( Reed et al . , 2014; Kleba et al . , 2010 ) . However , it remains unclear if Sca2 enables R . parkeri to escape detection or restriction by innate immunity . The actin-based motility factor ActA enables L . monocytogenes to avoid autophagy ( Yoshikawa et al . , 2009b; Yoshikawa et al . , 2009a ) , and the antimicrobial guanylate binding proteins ( GBPs ) inhibit Shigella flexneri actin-based motility ( Piro et al . , 2017 ) . We therefore sought to evaluate whether Sca2-mediated actin-based motility enables R . parkeri to evade innate immunity in vitro . We found that the sca2::Tn mutant grew similarly to WT bacteria in endothelial cells ( Figure 4—figure supplement 3a ) , consistent with previous reports in epithelial cells ( Reed et al . , 2014; Kleba et al . , 2010 ) . We also examined whether Sca2 contributed to R . parkeri survival or growth in bone marrow-derived macrophages ( BMDMs ) , which can restrict other R . parkeri mutants that grow normally in endothelial cells ( Engström et al . , 2019 ) . However , no significant difference in bacterial survival was observed between WT and sca2::Tn bacteria in BMDMs in the presence or absence of IFN-β ( Figure 4—figure supplement 3b ) . WT and sca2 mutant R . parkeri also elicited similar amounts of host cell death ( Figure 4—figure supplement 3c ) and IFN-I production ( Figure 4—figure supplement 3d ) . Moreover , we found that the anti-rickettsial factor GBP2 localized to the surface of sca2::Tn mutant R . parkeri at similar frequency as with WT bacteria in the presence or absence of IFN-β ( Figure 4—figure supplement 3e , f ) . Together , these data suggest that Sca2 does not significantly enhance the ability of R . parkeri to evade interferon-stimulated antimicrobial genes or inflammasomes in vitro . Because sca2 mutant R . parkeri showed no defect in eschar formation compared to WT , it remained unclear whether skin lesion formation in Ifnar1-/-;Ifngr1-/- mice was influenced by bacterial virulence factors . We therefore investigated i . d . infection with ompB::TnSTOP R . parkeri , which harbors both a transposon and a stop codon in ompB but no other mutations as determined by whole genome sequencing ( Engström et al . , 2019 ) . ompB mutant R . parkeri are severely attenuated in mice , as evaluated by measuring p . f . u . in organs of WT mice after i . v . infection ( Engström et al . , 2019 ) or by measuring lethality in Ifnar1-/-;Ifngr1-/- mice after i . v . infection ( Burke et al . , 2020 ) . In contrast with WT bacteria , i . d . infection of Ifnar1-/-;Ifngr1-/- mice with ompB::TnSTOP R . parkeri caused no lethality ( Figure 5a ) or weight loss ( Figure 5b ) . The ompB::TnSTOP mutant R . parkeri also caused significantly less severe skin lesions than WT bacteria ( Figure 5c ) . These findings suggest that Ifnar1-/-;Ifngr1-/- mice can be used as a model to identify bacterial genes important for eschar formation . There is currently no available vaccine to protect against SFG Rickettsia , which can cause severe and lethal human disease ( Osterloh , 2017; Dantas-Torres , 2007 ) , and identifying mouse models that develop protective immunity to R . parkeri would aid investigations into identifying live attenuated vaccine candidates . We therefore examined whether immunization with attenuated R . parkeri mutants would protect against subsequent re-challenge with a lethal dose of WT bacteria . Ifnar1-/-;Ifngr1-/- mice were immunized i . v . with 5 × 106 sca2::Tn or ompB::TnSTOP R . parkeri and 40 d later were re-challenged i . v . with 107 WT R . parkeri , which is approximately 10 times a 50% lethal dose ( LD50 ) ( Burke et al . , 2020 ) . All mice immunized with sca2 or ompB mutant R . parkeri survived , whereas all naïve mice succumbed to i . v . challenge by 6 d . p . i . ( Figure 6a ) . Upon i . v . rechallenge , mice immunized with ompB and sca2 mutants also did not lose significant weight ( Figure 6b ) or body temperature ( Figure 6c ) . We next examined whether immunized mice were protected from i . d . rechallenge . Ifnar1-/-;Ifngr1-/- mice were immunized with ompB or sca2 mutant R . parkeri by i . d . infection or with sca2 mutant R . parkeri by i . v infection and rechallenged 40 d later by i . d . infection with 105 WT bacteria . Immunized mice were protected from lethal disease ( Figure 6d ) and had less severe skin lesions than naïve mice ( Figure 6e ) . These data indicate that attenuated R . parkeri mutants elicit robust protective immune responses , and that Ifnar1-/-;Ifngr1-/- mice may serve as tools to develop live attenuated R . parkeri vaccine candidates . In this study , we find that IFN-I and IFN-γ redundantly protect inbred mice from eschar-associated rickettsiosis and disseminated disease by R . parkeri . Eschar formation is a hallmark clinical feature of human disease caused by R . parkeri ( Paddock et al . , 2008; Paddock et al . , 2004 ) , and thus these findings suggest that the striking difference between human and mouse susceptibilities to R . parkeri may be due to IFN signaling in the skin . Using this mouse model , we uncover a role for R . parkeri Sca2 in dissemination , for OmpB in skin lesion formation , and for both proteins in causing lethal disease . We further demonstrate that attenuated R . parkeri mutants elicit long-lasting immunity , revealing live attenuated vaccine candidates . Obligate cytosolic bacterial pathogens cause a variety of severe human diseases on six continents ( Bonell et al . , 2017; Abdad et al . , 2018 ) , and the animal model described here will facilitate future investigations into rickettsial virulence factors , the host response to infection , and the molecular determinants of human disease . Our observation that i . d . infection of DKO Ifnar1-/-;Ifngr1-/- mice causes eschar formation highlights the critical importance of interferons in restricting R . parkeri in mice and may reveal a key molecular determinant of human disease . IFN-γ restricts virulent rickettsial species in vitro and in mice , including R . prowazekii ( Turco and Winkler , 1983 ) , R . conorii ( Feng et al . , 1994; Li et al . , 1987; Manor and Sarov , 1990 ) , and R . australis ( Walker et al . , 2001 ) . Moreover , IFN-I has modest anti-rickettsial activity towards R . prowazekii , R . conorii , and R . rickettsii in cell lines in vitro ( Turco and Winkler , 1990; Colonne et al . , 2011; Hanson , 1991 ) . However , the contribution made by IFN-I in restricting Rickettsia species besides R . parkeri in mice remains unknown ( Osterloh , 2017 ) . Importantly , the gross similarities we observe between human eschars ( Paddock et al . , 2008; Herrick et al . , 2016 ) and skin lesions in Ifnar1-/-;Ifngr1-/- mice may indicate that the IFN response in humans is less well adapted to control R . parkeri than that in mice . Cytokine profiling and mRNA transcript analysis of human rickettsial infections reveals IFN-I and IFN-γ activation ( Cillari et al . , 1996; Jia et al . , 2020; de Sousa et al . , 2007 ) ; however , it remains unclear whether human interferon-stimulated genes ( ISGs ) are as protective as mouse ISGs . Future investigations into the ISGs that restrict R . parkeri in mouse versus human cells may improve our understanding of human susceptibility to SFG Rickettsia . Human eschars resulting from R . parkeri infection are characterized by necrosis of the epidermis , vasculitis of small- to medium-sized dermal vessels , fibrin thrombi , infiltration by macrophages and neutrophils , and infection of macrophage/mononuclear cells ( Paddock et al . , 2008; Herrick et al . , 2016 ) . Eschars elicited by R . parkeri in non-human primates revealed similar histological findings and also bacteria within macrophages and neutrophils ( Banajee et al . , 2015 ) . Similar histological findings were noted in guinea pigs after intraperitoneal infection with R . parkeri , as well as infection of mononuclear cells ( Paddock et al . , 2017 ) . In Ifnar1-/-;Ifngr1-/- mice , we observed necrosis , vasculitis , fibrin thrombi , and rickettsial staining that coincided primarily with infiltrating macrophages and neutrophils . We note that R . parkeri primary targeting of macrophages and neutrophils differs from cell types infected by R . rickettsii , which primarily targets endothelial cells ( Walker and Ismail , 2008 ) , and therefore infection of Ifnar1-/-;Ifngr1-/- mice with R . parkeri is likely not an appropriate model for lethal Rocky Mountain spotted fever disease . One contrasting finding in eschars of Ifnar1-/-;Ifngr1-/- mice as compared to humans is the number of bacteria observed - we find that the pathogen is numerous throughout the skin of Ifnar1-/-;Ifngr1-/- mice , but their numbers are fewer in reported human eschar biopsies ( Paddock et al . , 2008; Herrick et al . , 2016 ) . This discrepancy may be due to differences in infection kinetics or in interferon-mediated restriction of bacteria . Nevertheless , our pathological findings of infected interferon-receptor deficient mice are similar in many ways to the previous findings in humans , primates , and mice . Our results suggest that interferon receptor-deficient mice may be useful tools for modeling eschar-associated rickettsiosis at the cellular level . Investigating the IFN response in the skin may also lead to a better understanding of other arthropod-borne diseases . One example may be scrub typhus , caused by Orientia tsutsugamushi ( Rajapakse et al . , 2017 ) , a prevalent but poorly understood tropical disease endemic to Southeast Asia ( Bonell et al . , 2017; Kelly et al . , 2015; Richards and Jiang , 2020 ) . O . tsutsugamushi elicits eschar formation in humans , but inbred mice do not recapitulate eschar formation during O . tsutsugamushi infection ( Sunyakumthorn et al . , 2013 ) , similar to R . parkeri . A second example may be Borrelia burgdorferi , a tick-borne pathogen that causes a skin rash at the site of tick bite as a hallmark feature of Lyme disease ( Bratton et al . , 2008 ) , the most prevalent tick-borne disease in the United States ( Bratton et al . , 2008; Schwartz et al . , 2008 ) . Existing mouse models also do not recapitulate skin rash formation following B . burgdorferi infection ( Barthold et al . , 1990; Wang et al . , 2001 ) . Lastly , other rickettsial pathogens including R . conorii , R . typhi , and R . akari cause no or limited disease in WT C57BL/6 mice ( Osterloh , 2017; Anderson and Osterman , 1980; Eisemann et al . , 1984; Osterloh et al . , 2016 ) and it remains unclear if interferons redundantly protect mice from these pathogens . Further investigations into how IFNs protect the skin from arthropod-borne pathogens may reveal critical aspects of the innate immune response to zoonotic disease . Our study further highlights the utility of mouse models that mimic natural routes of infection . Our observation that i . d . infection can cause lethal disease with as few as 10 bacteria , ~ 10 , 000 fewer bacteria than i . v . infection ( Burke et al . , 2020 ) , suggests that R . parkeri may be highly adapted to reside in the skin . However , this model could be further improved by investigating the role for tick vector components in pathogenesis . Saliva from ticks , mosquitos , and sand flies enhances pathogenesis of arthropod-borne bacterial , viral , and parasitic pathogens ( Pingen et al . , 2017; Lestinova et al . , 2017; Šimo et al . , 2017 ) , and non-human primates inoculated with R . parkeri exhibit altered inflammatory responses when administered after tick-bite ( Banajee et al . , 2015 ) . This may suggest a potential role for tick vector components such as tick saliva in R . parkeri pathogenesis . Developing improved murine infection models that mimic the natural route of infection , including with tick saliva or the tick vector , is critical to better understand the virulence and transmission of tick-borne pathogens . Many Rickettsia species , as well as many facultative cytosolic pathogens including L . monocytogenes , undergo actin-based motility to spread from cell to cell . For L . monocytogenes , the actin-based motility factor ActA enables the pathogen to survive in vivo , as actA mutant bacteria are over 1 , 000-fold attenuated by measuring lethality ( Goossens et al . , 1992; Brundage et al . , 1993 ) and by enumerating bacteria in spleens and livers of mice after i . v . infection ( Auerbuch et al . , 2001; Le Monnier et al . , 2007 ) . However , the pathogenic role for actin-based motility in the Rickettsiae has remained unclear . We find that Sca2 is not required for intracellular survival in organs upon i . v . infection of Ifnar1-/-;Ifngr1-/- mice , but rather , is required for dissemination from skin to internal organs and lethality upon i . d . infection . Consistent with an important role for Sca2 in pathogenesis , a previous study reported that i . v . infection of guinea pigs with sca2 mutant R . rickettsii did not elicit fever ( Kleba et al . , 2010 ) . Our results suggest that Sca2-mediated actin-based motility by Rickettsia may facilitate dissemination in host reservoirs , although we cannot rule out other roles for Sca2 that do not involve actin assembly . R . prowazekii and R . typhi , which cause severe human disease , encode a fragmented sca2 gene ( Ngwamidiba et al . , 2005 ) , and undergo no or dramatically reduced frequency of actin-based motility , respectively ( Teysseire et al . , 1992; Heinzen et al . , 1993 ) . Although it remains unclear why some Rickettsia species lost the ability to undergo actin-based motility , Sca2 is dispensable for R . parkeri dissemination in the tick vector ( Harris et al . , 2018 ) , suggesting that actin-based motility may play a specific role in dissemination within mammalian hosts . We find that sca2 or ompB mutant R . parkeri elicit a robust protective immune response in Ifnar1-/-;Ifngr1-/- mice . These findings complement previous observations that sca2 mutant R . rickettsii elicits antibody responses in guinea pigs ( Kleba et al . , 2010 ) , and expands upon these findings by demonstrating protection from rechallenge and revealing additional vaccine candidates . There are currently limited vaccine candidates that protect against rickettsial disease ( Osterloh , 2017 ) . Identifying new vaccine candidates may reveal avenues to protect against tick-borne infections and aerosolized Rickettsia , which are extremely virulent and potential bioterrorism agents ( Walker , 2009 ) , as well as against Brill-Zinsser disease , caused by latent R . prowazekii ( Osterloh , 2017 ) . Future studies exploring whether attenuated R . parkeri mutants provide immunity against other Rickettsia species are warranted to better define the mechanisms of protection . These findings on immunity may also help develop R . parkeri as an antigen delivery platform . R . parkeri resides directly in the host cytosol for days and could potentially be engineered to secrete foreign antigens for presentation by major histocompatibility complex I . In summary , the mouse model described here will facilitate future investigations into numerous aspects of R . parkeri infection , including actin-based motility and immunity , and may serve as model for other arthropod-borne pathogens . R . parkeri strain Portsmouth was originally obtained from Dr . Christopher Paddock ( Centers for Disease Control and Prevention ) . To amplify R . parkeri , confluent monolayers of female African green monkey kidney epithelial Vero cells ( obtained from UC Berkeley Cell Culture Facility , tested for mycoplasma contamination , and authenticated by mass spectrometry ) were infected with 5 × 106 R . parkeri per T175 flask . Vero cells were grown in DMEM ( Gibco 11965–092 ) containing 4 . 5 g l–1 glucose and 2% fetal bovine serum ( FBS; GemCell ) . Infected cells were scraped and collected at five or 6 d . p . i . when ~90% of cells were rounded due to infection . Scraped cells were then centrifuged at 12 , 000 g for 20 min at 4°C . Pelleted cells were resuspended in K-36 buffer ( 0 . 05 M KH2PO4 , 0 . 05 M K2HPO4 , 100 mM KCl , 15 mM NaCl , pH 7 ) and dounced for ~40 strokes at 4°C . The suspension was then centrifuged at 200 g for 5 min at 4°C to pellet host cell debris . Supernatant containing R . parkeri was overlaid on a 30% MD-76R ( Merry X-Ray ) gradient solution in ultracentrifuge tubes ( Beckman/Coulter Cat 344058 ) . Gradients were centrifuged at 18 , 000 r . p . m . in an SW-28 ultracentrifuge swinging bucket rotor ( Beckman/Coulter ) for 20 min at 4°C . These ‘30% prep’ bacterial pellets were resuspended in brain heart infusion ( BHI ) media ( BD , 237500 ) , aliquoted , and stored at −80°C . Titers were determined by plaque assays by serially diluting the R . parkeri in six-well plates containing confluent Vero cells . Plates were spun for 5 min at 300 g in an Eppendorf 5810 R centrifuge and at 24 hr post infection ( h . p . i . ) ; the media from each well was aspirated , and the wells were overlaid with 4 ml/well DMEM with 5% FBS and 0 . 7% agarose ( Invitrogen , 16500–500 ) . At 6 d . p . i . , an overlay of 0 . 7% agarose in DMEM containing 2 . 5 % neutral red ( Sigma , N6264 ) was added . Plaques were then counted 24 hr later . For infections with ompB mutant bacteria , the ompBSTOP::Tn mutant was used , which contains a transposon and an upstream stop codon in ompB , as previously described ( Engström et al . , 2019 ) . For obtaining bone marrow , male or female mice were euthanized , and femurs , tibias , and fibulas were excised . Bones were sterilized with 70% ethanol and washed with BMDM media ( 20% FBS ( HyClone ) , 0 . 1% β-mercaptoethanol , 1% sodium pyruvate , 10% conditioned supernatant from 3T3 fibroblasts , in DMEM ( Gibco ) with 4 . 5 g l–1 glucose and 100 μg/ml streptomycin and 100 U/ml penicillin ) , and ground with a mortar and pestle . Bone homogenate was passed through a 70 μm nylon cell strainer ( Thermo Fisher Scientific , 08-771-2 ) for particulate removal . Filtrates were then centrifuged at 290 g in an Eppendorf 5810 R centrifuge for 8 min , supernatant was aspirated , and the pellet was resuspended in BMDM media . Cells were plated in 30 ml BMDM media in non-TC-treated 15 cm petri dishes at a ratio of 10 dishes per two femurs/tibias and incubated at 37°C . An additional 30 ml of BMDM media was added 3 d later . At 7 d the media was aspirated , 15 ml cold PBS ( Gibco , 10010–023 ) was added , and cells were incubated at 4°C with for 10 min . BMDMs were scraped from the plate , collected in a 50 ml conical tube , and centrifuged at 290 g for 5 min . PBS was aspirated , and cells were resuspended in BMDM media with 30% FBS and 10% DMSO at 107 cells/ml . One milliliter of aliquots was stored at –80°C for 24 hr in Styrofoam boxes and then moved to long-term storage in liquid nitrogen . HMEC-1 cells ( obtained from the UC Berkeley Cell Culture Facility and authenticated by short-tandem-repeat analysis ) were passaged two to three times weekly and grown at 37°C with 5% CO2 in DMEM containing 10 mM L-glutamine ( Sigma , M8537 ) , supplemented with 10% heat-inactivated FBS ( HyClone ) , 1 μg/ml hydrocortisone ( Spectrum Chemical , CO137 ) , 10 ng/ml epidermal growth factor ( Thermo Fisher Scientific , CB40001; Corning cat . no . 354001 ) , and 1 . 18 mg/ml sodium bicarbonate . HMEC media was prepared every 1–2 months and aliquoted and stored at 4°C . To prepare HMEC-1 cells for infection , cells were treated with 0 . 25% trypsin-EDTA ( Thermo Fisher Scientific ) ; the number of cells was counted using a hemocytometer ( Bright-Line ) , and 3 × 104 cells were plated into 24-well plates 48 hr prior to infection . To plate macrophages for infection , BMDM aliquots were thawed on ice , diluted into 9 ml of DMEM , centrifuged at 290 g for 5 min in an Eppendorf 5810 R centrifuge , and the pellet was resuspended in 10 ml BMDM media without antibiotics . 5 × 105 cells were plated into 24-well plates . Approximately 16 hr later , ‘30 % prep’ R . parkeri were thawed on ice and diluted into fresh BMDM media to either 106 p . f . u . /ml or 2 × 105 p . f . u . /ml . Media was then aspirated from the BMDMs and replaced with 500 µl media containing R . parkeri , and plates were spun at 300 g for 5 min in an Eppendorf 5810 R centrifuge . Infected cells were incubated in a humidified CEDCO 1600 incubator set to 33°C and 5% CO2 . Recombinant mouse IFN-β ( PBL , 12405–1 ) was added directly to infected cells after infection . For measuring p . f . u . , supernatants from infected BMDMs were aspirated , and each well was washed twice with 500 µl sterile milli-Q-grade water . After adding 1 ml of sterile milli-Q water to each well , macrophages were lysed by repeated pipetting . Serial dilutions of lysates were added to confluent Vero cells in 12-well plates . Plates were spun at 300 g using an Eppendorf 5810 R centrifuge for 5 min at room temperature and incubated at 33°C overnight . At ~16 h . p . i . , media was aspirated and replaced with 2 ml/well of DMEM containing 0 . 7% agarose and 5% FBS ( GemCell ) . At ~6 d . p . i . , 1 ml of DMEM containing 0 . 7% agarose , 1% FBS ( GemCell ) , 200 µg/ml amphotericin B ( Invitrogen , 15290–018 ) , and 2 . 5 % neutral red ( Sigma ) was added to each well . Plaques were then counted after 24 hr . Microscopy , LDH , and IFN-I experiments were performed as described ( Burke et al . , 2020 ) . Animal research was conducted under a protocol approved by the University of California , Berkeley Institutional Animal Care and Use Committee ( IACUC ) in compliance with the Animal Welfare Act and other federal statutes relating to animals and experiments using animals ( Welch lab animal use protocol AUP-2016-02-8426 ) . The University of California , Berkeley IACUC is fully accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care International and adheres to the principles of the Guide for the Care and use of Laboratory Animals ( National Research Council , 2011 ) . Mouse infections were performed in a biosafety level two facility . All animals were maintained at the University of California , Berkeley campus , and all infections were performed in accordance with the approved protocols . Mice were between 8 and 20 weeks old at the time of initial infection . Mice were selected for experiments based on their availability , regardless of sex . The sex of mice used for survival after i . d . infection and raw data for mouse experiments is provided in the Source Data for each figure . A statistical analysis was not performed to predetermine sample size prior to initial experiments . Initial sample sizes were based on availability of mice and the capacity to process or measure samples within a given time . After the first experiment , a Power Analysis was conducted to determine subsequent group sizes . All mice were of the C57BL/6J background , except for outbred CD-1 mice . All mice were healthy at the time of infection and were housed in microisolator cages and provided chow , water , and bedding . No mice were administered antibiotics or maintained on water with antibiotics . Experimental groups were littermates of the same sex that were randomly assigned to experimental groups . For experiments with DKO mice deficient in Ifnar1 and Ifngr1 , mice were immediately euthanized if they exhibited severe degree of infection , as defined by a core body temperature dropping below 90°F or lethargy that prevented normal movement . Single mutant Tlr4-/- ( Hoshino et al . , 1999 , Ifnar1-/- Müller et al . , 1994 , Ifngr1-/- Huang et al . , 1993 ) , DKO Ifnar1-/-;Ifngr1-/- , and WT C57BL/6J mice were previously described and originally obtained from Jackson Laboratories . CD-1 mice were obtained from Charles River . For genotyping , ear clips were boiled for 15 min in 60 µl of 25 mM NaOH , quenched with 10 µl Tris–HCl pH 5 . 5 , and 2 µl of lysate was used for PCR using SapphireAMP ( Takara , RR350 ) and gene-specific primers . Primers used were: Ifnar1 forward ( F ) : CAACATACTACAACGACCAAGTGTG; Ifnar1 WT reverse ( R ) : AACAAACCCCCAAACCCCAG; Ifnar1-/- R: ATCTGGACGAAGAGCATCAGG; Ifngr1 ( F ) : CTCGTGCTTTACGGTATCGC; Ifngr1 ( R ) : TCGCTTTCCAGCTGATGTACT; WT Tlr4 ( F ) : CACCTGATACTTAATGCTGGCTGTAAAAAG; WT Tlr4 ( R ) : GGTTTAGGCCCCAGAGTTTTGTTCTTCTCA; Tlr4-/- ( F ) : TGTTGCCCTTCAGTCACAGAGACTCTG; and Tlr4-/- ( R ) : TGTTGGGTCGTTTGTTCGGATCCGTCG . For mouse infections , R . parkeri was prepared by diluting 30%-prep bacteria into cold sterile PBS on ice . Bacterial suspensions were kept on ice during injections . For i . d . infections , mice were anaesthetized with 2 . 5% isoflurane via inhalation . The right flank of each mouse was shaved with a hair trimmer ( Braintree CLP-41590 ) , wiped with 70% ethanol , and 50 µl of bacterial suspension in PBS was injected intradermally using a 30 . 5-gauge needle . Mice were monitored for ~3 min until they were fully awake . No adverse effects were recorded from anesthesia . For i . v . infections , mice were exposed to a heat lamp while in their cages for approximately 5 min and then each mouse was moved to a mouse restrainer ( Braintree , TB-150 STD ) . The tail was sterilized with 70% ethanol , and 200 µl of bacterial suspension in sterile PBS was injected using 30 . 5-gauge needles into the lateral tail vein . Body temperatures were monitored using a rodent rectal thermometer ( BrainTree Scientific , RET-3 ) . For fluorescent dextran experiments , mice were intravenously injected with 150 µl of 10 kDa dextran conjugated with Alexa Fluor 680 ( D34680; Thermo Fisher Scientific ) at a concentration of 1 mg/ml in sterile PBS ( Glasner et al . , 2017 ) . As a negative control , mice with no R . parkeri infection were injected with fluorescent dextran . As an additional negative control , uninfected mice were injected intravenously with PBS instead of fluorescent dextran . At 2 hr post-injection , mice were euthanized with CO2 and cervical dislocation , doused with 70% ethanol , and skin surrounding the injection site ( approximately 2 cm in each direction ) was removed . Connective tissue between the skin and peritoneum was removed , and skin was placed hair-side-up on a 15 cm Petri dish . Skin was imaged with an LI-COR Odyssey CLx ( LI-COR Biosciences ) , and fluorescence was quantified using ImageStudioLite v5 . 2 . 5 . The skin from mice with no injected fluorescent dextran was used as the background measurement . Skin from mice injected with fluorescent dextran but no R . parkeri was normalized to an arbitrary number ( 100 ) , and R . parkeri-infected samples were normalized to this value ( R . parkeri-infected/uninfected × 100 ) . The number of pixels at the injection site area was maintained across experiments ( 7 , 800 for small area and 80 , 000 for the large area ) . All mice in this study were monitored daily for clinical signs of disease throughout the course of infection , such as hunched posture , lethargy , scruffed fur , paralysis , facial edema , and lesions on the skin of the flank and tail . If any such manifestations were observed , mice were monitored for changes in body weight and temperature . If a mouse displayed severe signs of infection , as defined by a reduction in body temperature below 90°F or an inability to move normally , the animal was immediately and humanely euthanized using CO2 followed by cervical dislocation , according to IACUC-approved procedures . Pictures of skin and tail lesions were obtained with permission from the Animal Care and Use Committee Chair and the Office of Laboratory and Animal Care . Pictures were captured with an Apple iPhone 8 , software v13 . 3 . 1 . For harvesting spleens and livers , mice were euthanized at the indicated pre-determined times and doused with ethanol . Mouse organs were extracted and deposited into 50 ml conical tubes containing 4 ml sterile cold PBS for the spleen and 8 ml PBS for the liver . Organs were kept on ice and were homogenized for ~10 s using an immersion homogenizer ( Fisher , Polytron PT 2500E ) at ~22 , 000 r . p . m . Organ homogenates were spun at 290 g for 5 min to pellet the cell debris ( Eppendorf 5810 R centrifuge ) . Twenty microliters of organ homogenates was then serial diluted into 12-well plates containing confluent Vero cells . The plates were then spun at 260 g for 5 min at room temperature ( Eppendorf 5810 R centrifuge ) and incubated at 33°C . To reduce the possibility of contamination , organ homogenates were plated in duplicate and the second replicate was treated with 50 µg/ml carbenicillin ( Sigma ) and 200 µg/ml amphotericin B ( Gibco ) . The next day , at approximately 16 h . p . i . , the cells were gently washed by replacing the existing media with 1 ml DMEM containing 2% FBS ( GemCell ) . The media were then aspirated and replaced with 2 ml/well of DMEM containing 0 . 7% agarose , 5% FBS , and 200 µg/ml amphotericin B . When plaques were visible at 6 d . p . i . , 1 ml of DMEM containing 0 . 7% agarose , 1% FBS , and 2 . 5% neutral red ( Sigma ) was added to each well , and plaques were counted at 24 h . p . i . For histology , spleens and livers were harvested from infected mice and immediately stored in 10% neutral buffered formalin ( Sigma HT501128 ) . Histology was performed by HistoWiz Inc ( http://histowiz . com ) using a standard operating procedure and fully automated workflow . Samples were processed , embedded in paraffin , and sectioned at 4 μm thickness . BOND Polymer Refine Detection ( Leica Biosystems ) was used according to manufacturer’s protocol . After staining with hematoxylin and eosin , sections were dehydrated and film coverslipped using a TissueTek-Prisma and Coverslipper ( Sakura ) . Whole slide scanning ( 40× ) was performed on an Aperio AT2 ( Leica Biosystems ) . Histological consultation was blindly performed by a pathologist . Immunohistochemistry was performed with a rabbit anti-Rickettsia I7205 antibody ( 1:500 dilution; gift from Ted Hackstadt ) . Shareable links for all histology and immunohistochemistry data are available upon request to the authors . Statistical parameters and significance are reported in the figure legends . For comparing two sets of data , a two-tailed Student’s T test was performed . For comparing two sets of in vivo p . f . u . data , Mann–Whitney U tests were used . For comparing two survival curves , log-rank ( Mantel–Cox ) tests were used . For comparing curves of two samples ( mouse health , weight , and temperature ) , two-way ANOVAs were used . For two-way ANOVAs , if a mouse was euthanized prior to the statistical end point , the final value that was recorded for the mouse was repeated until the statistical endpoint . For two-way ANOVAs , if a measurement was not recorded for a timepoint , the difference between values at adjacent time points was used . Data were determined to be statistically significant when p<0 . 05 . Asterisks denote statistical significance as: p<0 . 05; p<0 . 01; p<0 . 001; **p<0 . 0001 , compared to indicated controls . Error bars indicate standard deviation ( SD ) for in vitro experiments and standard error of the mean ( SEM ) for in vivo experiments . All other graphical representations are described in the figure legends . Statistical analyses were performed using GraphPad PRISM v7 . 0 .	Tick bites allow disease-causing microbes , including multiple species of Rickettsia bacteria , to pass from arthropods to humans . Being exposed to Rickettsia parkeri , for example , can cause a scab at the bite site , fever , headache and fatigue . To date , no vaccine is available against any of the severe diseases caused by Rickettsia species . Modelling human infections in animals could help to understand and combat these illnesses . R . parkeri is a good candidate for such studies , as it can give insight into more severe Rickettsia infections while being comparatively safer to handle . However , laboratory mice are resistant to this species of bacteria , limiting their use as models . To explore why this is the case , Burke et al . probed whether an immune mechanism known as interferon signalling protects laboratory rodents against R . parkeri . During infection , the immune system releases molecules called interferons that stick to ‘receptors’ at the surface of cells , triggering defense mechanisms that help to fight off an invader . Burke et al . injected R . parkeri into the skin of mice that had or lacked certain interferon receptors , showing that animals without two specific receptors developed scabs and saw the disease spread through their body . Further investigation showed that two R . parkeri proteins , known as OmpB or Sca2 , were essential for the bacteria to cause skin lesions and damage internal organs . Burke et al . then used R . parkeri that lacked OmpB or Sca2 to test whether these modified , inoffensive microbes could act as ‘vaccines’ . And indeed , vulnerable laboratory mice which were first exposed to the mutant bacteria were then able to survive the ‘normal’ version of the microbe . Together , this work reveals that interferon signalling protects laboratory mice against R . parkeri infections . It also creates an animal model that can be used to study disease and vaccination .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2021	Interferon receptor-deficient mice are susceptible to eschar-associated rickettsiosis
Immune cells infiltrating tumors can have important impact on tumor progression and response to therapy . We present an efficient algorithm to simultaneously estimate the fraction of cancer and immune cell types from bulk tumor gene expression data . Our method integrates novel gene expression profiles from each major non-malignant cell type found in tumors , renormalization based on cell-type-specific mRNA content , and the ability to consider uncharacterized and possibly highly variable cell types . Feasibility is demonstrated by validation with flow cytometry , immunohistochemistry and single-cell RNA-Seq analyses of human melanoma and colorectal tumor specimens . Altogether , our work not only improves accuracy but also broadens the scope of absolute cell fraction predictions from tumor gene expression data , and provides a unique novel experimental benchmark for immunogenomics analyses in cancer research ( http://epic . gfellerlab . org ) . Tumors form complex microenvironments composed of various cell types such as cancer , immune , stromal and endothelial cells ( Hanahan and Weinberg , 2011; Joyce and Fearon , 2015 ) . Immune cells infiltrating the tumor microenvironment play a major role in shaping tumor progression , response to ( immuno- ) therapy and patient survival ( Fridman et al . , 2012 ) . Today , gene expression analysis is widely used to characterize tumors at the molecular level . As a consequence , tumor gene expression profiles from tens of thousands of patients are available across all major tumor types in databases such as Gene Expression Omnibus ( GEO [Edgar et al . , 2002] ) or The Cancer Genome Atlas ( TCGA [Hoadley et al . , 2014] ) . Unfortunately , flow cytometry or immunohistochemistry ( IHC ) measurements to quantify the number of both malignant and tumor-infiltrating immune cells are rarely performed for samples analyzed at the gene expression level . Therefore , to correctly interpret these data in particular from an immuno-oncology point of view ( Angelova et al . , 2015; Gentles et al . , 2015; Hackl et al . , 2016; Li et al . , 2016; Linsley et al . , 2015; Rooney et al . , 2015; Şenbabaoğlu et al . , 2016; Zheng et al . , 2017 ) , reliable and carefully validated bioinformatics tools are required to infer the fraction of cancer and immune cell types from bulk tumor gene expression data . To this end , diverse bioinformatics methods have been developed . Some aim at estimating tumor purity based on copy number variation ( Carter et al . , 2012; Li and Li , 2014 ) , or expression data ( Ahn et al . , 2013; Clarke et al . , 2010; Quon et al . , 2013; Yoshihara et al . , 2013 ) , but do not provide information about the different immune cell types . Others focus on predicting the relative proportions of cell types by fitting reference gene expression profiles from sorted cells ( Gong and Szustakowski , 2013; Li et al . , 2016; Newman et al . , 2015; Qiao et al . , 2012 ) or with the help of gene signatures ( Becht et al . , 2016; Zhong et al . , 2013 ) . These approaches have been recently applied to cancer genomics data to investigate the influence of immune infiltrates on survival or response to therapy ( Charoentong et al . , 2017; Gentles et al . , 2015; Şenbabaoğlu et al . , 2016 ) or predict potential targets for cancer immunotherapy ( Angelova et al . , 2015; Li et al . , 2016 ) . However , none of these methods provides quantitative information about both cancer and non-malignant cell type proportions directly from tumor gene expression profiles . In addition , reference gene expression profiles used in previous studies have been mainly obtained from circulating immune cells sorted from peripheral blood and were generally based on microarrays technology . Finally , several of these approaches have not been experimentally validated in solid tumors from human patients . Here , we developed a robust approach to simultaneously Estimate the Proportion of Immune and Cancer cells ( EPIC ) from bulk tumor gene expression data . EPIC is based on a unique collection of RNA-Seq reference gene expression profiles from either circulating immune cells or tumor- infiltrating non-malignant cell types ( i . e . , immune , stromal and endothelial cells ) . To account for the high variability of cancer cells across patients and tissue of origin , we implemented in our algorithm the ability to consider uncharacterized , possibly highly variable , cell types . To validate our predictions in human solid tumors , we first analyzed melanoma samples with both flow cytometry and RNA-Seq . We then collected publicly available IHC and single-cell RNA-Seq data of colorectal and melanoma tumors . All three validation datasets showed that very accurate predictions of both cancer and non-malignant cell type proportions could be obtained even in the absence of a priori information about cancer cells . EPIC incorporates reference gene expression profiles from each major immune and other non-malignant cell type to model bulk RNA-Seq data as a superposition of these reference profiles ( Figure 1A , B ) . To tailor our predictions to recent gene expression studies , we first collected and curated RNA-Seq profiles of various human innate and adaptive circulating immune cell types ( Hoek et al . , 2015; Linsley et al . , 2014; Pabst et al . , 2016 ) ( CD4 T cells , CD8 T cells , B cells , NK cells , Monocytes and Neutrophils ) from a diverse set of patients analyzed in different centers ( see Materials and methods ) . Principal component analysis ( PCA ) of these data ( Figure 1C ) showed that samples clustered first according to cell type and not according to experiment of origin , patient age , disease status or other factors , suggesting that they could be used as bona fide reference expression profiles across different patients . Reference gene expression profiles for each major immune cell type were built from these RNA-Seq samples based on the median normalized counts per gene and cell type . The variability in expression for each gene was also considered when predicting the various cell proportions based on these reference profiles ( see Materials and methods and Supplementary file 1 ) . Immune cells differ in their gene expression profiles depending on their state and site of origin ( e . g . , blood or tumors ) ( Ganesan et al . , 2017; Speiser et al . , 2016; Zheng et al . , 2017 ) . To study the potential effect of these differences on our predictions , we established reference gene expression profiles of each major tumor-infiltrating immune cell type ( i . e . , CD4 T , CD8 T , B , NK , macrophages ) . We further derived reference profiles for stromal cells ( i . e . cancer-associated fibroblasts ( CAFs ) ) and endothelial cells . These reference gene expression profiles were obtained as cell type averages from the single-cell RNA-Seq data of melanoma patients from Tirosh and colleagues ( Tirosh et al . , 2016 ) , considering only samples from primary tumor and non-lymphoid tissue metastasis ( see Materials and methods and Supplementary file 2 ) . As for circulating immune cell data , principal component analysis of the tumor-infiltrating cells’ gene expression profiles showed that samples clustered first according to cell type ( Figure 1D and Figure 1—figure supplement 1 , see also results in [Tirosh et al . , 2016] ) . Reference gene expression profiles from each of the immune and other non-malignant ( i . e . , stromal and endothelial ) cell types were then used to model bulk gene expression data as a linear combination of m different cell types ( Figure 1B ) . To include cell types like cancer cells that show high variability across patients and tissues of origin , we further implemented in our algorithm the ability to consider an uncharacterized cell population . Mathematically this was done by taking advantage of the presence of gene markers of non-malignant cells that are not expressed in cancer cells . Importantly , we do not require our signature genes to be expressed in exactly one cell type , but only to show very low expression in cancer cells . The mRNA proportion of each immune and other non-malignant cell type was inferred using least-square regression , solving first our system of equations for the marker genes ( green box in Figure 1B , see Materials and methods ) . The fraction of cancer cells was then determined as one minus the fraction of all non-malignant cell types . Cell markers used in this work were determined by differential expression analysis based on our reference cell gene expression profiles as well as gene expression data from non-hematopoietic tissues ( see Materials and methods and Appendix 1—table 1 ) . Finally , to account for different amounts of mRNA in different cell types and enable meaningful comparison with flow cytometry and IHC data , we measured the mRNA content of all major immune cell types as well as of cancer cells ( Figure 1—figure supplement 2 ) and used these values to renormalize our predicted mRNA proportions ( see Materials and methods ) . We first tested our algorithm using three datasets comprising bulk RNA-Seq data from PBMC ( Hoek et al . , 2015; Zimmermann et al . , 2016 ) or whole blood ( Linsley et al . , 2014 ) , as well as the corresponding proportions of immune cell types determined by flow cytometry ( Figure 2A ) . These data were collected from various cancer-free human donors ( see Materials and methods ) . Overall , very accurate predictions were obtained by fitting reference profiles from circulating immune cells , considering either all cell types together ( Figure 2A ) or each cell type separately ( Figure 2—figure supplement 1 ) . When comparing with other widely used cell fraction prediction methods ( Becht et al . , 2016; Gong and Szustakowski , 2013; Li et al . , 2016; Newman et al . , 2015; Quon et al . , 2013; Zhong et al . , 2013 ) , we observed a clear improvement ( Figure 2B and Figure 2—figure supplement 1 ) . Of note , the very high correlation values can partly result from the broad range of different cell fractions in our data ( Figure 2A ) and we emphasize that these correlation values should only be used to compare methods tested on the same datasets ( Figure 2B ) . The root mean squared error ( RMSE ) , which is less dependent on such ‘outlier data points’ , shows also improved accuracy for EPIC compared to other methods ( Figure 2—figure supplement 1B ) . The renormalization by mRNA content , which had not been considered in previous approaches , appeared to be important for predicting actual cell fractions ( Figure 2C ) . Moreover , we observed that most other methods could also benefit from such a renormalization step , be it for the global prediction of all cell types together or for the predictions of each cell type ( Figure 2—figure supplement 2 ) . A conceptually related approach was developed by Baron et al . ( 2016 ) , but the rescaling was done on the gene expression reference profiles ( in their case of pancreatic cells ) based on the total number of transcripts per cell type and the prediction of the proportion of pancreatic cells from a bulk sample was carried out with CIBERSORT ( Newman et al . , 2015 ) . For comparison purpose , we implemented such an a priori renormalization step in EPIC as well , and observed similar results than with the a posteriori renormalization ( Figure 2—figure supplement 3 ) . With respect to the other methods , EPIC has two other main distinctive features: ( i ) it allows for a cell type without a known reference gene expression profile ( such a feature is also part of ISOpure [Quon et al . , 2013] ) and ( ii ) it integrates information about the variability in each signature gene from the reference gene expression profiles . The latter point slightly improves the prediction accuracy but to a lesser extent than the renormalization by mRNA ( Figure 2—figure supplement 3 ) . The former point cannot be tested directly in the blood samples as only cell types with known reference gene expression profiles are composing these samples . Therefore , to test the effect of including a cell type without a known reference profile , we removed all the T cell subsets from the reference gene expression profiles and predicted the proportion of the other cell types in the bulk samples , allowing for one uncharacterized cell type in EPIC . As expected , the results from EPIC including or not the T cell reference profiles remained nearly unchanged , while the other methods suffered from this , except for DSA ( Zhong et al . , 2013 ) which is based only on signature genes and not reference profiles ( Figure 2—figure supplement 4 ) . Such an advantage of EPIC is especially useful in the context of tumor samples where in general no reference gene expression profile is available for the cancer cells . To validate our predictions in tumors , we collected single cell suspensions from lymph nodes of four metastatic melanoma patients ( see Materials and methods ) . A fraction of the cell suspension was used to measure the different cell type proportions with flow cytometry ( CD4 T , CD8 T , B , NK , melanoma and other cells comprising mostly stromal and endothelial cells; Supplementary file 3A ) , and the other fraction was used for bulk RNA sequencing ( Figure 3—figure supplement 1 ) . EPIC was first run with reference profiles from circulating immune cells . We observed a remarkable agreement between our predictions and experimentally determined cell fractions ( Figure 3A ) . The high correlation value is possibly driven by the two samples containing about 80% of melanoma cells , but we stress that all predicted cell proportions fall nearly on the ‘y = x’ line . This clearly indicates that the absolute cell fractions were correctly predicted for all cell types , as confirmed by a low RMSE . Of note , the proportion of melanoma cells could be very accurately predicted even in the absence of a priori information about their gene expression . As a second validation , we compared EPIC predictions with IHC data from colon cancer ( Becht et al . , 2016 ) ( see Materials and methods ) . Although a limited number of immune cell types had been assayed in this study , we observed a significant correlation between cell proportions measured by IHC and our predictions , except for the CD8 T cells ( Figure 3B ) . As a third validation , we used recent single-cell RNA-Seq data from 19 melanoma samples ( Tirosh et al . , 2016 ) . We applied EPIC on the average expression profile over all single cells for each patient and compared the results with the actual cell fractions ( see Materials and methods ) . Here again , our predictions were consistent with the observed cell fractions , even for melanoma cells for which we did not assume any reference gene expression profile ( Figure 3C ) . Notably , the predicted cell fractions from melanoma cells as well as from all other immune cell types fall nearly on the y = x line , showing that not only the relative cell type proportions could be predicted but also the absolute proportions for all cell types . We next compared these predictions to those obtained with reference profiles from tumor- infiltrating cells , including also CAFs and endothelial cells ( Figure 4 ) . For the single-cell RNA-Seq data ( Figure 4C ) , we applied a leave-one-out procedure , to avoid using the same samples both to build the reference profiles and the bulk RNA-Seq data used as input for the predictions ( see Materials and methods ) . Overall , predictions did not change much compared to those based on circulating immune cell reference gene expression profiles ( Figure 4 ) , except for the IHC data where the predictions for CD8 T cells and macrophages clearly improved ( Figure 4B ) . Moreover , we could observe some differences between the results obtained from circulating immune cell reference gene expression profiles and those from tumor-infiltrating cell reference gene expression profiles , when considering the proportions from each cell type independently ( Figures 3–4 and Figure 4—figure supplement 1 ) : ( i ) predictions for CD8 T cells and macrophages improved in the datasets of primary tumors and non-lymph node metastases but not in the datasets from lymph node metastases; ( ii ) predictions for B and NK cells displayed similar accuracy based on the circulating cells or tumor-infiltrating cells profiles for all datasets; and ( iii ) CAFs and endothelial cells were not available for the blood-based reference profiles but the proportion of these cells could be well predicted based on the reference profiles constructed from tumor-infiltrating cells ( Figure 4 and Figure 4—figure supplement 1 ) . We took advantage of our unique collection of independent validation datasets to benchmark other methods for predictions of immune cell type fractions in human tumors . We first compared the results of EPIC and ISOpure ( Quon et al . , 2013 ) , which is the only other method that can consider uncharacterized cell types and therefore predict the fraction of cancer and immune cell types based only on RNA-seq data . EPIC displayed improved accuracy in all three datasets ( Figure 5A , and Figure 5—figure supplements 1–4 ) . To benchmark other methods , we then restricted our analysis to the predictions of the different immune cell types ( Figure 5B and Figure 5—figure supplements 1–4 ) . Predictions from EPIC were in general more accurate , especially when considering all cell types together . Nevertheless , when restricting the comparisons to relative cell type proportions , some methods like MCPcounter ( Becht et al . , 2016 ) and TIMER ( Li et al . , 2016 ) were quite consistent in their predictions across the various datasets and showed similar accuracy as EPIC for the cell types considered in these methods ( Figure 5—figure supplements 1–4 ) . Of note , MCPcounter could not be included in the global prediction comparison as this method returns scores that are not comparable between different cell types . Predictions from DSA ( Zhong et al . , 2013 ) were also quite accurate when available , but in multiple cases some cell type proportions returned by the method were simply equal to 0 in all samples ( Figure 5—figure supplements 1–4 – correlation values were replaced by NA in these cases ) . Immune , stromal and tumor purity scores ( Yoshihara et al . , 2013 ) based on gene set enrichment analysis were also correlated with the total fraction of immune cells , stromal cells and the fraction of cancer cells ( correlation was not significant for the tumor purity score – Figure 5C and Figure 5—figure supplement 5 ) . However , these correlations were considerably lower than those obtained with our approach ( Figure 5D and Figure 5—figure supplement 5 ) . Moreover , such scores are less quantitative and are thus more difficult to interpret with respect to actual cell type proportions . Next , we performed some explorative analysis to test if such cell fraction prediction methods could go further into the details of the T cell subsets . Based on the single-cell RNA-seq data from Tirosh and colleagues ( Tirosh et al . , 2016 ) , we identified CD4+ Treg and Thelper cells ( see Materials and methods ) , and built reference gene expression profiles for these cell types as we had done for the other cell types . As before , a leave-1-out procedure at the patient level was used to predict the proportions for these cell types based on the bulk samples constructed from this single-cell RNA-seq data . We observed significant correlations , but the R-values are lower than before ( Figure 5—figure supplement 6 ) , suggesting that we may be reaching the limits of cell fraction prediction methods for cell types that show lower abundance and substantial transcriptional similarity . Of note , CIBERSORT using either the LM22 signature or our newly derived signature did not perform better than EPIC . By combining RNA-Seq profiles of all major immune and other non-malignant cell types established from both circulating and tumor-infiltrating cells together with information about cell morphology ( i . e . , mRNA content ) and algorithmic developments to consider uncharacterized and possibly highly variable cell types , EPIC overcomes several limitations of previous approaches to predict the fraction of both cancer and immune or other non-malignant cell types from bulk tumor gene expression data . From an algorithmic point of view , EPIC takes advantage of the fact that cancer cells , in general , express no or only low levels of immune and stromal markers . Therefore the method can be broadly applied to most solid tumors , as confirmed by our validation in melanoma and colorectal samples , but it will not be suitable for hematological malignancies like leukemia or lymphoma . The accuracy of the predictions for some cell types might be sensitive to the origin or condition of the immune cells used for establishing reference profiles . For instance , we observed that CD8 T cells and macrophages from primary tumors and non-lymph node metastases samples were best predicted using the reference profiles from tumor-infiltrating cells . This may be explained by the fact that the reference profiles from circulating cells corresponded to resting CD8 T cells and monocytes as only few activated C8 T cells and no macrophages are circulating in blood . Overall , our results suggest that for primary tumors or non-lymphoid tissue metastases reference gene expression profiles from tumor-infiltrating immune cells are more appropriate , while for lymph node metastases , profiles from either circulating or tumor-infiltrating immune cells could be used . One known limitation of cell fraction predictions arises when some cell types are present at very low frequency ( Shen-Orr and Gaujoux , 2013 ) . Our results suggest that predictions of cell proportions are reliable within an absolute error of about 8% , as estimated by the root mean squared error ( Figure 3 and Figure 4—figure supplement 1B ) . These estimates are consistent with the lower detection limit proposed by other groups ( Becht et al . , 2016; Zhong et al . , 2013 ) and may explain why the relative proportions of NK cells , which are present at lower frequency in melanoma tumors ( Balch et al . , 1990; Sconocchia et al . , 2012 ) , could not be predicted with accuracy comparable to other cell types ( Figure 4—figure supplement 1 ) . While this may prevent applications of cell fraction predictions in some tumor types that are poorly infiltrated , many other tumors , like melanoma or colorectal cancer , display high level of infiltrating immune cells and the role of immune infiltrations on tumor progression and survival appears to be especially important in these tumors ( Clemente et al . , 1996; Fridman et al . , 2012; Galon et al . , 2006 ) . Another limitation of cell fraction predictions arises when considering cell types that show high transcriptional similarity . For instance , Treg in Figure 5—figure supplement 6 could not be well predicted neither by EPIC nor CIBERSORT . This can be understood because the gene expression profiles of Thelper and Treg are highly similar with only a few genes expressed differently between the two , making it harder for cell fraction prediction methods to work accurately . In addition , Treg were present at a proportion lower than 10% in all samples ( Supplementary file 3E ) . For such cases , gene set enrichment approaches , although less quantitative , may be more convenient , possibly combining them with the predictions obtained by EPIC for the main immune cell types . Our predictions for the fraction of uncharacterized cells may include non-malignant cells , such as epithelial cells from neighboring tissues in addition to cancer cells . Compared to recent algorithms that first predict tumor purity based on exome sequencing data , and later infer the relative fraction of immune cell types ( Li et al . , 2016 ) , the predictions of EPIC for immune and stromal cells are likely more quantitative because they can implicitly consider the presence of not only cancer cells but also other non-malignant cells for which reference profiles are not available . Moreover , EPIC does not require both exome and RNA-Seq data from the same tumor samples , thereby reducing the cost and amount of experimental work for prospective studies , and broadening the scope of retrospective analyses of cancer genomics data to studies that only include gene expression data . Recent technical developments in single-cell RNA-Seq technology enable researchers to directly access information about both the proportion of all cell types together with their gene expression characteristics ( Carmona et al . , 2017; Efroni et al . , 2015; Jaitin et al . , 2014; Singer et al . , 2016; Stegle et al . , 2015; Tirosh et al . , 2016 ) . Such data are much richer than anything that can be obtained with computational deconvolution of bulk gene expression profiles and this technology may eventually replace standard gene expression analysis of bulk tumors . Nevertheless , it is important to realize that , even when disregarding the financial aspects , single-cell RNA-Seq of human tumors is still logistically and technically very challenging due to high level of cell death upon sample manipulation ( especially freezing and thawing ) and high transcript dropout rates ( Finak et al . , 2015; Saliba et al . , 2014; Stegle et al . , 2015 ) . Moreover , one cannot exclude that some cells may better survive the processing with microfluidics devices used in some single-cell RNA-Seq platforms , thereby biasing the estimates of cell type proportions . It is therefore likely that bulk tumor gene expression analysis will remain widely used for several years . Our work shows how we can exploit recent single-cell RNA-Seq data of tumors obtained from a few patients to refine cell fraction predictions in other patients that could not be analyzed with this technology , thereby overcoming some limitations of previous computational approaches that relied only on reference gene expression profiles from circulating immune cells . In this work , we provide a detailed and biologically relevant validation of our predictions using actual tumor samples from human patients analyzed with flow cytometry , IHC and single-cell RNA-Seq . We note that the slightly lower agreement between our predictions and IHC data may be partly explained by the fact that the exact same samples could not be used for both gene expression and IHC analyses because of the incompatibility between the two techniques . Nevertheless , the overall high accuracy of our predictions indicates that infiltrations of major immune cell types can be quantitatively studied directly from bulk tumor gene expression data using computational approaches such as EPIC . EPIC can be downloaded as a standalone R package ( https://github . com/GfellerLab/EPIC/releases/tag/v1 . 1 ) ( Racle , 2017 ) and can be used with reference gene expression profiles pre-compiled from circulating or tumor-infiltrating cells , or provided by the user . EPIC is also available as a web application ( http://epic . gfellerlab . org ) where users can upload bulk samples gene expression data and perform the full analysis . EPIC has been written as an R package . It is freely available on GitHub ( https://github . com/GfellerLab/EPIC; copy archived on https://github . com/elifesciences-publications/EPIC ) for academic non-commercial research purposes . Version v1 . 1 of the package was used for our analyses . In addition to the R package , EPIC is available as a web application at the address: http://epic . gfellerlab . org . In EPIC , the gene expression of a bulk sample is modeled as the sum of the gene expression profiles from the pure cell types composing this sample ( Figure 1A , B ) . This can be written as ( Venet et al . , 2001 ) : ( 1 ) b=C × p Where b is the vector of all n genes expressed from the bulk sample to deconvolve; C is a matrix ( n x m ) of the m gene expression profiles from the different cell types; and p is a vector of the proportions from the m cell types in the given sample ( Figure 1B ) . Matrix C consists of m-1 columns corresponding to various reference non-malignant cell types whose gene expression profiles are known , and a last column corresponding to uncharacterized cells ( i . e . mostly cancer cells , but possibly also other non-malignant cell types not included in the reference profiles ) . EPIC assumes the reference gene expression profiles from the non-malignant cell types are well conserved between patients . Such a hypothesis is supported by the analysis in Figure 1C , D . The uncharacterized cells can be more heterogeneous between patients and EPIC makes no assumption on them . EPIC works with RNA-seq data , which is implicitly normalized . We use data normalized into transcripts per million ( TPM ) because it has some properties needed for the full cell proportion prediction to work , as will be shown in the next paragraphs . Therefore , instead of the raw data from Equation ( 1 ) , we are working with TPM-normalized data , which correspond to the following: ( 2 ) bi¯=106∑k=1nbklk⋅biliCij¯=106∑k=1nCkjlk⋅Cijli Where b¯ and C¯ are the TPM-normalized bulk sample and reference cell gene expression profiles respectively and li is the length of gene i . Using these , Equation ( 1 ) is rewritten to: ( 3 ) b¯=C¯×p¯ where ( 4 ) pj¯=∑k=1nCkjlk∑i=1nbili⋅pj This normalization guarantees the sum of the new proportions , p- , is equal to 1:∑i=1nb¯i=from eq . 2 ∑i ( 106∑kbklk⋅bili ) =106 and∑i=1nb¯i=from eq . 3 ∑i ( C¯×p¯ ) i=∑i=1n∑j=1mCij¯⋅p¯j=from eq . 2∑j[ ( 106∑kCkjlk⋅∑iCijli ) ⋅p¯j]=106⋅∑jp¯j ( 5 ) ⇒∑j=1mpj¯=1 In addition to p¯ and C¯ we also define p¯∗ and C¯∗ , which are the same except that they include the normalized proportions and profiles from the reference cell types only ( i . e . they have one less element and one less column than p¯ and C¯ respectively ) . Using these normalized quantities , EPIC then solves Equation ( 3 ) for a subset of ns equations corresponding to the ns signature genes ( S ) that are expressed by one or more of the reference cell types but only expressed at a negligible level in the uncharacterized cells ( Figure 1B ) . Previous computational work ( Clarke et al . , 2010; Gosink et al . , 2007 ) showed that the proportion from uncharacterized cells in bulk samples could indeed be inferred with help of genes not expressed by the uncharacterized cells . Importantly , cell-specific signature genes are well established and widely used in flow cytometry to sort immune cells . Thus , EPIC solves the following system of equations: ( 6 ) bi¯\|i∈S= ( ( C¯∗×p¯∗ ) i+C¯im⋅p¯m ) \|i∈S= ( C¯∗×p¯∗ ) i\|i∈S where the term corresponding to the uncharacterized cells ( m ) vanished thanks to the definition of the signature genes ( C¯im\|i∈S≅0 ) . The solution to Equation ( 6 ) can be estimated by a constrained least square optimization . EPIC takes advantage of the known variability for each gene in the reference profile , to give weights in the function to minimize: ( 7 ) fp¯=∑i∈Swib¯i-C¯×p¯i2 with constraints{p¯j∗≥0∑j=1m−1p¯j∗≤1 Here , the weights , wi , give more importance to the signature genes with low variability in the reference gene expression profiles . In EPIC , these weights are given by:wi=min⁡ui , 100∙medianui whereui=∑j=1m-1C¯ijV¯ij+ε V- is the matrix of the TPM-based variability of each gene for each of the reference cells ( Supplementary files 1–2 ) , ε is a small number to avoid division by 0 , and the term '100∙medianui' is used to avoid giving too much weight to few of the genes . Finally , thanks to Equation ( 5 ) , the proportion for the uncharacterized cells can be obtained by: ( 8 ) p¯m=1−∑j=1m−1p¯j Since we used normalized gene expression data , values of p¯ correspond actually to the fraction of mRNA coming from each cell type , rather than the cell proportions . As the mRNA content per cell type can vary significantly ( Figure 1—figure supplement 2 ) , the actual proportions of each cell type can be estimated as: ( 9 ) pj=α∙p-jrj where rj is the amount of RNA nucleotides in cell type j ( or equivalently the total weight of RNA in each cell type ) and α is a normalization constant to have ∑pj=1 . Another method , DeMix ( Ahn et al . , 2013 ) , starts from Equation ( 1 ) to estimate the proportion and gene expression profile from cancer cells in mixed samples . In this method the mixed sample is assumed to be composed only by two cell types: cancer cells ( without any a priori known gene expression profile ) and normal cells ( with known gene expression data , which can either come from tumor-matched or unmatched samples ) . This method was developed for microarray data and shows that it was important to use the raw data as input assuming it follows a log-normal distribution as is the case for microarray , instead of working with log-transformed data like most other methods did . DeMix estimates the variance of the gene expression in the normal samples and uses this in the maximum likelihood estimation to predict the cancer cell gene expression and proportions , using thus implicitly a gene-specific weight for each gene . EPIC was derived for RNA-seq data and is directly using linear ( non-log ) data . Notably , when solving the linear regression from Equation ( 7 ) in EPIC , we are not assuming any specific distribution for the gene expression and the weights we give to each gene are simply based on their interquartile range in the reference samples . Other measures of gene variability could however be given as input into EPIC's R-package . Contrary to DeMix , another important assumption performed in EPIC is that our signature genes are not expressed by the cancer cells , so that we can easily estimate the proportion of multiple non-malignant cell types composing the bulk samples . Patients agreed to donate metastatic tissues upon informed consent , based on dedicated clinical investigation protocols established according to the relevant regulatory standards . The protocols were approved by the local IRB , that is , the 'Commission cantonale d’éthique de la recherche sur l’être humain du Canton de Vaud' . Lymph nodes ( LN ) were obtained from stage III melanoma patients , by lymph node dissection that took place before systemic treatment . The LN from one patient was from the right axilla and the LNs from the other three patients were from the iliac and ilio-obturator regions ( Appendix 2—table 1 ) . Single cell suspensions were obtained by mechanical disruption and immediately cryopreserved in RPMI 1640 supplemented with 40% FCS and 10% DMSO . Single cell suspensions from four lymph nodes were thawed and used in parallel experiments of flow cytometry and RNA extraction . In order to limit the number of dead cells after thawing , we removed those cells using a dead cell removal kit ( Miltenyi Biotech ) . Proportions of CD4 T ( CD45+/CD3+/CD4+/Melan-A- ) , CD8 T ( CD45+/CD3+/CD8+/Melan-A− ) , NK ( CD45+/CD56+/CD3−/CD33−/Melan-A− ) , B ( CD45+/CD19+/CD3−/CD33−/Melan-A− ) and Melan-A expressing tumor cells ( Supplementary file 3A ) were acquired via flow cytometry using the following antibodies: anti-CD3 BV711 ( RRID:AB_2566035 ) , anti-CD4 BUV737 ( RRID:AB_2713927 ) , anti-CD8 PE-Cy5 ( RRID:AB_2713928 ) , anti-CD56 BV421 ( RRID:AB_11218798 ) , anti-CD19 APCH7 ( RRID:AB_1645728 ) , anti-CD33 PE-Cy7 ( RRID:AB_2713932 ) , anti-CD45 APC ( RRID:AB_314400 ) , anti-Melan-A FITC ( RRID:AB_2713930 ) and Fixable Viability Dye eFluor 455UV ( eBioscience ) . Data was acquired on a BD LSR II SORP flow cytometry machine ( BD Bioscience ) . Analysis was performed using FlowJo ( Tree Star ) . Cell proportions were based on viable cells only . In parallel total RNA was extracted using the RNAeasy Plus mini kit ( Qiagen ) following the manufactures’ protocol . Starting material always contained minimum 0 . 2 × 106 cells . RNA was quantified and integrity was analyzed using a Fragment Analyser ( Advanced Analytical ) . Total RNA from all samples used for sequencing had an RQN ≥7 . Libraries were obtained using the Truseq stranded RNA kit ( Illumina ) . Single read ( 100 bp ) was performed using an Illumina HiSeq 2500 sequencer ( Illumina ) . Post processing of the sequencing was done using Illumina pipeline Casava 1 . 82 . FastQC ( version 0 . 10 . 1 ) was used for quality control . RNA-seq reads alignment to the human genome , hg19 , and TPM quantification were performed with RSEM ( Li and Dewey , 2011 ) version 1 . 2 . 19 , using Bowtie2 ( Langmead and Salzberg , 2012 ) version 2 . 2 . 4 and Samtools ( Li et al . , 2009 ) version 1 . 2 . RNA-Seq data from this experiment have been deposited in NCBI's Gene Expression Omnibus ( Edgar et al . , 2002 ) and are accessible through GEO Series accession number GSE93722 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE93722 ) . Healthy donor peripheral blood was obtained through the blood transfusion center in Lausanne . PBMCs were purified by density gradient using Lymphoprep ( Axis-Shieldy ) . Mononuclear cells were stained in order to sort monocytes , B , T and NK cells using the following antibodies: CD14 FITC ( RRID:AB_130992 ) , CD19 PE ( RRID:AB_2716572 ) , CD3 APC ( RRID:AB_130788 ) , CD56 BV421 ( RRID:AB_11218798 ) and fixable live/dead near IR stain ( ThermoFisher Scientific ) . 1 × 106 live cells from each cell type were sorted using the BD FACS ARIA III ( BD Biosciences ) . Total RNA was extracted using the RNAeasy Plus mini kit ( Qiagen ) following the manufactures’ protocol and quantified using a Fragment Analyser ( Advanced Analytical ) . Values obtained are given in Figure 1—figure supplement 2A . The mRNA content for the cancer cells was estimated from the flow cytometry data described in the previous section from the four patients with melanoma . For this we used the forward scatter width , which is a good proxy of cell size and mRNA content ( Padovan-Merhar et al . , 2015; Tzur et al . , 2011 ) , and observed that cancer cells had similar amount of mRNA than B , NK and T cells ( Figure 1—figure supplement 2B ) . We thus used a value of 0 . 4 pg of mRNA per cancer cell . For the above datasets 2 and 3 , we obtained raw fastq files . RNA-seq reads alignment to the human genome , hg19 , and TPM quantification were performed with RSEM ( Li and Dewey , 2011 ) version 1 . 2 . 19 , using Bowtie2 ( Langmead and Salzberg , 2012 ) version 2 . 2 . 4 and Samtools ( Li et al . , 2009 ) version 1 . 2 . For the other datasets , we directly obtained the summary counts data from the respective studies without mapping the reads by ourselves , and we transformed these counts to TPM wherever necessary . Reference gene expression profiles of sorted immune cells from peripheral blood were built from the datasets 2 , 3 and 4 described in previous section . We verified no experimental biases were present in these data through unsupervised clustering of the samples , with help of a principal component analysis based on the normalized expression from the 1000 most variable genes ( Figure 1C ) . The median value of TPM counts was computed per cell type and per gene . Similarly , the interquartile range of the TPM counts was computed per cell type and gene , as a measure of the variability of each gene expression in each cell type . Values of these reference profiles are given in Supplementary file 1 ) . Granulocytes from dataset 4 and neutrophils from datasets 2 and 3 were combined to build the reference profile for neutrophils ( neutrophils constitute more than 90% of granulocytes ) . No reference profile was built for the myeloid dendritic cells as only few samples of these sorted cells existed and they were all from the same experiment . Monocytes are not found in tumors but instead there are macrophages , mostly from monocytic lineage , that are infiltrating tumors and that are not found in blood . For this reason , we also used the monocyte reference gene expression profile as a proxy for macrophages when applying EPIC to tumor samples . Such an assumption gave coherent results as observed in the results . We also built gene expression reference profiles from tumor-infiltrating cells . These are based on the single-cell RNA-Seq data from Tirosh and colleagues ( Tirosh et al . , 2016 ) described above . We only used the non-lymphoid tissue samples to build these tumor-infiltrating cell’s profiles , avoiding in this way potential ‘normal immune cells’ present in the lymph nodes and spleen . These reference profiles ( Supplementary file 2 ) were built in the same way as described above for the reference profiles of circulating immune cells , but based on the mean and standard deviation instead of median and interquartile range respectively , due to the nature of single-cell RNA-Seq data and gene dropout present with such technique . When testing EPIC with these profiles for the single-cell RNA-Seq datasets , for the samples of primary tumor and other non-lymph node metastases , a leave-one-out procedure was applied: for each donor we built reference cell profiles based only on the data coming from the other donors . EPIC relies on signature genes that are expressed by the reference cells but not by the uncharacterized cells ( e . g . , cancer cells ) . For each reference cell type , we thus built a list of signature genes through the following steps: All the differential expression tests were performed with DESeq2 ( Love et al . , 2014 ) . Appendix 1—table 1 summarizes the full list of signature genes per cell type . We compared EPIC's predictions with those from various cell fraction prediction methods . These other methods were run with the following packages ( using the default options when possible ) : For CIBERSORT , DeconRNASeq and ISOpure , when run based on our gene expression reference profiles , we used the reference profiles from peripheral blood immune cells for the predictions in blood and the reference profiles from tumor-infiltrating cells for the predictions in solid tumors . CAFs: cancer-associated fibroblasts; EPIC: acronym for our method to ‘Estimate the Proportion of Immune and Cancer cells’; GEO: Gene Expression Omnibus; IHC: immunohistochemistry; PCA: principal component analysis; RMSE: root mean squared error; TCGA: The Cancer Genome Atlas; TPM: transcripts per million .	Malignant tumors do not only contain cancer cells . Normal cells from the body also infiltrate tumors . These often include a variety of immune cells that can help detect and kill cancer cells . Many evidences suggest that the proportion of different immune cell types in a tumor can affect tumor growth and which treatments are effective . Researchers often study tumors by measuring the expression of genes , i . e . , which genes are active in tumors . However , the proportion of different cell types in the tumor is often not measured for tumors studied at the gene expression level . Racle et al . have now demonstrated that a new computer-based tool can accurately detect all the main cell types in a tumor directly from the expression of genes in this tumor . The tool is called “Estimating the Proportion of Immune and Cancer cells” – or EPIC for short . It compares the level of expression of genes in a tumor with a library of the gene expression profiles from specific cell types that can be found in tumors and uses this information to predict how many of each type of cell are present . Experimental measurements of several human tumors confirmed that EPIC’s predictions are accurate . EPIC is freely available online . Since the active genes in tumors from many patients have already been documented together with clinical data , researchers could use EPIC to investigate whether the cell types in a tumor affect how harmful it is or how well a particular treatment works on it . In the future , this information could help to identify the best treatment for a particular patient and may reveal new genes that cause malignant tumors to develop and grow .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "tools", "and", "resources", "cancer", "biology" ]	2017	Simultaneous enumeration of cancer and immune cell types from bulk tumor gene expression data
Mediator-associated kinases CDK8/19 are context-dependent drivers or suppressors of tumorigenesis . Their inhibition is predicted to have pleiotropic effects , but it is unclear whether this will impact on the clinical utility of CDK8/19 inhibitors . We discovered two series of potent chemical probes with high selectivity for CDK8/19 . Despite pharmacodynamic evidence for robust on-target activity , the compounds exhibited modest , though significant , efficacy against human tumor lines and patient-derived xenografts . Altered gene expression was consistent with CDK8/19 inhibition , including profiles associated with super-enhancers , immune and inflammatory responses and stem cell function . In a mouse model expressing oncogenic beta-catenin , treatment shifted cells within hyperplastic intestinal crypts from a stem cell to a transit amplifying phenotype . In two species , neither probe was tolerated at therapeutically-relevant exposures . The complex nature of the toxicity observed with two structurally-differentiated chemical series is consistent with on-target effects posing significant challenges to the clinical development of CDK8/19 inhibitors . The Mediator complex is a multi-subunit regulator of transcription in eukaryotes that transfers signals from DNA-bound transcription factors to the RNA polymerase II pre-initiation complex ( Allen and Taatjes , 2015; Yin and Wang , 2014; Poss et al . , 2013 ) . It also has a role in transcription elongation and pausing , and can influence chromatin structure , where it facilitates the formation of enhancer-promoter gene loops and is enriched at ‘super-enhancer’ regions ( Allen and Taatjes , 2015; Poss et al . , 2013; Whyte , 2013 ) . Mediator activity is regulated by the association with a four-subunit kinase module containing cyclin-dependent kinase 8 ( CDK8 ) , cyclin C ( CCNC ) and Mediator subunits MED12 and MED13 ( Allen and Taatjes , 2015; Poss et al . , 2013; Taatjes et al . , 2002 ) . As a kinase that reversibly associates with the Mediator , CDK8 is thought to regulate gene expression through phosphorylation of transcription factors and Mediator subunits ( Rzymski et al . , 2015 ) . Phosphorylation by CDK8 can directly alter transcription factor activity ( Bancerek et al . , 2013; Morris et al . , 2008; Zhao et al . , 2013 ) or mark factors for degradation ( Fryer et al . , 2004; Alarcón et al . , 2009; Zhao et al . , 2012; Li et al . , 2014 ) . The role of CDK8 , and the Mediator kinase module , in the control of transcription may not be unique as paralogs of CDK8 , MED12 and MED13 have been identified that may have distinct roles in vitro and in vivo ( Sato et al . , 2004; Tsutsui et al . , 2008; Galbraith et al . , 2013; Westerling et al . , 2007 ) . The biological function of CDK8 varies by cell type and response to different stimuli ( Allen and Taatjes , 2015; McCleland et al . , 2015 ) . This is particularly true in cancer , where CDK8 may function not only as an oncogene , but also as a tumor-suppressor depending on the cellular context ( McCleland et al . , 2015; Mitra et al . , 2006; Chattopadhyay et al . , 2010; Gu et al . , 2013; Firestein et al . , 2008 , 2010; Seo et al . , 2010; Adler et al . , 2012 ) . CDK8 may act as an oncogene in colorectal cancer where CDK8 is amplified , with copy number gains observed in ~60% of tumors ( Firestein et al . , 2010; Seo et al . , 2010 ) , and shRNA knockdown can reduce the growth of human colorectal cancer xenografts harbouring CDK8 gene amplification ( Firestein et al . , 2008; Adler et al . , 2012; Starr et al . , 2009 ) . Furthermore , CDK8 expression is reportedly required for growth of colorectal cancer xenografts and to maintain embryonic stem cells in an undifferentiated state ( Adler et al . , 2012 ) . Importantly , CDK8 expression transforms fibroblasts into a malignant phenotype , whereas expression of a kinase-dead mutant does not ( Firestein et al . , 2008 ) . An shRNA screen has also demonstrated a requirement for CDK8 in the activation of WNT signaling in colorectal cancer ( Firestein et al . , 2008 ) , suggesting that CDK8 and the Mediator kinase module may promote oncogenesis through activation of the canonical WNT pathway . Previously , we reported the discovery and optimization of a potent and selective 3 , 4 , 5-trisubstituted pyridine series of small-molecule inhibitors of WNT signaling from a cell-based pathway screen , and using a chemo-proteomic strategy we identified CDK8 and CDK19 as the primary molecular targets ( Dale et al . , 2015; Boyer , 2015 ) . Through further optimization we identified a potent , highly selective and orally bioavailable dual CDK8/19 ligand with excellent cell-based activity and pharmaceutical properties ( Mallinger et al . , 2016a ) . Subsequently , we discovered a second , chemically-distinct series of CDK8/19 ligands and optimization of pharmacological , pharmaceutical and pharmacokinetic properties identified a 3-methyl-1H-pyrazolo[3 , 4-b]pyridine , which also binds to CDK8/CCNC ( Czodrowski et al . , 2016 ) . With potent and selective exemplar compounds from these two structurally differentiated chemical series in hand together with corresponding inactive control compounds , we were well positioned to investigate the therapeutic potential of dual CDK8/19 modulation . Specifically , we set out to establish if these compounds had antiproliferative or antitumor activity and whether a therapeutic window could be identified in preclinical models that would justify the clinical development of these compounds . We identified two structurally differentiated , potent , selective and cell permeable chemical series , namely 3 , 4 , 5-trisubstituted pyridines and 3-methyl-1H-pyrazolo[3 , 4-b]pyridines , suitable for exploring the function of the Mediator complex-associated protein kinases CDK8 and CDK19 ( Figure 1A and Figure 1—source data 1 ) . In addition to two tool compounds , 1 ( CCT251545; Mallinger et al . , 2015 ) and 2 ( compound 42;Mallinger et al . , 2016a ) , that fulfill all of the criteria set out for chemical probes ( Frye , 2010 ) , the lead compounds from each of the chemical series , 3 ( CCT251921; Mallinger et al . , 2016a ) and 4 ( MSC2530818; Czodrowski et al . , 2016 ) had optimal pharmacological and pharmaceutical properties that made them suitable for further progression to preclinical studies ( Figure 1A and Figure 1—source data 1 ) . All four compounds had single digit nanomolar binding affinities for CDK8 and 19 , and were very highly selective with little evidence for off-target activity in extended protein kinase panels ( Figure 1—source data 1 ) . Our compounds also potently inhibited inducible ( 7dF3; Ewan et al . , 2010 ) or basal ( LS174T; Dale et al . , 2015; Mallinger et al . , 2015 ) WNT-pathway luciferase-reporter expression together with STAT1SER727 phosphorylation ‑ a target-engagement biomarker ‑ at low nanomolar concentrations ( Figure 1A and Figure 1—source data 1 ) ( Bancerek et al . , 2013; Dale et al . , 2015 ) . 10 . 7554/eLife . 20722 . 003Figure 1 . Optimised compounds for exploring CDK8 and CDK19 function . ( A ) Chemical structure and activity of compounds 2 , 3 and 4 ( n > 2 , mean ± s . d . ) . ( B ) Overlay of 3 ( grey; ocd 5HBJ ) and 4 ( pink; Pdb code: 5IDN ) bound to CDK8/CCNC . Key interactions ( yellow ) and residues are shown . Residues 23 – 39 and 359 – 361 have been cropped for clarity . ( C ) Amino acid sequence alignment for human CDK8 and CDK19 . Red , sequence differences; yellow , ATP binding; green , CCNC binding; gray , activation loop; blue , inhibitor binding . ( D ) Luciferase activity in COLO205-cl4 cells containing a TCF/LEF reporter gene construct following 24 hr compound treatment ( n = 4 , mean ± s . d . ) . ( E ) Colony assay . Plates were seeded with LS513 or insensitive RKO cells and treated for 14 d . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 00310 . 7554/eLife . 20722 . 004Figure 1—source data 1 . Properties of CDK8/19 ligands and their effects on reporter expression and cell proliferation in a human colorectal cancer cell line panel . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 00410 . 7554/eLife . 20722 . 005Figure 1—figure supplement 1 . Effect of CDK8 and CDK19 shRNA and siRNA treatment in CDK8-amplified human colorectal cancer cell lines . ( A ) WNT pathway reporter activity , cell viability and CDK8/19 transcript levels in COLO205-cl4 TCF/LEF reporter cells expressing either an inducible CDK8 shRNA , a constitutive CDK19 shRNA , an inducible CDK8 plus constitutive CDK19 shRNA or a non-targeting constitutive ( GIPZ ) or inducible ( TRIPZ ) control shRNA . Reporter activity and viability were measured following 8 d 1µg/ml Dox induction ( mean ± s . e . m , n = 3 ) . CDK8 , CDK19 , p-STAT1SER727 and STAT1 levels ( B ) and HT29 cell viability ( C ) following 5 d treatment with CDK8 and/or CDK19 siRNA ( Mock = no siRNA , Death = positive control siRNA , Non-coding = negative control siRNA ) . In B , β-actin was used as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 00510 . 7554/eLife . 20722 . 006Figure 1—figure supplement 2 . Comparison of CDK8 and CDK19 gene copy number or protein expression with sensitivity to treatment with compound . Data from Figure 1—source data 1 for CDK8 gene copy number or protein expression , or CDK19 protein expression , were compared with the effects of 1 , 3 and 4 on 14 d colony growth assay . Pearson r2 correlation values are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 006 A comparison of the co-crystal structure of CDK8/CCNC with 3 or 4 showed that both molecules adopt a Type I binding mode and make similar contacts with active site residues ( Figure 1B ) . Compound 3 binds in a twisted conformation , as previously described for 1 , with the indazole substituent at C5 of the pyridine ring forming a pi-cation interaction with Arg356 ( Figure 1B ) ( Mallinger et al . , 2015 , 2016a ) . Compound 4 forms similar interactions with the hinge region and with the catalytic Lys52 to those observed for compound 3 , and its p-chlorophenyl substituent occupies the same region as the indazole subsituent of compound 3; however , the scaffold architecture of the two compounds is entirely different ( Figure 1B ) . Throughout our studies on both chemical series , we observed a strong correlation between the compounds’ affinities for both CDK8 and CDK19 , suggesting that selective inhibition of CDK8 versus CDK19 is likely to be a significant challenge ( Dale et al . , 2015; Mallinger et al . , 2016a ) . This reflects the high sequence similarity between CDK8 and CDK19 ( Figure 1C ) . We also tested selected compounds from a further three chemical series that we identified from the literature and again could not detect any substantial selectivity for CDK8 versus CDK19 ( Figure 1—source data 1 ) . We confirmed compound activity in a range of in vitro assays using human colorectal cancer cell lines , including some with an increased CDK8 gene copy number ( Figure 1—source data 1 ) . All four compounds potently inhibited a WNT-dependent reporter in all of the cell lines tested , but did not inhibit a WNT-independent housekeeping EEF1A1 promoter-reporter construct in the negative control RKO colorectal cancer cell line , which expresses low levels of beta-catenin ( Figure 1D and Figure 1—source data 1 ) ( Mallinger et al . , 2015; Dale et al . , 2015 ) . Weakly-active negative-control compounds from the 3 , 4 , 5-trisubstituted pyridine and 3-methyl-1H-pyrazolo[3 , 4-b]pyridine series ( compounds 5 and 6 respectively ) did not inhibit reporter gene expression or STAT1SER727 phosphorylation ( Figure 1—source data 1 ) . Given the potent inhibition of reporter activity we were surprised by the lack of effect of our potent inhibitors on tumor cell proliferation after standard 4 d continuous exposure conditions ( Figure 1—source data 1 ) . However , this was consistent with a reported lack of antiproliferative effects for a different chemical series in a single colorectal cancer cell line ( Koehler et al . , 2016 ) . Silencing of CDK8 and/or CDK19 by shRNA in CDK8-amplified COLO205 cells also had no effect on viability , despite evidence for inhibition of reporter output or target gene expression ( Figure 1—figure supplement 1A ) . Knockdown was more effective when we used CDK8 and/or CDK19 siRNA in CDK8-amplified HT29 cells , but again we saw no significant effect on viability after 5 d exposure to siRNA , despite near complete inhibition of STAT1SER727 phosphorylation by the CDK8 siRNA ( Figure 1—figure supplement 1B–C ) . In contrast , a 14 d colony growth assay revealed a significantly similar antiproliferative effect for the lead compounds from both chemical series ( p<0 . 001 for all comparisons with 1 , 3 and 4; Figure 1—figure supplement 2 ) , which was not observed for the negative-control compounds 5 and 6 . However , no compounds showed colony growth inhibition in the negative control RKO colorectal cancer cell line ( Mallinger et al . , 2015; Dale et al . , 2015 ) . In this assay , we found three beta-catenin mutant ( LS513 , LS180 , LS174T ) and an APC mutant ( SW620 ) colorectal cell lines to be most sensitive to treatment ( Figure 1E and Figure 1—source data 1 ) . The association between beta-catenin mutation and sensitivity to compound treatment in the colony assay did not reach significance ( p>0 . 05 ) . The lack of response of the RKO cells suggested that colony growth in this line did not require beta-catenin or CDK8 and contrasted with the APC or beta-catenin mutant lines where colony growth appeared to be dependent on beta-catenin-regulated transcription that also required CDK8/19 . Overall , there was no significant correlation between compound activity in TCF-reporter , phospho-biomarker , colony growth assays and either CDK8/19 protein levels or gene copy number ( Figure 1—figure supplement 2 and Figure 1—source data 1 ) . Next , we determined if our two series of compounds had antitumor activity in vivo in human colorectal cancer xenograft mouse models . Previously , we demonstrated that compound 1 inhibited TCF/LEF-reporter gene expression and reduced STAT1SER727 phosphorylation by >80% . This translated into tumor growth inhibition following oral dosing of 1 in mice bearing established COLO205 or SW620 colorectal cancer cell xenografts ( Mallinger et al . , 2015; Dale et al . , 2015 ) . We also found evidence for a significant , dose-dependent , reduction in tumor growth in HCT116 human colorectal cancer cell line xenografts as well as significant tumor growth inhibition ( TGI = 81%; p<0 . 001 ) at 70 mg/kg in an LS513 human colorectal cancer xenograft model with concomitant reduction of p-STAT1SER727 at 6 hr ( Figure 2—figure supplement 1 and Figure 2—source data 1 ) . For the lead compounds 3 and 4 , we modelled the inhibition of STAT1SER727 phosphorylation using multiple sets of experimentally-derived data from HCT116 and SW620 human tumor xenografts ( Figure 2 and Figure 2—figure supplement 2A–C ) . Detailed analysis , initially in HCT116 tumor xenografts , showed that maximal inhibition of STAT1SER727 phosphorylation required treatment with ≥5 mg/kg compound 3 and that higher concentrations prolonged the period of maximal inhibition ( Figure 2—figure supplement 2A–C ) . In SW620 tumor xenografts biomarker inhibition was rapidly achieved following treatment with 3 or 4 and could be maintained for approximately 10 hr after a single treatment with 30 mg/kg of 3 , while 30 or 100 mg/kg of 4 prolonged the period of inhibition of STAT1SER727 phosphorylation so that , unlike 3 , multiple dosing with 4 prevented recovery of biomarker to control levels between treatments ( Figures 2A , 3D–E , Figure 2—figure supplement 2D and Figure 2—figure supplement 3B ) . Both lead compounds 3 and 4 exhibited reproducible , dose-dependent antitumor activity in SW620 tumor xenografts ( Figure 2D–E , Figure 2—figure supplement 3A–B and Figure 2—source data 1 ) and also in an additional LS1034 colorectal tumor model that responded to our compounds in the in vitro clonogenic assay ( Figure 2—figure supplement 3C–E , Figure 2—figure supplement 4 and Figure 2—source data 1 ) . While 5 mg/kg compound 3 induced maximal inhibition of STAT1SER727 phosphorylation in SW620 human tumor xenografts , its effects were short-lived , and were not sufficient to translate into antitumor activity ( Figure 2—figure supplement 2D and Figure 2—source data 1 ) . However , higher 30 mg/kg doses of compound 3 prolonged the maximal inhibition of STAT1SER727 phosphorylation and had a significant ( p<0 . 01 ) antitumor effect ( Figure 2—figure supplement 3A–B and Figure 2—source data 1 ) . As with 3 , 10 mg/kg of 4 inhibited the pathway biomarker for 6 hr , but was not sufficient for antitumor activity in SW620 xenografts ( Figure 2A and Figure 2—source data 1 ) . However , repeated dosing of 4 at 50 or 100 mg/kg prolonged the inhibition of STAT1SER727 phosphorylation and gave evidence of antitumor activity ( Figure 2D–E and Figure 2—source data 1 ) . Hence , the antitumor activity of 3 and 4 against colorectal cancer xenografts showed clear dose-dependence , with a requirement for prolonged pathway inhibition in order to significantly reduce tumor growth . 10 . 7554/eLife . 20722 . 007Figure 2 . Target inhibition and antitumor activity of CDK8/19 ligands 3 and 4 in established human colorectal cancer cell line xenografts . ( A ) Level of p-STAT1SER727 in SW620 human colorectal cancer xenografts following a single dose of 4 , relative to the p-STAT1SER727 level in vehicle-treated mice . Significance was determined by Kruskal-Wallis test and Dunn’s post-test ( p=<0 . 001; ) . ( B , C ) Modelling of experimental data , including data from ( A ) and Figure 2—figure supplement 2D , of STAT1SER727 phosphorylation following ( B ) single or ( C ) twice daily doses of 30 mg/kg 3 ( black ) or 30 and 100 mg/kg 4 ( orange and purple ) . ( D ) Volume of SW620 xenografts in mice treated with 4 or a vehicle control . ( E ) Level of p-STAT1SER727 , relative to control , in SW620 tumor xenografts at the stated time following the final dose of 4 ( from D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 00710 . 7554/eLife . 20722 . 008Figure 2—source data 1 . Details of human colorectal cancer cell line xenograft studies . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 00810 . 7554/eLife . 20722 . 009Figure 2—figure supplement 1 . Differential antitumor activity in human LS513 colorectal cancer xenografts treated with CDK8/19 ligand 1 . Antitumor activity in ( A ) HCT116 and ( B ) LS513 colorectal cancer xenografts treated with 1 . ( C ) target engagement in LS513 colorectal cancer xenografts treated with 1 . STAT1 and p-STAT1SER727 levels 2 or 6 hr after the final dose . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 00910 . 7554/eLife . 20722 . 010Figure 2—figure supplement 2 . Pharmacodynamic profiling of CDK8/19 ligand 3 in human colorectal cancer xenografts . ( A , B ) Level of p-STAT1SER727 in HCT116 colorectal cancer xenografts in mice treated with single doses of 3 . Significance was determined by Kruskal-Wallis test and Dunn’s post-test ( p=<0 . 001 , **p=<0 . 0001 ) . ( C ) Simulation of PD dose-dependence of 3 in HCT116 colorectal cancer xenografts . ( D ) Level of p-STAT1SER727 in SW620 colorectal cancer xenografts after the mice were treated with a single dose of 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01010 . 7554/eLife . 20722 . 011Figure 2—figure supplement 3 . Antitumor activity and target engagement in human colorectal cancer xenografts treated with CDK8/19 ligands 3 and 4 . ( A , B ) Antitumor activity ( A ) and target engagement ( B ) in SW620 colorectal cancer xenografts treated with 3 . ( C , D ) Antitumor activity ( C ) and target engagement ( D ) in LS1034 colorectal cancer xenografts treated with 3 . Antitumor activity ( E ) in LS1034 colorectal cancer xenografts treated with 4 . In ( E ) p-STAT1SER727 was decreased by 58% 2 hr post-treatment ( p-STAT1SER727 values normalised to total STAT1 levels; control: 1719 ± 174; CMPD 4: 719 ± 46; ( n = 4 , mean ± s . e . m ) . p-STAT1SER727 and total STAT1 levels were determined at the specified times after the final dose . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01110 . 7554/eLife . 20722 . 012Figure 2—figure supplement 4 . LS1034 colony assay . ( A ) Growth of LS1034 cell colonies treated for 14 d with 350 nM test compounds 1 , 3 and 5 ( n = 3 , mean ± s . d . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01210 . 7554/eLife . 20722 . 013Figure 3 . In vitro and in vivo activity of CDK8/19 ligands in patient-derived tumour xenograft models . ( A ) GI50 values for 2 in PDX soft agar colony cultures . ( B ) Exemplar dose-response profiles for selected colorectal cancer clonogenic assays treated with 2 . ( C ) Volume of human colorectal cancer CXF 1034 ( CTNNB1MUT , PIK3CAMUT , PTENMUT ) PDXs in mice treated with vehicle , 3 and / or irinotecan ( mean values ± s . e . m . , n = 10 per cohort ) . Tumor volume was significantly different ( p=<0 . 001; 2 way ANOVA and Tukey’s multiple comparison test ) in mice receiving the combination treatment , compared with the monotherapy groups . ( D ) Level of p-STAT1SER727 , relative to control , in CXF 1034 xenografts in mice treated with 3 measured 1 hr after the final dose ( p=<0 . 0001 , Mann-Whitney test; mean values ± s . d . , n = 10 per cohort ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01310 . 7554/eLife . 20722 . 014Figure 3—figure supplement 1 . In vivo activity of CDK8/19 ligand 3 in PDXs . Volume of PDXs treated with vehicle , 3 and/or irinotecan/oxaliplatin , 0 ( n = 10 , mean ± s . e . m ) . Xenografts tested: A , CFX 883 ( KRASMUT , CTNNB1MUT , PTENMUT ) ; C , CFX 1729 ( p53MUT ) ; E , CFX 280 ( KRASMUT , APCMUT , p53MUT ) ; F , CFX 609 ( KRASMUT , APCMUT , p53MUT ) ; G , CFX 1753 ( NRASMUT , PIK3CAMUT , APCMUT ) . Ratio of p-STAT1SER727 to total STAT1 , expressed relative to the vehicle control , in ( B ) CFX883 and ( D ) CFX1729 xenografts treated with 3 . p-STAT1SER727 and STAT1 levels were assayed 1 hr after the last dose ( n = 10 , mean ± s . d . ) and a statistically significant difference detected ( p=<0 . 0001 , Mann-Whitney two-tailed test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01410 . 7554/eLife . 20722 . 015Figure 3—figure supplement 2 . Pharmacodynamic and antitumor activity of 3 and 4 in AML models . ( A ) Percentage of AML Nomo-1 cells detected in the peripheral blood of NOD/Shi-SCID/IL-2Rγnull mice treated with 10 mg/kg po bid 3 or a vehicle control ( p=<0 . 001 , Kruskal-Wallis and Dunn’s post test ) . ( B ) Percentage of AML cells detected in the spleen and bone marrow of mice from A after 21 d treatment with 3 or the vehicle control ( p=<0 . 05 ) . ( C ) Antitumor activity in MV-4-11 AML xenografts treated with po 5/7 bid , 2/7 qd 4 or the vehicle control , relative to day 0 . ( D ) p-STAT1SER727 to STAT1 ratio in MV-4-11 AML xenografts from C treated with 4 or the vehicle control . Protein levels were determined 2 and 6 hr after the last dose was administered ( p<0 . 001 , Mann-Whitney two-tailed test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 015 To follow up our standard xenograft experiments , we next tested compound activity in human patient-derived tumor xenograft models ( PDXs ) . Firstly , we determined if our compounds were active in in vitro soft agar clonogenic assays , using cells derived from 89 distinct PDX models from different tissue types . In this assay , colony growth was inhibited , but only by up to 50% , in one-third of PDX cell cultures treated with tool compound 2 ( Figure 3A–B ) . Six of the more sensitive colorectal PDX tumor models were selected for in vivo monotherapy either with the lead compound 3 alone , or combined with a standard of care drug ( irinotecan or oxaliplatin ) . Only two of these colorectal cancer PDX tumors models , CFX 883 and CFX 1753 , responded to monotherapy with 3 ( p<0 . 0001; Figure 3—figure supplement 1A ) . Importantly , the combination of 3 with standard of care therapies was only of statistically significant benefit in one colorectal tumor model , CFX 1034 ( p<0 . 001; Figure 3C and Figure 3—figure supplement 1 ) . Examination of p-STAT1SER727 levels indicated near maximal CDK8 inhibition in the three tumors where p-STAT1SER727 was tested ( p<0 . 0001; Figure 3 and Figure 3—figure supplement 1B and D ) , indicating that the failure of 3 to slow PDX growth ( Figure 3—figure supplement 1C and F ) was unlikely to be due to a lack of target engagement . The most CDK8/19 inhibitor sensitive cell model in the soft agar assays was one of the two acute myeloid leukaemia ( AML ) cell lines , Nomo-1 , included in the cell panel . The Nomo-1 line was particularly sensitive to treatment with compound 2 ( Figure 3A ) with an 11 nM GI50 . Subsequently , in a Nomo-1 systemic in vivo model , treatment with 3 led to a potent reduction in circulating tumor cells ( Figure 3—figure supplement 2A–B ) . An additional subcutaneous MV-4-11 AML xenograft also responded to monotherapy ( TGI = 100% ) with 10 mg/kg po qd 3 , and also with 4 which showed evidence of near maximal target inhibition , as determined using p-STAT1SER727 at 2 hr post-treatment ( Figure 3—figure supplement 2C–D ) . Having demonstrated target engagement and associated antitumor activity in vivo , we further explored the molecular response of tumors treated with compounds 1 and 3 ( Figure 4—source data 1 ) . We initially identified 278 transcription factor-associated genesets that were enriched in genes whose expression was significantly altered following in vivo treatment of COLO205 or SW620 human tumor xenografts . The altered expression of 121/278 of these transcription factor-associated genesets was identified following treatment of both xenografts ( Figure 4A ) . Of note we also identified 185 transcription factor-associated genesets that were significantly enriched in sets of genes associated with super-enhancers; of these , 2/3rds were shared with the genesets whose expression was altered by compound treatment ( Figure 4A ) . These common transcription factor genesets included transcription factors known to be regulated by CDK8 , such as TCF4 , SMADs , STATs , c/EBP and HIF1A ( Figure 4—source data 1 ) ( Bancerek et al . , 2013; Morris et al . , 2008; Zhao et al . , 2013; Fryer et al . , 2004; Alarcón et al . , 2009; Zhao et al . , 2012 ) . The genesets shared between both treatment and super-enhancers also encompassed transcription factors ( NANOG , OCT3/4 and SOX2 ) required for stem cell pluripotency ( Figure 4—source data 1 ) ( Whyte et al . , 2013 ) . 10 . 7554/eLife . 20722 . 016Figure 4 . Microarray gene expression profiling following in vivo treatment of human colorectal cancer xenografts with CDK8/19 ligands . Mice were treated with 70 mg/kg po 1 ( SW620 and COLO205 ) , 20 mg/kg po 3 ( COLO205 ) . ( A ) Venn plots of transcription factor-associated genesets or those encoding or regulating pathways enriched in genes whose expression was significantly altered by treatment ( Supplementary Dataset ) . ( B ) GSEA of CDK8 or BRD4-associated super-enhancer genes in treated human tumor xenografts . ( C ) Scatterplot of false discovery rate ( FDR-q ) versus normalized enrichment score ( NES ) for indicated gene sets evaluated by GSEA ( n=10 , 218 ) , signatures include those from MSigDB , dbSUPER and the ChIP-seq data from Pelish and colleagues ( Pelish et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01610 . 7554/eLife . 20722 . 017Figure 4—source data 1 . Geneset expression analysis of microarray data following in vivo treatment of SW620 or COLO205 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01710 . 7554/eLife . 20722 . 018Figure 4—figure supplement 1 . Microarray gene expression profiling following in vivo treatment of colorectal cancer cell line xenografts . Xenografts were treated with 70 mg/kg po 1 ( SW620 and COLO205 ) and 20 mg/kg po 3 ( COLO205 ) . Scatterplots of false discovery rate ( FDR-q ) versus normalized enrichment score ( NES ) for indicated gene sets evaluated by GSEA ( n=10 , 218 ) , signatures include those from MSigDB , dbSUPER and the ChIP-seq data from Pelish and colleagues ( Pelish et al . , 2015 ) . Scatter plots highlighting A , BMP2 and bone remodelling genesets or B , genesets regulated by the polycomb group complex ( EED , PRC2 or SUZ12 ) or transcription factors associated with pluripotency ( NANOG , OCT4 or SOX2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 018 A similar analysis , but this time of genes encoding specific pathway components , found 85 genesets were significantly enriched in the pool of genes whose expression was modulated by compound treatment ( Figure 4A ) ; 15 of which were found in both colorectal cancer xenograft experiments . A third of these were geneset-encoded gene products associated with WNT signaling , while others encoded components of individual pathways involved in development , inflammation or osteoblast biology ( Figure 4—source data 1 ) . Five of the genesets , common to both tumor models , all comprising WNT pathway components , were enriched with super-enhancer associated genes . In addition , we used geneset enrichment analysis ( GSEA ) to compare the compound-treated profiles with sets of genes associated with super-enhancers identified from published ChIP-seq datasets ( Pelish et al . , 2015 ) , and genesets from MSigDB ( www . broadinstitute . org/msigdb ) . We found that treatment of both tumor models resulted in a consistent and significant modulation of genes in the vicinity of MED1 and CDK8-associated super-enhancers identified from ChIP-seq datasets ( Figure 4B–C and Figure 4—source data 1 ) . Interestingly , super-enhancers defined by a BRD4 ChIP-seq in lymphoid cells did not show the same significant modulation as the MED1/CDK8-defined super-enhancers ( Figure 4B–C ) . The expression of many of the MSigDB genesets were positively or negatively correlated with the compound-treated samples and included genesets encoding gene products associated with , or regulated by , developmental , immunological or inflammatory pathways ( Figure 4—source data 1 ) . Some patterns were common to all treated samples , for example the genesets with decreased expression associated with stem cell pluripotency ( SOX2 , NANOG and OCT4 ) were modulated by treatment in both of the tumor models and also with both 1 and 3 ( Figure 4—figure supplement 1 ) . In contrast , other genesets exhibited cell line-specific regulation , for example the expression of bone morphogenic protein 2 ( BMP2 ) -regulated genes and others associated with bone remodelling were inhibited by treatment of COLO205 xenografts , but were activated in SW620 xenografts ( Figure 4—figure supplement 1 ) . Overall , the altered levels of transcripts from genes influenced by transcription factors or super-enhancers known to be regulated by the Mediator complex and CDK8 are consistent with our compounds affecting a broad range of CDK8/19-regulated gene transcription ( Mallinger et al . , 2015 ) . The genes and genesets identified suggested that effects on stem cells , bone , immunology and inflammation might influence compound tolerability ( Figure 4—figure supplement 1 ) . Among the profiles identified by GSEA were those associated with the regulation of normal and tumor stem cell populations ( Figure 4—source data 1 ) . We used a mouse model expressing doxycycline ( Dox ) -inducible activated beta-catenin ( Jardé et al . , 2013 ) to explore the effect of the tool compound 1 on an oncogenically-activated stem cell compartment . We separated the crypt-cell compartment from intestinal epithelial cells using a panel of cell surface markers ( Figure 5—figure supplement 1 and Figure 5—source data 1; Wang et al . , 2013 ) . GRP78 staining was then used to separate stem-like and transit amplifying-like ( TA ) cells , the success of which was confirmed by RT-PCR ( Figure 5A and Figure 5—source data 1 ) . 10 . 7554/eLife . 20722 . 019Figure 5 . Treatment with CDK8/19 ligand 1 reduces the hyperplastic crypt stem cell population . Gene expression , measured by RT-PCR , in the intestinal epithelial stem and TA cells isolated from mice expressing a Dox-inducible activated β-catenin transgene . ( A ) Transcript abundance , relative to control , in stem and TA cells following induction with 2 mg/ml Dox . ( B ) Abundance of different cell types following treatment with Dox and compound 1 . ( C ) Proportion of stem versus TA cells following treatment . ( D ) Fold changes in transcript abundance in stem ( D ) and TA ( E ) cells following treatment . All data are mean values ± s . d . , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 01910 . 7554/eLife . 20722 . 020Figure 5—source data 1 . Antibodies and PCR primers used for analysis of mouse intestinal epithelial cells . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 02010 . 7554/eLife . 20722 . 021Figure 5—figure supplement 1 . Analysis of stem and TA cells isolated from the hyperplastic crypts of mice expressing a Dox-inducible activated β-catenin transgene . ( A ) Experimental design . ( B ) FACS approach used to isolate the stem and TA cells . ( C ) Abundance of beta-actin control transcripts following treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 021 High-level induction of beta-catenin with 2 mg/mL Dox significantly increased the percentage of stem cells ( GRP78− ) in the crypt cell compartment , whereas lower-level induction with 0 . 1 mg/mL Dox was associated with a larger transit amplifying ( GRP78+ ) cell population , as previously described ( Figure 5B–C ) ( Hirata et al . , 2013 ) . The complete withdrawal of Dox inhibited WNT signaling in the stem cell population , confirmed by a change in the abundance of transcripts from the WNT-regulated genes Axin2 and Car4 ( Figure 5D–E ) . However , Dox withdrawal did not significantly alter the ratio of stem to TA cells ( Figure 5B–C ) . In contrast , reducing the degree of Dox-induction from 2 to 0 . 1 mg/ml shifted the population towards an increased proportion of TA cells ( Figure 5B–C ) . Consistent with the shift from stem to TA cell distribution , the gene expression pattern in the stem cell fraction was similar to Dox-removal , but was different from the TA cell population . Previously , we demonstrated that treatment with >37 . 5 mg/kg compound 1 was sufficient to inhibit WNT signaling in this model ( Dale et al . , 2015 ) . Here , we found that treatment with 1 at 75 mg/kg x 3 over 24 hr resulted in a gene expression pattern in the stem cell fraction that was similar to both Dox-removal and reduction of Dox from 2 . 0 to 0 . 1 mg/ml ( Figure 5D ) . In contrast , the gene expression in the TA fraction following treatment with 1 was similar to the effect of reducing the level of Dox from 2 . 0 to 0 . 1 mg/ml , but was different from the Dox withdrawal condition ( Figure 5E ) . This observation was consistent with the observation that treatment with 1 resulted in a shift in the population distribution from stem cell to TA similar to that seen by reducing the level of Dox from 2 . 0 to 0 . 1 mg/ml ( Figure 5B–C ) . This implies that the inhibition of CDK8/19 by 1 reduces , rather than eliminates , WNT signaling in the oncogenically-activated stem cell compartment and it is this that alters the proportion of stem cells to proliferative TA cells in the hyperplastic crypt . GSEA also indicated that genes encoding products associated with the bone environment , such as genes regulated by BMP2 , were preferentially affected by compound treatment . Having already determined the effect of 1 on the intestinal crypt stem cell population ( Figure 5 ) , we next investigated the effect of CDK8/19 inhibition on a bone progenitor cell model . We reasoned that , potentially , CDK8/19 inhibition could affect the ability of stem cells in the bone marrow to self-renew through inhibition of the WNT pathway or via a BMP-dependent signaling mechanism that requires SMAD-regulated transcription . SMAD is a transcriptional target of CDK8 ( Alarcón et al . , 2009 ) and both SMADs1-5 and BMP2-regulated gene expression were identified as significant genesets ( Figure 4—source data 1 ) . We treated mouse KS483 osteoprogenitor cells with LGK974 , a Porcupine inhibitor that inhibits WNT signaling , and observed reduced secretion of Procollagen type I N-terminal propeptide ( PINP ) , an organic component of bone , and also reduced deposition of calcium , an inorganic component of bone ( Figure 6A–B ) ( Dang et al . , 2002 ) . In contrast , compound 3 stimulated PINP secretion at low concentrations and inhibited PINP secretion at higher concentrations ( Figure 6A ) , while calcium deposition was inhibited in a concentration-dependent manner ( Figure 6B ) . These data indicate that our lead compound 3 adversely affects bone development in an in vitro model and that its effects are distinct from a specific inhibitor of WNT signaling . 10 . 7554/eLife . 20722 . 022Figure 6 . Effect of CDK8/19 chemical ligands on bone development and the immune response in model systems . Mouse KS483 osteoprogenitor cells were treated with LGK974 ( red ) or compound 3 ( black ) for 13 d and bone matrix formation determined by measuring ( A ) N-terminal propeptide of type I procollagen ( PINP ) and ( B ) calcium , in the external medium ( mean ± s . d . , n = 6 ) . Blue region , level following 50 ng/ml BMP-2 ( positive control ) ; red region , basal level . ( C ) Heat map showing the 10 biomarkers most affected by compound treatment in cell co-culture models . Data are log2 ratios of biomarker levels following compound treatment relative to control ( range: −0 . 874 to 0 . 396 ) . Blue , decreased ratio; red , increased ratio . ( D ) Plasma IL-12 levels in Wistar rats treated with 5 – 20 mg/kg po qd 3 . Two rats per cohort . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 02210 . 7554/eLife . 20722 . 023Figure 6—source data 1 . Culture conditions and data from CDK8/19 ligand profiling in the culture/co-culture cell model panel . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 02310 . 7554/eLife . 20722 . 024Figure 6—figure supplement 1 . Effect of CDK8/19 ligands 1-4 on 12 culture/coculture cell models . ( A ) Log2 ratio of cell model biomarkers plotted relative to levels under control conditions . The gray area indicates the control envelope ( >95% confidence ) . Points outside this area were considered significant . The ten markers consistently altered by all four compounds are indicated on the x-axis . ( B ) Cell types used and ( C ) biology covered by the 12 culture/co-culture models . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 024 Genesets associated with the immune and inflammatory response were selectively affected by our CDK8/19 ligands , prompting us to examine the effects of our compounds on immune and inflammatory responses ( Figure 4—source data 1 ) . In 12 single or co-culture in vitro tissue models we assayed 163 clinically relevant extracellular biomarkers including cytokines , chemokines , membrane receptors , matrix components , and proteases ( Figure 6—figure supplement 1 and Figure 6—source data 1 ) ( Berg et al . , 2010 ) . All four compounds ( 1–4 ) elicited a similar biomarker response across the cell line panel ( median r = 0 . 812; range: 0 . 630 – 0 . 924 ) , indicating that the observed changes were the result of CDK8/19 inhibition rather than off-target effects . Comparing the effects of our compounds with a proprietary database of >3000 approved drugs and experimental agents failed to find any close matches ( Berg et al . , 2010 ) . The levels of interleukin 17A and 17F usually rise and fall together ( Melton et al . , 2013 ) , but treatment with our compounds elicited an unusual split response , with an increase in IL-17A and a corresponding decrease in IL-17F ( BT condition: Figure 6C , Figure 6—figure supplement 1 and Figure 6—source data 1 ) . Overall , the pattern observed across the biomarkers was consistent with the effects of our compounds on inflammation ( VCAM-1 , sPGE2 and IL-8 ) , immunomodulation ( IL-17A , IL-17F , HLA-DR and IgG ) and tissue remodelling ( uPAR ) ( Figure 6C and Figure 6—source data 1 ) . In additional studies in rats treated with 3 we measured the levels of ten plasma cytokines associated with the Th1/Th2 immune response . Of the cytokines measured , only IL-12 , a proinflammatory and proimmunestimulatory cytokine , increased significantly ( p<0 . 005 , Figure 6D ) . Compounds from both chemical series were sufficiently well tolerated in mice to enable antitumor experiments to be conducted with a dosage regimen that resulted in near-maximal inhibition of tumor STAT1SER727 phosphorylation , showing that CDK8/19 activity was repressed for prolonged periods ( Figure 2 , Figure 2—figure supplements 1–3 , Figure 3 , Figure 3—figure supplement 1 . However , there was some evidence for body weight loss that sometimes necessitated short dosing breaks ( Figure 2—source data 1 ) . Given this sporadic body weight loss in mice and our evidence that the compounds had effects on bone development , altered immune/inflammatory profiles and stem cell differentiation in vitro or in vivo experimental models , we carried out detailed toleration studies . Our aim was to determine if there was a therapeutic window for the compounds , using PK/PD and efficacy data determined in our mouse models as a guide . Following daily doses of compound 3 or 4 for 14 days in rats we detected significant , adverse alterations in multiple organ systems and tissues ( Table 1 ) . In these toleration studies , we demonstrated extensive ( 80% ) inhibition of STAT1SER727 phosphorylation 6 hr post-treatment at all doses of 3 tested ( Figure 7—figure supplement 1 ) . The lowest doses administered resulted in plasma concentrations below or equivalent to the plasma exposures achieved at efficacious doses in our experimental human tumor xenograft efficacy models , suggesting the lack of a clear therapeutic window ( Table 1 ) . Consistent with our in vitro observations of the effects on bone maturation , the rat studies revealed two different , paradoxical effects on bone: an inhibitory effect resulting in dysplasia of the growth plate , a decrease in the proliferative zone and false endochondral ossification , and an activating effect resulting in proliferation of irregular woven bone in the bone cavity and below the periosteum ( Table 1 and Figure 7 ) . Other adverse pathomorphological findings included necrotic and apoptotic cell lesions in the exocrine pancreas , gastrointestinal mucosa ( stomach and duodenum ) , male reproductive tract ( Figure 7 ) , mammary gland , skin ( hair follicle ) , heart ( valvular interstitial cells ) and lymphatic tissues ( thymus , spleen , lymph nodes ) . Proliferative lesions were found in the lungs ( bronchiolar epithelium and smooth muscle cells ) , liver ( bile ducts and smooth muscle cells of hepatic arteries ) , thymus ( epithelium free areas ) , mammary gland , male reproductive system and heart ( valvular interstitial cells ) ( Table 1 ) . 10 . 7554/eLife . 20722 . 025Figure 7 . Examples of degenerative and proliferative lesions in rats treated with CDK8/19 ligands 3 or 4 . ( A ) Intact proliferative zone in the bone growth plate of a control rat . ( B ) Dysplastic proliferative zone , showing disturbance of regular endochondrial ossification , from a rat treated with 20 mg/kg 3 . Scale bar in A and B = 100 µm . ( C ) Intact epididymides , with epididymal cells , isolated from a control rat . ( D ) Epididymides with epithelial hyperplasia ( distal corpus ) isolated from a rat treated with 60 mg/kg 4 . Scale bar in C and D = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 02510 . 7554/eLife . 20722 . 026Figure 7—figure supplement 1 . Pharmacodynamic response in Wistar rats treated with CDK8/19 ligand 3 . ( A ) Immunoblot showing p-STAT1SER727 and STAT1 levels in lysates made from rat spleens harvested 2 , 6 or 24 hr after the final dose of 3 . ( B ) Ratio of p-STAT1SER727 to STAT1 in data from A . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 02610 . 7554/eLife . 20722 . 027Table 1 . CDK8/19 ligands 3 and 4 adversely affect multiple organs in rats and dogs . Wistar rats ( 5 male and 5 female per cohort ) or Beagle dogs ( 2 male and 2 female per cohort for 3 and 1 male and 1 female for 4 ) received a daily oral dose of 3 or 4 for 14 days . In the rat study of 4 , all animals were prematurely culled at 60 mg/kg and one male and female at 20 mg/kg , as a result of compound toxicity . In the dog studies , all animals were prematurely culled in the study of 3 and one female following exposure to 4 as a result of toxicity . The most severely affected organs are indicated in bold . The fold efficacious dose was calculated from a plasma PK measurement of compound exposure in satellite animals run in parallel to the tolerability study and compared to exposures at efficacious doses in human tumour xenograft models in mice ( m – male and f – female ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20722 . 027RatDogLow doseMid doseHigh doseLow doseCMPD 3 ( mg/kg ) 510205Target organsBone , bone , marrow , heart , liver , lung , lymph nodes , pancreas , reproductive tract ( m ) , spleen , thymus . Bone , bone marrow , heart , liver , lung , lymph nodes , pancreas , reproductive tract ( m and f ) , spleen , thymus . Bone , bone marrow , heart , liver , lung , lymph nodes , pancreas , reproductive tract ( m and f ) , spleen , thymus . Bone marrow , gastrointestinal mucosa , heart , lymphatic systemFold of efficacious dose; 10 mg/kg~0 . 3 ( m ) – 1 . 3 ( f ) ~0 . 5 ( m ) – 2 ( f ) ~1 ( m ) – 5 ( f ) ~0 . 3 ( m ) – 0 . 3 ( f ) CMPD 4 ( mg/kg ) 10206020Target organsBone , bone marrow , intestines , liver , lung , lymph nodes , mammary gland , pancreas , reproductive tract ( m and f ) , skin , spleen , stomach , thymus . Bone , bone marrow , heart , intestines , liver , lung , lymph nodes , mammary gland , pancreas , reproductive tract ( m and f ) , skin , spleen , stomach , thymusBone , bone marrow , brain , heart , intestines , liver , lung , lymph nodes , mammary gland , pancreas , reproductive tract ( m and f ) , skin , spleen , stomach , thymusBone marrow , heart , Intestines , lymphatic systemFold of efficacious dose; 10 mg/kg 30 mg/kg~0 . 9 ( m ) – 2 . 4 ( f ) ~0 . 3 ( m ) – 0 . 8 ( f ) ~3 . 9 ( m ) – 5 . 7 ( f ) ~1 . 3 ( m ) – 1 . 9 ( f ) ~10 . 8 ( m ) – 23 . 1 ( f ) ~3 . 6 ( m ) – 7 . 7 ( f ) ~22 ( m ) – 46 ( f ) ~7 ( m ) – 15 ( f ) Follow-up studies in dogs indicated a similar , widespread adverse safety profile at therapeutically relevant exposures of 3 and 4 ( Table 1 ) . Since these pathological effects were seen with two highly selective , but structurally distinct CDK8/19 inhibitors in both rats and dogs , we conclude that the adverse effects of treatment are the direct result of inhibition of CDK8 and/or CDK19 . These observations indicate that the clinical development of either series of CDK8/19 inhibitors , or other chemotypes with similar profiles , would be extremely challenging . CDK8 may reportedly act both as an oncogene and as a tumor suppressor , but until recently , the absence of a potent and selective inhibitor of CDK8 has restricted many functional studies to genetic inhibition using shRNA or siRNA ( Mitra et al . , 2006; Chattopadhyay et al . , 2010; Gu et al . , 2013; Firestein et al . , 2008 , 2010; Seo et al . , 2010; Adler et al . , 2012; Starr et al . , 2009 ) . The potential difference between loss of protein and inhibition of enzymatic activity is highlighted by a recent study with the natural product CDK8/19 inhibitor cortistatin A that found a treatment-induced gene expression profile distinct from the profile resulting from CDK8/19 shRNA knockouts in the same cell line ( Poss et al . , 2016 ) . Also conscious of the potentially opposing context-dependent roles of CDK8 in tumor development , we set out to clarify the therapeutic potential of targeting CDK8/19 . Using two structurally-distinct series of potent and highly selective ligands that we discovered , we explored the consequences of CDK8/19 targeting in vitro and in vivo and investigated tolerability to determine if there was a therapeutic window ( Mallinger et al . , 2015 , 2016a; Czodrowski et al . , 2016 ) . Despite reproducible inhibition of TCF/LEF reporter activity and STAT1SER727 phosphorylation by both series of compounds , we were only able to detect modest , though generally significant , antiproliferative or antitumor effects in vitro or in vivo . In follow-up studies on cell cultures derived from PDX tumors , and selected PDXs in vivo , we found that compounds from the 3 , 4 , 5-trisubstituted pyridine series had only modest effects in vitro and little effect in vivo despite maximal inhibition of STAT1SER727 phosphorylation . The pleiotropic nature of CDK8/19 function , influencing the activity of multiple specific transcription factors and also super-enhancers , may make identification of biomarkers for particular cancer cell types that are especially sensitive to CDK8/19 inhibitors challenging; despite this both series of compounds did show greater potency in a systemic or subcutaneous model of human AML , similar to that reported for the natural product inhibitor of CDK8/19 , cortistatin A ( Pelish et al . , 2015 ) . We found that the in vivo activity of our CDK8/19 inhibitory compounds was associated with modulation of gene expression regulated by transcription factors that are CDK8 substrates . Moreover , Pelish and colleagues ( Pelish et al . , 2015 ) recently demonstrated that in AML cells CDK8 is associated with gene super-enhancers and that pharmacological inhibition of CDK8 activated super-enhancer output . Here , we also found that pharmacological inhibition of CDK8 with our potent and selective chemical probes in colorectal cancer models resulted in gene expression profiles consistent with increased super-enhancer activity . Genes whose expression were altered by the compounds included those encoding products associated with bone development , stem cell biology , immunology and inflammation . Extensive follow up experiments in in vitro and in vivo models demonstrated effects on bone , stem cell differentiation and response of immune cells to different stimuli . Some of the effects were unique , for example concentration-dependent stimulation or inhibition of bone matrix production by osteoprogenitor cells in vitro that was not observed with a WNT-pathway inhibitor . CDK8 activity maintains embryonic stem cells in an undifferentiated , pluripotent state and colorectal tumors in a de-differentiated state ( Adler et al . , 2012 ) . We found that complete inhibition of CDK8/19 in the presence of activated oncogenic beta-catenin mimicked the effect of reducing , rather than completely abrogating , WNT-signaling . CDK8/19 inhibition resulted in a shift from a stem cell to a predominantly TA cell phenotype . This response may also be linked to super-enhancer activation , as expression of Myc , a super-enhancer-regulated gene ( Lovén et al . , 2013 ) , was repressed in the stem cell population , but elevated in the TA cell population following compound treatment . The example of Myc illustrates the potential complexity of the response to CDK8 inhibition , as Myc expression will potentially be repressed through loss of CDK8 activity required by beta-catenin , but Myc expression may also be promoted through super-enhancer activation . This suggests that cellular context will have a major impact on the transcriptional response to CDK8 inhibition . The final key aim of our studies was to investigate tolerability to CDK8/19 inhibition and identify a possible potential therapeutic window for compounds 3 and 4 . In rat and dog tolerability studies , we found that 3 and 4 produced unusually extensive , but similar , adverse effects in a wide range of tissues . Given our observations , that are consistent with the reported role of CDK8 in repressing super-enhancer activity ( Pelish et al . , 2015 ) , and the potentially key roles of super-enhancers as master controllers of cell identity and function ( Whyte et al . , 2013 ) , the breadth , degree and depth of adverse effects is perhaps not surprising . We detected elevation of the mainly proinflammatory and proimmunestimulatory associated cytokine , IL-12 , in rats treated with 3 . Elevated IL-12 may result from decreased STAT1SER727 following CDK8 inhibition , as bone marrow macrophages from STAT1S727A mutant mice exhibit elevated IL-12 and Cox-2 expression following activation ( Schroder et al . , 2007 ) . Cox-2 induction is associated with an increased production of PGE2 , which we detected in vitro , that may also explain the unusual split response of increased IL-17A and decreased IL-17F observed in our in vitro experiments , since PGE2 , or agonists of the EP receptor for PGE2 are among the few stimuli that induce this split effect ( Melton et al . , 2013 ) . We concluded that the multiplicity of preclinical pathomorphological lesions would make monitoring and controlling toxicity in a clinical study very challenging , especially as no clear safety window could be established from our studies . Our data strongly suggest that the adverse effects are target related as they are detected with two chemically distinct series of potent and selective CDK8/19 ligands and indicate that dual pharmacological CDK8/19 inhibition is not tolerated in rats or dogs at exposure levels that correspond to therapeutically relevant exposures . Even in the more sensitive xenograft models of AML , there was no clear therapeutic window , as these tumors responded only when STAT1SER727 phosphorylation was continuously inhibited and equivalent exposures were not tolerated in our in vivo rat and dog toleration studies . Protein kinase inhibitors in clinical development or approved often have off-target activity or in some cases can be intentionally multi-targeted to have inhibitory activity against multiple protein kinases . Our observations are important for both scenarios as we identify CDK8/19 as potential ‘anti-targets’ to be avoided and we recommend screening against these protein kinases when establishing the safety profile of lead compounds and development candidates as well as assessing the quality of chemical probes . As described earlier , the natural product cortistatin A was recently reported by Pelish and colleagues as a CDK8/19 inhibitor with specificity , potency , favourable pharmacokinetics that would make it a useful in vitro and in vivo probe for the Mediator kinases and as a promising lead for development of therapeutics ( Pelish et al . , 2015 ) . Here , we employed two different series of potent and selective CDK8/19 inhibitors , with paired negative controls , that are much less challenging to synthesise compared to cortistatin A ( Pelish et al . , 2015 ) . Our synthetic compounds have optimal pharmacological and pharmaceutical properties with single digit nM affinities for CDK8/19 and low nanomolar activity against promoter and STAT1SER727 reporter assays and are as potent as cortistatin A , as well as exhibiting very high selectivity in broad kinome profiling . As described by Pelish and colleagues for cortistatin A , we found similar induction of super-enhancer-regulated genes with our compounds , although note that in our case we profiled gene expression in tumours treated in vivo rather than in vitro ( Pelish et al . , 2015 ) . Similar to cortistatin A our compounds were also tolerated in mice and likewise we also found AMLs to be highly sensitive in vivo models ( Pelish et al . , 2015 ) . However , in our study we also evaluated our two series in dedicated tolerability studies in rat and dog that revealed toxicity not apparent in mouse studies . The detailed toxicity profile of cortistatin A was not reported by Pelish and colleagues , but given the similarities in the results from in vitro and in vivo studies between cortistatin A and our two series we predict that cortistatin A and other CDK8/19 inhibitors would exhibit similar toxicity . It remains to be seen if toxicity could be avoided if CDK8/19 inhibitors were administered intermittently as part of a combination therapy . For example , CDK8/19 inhibitors might modulate antitumor immunotherapy by inactivating STAT1 and stimulating tumor surveillance by NK cells ( Putz et al . , 2013 ) . It is also unclear if toxicity could be avoided by selectively targeting CDK8 or CDK19 alone . During the optimisation of both the 3 , 4 , 5-trisubstituted pyridine and 3-methyl-1H-pyrazolo[3 , 4-b]pyridine series , we were unable to separate CDK8 from CDK19 affinity ( Mallinger et al . , 2015 , 2016a; Czodrowski et al . , 2016 ) . This is also true for two additional distinct chemical series that we identified ( Schiemann et al . , 2016; Mallinger et al . , 2016b ) , and also for a further three chemotypes we profiled from the literature , all of which demonstrated balanced CDK8 and CDK19 affinity in our hands . Similarly , the natural product cortistatin A cannot selectively distinguish between CDK8 and 19 ( Pelish et al . , 2015 ) . The inability to selectively target CDK8 or CDK19 is likely due to the very high degree of sequence similarity around the active site of these two kinases ( Figure 1c ) , suggesting that strategies to selectively target either of these paralogs will require a different approach , such as allosteric modulation or selective degradation . The recent identification of selective inhibitors of the transcriptional kinases CDK7 or CDK12/13 has fuelled interest in the clinical development of inhibitors of these targets ( Kwiatkowski et al . , 2014; Wang et al . , 2015; Zhang et al . , 2016; Christensen et al . , 2014; Chipumuro et al . , 2014 ) . This is of relevance here as inhibition of these kinases has also been reported to affect super-enhancer-associated gene expression in T cell acute lymphoblastic leukaemia , B cell chronic lymphocytic leukaemia , MYCN-driven tumours , small cell lung cancer and triple negative breast cancer models where anti-tumour activity is observed ( Kwiatkowski et al . , 2014; Wang et al . , 2015; Zhang et al . , 2016; Christensen et al . , 2014; Chipumuro et al . , 2014 ) . In adult CDK7 conditional knockout mice , effects were seen in tissues with a high cell turnover that were at least in part due to a depleted stem cell population resulting from a loss of CDK1 and CDK2 activation or reduced super-enhancer-associated gene expression ( Ganuza et al . , 2012 ) . Super-enhancers are frequently found associated with genes whose products control the pluripotent state or define cell identity and this may make stem cell populations particularly vulnerable to therapeutic interventions that interfere with super-enhancer associated gene expression , including inhibitors of the transcription-regulating CDKs ( Whyte et al . , 2013 ) . In an in vivo mouse model of an oncogenically-activated stem cell compartment we found our CDK8/19 inhibitors altered the proportion of stem cells to proliferative TA cells that may in part be due to super-enhancer activation . This raises the hypothetical possibility that , similar to AML cells ( Pelish et al . , 2015 ) , the stem cell compartment requires a precise ‘dosage’ of super-enhancer activity and that activation or inhibition of super-enhancer activity will negatively impact the stem cell compartment . The selective CDK7 tool inhibitors are reported to be tolerated in mice and so predicted to be non-toxic ( Kwiatkowski et al . , 2014; Wang et al . , 2015; Christensen et al . , 2014; Chipumuro et al . , 2014 ) ; however , our CDK8/19 inhibitors were also tolerated in mice and the toxicity was not revealed until detailed tolerability studies were performed in other species . Thus the true toxicity profile of other transcriptional CDK inhibitors may not be revealed until dedicated tolerability studies are performed using selective compounds with optimised pharmaceutical properties . In summary , we have discovered and made available two chemically distinct series of potent selective chemical probes and appropriate inactive control compounds that can be used to further explore the function of Mediator-associated kinases CDK8/19 and their role in human disease both in vitro and in vivo . These compounds will also be of particular value for exploring the regulation of super-enhancer activity in development and disease . However , on the basis of the complex toxicological profile and an inability to define a clear therapeutic window , we have decided against the further clinical development of our compounds and suggest caution when considering the clinical applicability of other CDK8/19 inhibitors . We also advise incorporating the profiling CDK8/19 as anti-targets in drug discovery and chemical probe projects aimed at other kinases . Compounds were prepared as described ( Mallinger et al . , 2015 , 2016a; Czodrowski et al . , 2016 ) or resynthesised by published routes . In vitro binding of compounds to CDK8 was determined using FRET-based Lanthascreen binding competition with a dye-labeled ATP competitive tracer assay . Alternatively , we used a reporter displacement assay provided by Proteros Biostructures GmbH ( Germany ) for CDK8 or CDK19 as described previously ( Mallinger et al . , 2016a ) . The human CDK8–CCNC complex was expressed , purified and crystallized as described previously ( Dale et al . , 2015 ) . Crystals were back-soaked for different times and concentrations of ligand before being selected for structure determination ( Dale et al . , 2015 ) . Only authenticated and mycoplasma-free cell lines were used in this study . All cancer cell lines used in this study ( COLO205 - RRID:CVCL_0218; DLD1 - RRID:CVCL_0248; HT29 - RRID:CVCL_0320; LS174T - RRID:CVCL_1384; LS180 - RRID:CVCL_0397; LS513 - RRID:CVCL_1386; RKO - RRID:CVCL_0504; SW620 - RRID:CVCL_0547; SW837 - RRID:CVCL_1729; SW948 - RRID:CVCL_0632 ) were obtained from the ATCC ( LGC Promochem , UK ) , were regularly tested and confirmed as mycoplasma-free ( Lonza , UK ) and were authenticated by short tandem repeat ( STR ) analysis profiling . Multiplex amplification of genomic loci Penta E , D18S51 , D21S11 , TH01 , D3S1358 , FGA , TPOX , D8S1179 , vWA , Amelogenin , Penta D , CSF1PO , D16S539 , D7S820 , D13S317 , and D5S818 is performed using a PowerPlex 16 HS ( Promega , Madison , Wisconsin , USA ) . STR sequences were analyzed on an Applied Biosystems 3500xL Genetic Analyzer ( Thermo Fisher Scientific Life Technologies , UK ) and compared to different cell line reference databases . Soft agar PDX cell cultures were run at Oncotest ( Germany ) . Cell lines transduced with a TCF/LEF fLUC reporter were generated and assayed as described previously ( Mallinger et al . , 2015; Dale et al . , 2015 ) . Inducible shRNA knockout models were established in the Colo205-F1921 subline carrying the TCF/LEF fLUC reporter have been described previously ( Dale et al . , 2015 ) . HT29 cells ( 5 × 104/ml ) were reverse transfected in 6 well plates with 3 . 75 μl/ml LipofectamineRNAiMAX ( ThermoFisher Scientific , UK ) in a 2 ml final volume with 50 nM pooled siRNA targeting CDK8 , CDK19 or a non-silencing control siRNA ( Quiagen cat . GS1024 , GS23097 and 1027280 respectively ) . Viability was determined by resazurin staining ( R&D systems , UK ) for 30 min 5 d post-transfection . Bone formation assays were conducted by Pharmatest ( Finland ) in mouse KS483 osteoprogenitor cells cultured for 13 d with LGK974 or 3 . The amount of PINP secreted into the culture medium , and calcium deposition , were measured by ELISA ( Roche Diagnostics , UK ) . Seven different primary cells types were exposed to different stimuli as single or co-cultures and the response of relevant extracellular biomarkers assayed . Biomarker assays were run at DiscoveRx ( Fremont , California , USA ) ( Figure 6—source data 1 ) . Cultures were treated with DMSO ( vehicle ) or a dilution series of 1 – 10000 nM of test compounds . Profiles were compared to a proprietary database of compounds previously profiled in this system ( Berg et al . , 2010 ) . In the UK , all animal work was conducted in accordance with the National Institute for Cancer Research guidelines ( Workman et al . , 2010 ) , with the research programme and procedures approved by the local Animal Welfare and Ethical Review Boards and subject to UK Government Home Office regulations ( Licence PPL 70/7635 & PPL 30/3279 ) . In Germany the animal work was carried out in accordance with the German Law on the Protection of Animals ( Article 8a ) and the pertaining files at the local animal welfare authorities in Darmstadt and Freiburg bear the references DA/375 , DA4/1003 , DA4/1004 and G13/13 respectively . The studies were designed in accordance with presently valid international study guidelines ( e . g . ICH guideline M3 R2 ) and performed in compliance with animal health and welfare guidelines . Total RNA was extracted from xenograft tumors using a MagNa Pure 96 high-throughput robotic workstation ( Roche Diagnostics , UK ) and analyzed by microarray expression profiling ( Dale et al . , 2015 ) . Purified , labeled cDNA products were hybridized to 8 × 60K human microarrays ( Agilent ) and analyzed using Genespring ( Agilent Santa Clara , California , USA , RRID:SCR_009196 ) . Significantly differentially expressed genes were investigated for enrichment in terms of particular pathways or potential transcription factor regulation using the Metacore software ( Thomson Reuters , New York City , New York , USA , RRID:SCR_008125 ) . Lists of super-enhancer-associated genes were derived from published ChIP-seq datasets ( dbSUPER; http://bioinfo . au . tsinghua . edu . cn/dbsuper/ and Pelish et al . , 2015 ) . GSEA was performed on the super-enhancer gene lists and also genesets from MSigDB ( www . broadinstitute . org/msigdb ) . The links for the signatures used for the GSEA software are available in the Datasets section . Users are required to register to view the MSigDB gene sets and/or download the GSEA software . Microarray data are available on the NCBI Gene Expression Omnibus ( GEO; http://www . ncbi . nlm . nih . gov/geo/ ) website under accession number GSE80472 . Levels of p-STAT1SER727 and total STAT1 were quantified from cell or tumor lysates by immunoblotting , luminex or electrochemiluminescent ELISA as previously described in detail in ( Mallinger et al . , 2015; Dale et al . , 2015; Mallinger et al . , 2016a; Czodrowski et al . , 2016 ) . Proteins were also detected using an automated capillary immunoassay system ( Protein Simple ) with antibodies specific for CDK8 ( Cell Signaling , Danvers Massachusetts , USA #4106 , RRID:AB_1903936 ) , CDK19 ( Sigma-Aldrich , UK , HPA007053 , RRID:AB_1233803 ) phospho-STAT1SER727 ( Cell Signaling , #8826 ) , total STAT1 ( Santa-Cruz Biotechnology , Dallas , Texas , USA , #346 , RRID:AB_632435 ) , and B-actin ( Cell Signalling , #4970 , RRID:AB_2223172 ) , subsequently immunodetected using a horseradish peroxidase ( HRP ) -conjugated secondary antibody and chemiluminescent substrate . Lithium-heparin-plasma ( 120 µL ) was taken from each non-fasted animal on day 1 pre-dose and on days 3 and 14 pre-dose and also 2 hr after the last treatment . The rat Th1/Th2 multiPlex bead immunoassay panel ( Invitrogen ) was used to generate calibration curves and to measure cytokine ( IL-1α , IL-1β , IL-2 , IL-4 , IL-6 , IL-10 , IL-12 , IFNγ , gmCSF and TNFα ) levels ( Luminex , Austin , Texas , USA ) .	Healthy cells in the human body can become cancerous if they gain genetic mutations that allow them to rapidly grow and divide . Some types of cancer respond better to drug treatments than others and tumors often develop resistance to a particular drug treatment after a while . Because of this , researchers are always searching for new molecules to develop into anticancer drugs . Recently , a team of researchers identified some small molecules that could inactivate two closely related proteins called CDK8 and CDK19 . CDK8 is essential for the WNT signaling pathway – which enables cells to communicate with one another – and has been extensively studied in various cancers . Previous studies indicate that this protein can either promote or inhibit the growth of tumors , depending on the type and stage of the cancer . Furthermore , CDK8 regulates a type of molecular switch called a “super-enhancer” , which controls the activity of many genes . In contrast , the role of CDK19 in cells was not as well understood . Here Clarke , Ortiz-Ruiz et al . investigated whether two different classes of small molecules that target CDK8 and CDK19 ( referred to as “prototype CDK8/19 drugs” ) could inhibit the growth of cancers , and whether they have any harmful side effects on healthy cells . For the experiments , human cancer cells were implanted into mice . Treating these mice with prototype CDK8/19 drugs inhibited the activity of CDK8 and CDK19 in the cancer cells and slowed the growth of colorectal tumors . A type of blood cancer called acute myeloid leukaemia was particularly sensitive to the drugs . However , Clarke , Ortiz-Ruiz et al . also observed that the prototype drugs altered the activity of many genes with roles in healthy tissues such as immune , bone and stem cells . Further experiments in mice and cells grown in the laboratory confirmed that these prototype drugs have adverse effects on healthy intestinal and bone marrow stem cells and trigger changes to immune cells . These concerning side effects were also evident when the prototype drugs were tested in rats and dogs . Furthermore , the experiments indicate that there is not a suitable range of doses of these drugs in which the therapeutic benefits outweigh the toxic side effects . Clarke , Ortiz-Ruiz et al . conclude that the clinical development of CDK8/19 drugs will be extremely challenging and that their prototype drugs would not currently be suitable for use as cancer treatments . However , the small molecules they describe will be important probes in research to study exactly how CDK8/19 regulate gene activity in both healthy cells and cancers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2016	Assessing the mechanism and therapeutic potential of modulators of the human Mediator complex-associated protein kinases
APC , a key negative regulator of Wnt signaling in development and oncogenesis , acts in the destruction complex with the scaffold Axin and the kinases GSK3 and CK1 to target βcatenin for destruction . Despite 20 years of research , APC's mechanistic function remains mysterious . We used FRAP , super-resolution microscopy , functional tests in mammalian cells and flies , and other approaches to define APC's mechanistic role in the active destruction complex when Wnt signaling is off . Our data suggest APC plays two roles: ( 1 ) APC promotes efficient Axin multimerization through one known and one novel APC:Axin interaction site , and ( 2 ) GSK3 acts through APC motifs R2 and B to regulate APC:Axin interactions , promoting high-throughput of βcatenin to destruction . We propose a new dynamic model of how the destruction complex regulates Wnt signaling and how this goes wrong in cancer , providing insights into how this multiprotein signaling complex is assembled and functions via multivalent interactions . The Wnt signaling pathway is one of the most critical in animals , controlling both cell proliferation and fate ( Clevers and Nusse , 2012 ) . Deregulated Wnt signaling plays roles in several cancers ( Polakis , 2007; Kandoth et al . , 2013 ) ; most strikingly , mutations in Adenomatous polyposis coli ( APC ) , a key negative regulator of Wnt signaling , initiate ∼80% of colon cancers . Although APC was identified in 1991 , its mechanistic role in Wnt regulation remains mysterious . Wnt signaling regulates levels of the transcriptional co-activator βcatenin ( βcat; Clevers and Nusse , 2012 ) . Like other powerful signaling pathways driving development and oncogenesis , animals evolved dedicated machinery to keep the Wnt pathway off in the absence of signal . In the OFF-state , βcat levels are kept low by the action of the multiprotein destruction complex , which includes APC ( Figure 1A ) , the scaffold Axin ( Figure 1A ) and the kinases GSK3 and CK1 . In the current model , βcat is recruited into the destruction complex by binding APC or Axin and sequentially phosphorylated by CK1 and GSK3 . This creates a binding site for the E3-Ligase SCFβTrCP and βcat is ubiquitinated and destroyed . Wnt ligand binding to the Wnt receptor inhibits the destruction complex , through a series of events whose order and relative importance remains unclear ( MacDonald and He , 2012 ) . βcat levels rise and it activates Wnt target genes . 10 . 7554/eLife . 08022 . 003Figure 1 . APC2's Arm rpts provide a second means of interacting with the Axin complex . ( A ) Fly APC2 and Axin . ( B ) Constructs used . hAPC1-1338 = the endogenous truncated hAPC1 in SW480 cells . ( C and D ) SW480 cells coexpressing GFP-APC2Arm rpts only and Axin-RFP , which localize adjacent to one another ( arrows ) . ( D ) Insets = box in ( C ) . ( E ) Known and novel APC:Axin interaction sites ( top ) and Axin constructs ( bottom ) . ( F and G ) IPs from SW480 cells . ( F ) APC's Arm rpts coIP with Axin's middle region that contains the GSK3 binding site . Axin's DIX domain was weakly detected . The βcat binding site fragment was not detected in either immunoblots or immunofluorescence ( not shown ) , suggesting rapid degradation . ( G ) IP of endogenous hAxin1 or control Ig . Truncated endogenous hAPC1-1338 coIPs with hAxin1 at endogenous levels . ( H ) GFP-hAPC1Arm rpts only and hAxin1-RFP colocalize in puncta ( arrows ) . ( I ) Insets = box in ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 00310 . 7554/eLife . 08022 . 004Figure 1—figure supplement 1 . Assessing levels of over-expression of Axin and APC . ( A ) When overexpressed in SW480 cells , fly Axin forms puncta . ( B ) Plot of immunofluorescence intensities in SW480 cells transfected with GFP-APC2 , or Axin-RFP , or GFP-APC2 + Axin–RFP , and stained for βcat via antibody . βcat intensities of transfected cells were normalized to adjacent untransfected cells and plotted against the GFP , or RFP , or sum of GFP and RFP intensities . 10 cells were measured each time in 3 independent experiments . ( C ) Measuring the levels of Axin overexpression in SW480 cells . Immunoblot analysis of cells transfected with human Axin1-GFP ( hAxin1-GFP ) or fly Axin-GFP with the indicated antibodies . γ-tubulin was used as a loading control . ( C1 , C2 ) Equal volumes of cell lysate were loaded . ( C3 ) 10% the amount of hAxin1-GFP lysate was loaded . ( C4 ) 1% of the amount of hAxin1-GFP lysate was loaded . One representative immunoblot of 3 independent experiments . Details of calculations used are in the Results and Materials and methods—full data is in Table 1 . ( D ) Measuring the levels of APC2 overexpression in SW480 cells . Immunoblot analysis of cells transfected with Flag-tagged truncated human APC1-1338 ( see Figure 1B ) or fly APC2 with the indicated antibodies . γ-Tubulin was used as a loading control . ( D1 , D2 ) Equal volumes of cell lysate were loaded . ( D3 ) 10% the amount of Flag-flyAPC2 lysate loaded . ( D4 ) 10% the amount of Flag hAPC1-1338 lysate was loaded . One representative immunoblot of 3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 00410 . 7554/eLife . 08022 . 005Figure 1—figure supplement 2 . Human APC's Arm repeat domain colocalizes with Axin . ( A ) Diagrams of fly APC2 and APC2Arm rpts only . ( B ) SW480 cells expressing GFP-APC2Arm rpts only . APC2Arm rpts only forms cytoplasmic puncta and is unable to reduce βcat levels ( insets = box in B ) . ( C ) Diagrams of hAPC1 and hAPC1Arm rpts only mutant . ( D ) GFP-hAPC1Arm rpts only expressed in SW480 cells . hAPC1Arm rpts also forms cytoplasmic puncta and is unable to reduce βcat levels ( insets = box in D ) . ( E ) GFP-hAxin1 expressed in SW480 cells forms cytoplasmic puncta , and does reduce βcat levels , but βcat accumulates in the puncta ( insets = box in D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 005 Despite this attractive model , several key issues remain . APC and Axin are destruction complex core components . Both are protein interaction hubs , combining folded domains with intrinsically disordered regions carrying peptide motifs that bind protein partners . APC was initially thought to be the scaffold templating βcat phosphorylation but subsequent work revealed that Axin plays this role ( Hart et al . , 1998; Ha et al . , 2004 ) . Axin has binding sites for βcat , APC , GSK3 , CK1 and PP2A , and self-polymerizes via its DIX domain ( Figure 1A ) , which allows it to locally increase the concentration of proteins mediating βcat phosphorylation ( Ikeda et al . , 1998; Fagotto et al . , 1999; Liu et al . , 2002; Fiedler et al . , 2011 ) . Thus while APC is essential for destruction complex function , its role in the complex remained a mystery . APC combines an N-terminal Armadillo repeat domain ( Arm rpts ) with a more C-terminal intrinsically disordered region; each contains binding sites for multiple partners ( Figure 1A ) . The Arm rpts bind cytoskeletal regulators ( Nelson and Nathke , 2013 ) , and also help regulate Wnt signaling but the relevant binding partner ( s ) for this is unclear ( Roberts et al . , 2012 ) . APC's SAMP repeats bind Axin's RGS region , helping mediate destruction complex assembly . Both the Arm rpts and SAMPs are essential for APC function . APC also has multiple βcat binding sites , the 15 and 20 amino acid repeats ( 15Rs and 20Rs ) , which have different affinities for βcat , and act redundantly to sequester βcat and fine tune Wnt signaling ( Liu et al . , 2006; Roberts et al . , 2011 ) . However , surprisingly , these binding sites are not essential for APC's mechanistic role in βcat destruction ( Yamulla et al . , 2014 ) . Recent work led to significant insights into how Wnt signaling turns off the destruction complex . Axin turnover and Dvl-Axin hetero-polymerization may inhibit destruction complex function by disassembling the destruction complex ( Schwarz-Romond et al . , 2007; Fiedler et al . , 2011 ) . However , recent studies revealed that the destruction complex remains intact for some time after Wnt signals are received , and can continue to phosphorylate βcat ( Hernandez et al . , 2012; Li et al . , 2012b; Kim et al . , 2013 ) . Thus these studies suggest that Wnt signaling leads to either reduction in the rate of βcat phosphorylation by the destruction complex , or that it inhibits a key regulated step after phosphorylation , perhaps transfer to the E3 ligase , thus turning down an intact destruction complex , leading to high levels of βcat . In contrast , less attention has been devoted to the active destruction complex when Wnt signaling is off , the state most relevant to what happens in colon tumors . Strikingly , these tumors lack wild-type APC but instead invariably express a truncated APC protein retaining some but not all functional domains—the reason for this remains controversial ( Albuquerque et al . , 2002 ) . Further , while tumor cells do not effectively target βcat for destruction , it is less clear at what step destruction is blocked by APC truncation . In the simplest model , mutant destruction complexes wouldn't template βcat phosphorylation , but Axin's ability to do this in vitro in APC's absence cast doubt on this ( Ha et al . , 2004 ) . Further , cancer cells with truncated APC have high levels of phosphorylated βcat , suggesting that the step blocked is after βcat phosphorylation ( Yang et al . , 2006; Kohler et al . , 2008 ) . One intriguing study proposed that APC acts in the destruction complex to protect βcat from dephosphorylation and then targets it to the E3 ligase ( Su et al . , 2008 ) . However phospho-βcat is elevated in APC-mutant cell lines , despite loss of APC function suggesting that protecting βcat from dephosphorylation is not APC's only role in the destruction complex ( Yang et al . , 2006 ) . Thus APC's key mechanistic function in the destruction complex remains largely unknown . In exploring APC's mechanism of action when Wnt signaling is inactive , we recently focused on two conserved binding sites in APC , 20R2 ( R2 ) and motif B ( B; this is also known as the catenin interaction domain = CID; Kohler et al . , 2009; Roberts et al . , 2011; Schneikert et al . , 2014; Choi et al . , 2013; Figure 1A ) . R2 is related to the 20R βcat binding sites , but lacks a key interacting residue and cannot bind βcat ( Liu et al . , 2006 ) . B is immediately adjacent to R2—its function was initially unknown , though it was recently shown to bind α-catenin , and thus play a role in Wnt regulation ( Choi et al . , 2013 ) . Strikingly , although other 20Rs are individually dispensable , both R2 and B are essential for the destruction complex to target βcat for destruction ( Kohler et al . , 2009; Roberts et al . , 2011; Schneikert et al . , 2014 ) . Our data further suggested that R2/B negatively regulates APC/Axin interaction , a somewhat surprising role for an essential part of the destruction complex . This prompted us to broaden our analysis . Most destruction complex models , even those that consider the kinetics of initial destruction complex assembly ( Lee et al . , 2003 ) , portray it as a static entity that binds , phosphorylates and hands off βcat . Recent work prompted us to consider an alternative hypothesis , viewing the destruction complex as a complex multiprotein entity whose assembly , structure and dynamics are key to its function in maintaining low βcat levels when Wnt signaling is off . To test this hypothesis , we used FRAP , super-resolution microscopy and other approaches to explore the structure and dynamics of the destruction complex . This provided new insights into APC's mechanism of action , providing evidence that it plays two roles inside the active destruction complex: ( 1 ) APC promotes efficiency of the destruction complex by enhancing complex assembly through two separate APC:Axin interaction sites , one of which is novel , ( 2 ) the novel APC:Axin interaction is dynamic , and regulation of this interaction by GSK3 acting via motifs R2 and B is essential to send phosphorylated βcat to destruction . More broadly , our data also provide insights into how intrinsically disordered regions assist in the assembly and dynamics of multiprotein signaling complexes . Our goal is to define APC's mechanistic role in βcat regulation when Wnt signaling is off . In recent work , we discovered that two motifs in APC's intrinsically unstructured region , R2 and B , are essential for promoting βcat destruction in human cells and in Drosophila ( Roberts et al . , 2011 ) . Here we sought to define the mechanism by which these motifs and APC itself act , using as a model SW480 colon cancer cells . These cells have high βcat levels , as they lack wildtype human APC1 ( hAPC1 ) and instead express a truncated APC1 protein ending before the mutation cluster region ( MCR; Figure 1B ) . SW480 cells also express human APC2 ( Maher et al . , 2009 ) , but this is not sufficient to help mediate βcat destruction . We express in these cells fly APC2 , the homolog of hAPC1 , a full length APC that shares all conserved regions important for Wnt regulation with hAPC1 but is significantly smaller in size ( Figure 1B ) . Fly APC2 effectively reduces βcat levels in SW480 cells ( Roberts et al . , 2011 ) , and thus can interact with all human destruction complex proteins needed to target βcat for degradation . There is abundant evidence that the functional destruction complex is a multimer of the individual destruction complex proteins . One important underpinning of this idea is that Axin oligimerizes via self-polymerization of the DIX domain , and this multimerization is critical for its Wnt regulatory function ( Kishida et al . , 1999; Schwarz-Romond et al . , 2007 ) . Endogenous Axin forms small puncta in cultured cells and when overexpressed these puncta become more prominent , in a DIX-domain dependent fashion ( Fagotto et al . , 1999; Faux et al . , 2008; Figure 1—figure supplement 1A ) . The level of Axin over-expression needed to trigger more prominent Axin puncta is not dramatic—for example , treatment of SW480 cells with tankyrase inhibitors increased levels of AXIN1 3-5x and AXIN2 5-20x and this was sufficient to trigger formation of Axin puncta ( de la Roche et al . , 2014 ) . Axin puncta are dynamic multiprotein complexes that can recruit APC and other destruction complex proteins , and previous data from many labs are consistent with the idea that the puncta can serve as useful models of the smaller endogenous destruction complexes , based on correlations between puncta formation , dynamics , and function in βcat destruction ( e . g . Faux et al . , 2008; Fiedler et al . , 2011 ) . We and others previously identified the key structural domains of APC and Axin that are essential for destruction complex function and βcat destruction ( e . g . Roberts et al . , 2011 ) . Our current goal was to define how these proteins' domains function together to facilitate APC and the destruction complex's mechanisms of action . To do so , we used the APC:Axin puncta formed in SW480 cells as a visible and thus measurable read-out to study mechanisms underlying destruction complex structure , assembly , dynamics and function . Our experiments and those of many earlier investigators used transfected human or in our case fly proteins to study Wnt signaling in cultured mammalian cells ( e . g . , Bilic et al . , 2007; Fiedler et al . , 2011; Kim et al . , 2013 ) . This strategy likely leads to both variable expression levels between cells and elevated expression relative to endogenous protein . We first investigated variation from cell to cell within a transfection , by quantitating whole cell fluorescence of GFP- or RFP-tagged APC2 or Axin and investigating whether different levels of Axin or APC2 expression altered the ability to down-regulate βcat levels ( Figure 1—figure supplement 1B ) . There was a substantial range of APC2 or Axin expression levels among cells ( 5- to 10-fold ) . Importantly , βcat levels were substantially reduced at all levels of APC2 or Axin expression , even the lowest levels assessed—this was true for Axin alone , APC2 alone , or Axin plus APC2 ( Figure 1—figure supplement 1B ) . In all cases , ability to reduce βcat levels was somewhat diminished at the highest levels of expression ( Figure 1—figure supplement 1B ) —this may be because at very high expression levels , the transfected protein forms non-functional complexes with only a subset of the destruction complex proteins , as was previously suggested ( Lee et al . , 2003 ) . We next used immunoblotting to get order of magnitude estimates for the level of expression of our transfected constructs relative to the endogenous proteins . We describe the procedure used in detail in the Materials and methods . We began with Axin , to determine the level of over-expression of fly Axin vs endogenous human Axin . Since fly Axin is not recognized by hAxin1 antibodies we did this in two steps , first comparing levels of tagged human Axin1 vs Drosophila Axin using antibodies against the epitope tag , and then comparing the levels of the transfected human Axin1 vs the endogenous Axin1 protein in SW480 , using antibodies against human Axin1 ( Figure 1—figure supplement 1C ) . We then used these ratios and the transfection efficiencies to calculate the average ratio of Drosophila Axin:endogenous hAxin1 . Transfected Axin accumulated at roughly 80- to 120-fold that of endogenous hAxin1 ( Table 1 ) . For APC , estimating ‘overexpression’ was more problematic , as SW480 cells do not accumulate wild-type APC1—instead they accumulate a truncated APC1 ending at amino acid 1338 ( Figure 1B ) . We thus used a similar two-step process , comparing levels of tagged human APC1 cloned so as to mimic the truncated APC1 seen in SW480 cells vs tagged Drosophila APC2 using antibodies to the epitope , and then levels of tagged truncated human APC1 vs that of the endogenous truncated APC1 protein ( Figure 1—figure supplement 1D ) . This ratio was roughly 1400–2400 ( Table 1 ) . We suspect this is an over-estimate of the relative ratio of Drosophila APC2 to normal levels of hAPC1 in a wild-type colon cell , as the truncated APC1 protein present in SW480 cells would accumulate at lower levels than wild-type APC1 if it is subjected to nonsense-mediated mRNA decay , like many other proteins with early stop codons and like other truncated APC1 alleles ( Castellsagué et al . , 2010; Popp and Maquat , 2013 ) . Further , we also may need to reduce the ratio by a further factor of two as it is probable SW480 cells carry only one copy of the truncated APC allele , as most colorectal tumors either have the second allele mutated early enough to not produce a truncated protein or have lost the second allele by deletion ( Christie et al . , 2013 ) . Regardless , it is important to remember that while the puncta provide a useful and visualizable model of the destruction complex , we are examining over-expressed proteins . 10 . 7554/eLife . 08022 . 006Table 1 . Quantitation of relative expression levels of transfected versus endogenous APC and AxinDOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 006Summary Axin overexpressionExperimentIIIIIIRatio GFP-FlyAxin to GFP-hAxin11 . 040 . 110 . 48Ratio GFP-hAxin1 to endo hAxin133 . 87431 . 6762 . 82Transfection efficiency30%42%36%Fold overexpression level of GFP-FlyAxin to endo hAxin1117 . 41113 . 0683 . 76Summary APC2 overexpresssionExperimentIIIIIIRatio Flag-FlyAPC2 to Flag-hAPC1-133817 . 31138 . 4528 . 02Ratio Flag-hAPC1-1338 to endo hAPC1-133825 . 175 . 8614 . 69Transfection efficiency30%33%25%Fold overexpression level of Flag-FlyAPC2 to endo hAPC1-13381452 . 302458 . 541646 . 46 To probe destruction complex structure and dynamics , we first need to define mechanisms of complex assembly . In the current model , destruction complex assembly is mediated by interaction of APC2's SAMPs and Axin's RGS domain ( Fagotto et al . , 1999; Figure 1A ) . Consistent with this , deleting the SAMPs disrupted APC2 recruitment into Axin puncta . However a shorter APC2 mutant , lacking both the SAMPs and R2 and B and thus resembling the truncated hAPC1 in tumors ( Figure 1B ) , did co-localize with Axin ( Roberts et al . , 2011 ) . We further found that deleting R2 and/or B , both of which are essential for APC function in flies and mammals , restored colocalization of Axin and APC2∆SAMPs ( Roberts et al . , 2011 ) . Thus a second means of mediating APC2:Axin interactions must exist , that is separate from the known SAMP:RGS interaction . To identify the protein sequences mediating this putative R2/B regulated mechanism by which APC2 and Axin can interact , we truncated APC2 from its C-terminus , and assessed APC2 mutant colocalization with Axin . APC2's Arm rpts ( Figure 1B ) formed cytoplasmic puncta when expressed alone ( Figure 1—figure supplement 2A , B ) , consistent with APC2's ability to self-oligimerize ( Kunttas-Tatli et al . , 2014 ) . When co-expressed with Axin , the Arm rpts and Axin puncta associate , suggesting the Arm rpts are the Axin association site ( Figure 1C , D arrows; intriguingly they often did not precisely colocalize , in contrast with the full length proteins ( Roberts et al . , 2011 ) ) . Next we defined where on Axin the Arm rpts associate , by expressing Axin fragments and conducting co-immunoprecipitation ( co-IP; Figure 1E , F ) . Both full length Axin and a region including the GSK3 binding site co-IPed with APC's Arm rpts ( Figure 1E , F ) . Axin's DIX domain was weakly detected in co-IPs , but this may occur by polymerization with full-length endogenous human Axin1 ( hAxin1 ) . Thus fly APC2's Arm rpts can mediate association with Axin's middle region . This was exciting , as it may explain the Arm rpts essential role in Wnt regulation ( McCartney et al . , 2006; Roberts et al . , 2012 ) . To determine if this second APC:Axin interaction mechanism is conserved in humans and whether it occurs between proteins expressed at endogenous levels , we used SW480 cells , in which endogenous hAPC1 is truncated after 20R1 at 1338aa , thus lacking the SAMPs , R2 and B ( Figure 1B ) . hAPC1-1338aa co-IPed with endogenous hAxin1 ( Figure 1G ) . To verify that association was through the Arm rpts , we tested whether hAPC1's Arm rpts alone ( Figure 1B ) colocalized with hAxin1 in SW480 cells . hAPC1Arm rpts and hAxin1 each formed cytoplasmic puncta ( Figure 1—figure supplement 2C–E ) . When coexpressed , they colocalized in cytoplasmic puncta ( Figure 1H , I arrows ) , suggesting both hAPC1 and fly APC2 can associate with the Axin complex in two ways: the known interaction via the SAMPs and this novel interaction via the Arm rpts . Given this new data on destruction complex assembly , we next explored the structures assembled by Axin vs Axin plus APC , using the puncta formed in SW480 cells as a visualizable destruction complex model . While current data suggest Axin polymerization and APC recruitment are essential for destruction complex function , previous microscopy provided only limited information about the internal structure of Axin–APC complexes in vivo . In these images , Axin and APC colocalized in puncta without resolvable internal structure ( e . g . , Mendoza-Topaz et al . , 2011; Roberts et al . , 2011; Figure 2A , B ) . Recent advances increased the resolution possible with light microscopy . We thus used the superresolution approach Structured Illumination Microscopy ( SIM ) to visualize APC:Axin complexes , revealing a striking and previously undescribed internal structure . Cytoplasmic complexes formed by Axin alone appear to consist of Axin cables/sheets , assembling into hollow structures ( Figure 2C–F; Videos 1–3 show 3D reconstructions of these three puncta ) . Interestingly , coexpressing APC2 substantially altered the structure of many complexes , increasing the length/complexity of Axin cables ( Figure 2G , H–J = three representative puncta; Videos 4–6 show 3D reconstructions ) , with APC2 and Axin cables intertwined and in some puncta APC bridging Axin cables ( Figure 2H–J , arrows ) . 10 . 7554/eLife . 08022 . 007Figure 2 . Axin and APC2 form structured macromolecular complexes in vivo . SW480 cells . ( A ) Confocal image , GFP-APC2 and Axin-RFP . APC2 is recruited into Axin puncta . ( B ) Closeups , showing failure to resolve internal structure . ( C–J ) SIM super-resolution . ( C ) Axin-RFP alone . ( D–F ) Closeups , Axin complexes from different cells . D = punctum boxed in C . Axin cables assemble into spheres/sheets . ( G ) Cell coexpressing GFP-APC2 and Axin-RFP . ( H–J ) Closeups of APC2:Axin complexes from different cells . H = punctum boxed in G . Axin cables increase in complexity and APC2 forms cables intertwined with Axin ( arrows ) . ( K ) Analysis of confocal images . Complexes formed by APC2 and Axin average nearly twice the cross-sectional area of complexes formed by Axin alone ( left ) . Axin-expressing cells have twice as many complexes as cells coexpressing APC2 + Axin ( right ) . Student's t-test . ( L ) Puncta volume in Axin expressing cells ( n = 3 ) vs APC2 + Axin expressing cells ( n = 11 ) showing volumes across puncta population . Volume differences expressing cells are consistent with area quantification in ( K ) . ANOVA-Bonferroni was used . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 00710 . 7554/eLife . 08022 . 008Video 1 . 3D reconstruction of SIM superresolution image of Axin-RFP expressed in SW480 cells ( see also Figure 2D ) . Volume view from Imaris 5 . 5 was used for reconstruction . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 00810 . 7554/eLife . 08022 . 009Video 2 . 3D reconstruction of SIM superresolution image of Axin-RFP ( see also Figure 2E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 00910 . 7554/eLife . 08022 . 010Video 3 . 3D reconstruction of SIM superresolution image of Axin-RFP ( see also Figure 2F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 01010 . 7554/eLife . 08022 . 011Video 4 . 3D reconstruction of SIM superresolution image of GFP-APC2 and Axin-RFP expressed in SW480 cells ( see also Figure 2H ) . Volume view from Imaris 5 . 5 was used for reconstruction . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 01110 . 7554/eLife . 08022 . 012Video 5 . 3D reconstruction of SIM superresolution image of GFP-APC2 and Axin-RFP ( see also Figure 2I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 01210 . 7554/eLife . 08022 . 013Video 6 . 3D reconstruction of SIM superresolution image of GFP-APC2 and Axin-RFP ( see also Figure 2J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 013 These data suggest APC2 may enhance assembly of Axin oligomers , stimulating formation of larger complexes . To test this hypothesis , we quantified puncta cross-sectional area in Axin vs APC + Axin expressing cells , making z-projections of transfected cells and using the ImageJ particle analyzer . This revealed APC2:Axin puncta average almost twice the cross-sectional area of puncta formed by Axin alone ( Figure 2K , left ) . We confirmed this by comparing the volume of Axin and APC:Axin complexes , as measured from our SIM data—once again APC + Axin puncta were on average significantly larger than those assembled from Axin alone , and the largest APC + Axin puncta were larger than any of the Axin puncta ( Figure 2L ) . When we compared the average number of complexes in Axin alone vs APC2 + Axin expressing cells , we found twice as many complexes in cells expressing Axin alone ( Figure 2K , right ) , suggesting that in the absence of functional APC , Axin forms more numerous but smaller puncta . These data support the hypothesis that APC2 induces changes in the structure of Axin complexes formed in vivo , enhancing Axin assembly into higher order complexes . The effect of APC on destruction complex structure might suggest APC helps regulate Axin turnover within the destruction complex . Previous analyses suggested that Axin dynamics within the destruction complex puncta correlate with function . Axin puncta are dynamic protein assemblies , and the Wnt effector Dishevelled significantly increases Axin dynamics in puncta , suggesting it negatively regulates Axin self-association ( Schwarz-Romond et al . , 2007 ) , consistent with its known negative regulatory role in destruction complex function . To directly assess APC2 and Axin dynamics in the destruction complex and their influence on one another , we used Fluorescent Recovery After Photobleaching ( FRAP ) to measure dynamics of both APC2-GFP and Axin-RFP ( Figure 3A–D ) . We assessed both recovery ( mobile ) fraction and t1/2 , using unbleached puncta as controls . Recovery fraction assesses what percentage of molecules in a complex turnover in the experimental time frame , and t1/2 reflects the rate at which the dynamic fraction is exchanged—they are not necessarily dependent on one another . We found APC2 is also a dynamic component of destruction complex puncta , reaching a recovery plateau of 40% and a t1/2 of 150 s ( Figure 3B ) ; however , APC2 is not as dynamic as Axin ( Figure 3C ) . 10 . 7554/eLife . 08022 . 014Figure 3 . APC2 stabilizes Axin complexes and promotes efficient βcat destruction . ( A ) Stills , FRAP movie , SW480 cells transfected with GFP-APC2 ( shown ) and Axin-RFP . Inset = magnified APC2 signal in punctum . ( B ) APC2 recovers to ∼40% when in Axin puncta . Recovery curve ( red ) ; unbleached control ( blue ) . ( C ) Axin expressed alone plateaus at ∼80% . ( D and E ) Axin is stabilized when coexpressed with APC2 . ( F ) Total cell βcat fluorescent intensity normalized to untransfected cells ( = 100% ) . APC2 or APC2 + Axin expression lead to stronger βcat reduction than Axin alone . ( G–I ) Indicated constructs expressed in SW480 cells . Insets = regions boxed . ( G ) GFP-Axin forms puncta and reduces βcat levels in this hAPC1 mutant cell line . βcat is detectable in puncta ( arrows ) . ( H ) GFP-APC2 expressed alone is dispersed throughout the cell and βcat levels are low overall and in puncta . ( I ) Axin-RFP + GFP-APC2 coexpressed . βcat is reduced in APC2:Axin puncta ( arrow ) relative to puncta with Axin alone ( G ) . ( J ) GFP-APC2 is recruited into puncta formed by human hAxin1-RFP . ( K ) Endogenous human hAxin1 co-IPs from SW480 cells with transfected Flag-APC2 . Untransfected cells serve as a negative control . ( L and M ) Phospho-S33/37-βcat levels are more reduced when either APC2 or APC2 + Axin are expressed relative to Axin alone . ( L ) Immunoblot , transiently transfected SW480 cell extracts , centrifuged at 1000 rpm . ( M ) Quantification , phospho-S33/37-βcat protein levels from ( L ) and 2 replicates . Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 01410 . 7554/eLife . 08022 . 015Figure 3—figure supplement 1 . While proteasome inhibition reduces βcat destruction and causes βcat to detectably accumulate in APC2 + Axin puncta , it does not abolish the ability of APC2 to enhance Axin function in this regard . ( A and B ) SW480 cells transfected with APC2 and Axin and treated with either ethanol as a control or with the proteasome inhibitor MG132 . MG132 treatment elevates overall βcat levels and leads detectable accumulation in APC2 + Axin puncta ( yellow arrows ) . Red arrows indicate untransfected cells . ( C ) Quantitation of total cell βcat fluorescent intensity normalized to untransfected cells ( = 100% ) . MG132 treatment increases βcat levels in both Axin and APC2 + Axin treated cells . However , APC2 + Axin expression still leads to stronger βcat reduction than Axin alone . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 015 To test the hypothesis that APC stabilizes assembly of Axin monomers into the multimeric destruction complex , we compared Axin dynamics in puncta containing Axin alone with those containing Axin plus APC2 . Interestingly , Axin expressed alone was quite dynamic , with a recovery plateau of almost 90% and a t1/2 = 150 s ( Figure 3C , E ) . However , when Axin was coexpressed with APC2 , Axin dynamics were significantly reduced ( recovery plateau = 40% and t1/2 = 300 s; Figure 3D , E ) , suggesting APC stabilizes Axin assembly within puncta . Axin cannot target βcat for destruction in APC's complete absence , even when Axin is overexpressed ( Mendoza-Topaz et al . , 2011 ) . However , in SW480 cells , which express both truncated hAPC1 and endogenous hAPC2 ( Maher et al . , 2009 ) , Axin overexpression can increase βcat destruction ( Nakamura et al . , 1998 ) . If APC's role in the active destruction complex is to stabilize Axin assembly and thus the destruction complex scaffold , then APC should enhance Axin's ability to target βcat for destruction . We thus examined whether Axin alone fully restores βcat destruction , or whether adding APC2 , either to endogenous wild-type Axin or co-expressed with fly Axin , further facilitates this . We began by measuring βcat fluorescence intensity in z-projections of cells transfected with either APC2 + Axin or Axin alone , using untransfected cells as internal controls ( Figure 3F–H ) . Interestingly , while Axin reduced total βcat levels ( Figure 3F , G ) , βcat was further reduced in cells expressing both APC2 and Axin ( Figure 3F , I ) suggesting that APC2 promotes more effective destruction complex activity . Cells expressing APC2 ‘alone’ also had strong βcat reduction ( Figure 3F , H ) , presumably due to interaction with endogenous human Axin . We thus tested whether fly APC2 can and does interact with human Axin . Fly GFP-APC2 colocalizes in puncta with exogenous hAxin1-RFP ( Figure 3J ) , and more importantly , endogenous hAxin1 coIPs with Flag-APC2 expressed in SW480 cells ( Figure 3K ) . One further caveat is that we were assessing the ability of APC2 and Axin to promote βcat destruction after over-expression . To determine if differing levels of over-expression might explain the differences between Axin and APC2 + Axin ( or ‘APC2 alone’ ) , we quantitated level of Axin or APC2 expression in a given cell by measuring levels of GFP/RFP fluorescence , and in parallel assessed levels of βcat in that cell ( via fluorescence intensity ) . Strikingly , βcat levels were more effectively reduced by APC2 + Axin or by ‘APC2 alone’ than by Axin at all levels of expression assessed . As noted above , in all cases , ability to reduce βcat levels was somewhat diminished at the highest levels of expression ( Figure 1—figure supplement 1B ) —this may be because at very high expression levels , the transfected protein forms non-functional complexes with only a subset of the destruction complex proteins , as was previously suggested ( Lee et al . , 2003 ) . Together these data support the hypothesis that APC2 enhances Axin's ability to promote βcat destruction . Interestingly , in Axin-alone transfected cells , much of the excess βcat that accumulated ( Figure 3F ) was in Axin puncta ( Figure 3G′ , inset , arrow ) . In contrast , puncta in cells co-expressing APC2 and Axin had almost undetectable βcat levels ( Figure 3I′ , inset ) . We thus hypothesized that APC enhances Axin's ability to promote βcat exit from the destruction complex , and thus the destruction complex's βcat throughput . To further explore this , we examined phospho-βcat levels . SW480 cells , like other colon cancer cell lines with hAPC1 truncated before the Mutation Cluster Region ( MCR; Figure 1B ) , have high phospho-βcat levels ( Yang et al . , 2006 ) . This suggests that while Axin can facilitate βcat phosphorylation in these cells , Axin is less efficient at targeting βcat for destruction in the absence of wild-type APC1 , and thus phosphorylated βcat accumulates . Strikingly , expressing APC2 alone or APC2 + Axin dramatically decreased phospho-serine 33/37 βcat ( to ∼20% that in untransfected cells ) , while Axin alone reduced phospho-βcat to only 60% ( Figure 3L , M ) . These data further support the hypothesis that APC2 enhances Axin's ability to promote βcat destruction . One potential caveat to the increased accumulation of βcat in puncta of cells expressing Axin alone vs those expressing APC2 plus Axin is that the former cells may simply have higher overall levels of βcat , thus resulting in higher accumulation in puncta . To address this , we inhibited βcat destruction using the proteasome inhibitor MG132 . As others have previously observed ( Sadot et al . , 2002 ) , proteasome inhibition elevates βcat levels . Proteasome inhibition elevates βcat levels both in cells expressing Axin alone and in those expressing APC2 plus Axin ( Figure 3—figure supplement 1C ) . Strikingly , this allows it to accumulate in puncta even in cells expressing Axin + APC2 ( Figure 3—figure supplement 1A vs B ) . However , cells expressing Axin still accumulate significantly higher levels of βcat than those expressing Axin plus APC2 ( Figure 3—figure supplement 1C ) . Together , these data are consistent with a model in which APC increases βcat throughput of the destruction complex by stabilizing Axin assembly . Since APC associates with Axin via two regions , the Arm rpts and SAMP motifs ( Figure 1 ) , we hypothesized each interaction helps stabilize destruction complex assembly . To test this , we first measured APC2 dynamics when either the Arm rpts or SAMPs were individually deleted ( Figure 4A ) . Deleting either region increased APC2 dynamics in Axin puncta; APC2ΔArm and APC2ΔSAMPs turnover reached higher plateaus ( Figure 4B; 80–90% vs 40% for wild-type ) in shorter times ( Figure 4B; t1/2 wildtype APC2 150 s; APC2ΔArm 75 s; APC2ΔSAMPs <25 s ) . Thus , APC2 needs both the Arm rpts and the SAMPs to stably associate with Axin complexes . Next we examined Axin dynamics in the presence of each APC2 mutant . Both the Arm rpts and SAMPs were required to stabilize Axin in destruction complexes , since Axin coexpressed with either APC2ΔArm or APC2ΔSAMPs exhibited the fast dynamics characteristic of Axin expressed alone ( Figure 4C ) . Thus APC2 stabilizes APC:Axin complexes through multivalent interactions mediated by the Arm rpts and SAMPs . 10 . 7554/eLife . 08022 . 016Figure 4 . APC2’s Arm rpts and SAMPs each are required to stabilize APC2:Axin complexes . ( A ) APC2 mutants . ( B and C ) FRAP analyses , SW480 cells . ( B ) APC2 needs both the Arm rpts and SAMPs to robustly associate with Axin puncta . Student's t-test . ( C ) Axin stabilization by APC2 is abolished when either APC2's Arm rpts or SAMPs are deleted . ( D , K ) SIM super-resolution images , SW480 cells expressing indicated constructs . ( D–F ) GFP-APC2ΔArm and Axin-RFP . ( E–F ) Close-ups , X–Y slice . Axin structure resembles complexes formed by Axin alone . ( G and H ) Axin-RFP puncta and APC2:Axin puncta for comparison . ( I–K ) GFP-APC2ΔSAMPs and Axin-RFP . ( J , K ) Close-ups . Axin does not form a complex internal structure when APC2 ΔSAMPs is expressed . ( L ) Puncta volume in SIM images of Axin ( n = 3 cells ) , APC2 + Axin ( n = 11 ) , APC2ΔArm + Axin ( n = 9 ) , and APC2ΔSAMPs + Axin ( n = 5 ) expressing cells . Deleting either the Arm rpts or the SAMPs inhibits APC2's ability to enhance puncta volume . ANOVA-Bonferroni . ( M ) Puncta area , confocal images . Area differences are consistent with volumes in ( L ) . Student's t-test . ( N ) Puncta number , confocal images . Deleting Arm rpts or SAMPs in APC2 fails to decrease number of APC2:Axin puncta as does wildtype APC2 . ( O ) APC mutants lacking the Arm rpts or SAMP motif show decreased ability to reduce βcat levels in SW480 cells . Quantification , total cell βcat fluorescent intensity normalized to untransfected cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 016 Our SIM imaging suggested APC stabilization of Axin complexes altered their substructure . We thus tested whether both Axin interaction sites were essential for the effects on the structure of Axin puncta . APC2ΔArm , which retains the SAMPs , closely colocalized with Axin , even at SIM resolution , rather than forming filaments of its own within puncta , like wild-type APC2 ( Figure 4D–F vs Figure 4H ) . Further , Axin within these puncta remained simple in structure , similar to puncta formed by Axin alone ( Figure 4G ) . APC2ΔSAMPs , which is much less tightly associated with Axin by either confocal localization or coIP ( Hart et al . , 1998; Roberts et al . , 2011 ) , did not strongly colocalize with Axin , instead forming a diffuse network surrounding Axin puncta . Importantly , APC2ΔSAMPs did not alter Axin structure within puncta as visualized by SIM . In the presence of APC2∆SAMPs , Axin puncta retained the simpler structure of those formed by Axin ( Figure 4I–K ) . Further , neither APC2ΔArm nor APC2ΔSAMPs increased Axin puncta size or reduced Axin puncta number ( Figure 4L–N ) . Thus APC's ability to stabilize destruction complexes and stimulate growth of Axin cables requires both sites mediating Axin complex interaction , the Arm rpts and SAMPs . Both sites are also required to allow APC2 to efficiently downregulate βcat levels; APC2∆SAMPs did not stimulate βcat destruction below Axin-alone mediated levels , while APC2∆Armrpts could not downregulate βcat levels ( Figure 4O ) . Since both APC2's Arm rpts and the SAMPs are essential for Wnt regulation in Drosophila ( Roberts et al . , 2011 , 2012 ) , this suggests that APC2's ability to stabilize destruction complex assembly through its multivalent interactions is critical for destruction complex throughput of βcat . Our earlier steady state analysis revealed that APC2 motifs R2 and B antagonized APC:Axin interaction when the SAMPs were deleted . Since both R2 and B are essential for APC function in targeting βcat for destruction in vivo , we extended our work to determine whether and how R2 and B affect the dynamics of APC:Axin interactions . Our data above reveal the new Axin complex-association site is in APC's Arm rpts ( Figure 1 ) . Since removing R2/B restored APC/Axin interaction even in the absence of the SAMPs ( Figure 5—figure supplement 1A–E ) , we hypothesized R2/B negatively regulates APC2 Armrpt:Axin interaction and that release of this interaction is essential to allow phosphorylated βcat to be moved on to destruction . Our hypothesis predicted deleting either R2 or B should stabilize APC2:Axin interaction , by enabling APC2's Arm rpts to also productively mediate association with Axin . To test this we assessed how deleting R2 or B ( Figure 5A ) affected APC2 dynamics by FRAP . Deleting either R2 or B had the predicted effect , decreasing APC2 turnover rate dramatically ( Figure 5B ) , suggesting R2 and B regulate APC2 dynamics in the active destruction complex . While deleting R2 and B strongly stabilized APC2 in the complex , it did not significantly diminish APC2's ability to stabilize Axin , destruction complex size and structure ( Figure 5—figure supplement 1F–K ) . 10 . 7554/eLife . 08022 . 017Figure 5 . R2 and B regulate APC2 dynamics in the destruction complex , and regulate βcat removal from the destruction complex . ( A ) APC2 mutants . ( B–F ) SW480 cells transfected with Axin-RFP and indicated GFP-APC2 constructs ( B ) FRAP assay . Deleting either R2 or B slows APC2's turnover in Axin puncta . ( C–E ) Axin-RFP , GFP-APC2 constructs , βcat ( inset = boxes ) . ( C ) βcat is essentially undetectable in APC2:Axin puncta ( arrows ) . ( D ) βcat accumulates in APC2ΔR2:Axin puncta ( arrows ) . ( E ) βcat strongly accumulates in APC2ΔB:Axin puncta ( arrows ) . ( F ) Deletion of R2 or B impair the ability of APC2 to aid Axin in reducing βcat fluorescent intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 01710 . 7554/eLife . 08022 . 018Figure 5—figure supplement 1 . Colocalization of APC2's Arm repeat domain with Axin is controlled by R2 and B and APC2 without R2 or region B still stabilizes Axin complexes . ( A ) Diagram of APC2 constructs . ( B ) GFP-APC2 and Axin-RFP colocalize with one another in SW480 cells and reduce βcat levels ( inset = box in B ) . ( C ) Deleting APC2's SAMPs reduces colocalization with Axin ( inset = box in C ) . ( D ) Deleting R2 from APC2ΔSAMPs enhances colocalization with Axin ( inset = box in D ) . ( E ) APC2ΔSAMPs lacking B colocalizes strongly with the Axin ( inset = box in E ) . ( F ) Puncta area of APC2ΔR2 + Axin and APC2ΔB + Axin is similar to wildtype APC2 + Axin and differs from puncta formed by Axin alone . Analysis of confocal images , total of n = 30 cells . Student's t-test was used . ( G ) Wildtype APC2 , APC2∆R2 and APC2∆B all reduce Axin puncta number . Quantitation of puncta number of confocal images used in ( F ) . ( H ) Comparison of puncta volumes using SIM images reveals that APC2ΔR2 + Axin and APC2ΔB + Axin form puncta that are in between those formed by Axin alone and those formed by APC2 + Axin puncta . Volumes across puncta population . Axin ( n = 3 ) , APC2 + Axin ( n = 11 ) , APC2ΔR2 + Axin ( n = 6 ) , and APC2ΔB + Axin ( n = 9 ) . ANOVA-Bonferroni was used . ( I , J ) APC2ΔR2 + Axin and APC2ΔB + Axin puncta maintain complex internal structure . Closeups of puncta of SIM high resolution images of APC2ΔB + Axin ( I ) and APC2ΔR2 + Axin expressed in SW480 cells . ( K ) Deleting R2 or B does not significantly diminish the ability of APC2 to stabilize Axin in the destruction complex . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 018 We next asked which step in the cycle of destruction complex function is blocked by removing R2 or B , and thus altering APC dynamics . To do so , we examined if βcat was retained in puncta containing these mutants . βcat was nearly undetectable when APC2 and Axin are coexpressed ( Figure 5C , arrows ) . Interestingly , deleting either R2 or B led to βcat accumulation in APC2:Axin complexes ( Figure 5D , E , arrows ) . Deleting R2 also abolished APC2's ability to enhance the reduction of βcat levels given by Axin alone ( Figure 5F ) , suggesting that APC2 without R2 is not functional and therefore inhibited in its ability to promote βcat destruction via Axin complexes . βcat levels increased even further when motif B was deleted ( Figure 5F ) suggesting that APC2∆B may interfere with βcat degradation via Axin . These data suggest that R2 and B regulate APC2 dynamics in the destruction complex and that APC2 lacking them cannot effectively support Axin in targeting βcat for destruction . R2 or B thus regulate APC2:Axin interactions . We next determined which Axin association site , APC2's Arm rpts or SAMPs , was regulated . Deleting the SAMPs substantially reduced APC2 recruitment into Axin complexes while further deleting either R2 or B restored strong Axin:APC2∆SAMPs colocalization ( Figure 5—figure supplement 1 ) ; in APC2∆SAMPs the only remaining means of interacting with the Axin complex was via the Arm rpts , suggesting Arm rpts:Axin association is regulated by R2/B . In contrast , deleting APC2's Arm rpts ( Figure 6—figure supplement 1A ) did not reduce Axin colocalization ( Figure 6—figure supplement 1B arrows ) ; thus not surprisingly deleting either R2 or B in APC2ΔArm did not further alter this ( Figure 6—figure supplement 1A , C , D arrows ) . To directly test the hypothesis that R2 and B regulate dynamics of Arm rpts:Axin association , we returned to our FRAP assay . Consistent with effects on colocalization , deleting R2 or B ( Figure 6A ) decreased APC2ΔSAMPs dynamics ( Figure 6B; t1/2 increased significantly for both mutants; intriguingly recovery fraction was only significantly affected by deleting R2 , perhaps reflecting the unaltered destruction complex structure in these mutants ) . Co-IPs extended the FRAP results , revealing more stable APC2ΔSAMPs:Axin interaction when R2 was deleted ( Figure 6C; quantified in Figure 6D; the change after deleting motif B and the SAMPs did not reach statistical significance ) . Thus without R2 , APC2's Arm rpts associate more robustly with Axin complexes , slowing APC2 dynamics . In contrast , SAMPs:Axin interaction was not regulated by R2 or B: APC2ΔArm ( in which the only remaining interaction with Axin was via the SAMPs ) , APC2ΔArmΔR2 , and APC2ΔArmΔB all had similar recovery plateaus and t1/2 ( Figure 6—figure supplement 1E ) , and co-IPs showed no difference in APC2:Axin association among these constructs ( Figure 6—figure supplement 1F , G ) . In fact , APC2ΔR2 and APC2ΔB , which retain both the Arm rpts and SAMPs , coIPed with Axin as or more robustly than wild-type APC2 ( Figure 6—figure supplement 1H , I ) consistent with enhanced APC:Axin interaction through increased Arm rpts:Axin association when R2 and B were deleted . Thus APC2 and Axin have two distinct interaction interfaces with different properties: strong association via the SAMPs is independent of R2/B and a second interaction via APC2's Arm rpts is controlled by R2/B . 10 . 7554/eLife . 08022 . 019Figure 6 . Association of Axin with APC2's Arm rpts is controlled by R2 and B . ( A ) APC2 constructs . ( B ) Deleting either R2 or B in APC2ΔSAMPs slows APC2 recovery time and reduces recovery fraction . FRAP assay , SW480 cells transfected with GFP-tagged APC2 constructs and Axin-RFP . ( C ) IPs of indicated constructs . Deleting the SAMPs substantially decreases APC2:Axin coIP , but this is partially restored by deleting R2 . ( D ) Quantification , >2 replicates , normalized to Axin pull down . Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 01910 . 7554/eLife . 08022 . 020Figure 6—figure supplement 1 . Binding of APC2's SAMP motif to Axin is not regulated by R2 and B . ( A ) Diagram of APC2 constructs . ( B ) GFP-APC2ΔArm colocalizes with Axin-RFP in SW480 cells ( inset = box in B ) . βcat is detectable in APC2ΔArm:Axin complexes . ( C ) Deleting R2 does not alter colocalization of APC2ΔArm with Axin ( inset = box in C ) . βcat remains detectable in the APC2ΔArmΔR2:Axin puncta . ( D ) APC2ΔArm lacking B colocalizes with Axin ( inset = box in D ) . βcat is detectable in the APC2ΔArmΔB:Axin complexes . ( E ) FRAP assay of GFP-APC2ΔArm mutants with Axin-RFP in SW480 cells . Deletion of either R2 or B does not alter turnover of APC2ΔArm . Student's t-test was used . ( F ) Co-Immunoprecipitations of GFP-APC2ΔArm mutants with Flag-Axin . IP of Axin via anti-Flag antibody . APC2 binding to Axin via the SAMP motifs is not altered when R2 or B are deleted . Representative blot of 3 independent experiments . ( G ) Quantification Co-IP of indicated APC2 mutants with Flag-Axin ( ( F ) and 2 replicates ) , normalized to coIP with wild-type APC2 . Binding via the SAMPs is not altered when R2 or B are deleted . Student's t-test was used . ( H ) Co-IP of indicated APC2 mutants with Flag-Axin . Deletion of either R2 or B in full length APC2 increases its association with Axin . Representative blot of 3 independent experiments . ( I ) Quantification , Co-IP of indicated APC2 mutants with Flag-Axin ( ( H ) and 2 replicates ) , normalized to coIP with wild-type APC2 . Student's t-test was used . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 020 The known essential role of R2 and B in the active destruction complex in targeting βcat for destruction suggested their ability to regulate APC:Axin interaction is critical . We thus further explored the mechanism by which they regulate destruction complex dynamics and function , by comparing sequences of these adjacent motifs in mammalian and fly APCs ( Figure 7A ) . Strikingly , threonines and serines were among the most highly conserved residues . In the destruction complex , GSK3 and CK1 phosphorylate not only βcat but also Axin and other sites on APC ( Ikeda et al . , 1998; Yamamoto et al . , 1999; Ha et al . , 2004 ) . Motif B has multiple serines matching GSK3's phosphorylation consensus and one match to the CK1 consensus , while R2 has several matches to both GSK3 and CK1 consensuses ( Figure 7A ) . GSK3 kinase plays multiple roles in promoting destruction complex activity , phosphorylating βcat to target it to the E3 ligase and also regulating the destruction complex by phosphorylating the βcat binding sites on APC , increasing their affinity , and phosphorylating Axin . We hypothesized GSK3 also phosphorylates R2 and/or B , to trigger release of APC2's Arm rpts from Axin . 10 . 7554/eLife . 08022 . 021Figure 7 . Axin:APC2 Arm rpts association is regulated by GSK3 . ( A ) R2 and B of Drosophila dAPC2 , mouse mAPC1 and human hAPC1 . Potential CK1 ( orange ) and GSK3 ( green ) phosphorylation sites . ( B ) APC2 mutants . ( C–E ) SW480 cells expressing GFP-APC2ΔSAMPs and Axin-RFP . Insets = boxes in C–E ( C ) Control ( Ethanol treated ( EtOH ) ) . Deleting the SAMPs reduces APC2:Axin colocalization ( arrows ) . ( D ) 2 μM BIO enhances APC2ΔSAMPs recruitment into Axin puncta . ( E ) Increasing BIO to 4 μm further boosts APC2ΔSAMPs recruitment into Axin puncta . ( F ) Quantification of ( C–E ) . ( G ) CoIP of APC2∆SAMPs with Axin in SW480 cells ± LiCl or BIO . Full length APC2 is a control . Deleting the SAMPs drastically reduces coIP but GSK inhibition partially restores this . ( H ) Quantification of coIP in G , >2 replicates , normalized to Axin . ( I ) FRAP assay , SW480 cells transfected with Axin-RFP + GFP-APC2 . GSK3 inhibition decreases APC2 dynamics . ( J ) GSK inhibition also slows APC2ΔSAMPs dynamics . ( K ) Deleting R2/B in APC2ΔSAMPs abolishes effect of GSK3 inhibition on dynamics . Student's t-test . ( L ) GSK3 inhibition with either BIO or LiCl does not further increase the size or decrease the number of APC plus Axin puncta , nor does treatment with the proteasome inhibitor MG132 . Puncta area and puncta number , from confocal images . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 02110 . 7554/eLife . 08022 . 022Figure 7—figure supplement 1 . GSK3 regulates association of APC2's Arm repeats with the Axin complex . ( A , B , D and E ) GFP-APC2 and Axin-RFP expressed in SW480 cells . βcat stained via antibody . ( A ) APC2 and Axin strongly colocalize in cytoplasmic puncta in cells treated with EtOH as control for BIO . ( Inset = box in A ) βcat levels in APC2:Axin complexes are very low . ( B ) GSK3 inhibition via BIO does not alter colocalization of APC2 with Axin , but leads to βcat accumulation in APC2:Axin puncta ( inset = box in B ) . ( C ) Quantification of total βcat fluorescent intensity in ( A and B ) . ( D ) In control cells for LiCl treatment , βcat levels in APC2:Axin complexes are very low ( inset = box in D ) . ( E ) βcat accumulates in APC2:Axin complexes when GSK3 is inhibited via LiCl ( inset = box in E ) . ( F ) Quantification of total βcat fluorescent intensity in ( D and E ) . ( G ) GFP-APC2ΔSAMPs is diffuse and only associates weakly with Axin ( inset = box in G ) in control cells for LiCl treatment . ( H ) GSK3 inhibition with LiCl enhances association of APC2ΔSAMPs with Axin ( inset = box in H ) . ( I ) Quantification of APC2ΔSAMPs colocalization with Axin ( G vs H ) . Student's t-test was used . ( J ) FRAP assay in SW480 cells transfected with Axin-RFP and GFP-APC2ΔArm . GSK3 inhibition via BIO does not alter APC2's turnover rate when association is mediated by the SAMPs . Student's t-test was used . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 022 To test this we first determined whether blocking GSK3 activity using LiCl , a well-known GSK3 inhibitor ( Klein and Melton , 1996 ) , or BIO , a very specific GSK3 inhibitor ( Meijer et al . , 2003 ) affected APC2∆SAMPs:Axin interaction ( we verified GSK3 inhibition by assessing βcat accumulation in APC2:Axin puncta; Figure 7—figure supplement 1A–F; Stambolic et al . , 1996 ) . APC2∆SAMPs ( Figure 7B ) is only weakly recruited into Axin puncta ( Figure 7C ) . If GSK3 phosphorylation of R2/B antagonizes Arm rpts:Axin association , inhibiting GSK3 should increase colocalization of APC2ΔSAMPs with Axin , as did deleting R2 or B ( Figure 5—figure supplement 1 ) . Strikingly , while APC2∆SAMPs is largely diffusely cytoplasmic ( Figure 7C ) , BIO treatment strongly increased Axin:APC2ΔSAMPs colocalization ( Figure 7D arrows ) , in a concentration dependent manner ( Figure 7E ) . Consistent with this , only 16% of untreated control cells had APC2ΔSAMPs:Axin colocalization ( Figure 7F ) , while inhibiting GSK3 with BIO increased colocalization to 75% of cells ( Figure 7F ) . LiCl also led to robust Axin:APC2ΔSAMPs colocalization ( Figure 7—figure supplement 1G vs Figure 7—figure supplement 1H; quantified in Figure 7—figure supplement 1I ) . We used coIP to verify that APC2ΔSAMPs associates more robustly with Axin upon GSK3 inhibition . Deleting the SAMPs drastically reduced APC2:Axin coIP ( Figure 7G; quantified in Figure 7H ) . GSK3 inhibition by either LiCl or BIO increased APC2ΔSAMPs coIP with Axin ( Figure 7G , H ) . Thus GSK3 inhibition stabilizes steady state Axin:APC2∆SAMPs association , as did deleting R2 or B , consistent with our model that phosphorylating these motifs normally antagonizes Axin:Arm rpts association . The hypothesis that association of APC2's Arm rpts with Axin is regulated by GSK3 also predicts inhibiting GSK3 should affect APC2 dynamics in the destruction complex , as did deleting R2 or B . Consistent with this , inhibiting GSK3 dramatically decreased APC2's dynamics ( Figure 7I; plateau reduced from 40% to 10%; t1/2 increased from 150 to >1000 s ) . This suggests APC2 turnover in Axin complexes is regulated by GSK3 . Our model further predicts that GSK3 regulates APC2 dynamics by regulating the APC2Arm rpts:Axin association . Thus GSK3 inhibition should stabilize APC2ΔSAMPs:Axin interactions and reduce APC2ΔSAMPs dynamics . As predicted , APC2ΔSAMPs rapid turnover was dramatically decreased by GSK3 inhibition ( Figure 7J ) . In contrast , GSK3 inhibition had no effect on APC2∆Arm turnover ( Figure 7—figure supplement 1J ) . These data suggest that GSK3 activity promotes release of APC2's Arm rpts from the Axin complex . Our data are consistent with GSK3 acting via phosphorylating R2 and/or B , but could also be mediated via other effects of GSK3 . To assess this , we examined whether inhibiting GSK3 affected turnover of an APC2ΔSAMPs mutant lacking both R2 and B ( Figure 7B ) , reasoning this would assess the effect of GSK3 on regulation of the Axin:APC2Arm rpts interaction . Strikingly , the recovery plateau and t1/2 of APC2ΔR2/BΔSAMPs were insensitive to GSK3 inhibition ( Figure 7K ) , in contrast to APC2∆SAMPs . These data are consistent with GSK3 affecting APC2 residence time in the destruction complex through R2/B . However , deleting R2/B and blocking GSK3 activity had different effects on APC2 recovery fraction , suggesting that not all effects of GSK3 are mediated through R2/B . GSK3 phosphorylates other targets in the destruction complex , and thus it is very likely GSK3 inhibition has additional means of altering destruction complex dynamics . Further , an unphosphorylated R2/B motif may affect APC's dynamics differently than deleting R2/B . Taken together , however , our data suggest that GSK3 acts in part through R2 and B to promote release of APC2's Arm rpts from the Axin complex . If GSK3 inhibition stabilizes the interaction of APC2 and Axin , and as noted above , APC2 also can stabilize Axin in puncta , increasing their size ( and thus decreasing puncta number ) , then GSK3 inhibition might synergize with APC , further increasing the size of Axin puncta . We thus inhibited GSK3 with either BIO or LiCl in cells co-expressing Axin and APC2 and examined both puncta size and number . GSK3 inhibition did not further increase puncta size or further decrease puncta number ( Figure 7L ) . These data are consistent with our analysis above of APC2∆R2 and APC∆B—these mutations also stabilize APC2 in the destruction complex ( Figure 5B ) but do not further increase puncta size or decrease puncta number ( Figure 5—figure supplement 1F–J ) . We also tested whether proteasome inhibition might trigger enlargement of the APC + Axin complexes—this also did not further increase puncta size or further decrease puncta number ( Figure 7L ) . Thus , while APC2 can stabilize Axin complexes , inhibition of GSK3 or the proteasome does not synergize with this . We hypothesized that R2/B phosphorylation by GSK3 is a key step in APC2's mechanism to target βcat for destruction . Thus we tested whether R2/B can be phosphorylated by GSK3 . A GST-fusion containing just R2 and B from either fly APC2 or human APC1 can be phosphorylated in vitro by human GSK3 ( Figure 8A ) . Human R2/B was strongly phosphorylated , whereas fly R2/B was more weakly phosphorylated . Thus the potential phosphorylation sites in R2/B can be phosphorylated by GSK3 , possibly at the GSK3 consensus sites ( Figure 8B ) . 10 . 7554/eLife . 08022 . 023Figure 8 . Mutating putative phosphorylation sites in B disrupts APC2 function . ( A ) R2/B of human or fly APCs can be phosphorylated by human GSK3 . In vitro kinase assay , GSK3 substrate peptide ( positive control , left panels ) , GST-tagged humanAPC1R2/B or fly APC2R2/B fragments . GST was a negative control . Left of each pair: Coomassie stained gel , right: Phosphorylation detected using P32 . Asterisks ( * ) indicate Coomassie-stained bands that align with P32-labeled proteins . In the presence of GST alone , GSK3 autophosphorylates ( lane 2 ) . HumanAPC1R2/B is strongly phosphorylated ( lane 4 ) while fly APC2R2/B was more weakly phosphorylated ( lane 6 ) . Representative of two experiments . ( B ) R2 and B of Drosophila dAPC2 , mouse mAPC1 and human hAPC1 . Potential CK1 ( orange ) and GSK3 ( green ) phosphorylation sites . Red Arrow in R2 = serine mutated to alanine in mutant assessed in panel E . Red arrows in B = serines mutated to aspartic acid or alanine in APC2AA or APC2DD . Magenta arrowheads = additional serines mutated in 4 serine and six serine mutations ( data not shown ) . ( C ) APC2DD ( 2 serines in B changed to aspartic acid; arrows in B ) effectively reduces βcat levels in SW480 cells ( arrow vs arrowhead ) . Inset = box in A . ( D ) APC2AA ( 2 serines in B changed to alanine , arrows in B ) is unable to target βcat for destruction ( arrow vs arrowhead ) . ( E ) APC2 R2S->A ( single serine in R2 changed to alanine , arrow in B ) is unable to target βcat for destruction ( arrow vs arrowhead ) . ( F ) Quantification , total βcat fluorescent intensity . Student's t-test . ( G ) FRAP , Axin-RFP + GFP-APC2 constructs . APC2AA reaches a lower recovery plateau than either wild-type APC or APC2DD . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 023 R2 and B are essential for APC2's function in βcat degradation . Our hypothesis suggests phosphorylation of R2/B promotes release of APC2's Arm rpts from the Axin complex , and that this would be essential for the catalytic cycle of the destruction complex—thus mutating these putative GSK3 phosphorylation sites in motif B would reduce APC2 function in helping mediate βcat destruction . To begin to test this , we replaced 2 ( APC2AA; Figure 8B , red arrows ) , 4 or 6 ( Figure 8B , magenta arrowheads ) conserved serines in B that match the GSK3 consensus with alanine , to prevent phosphorylation . All reduced function in downregulating βcat levels ( see below and data not shown ) . We thus focused on the least altered of these , the mutant that replaced the more N-terminal two serine residues with alanine ( APC2AA; Figure 8B , red arrows ) , thus preventing phosphorylation . We also created a mutant that replaced these same two residues with aspartic acid , creating a phosphomimetic APC2 ( APC2DD ) . Strikingly , while APC2DD effectively reduced βcat levels ( Figure 8C , F ) , APC2AA was unable to do so . APC2AA cells accumulated βcat at levels as high or higher than adjacent untransfected cells ( Figure 8D , F ) , suggesting these two amino acid missense mutations substantially reduced APC2 function . APC2DD substantially reduced βcat levels ( to ∼30% ) , but was not quite as effective as wildtype APC2 ( Figure 8F ) . This statistically significant difference may suggest dephosphorylation of these residues is also required for full APC2 function . We also found mutating a single conserved serine residue in R2 to alanine ( APC2 R2S->A ( =APC2S660A ) ; Figure 8B ) strongly diminished APC2 function in reducing βcat levels ( Figure 8E , F ) . Together , these data are consistent with a model in which phosphorylation of conserved serines in APC2 motifs R2 and B are important for APC's function in the destruction complex to target βcat for degradation . Finally , we tested whether the first of these putative phosphomutants affected APC2 dynamics in the FRAP assay . Since GSK3 inhibition slowed APC's dynamics , we predicted APC2AA should have a lower turnover rate than wildtype APC or APC2DD . We saw a subtle but statistically significant reduction in APC2AA recovery fraction ( Figure 8G ) . However this was not nearly as dramatic as that of deleting R2 or B ( Figure 5B ) ; perhaps this due to the fact that we only altered two of several potential phosphorylation sites . Together these data are consistent with the idea that phosphorylation of conserved serine residues in R2/B regulates APC's function in the destruction complex , but since the effect on dynamics was substantially less dramatic than that of deleting R2 or B , it suggests other residues in R2 and B may also contribute to regulating APC2 dynamics . Further , it is clear that GSK3 has other effects on the complex , complicating interpretation of its inhibition . It will be important to examine this in the future . To test the role of these two putative phosphorylation sites in APC function in an in vivo context where we can examine both cells receiving and not receiving Wnt signals , we turned to Drosophila , where we can express mutant APC2 under control of the endogenous APC2 promoter in the complete absence of all endogenous APC function , using embryos maternally and zygotically APC2 APC1 double mutant . We generated transgenics expressing APC2DD or APC2AA using the endogenous APC2 promotor ( as in Roberts et al . , 2011 ) , and crossed them into the APC2 APC1 mutant background . All progeny were maternally APC2 APC1 mutant and 50% of progeny were also zygotically mutant , while the other 50% were paternally rescued ( we cannot generate homozygous double mutant males ) . In the absence of the transgene , 43% of progeny hatch ( Figure 9A ) , consistent with ∼50% zygotic rescue . A wild-type APC2 transgene expressed in APC2 APC1 mutants led to 95% survival , comparable to wild-type flies ( 93–99%; Figure 9A ) . APC2DD was as effective as wild-type APC2 at restoring embryonic viability ( 96% survival; Figure 9A ) . In contrast , APC2AA expressed in APC2 APC1 mutants only weakly rescued embryonic lethality ( 64% viability; Figure 9A ) . 10 . 7554/eLife . 08022 . 024Figure 9 . Blocking potential phosphorylation at 2 conserved serines in B disrupts APC2 function in the fly . APC2DD and APC2AA ( Figure 8B ) were expressed with the endogenous APC2 promoter in APC2 APC1 maternal/zygotic double mutants . ( A ) APC2 APC1 maternal/zygotic double mutants die as embryos ( 50% of embryos are zygotically rescued ) . APC2DD rescues embryonic viability as well as wildtype APC2 . In contrast , APC2AA has only weak rescue ability . ( B ) Cuticles . ( B1 ) Wildtype . Note pattern of anterior denticles ( Wnt inactive ) and posterior naked cuticle ( Wnt active ) . ( B2 ) Loss of APC2 and APC1 leads to denticle loss and expanded naked cuticle . ( B3 ) Wildtype APC2 fully restores Wnt regulated cell fates of alternating denticles and naked cuticle . ( B4 ) APC2DD similarly restores cell fates . ( B5–B7 ) APC2AA largely fails to restore Wnt-regulated cell fates , and thus most cells secrete naked cuticle . Images = range of rescue ability . ( C–I ) βcat ( fly Armadillo ( Arm ) ) levels . Stage 9–10 embryos . ( J–N ) Close-ups of C–I . ( C , J ) Wildtype . Striped pattern of βcat indicative of Wg ( fly Wnt ) active and Wg inactive regions . ( D ) Loss of APC1 and APC2 leads to uniform very high levels of βcat . ( E ) WT APC2 restores normal βcat regulation , with higher levels in cells receiving Wg signal , and lower levels in other cells . However , Wg signal does not elevate βcat levels to those seen in embryos lacking functional APC . ( F and K ) APC2DD also rescues normal βcat regulation . ( G–I and L–N ) APC2AA restores some Wnt responsiveness , but βcat levels are elevated in all cells and especially elevated in a subset of cells receiving Wnt signal . ( O ) Quantification , embryos blind-scored . DOI: http://dx . doi . org/10 . 7554/eLife . 08022 . 024 We then examined rescue of cell fates . In wildtype embryos , a row of cells in each body segment expresses the Wnt homolog Wingless ( Wg ) , thus regulating cell fate . In cells not receiving Wg signal , the destruction complex effectively destroys βcat that is not sequestered in cadherin-catenin complexes at the cell membrane , and cells choose anterior fates and secrete cuticle covered with denticles ( Figure 9B1 ) . In contrast , cells receiving Wg signal accumulate cytoplasmic and nuclear βcat , choose posterior fates , and secrete naked cuticle ( Figure 9B1 ) . Maternal/zygotic APC2 APC1 mutants cannot destroy βcat and thus all cells accumulate extremely high levels of βcat , resulting in a smaller embryo in which all surviving cells choose posterior fates and secrete naked cuticle ( Figure 9B2; Ahmed et al . , 2002; Akong et al . , 2002 ) . Like our wildtype APC transgene ( Figure 9B3 ) , APC2DD fully restored alternating anterior ( denticle ) and posterior ( naked cuticle ) fates , resembling wild-type ( Figure 9B4 ) . In contrast , APC2AA largely failed to restore cell fates . Most embryos lost nearly all denticles ( Figure 9B5–7 ) , thus resembling embryos expressing an indestructible form of βcat ( Pai et al . , 1997 ) . This suggested APC2AA has severely reduced regulatory function , though it is not completely dead . The final test was to examine how well these APC2 mutants restored βcat destruction . In wildtype ( Figure 9C , J ) , Wg signal expressed in segmental stripes turns down destruction complex activity , leading to successive stripes of cells with only cortical βcat ( destruction complex on ) or with elevated cytoplasmic and nuclear βcat ( destruction complex turned down ) . In contrast APC2 APC1 maternal/zygotic mutants have exceptionally high βcat levels in all cells ( Figure 9D ) . We scored mutant embryos blinded to genotype . Consistent with its rescue of viability and cell fate , APC2DD fully rescued βcat destruction ( Figure 9E vs Figure 9F , K , O; 30/30 APC2DD embryos were scored as wild-type ) . In contrast , APC2AA had substantially reduced destruction complex function ( Figure 9O; 14/29 embryos scored as mutant , consistent with 50% zygotic rescue ) . Maternal/zygotic APC2 APC1 double mutants expressing APC2AA did not totally lose destruction complex function or the ability to respond to Wg signal . Stripes of cells with relatively reduced βcat levels were still present , but overall levels of βcat were substantially elevated and interestingly , levels were especially elevated in a subset of cells receiving Wg signal ( Figure 9G–I , L–N ) . Together , these data suggest that these two putative phosphorylation sites in B are critical for APC2 function in vivo . While textbook diagrams often depict the destruction complex as a four protein 1:1:1:1 Axin: APC:GSK3:CK1 complex , many lines of data suggest the functional destruction complex is a large multimeric complex ( e . g . , Fiedler et al . , 2011 ) . One mechanism involved was already known: the DIX domain of Axin polymerizes in a head-to-tail fashion , in which beta-sheet 2 of one DIX domain interacts with beta-sheet 4 of the next monomer , thus forming filaments , and this polymerization is essential for destruction complex function ( Schwarz-Romond et al . , 2007; Fiedler et al . , 2011 ) . These Axin polymers are responsible for assembly of the cytoplasmic puncta we use as a model . Previous and current data from a number of labs are consistent with the idea that the puncta serve as useful models of the smaller endogenous destruction complexes ( Faux et al . , 2008 ) , based on correlations between puncta formation , dynamics , and function in βcat destruction . We used the puncta to visualize assembly and dynamics of Axin:APC complexes in parallel with functional studies in colon cancer cells and Drosophila to define APC's role in destruction complex assembly and function . It is important to note that our data in SW480 cells involve significant over-expression—it will be useful in the future to examine destruction complex structure and function at endogenous levels , perhaps by tagging endogenous loci using CRISPR . APC is absolutely essential for the destruction complex to reduce βcat levels when Wnt signaling is off ( Mendoza-Topaz et al . , 2011 ) , but the mechanism by which it acts remained unclear . Our FRAP and super-resolution microscopy data support a model in which one role of APC is to promote/stabilize Axin self-assembly and slow Axin turnover in the destruction complex , thus increasing destruction complex multimerization and its ability to process βcat . We found APC does so by interacting with Axin via two different kinds of interaction sites: the known direct interaction with the SAMPs , and a novel interaction via APC's Arm rpts , which may be direct or indirect . Both interactions are critical for targeting βcat for destruction since APC without either SAMPs or Arm rpts cannot reduce βcat levels effectively . Almost all truncations in colon cancers remove the SAMPs ( Kohler et al . , 2008 ) . Earlier data from our lab also implicated APC's Arm repeats in Wnt regulation—our new results provide a mechanistic basis for this effect . APC2:Axin complexes were previously only resolved as co-localized spots . We provide the first glimpses inside the destruction complex . Super-resolution microscopy revealed that Axin puncta consist of Axin cables/sheets , which we hypothesize are bundled Axin polymers , assembled by the previously observed DIX domain polymerization ( Schwarz-Romond et al . , 2007 ) . Consistent with our observation that APC2 promotes growth and reduces dynamics of Axin complexes , APC2 cables intertwine with and bridge Axin cables . Both Axin interaction sites are required for these effects . Stimulating growth of Axin complexes via destruction complex stabilization by APC2 may be essential when proteins are expressed at endogenous levels , increasing local concentrations of all destruction complex components , accounting for the highly efficient βcat destruction observed in the presence of APC . It will be interesting to explore the nature of the protein network involved at even higher resolution . As noted above , thus far we have visualized complexes of APC2 and Axin expressed at significantly elevated levels—it will be important to verify and extend these studies to the more modest size complexes found in vivo during normal development , using CRISPR to tag endogenous loci . APC mutations in tumors do not eliminate APC; instead the N-terminus , Arm rpts and some βcat binding sites remain ( Kohler et al . , 2008 ) . Truncations cluster in the mutation cluster region ( MCR ) , suggesting this region is critical . R2 and B , which are removed by truncations in the MCR , are essential for βcat downregulation ( Kohler et al . , 2009; Roberts et al . , 2011 ) . In the textbook model , the destruction complex phosphorylates βcat and thus targets it to an E3 ligase . It was thus surprising that colon cancer cells with truncations disrupting R2/B have high levels of phosphorylated βcat , in contrast to tumors retaining R2/B ( Yang et al . , 2006 ) . Why do cells with non-functional APC have high levels of phosphorylated βcat ? Our data , together with earlier work , suggest Axin , the scaffold of the destruction complex , can mediate βcat phosphorylation even in cells with truncated APC , like the SW480 cells we use as a model , in which R2 and B are lost . Overexpressing Axin reduced total βcat levels , suggesting Axin can partially compensate for APC truncation , but phospho-βcat levels remained elevated inside the destruction complex . Interestingly , introducing APC2 reduced βcat accumulation in puncta and reduced phospho-βcat levels . These data suggest that without functional APC , Axin can mediate βcat phosphorylation , but transfer of βcat out of the destruction complex toward destruction is less efficient . It remains to be determined whether the accumulated phospho-βcat is actively transferred by Axin to the E3-ligase in the absence of functional APC , or if it is passively transferred due to a substantial increase in phospho-βcat . Thus while Axin can template βcat phosphorylation and can , at least in the presence of the truncated APC1 present in tumor cells , send it on to destruction , our data suggest APC promotes the rate at which βcat is transferred out of the destruction complex and sent to the proteasome . Future use of photoactivatible βcat constructs will further clarify this . Based on our model APC mediated βcat transfer is only possible when R2 and B are maintained in the truncated APC ( Figure 10 ) . R2 and B are essential for function in the absence of endogenous APC function in vivo in Drosophila ( Roberts et al . , 2011 ) . In SW480 cells our data suggest they are essential for APC2 to further simulate the rate of βcat destruction mediated by Axin transfection . Consistent with this , tumor cells that retain R2/B in the truncated APC have low levels of phospho-βcat ( Yang et al . , 2006 ) , suggesting APC is still able to facilitate βcat transfer out of the destruction complex due to the presence of the Arm rpts , R2 and B ( although the transfer would be less efficient due to loss of the SAMPs , the Axin binding sites ) . In contrast , truncated APC mutants that disrupt R2 and B function can associate with the Axin complex via the Arm rpts , but would not be able to assist in βcat transfer out of the destruction complex . The development of CRISPR knockout technology will allow future examination of the importance of truncated APC1 , as well as the endogenous APC2 and Axin expressed in colon cancer cells in destruction complex assembly and function . APC's Arm rpts bind cytoskeletal regulators ( Nelson and Nathke , 2013 ) , but their mechanism of action in Wnt signaling remained unclear . The fact that overexpressing hAPC1 fragments lacking the Arm rpts in SW480 cells rescued Wnt regulation initially suggested the Arm rpts were dispensable ( Rubinfeld et al . , 1997 ) . However these fragments were only tested in the presence of the truncated endogenous hAPC1 in these cells , which retains the Arm rpts . Thus hAPC1 fragments without Arm rpts may work with endogenous truncated APC , restoring partial function . In contrast , in flies , in the complete absence of endogenous APC , APC2 requires its Arm rpts for Wnt regulation ( McCartney et al . , 2006; Roberts et al . , 2012 ) . Our data provide the first mechanistic role for the Arm rpts in Wnt function , demonstrating they act as a regulated Axin interaction site , and revealing that this interaction is conserved in humans . Whether this interaction is direct or indirect remains to be determined . What then is the mechanism by which APC facilitates βcat destruction ? Two conserved APC motifs , R2 and B , are essential to target βcat for degradation ( Kohler et al . , 2009; Roberts et al . , 2011 ) . Deleting either leads to βcat accumulation in the destruction complex , suggesting R2 and B are critical for destruction complex throughput of βcat . Our colocalization , FRAP , and co-IP assays further suggest R2/B controls APC2:Axin association via APC2's Arm rpts . These data are consistent with a model in which APC2's R2 and B trigger an intramolecular conformational change in APC2 , releasing the Arm rpts from association with the Axin complex ( Figure 10 ) . In contrast , the SAMPs bind Axin independently of R2/B . In our model binding via the SAMPs would keep APC associated with Axin complexes when the Arm rpts are released , maintaining a functional destruction complex and facilitating βcat transfer to the E3-ligase . However , as noted above , it is also possible GSK3 may regulate the release of a complex of APC and phospho-βcat , allowing APC to shield phospho-βcat from dephosphorylation ( Su et al . , 2008 ) on the way to the E3-ligase . Motifs R2 and B include highly conserved serine/threonines matching the GSK3 and CK1 consensuses , and this region is phosphorylated by GSK3 in vitro . GSK3 increases APC2 dynamics , destabilizing the Arm rpts:Axin association via a mechanism that requires R2 and B . Strikingly , mutating two conserved serines in B to alanine blocked APC2's ability to reduce βcat levels , while a parallel phosphomimetic mutant did not disrupt APC2 function . We saw a similar reduction in APC2 function after mutating a conserved serine residue in R2 . Consistent with our data , CK1epsilon phosphorylation of hAPC1 R2 occurs in an Axin-dependent fashion , and site directed mutagenesis blocking phosphorylation of two conserved serines in R2 ( hAPC1 S1389 and S1392; distinct from and just C-terminal to the residue we mutated in APC2 R2 ) reduced the ability of a human APC1 fragment to down regulate Wnt signaling ( Rubinfeld et al . , 2001 ) , further suggesting R2 phosphorylation also is important for APC function . Our data are consistent with a model in which GSK3 phosphorylation of R2 and B could be one major regulatory step in APC2's dynamic cycle in the destruction complex , triggering release of the Arm rpts from the Axin complex , and thus allowing APC2 to promote βcat release for destruction ( Figure 10 ) . However , our data also show that mutating two residues in motif B to prevent their phosphorylation had only a subtle effect on APC2 dynamics . This may suggest additive roles for multiple phosphorylated residues , or may suggest the connection between phosphorylation , dynamics and function is more complex . Further , GSK3 can phosphorylate most of the other proteins in the destruction complex , and thus it clearly plays multiple roles in its function . Phosphorylation of R2/B itself could trigger release of APC's Arm rpts from the Axin complex , or alternately , phosphorylation may create a binding site for a binding partner that carries out this function . Intriguingly , a recent study found that α-catenin co-IPs with motif B of hAPC1 , and suggests α-catenin is an essential player in destruction complex function ( Choi et al . , 2013 ) . Perhaps R2/B phosphorylation by GSK3 regulates association of α-catenin with APC , and its binding induces alterations in APC:Axin interactions . It will be exciting to test this hypothesis . The multiple potential phosphorylation sites in R2/B , the hypomorphic nature of our double point mutant in Drosophila vs the complete loss of function after deleting either R2 or B ( Roberts et al . , 2011 ) , and the residual function of our two-residue phosphomimetic mutant may suggest different combinations of phosphorylated residues in R2 and B help tune and regulate APC function in the destruction complex . Our results also suggest GSK3 is likely to affect destruction complex structure and dynamics via several mechanisms , consistent with its known role in phosphorylating Axin and other sites in APC . Mathematical modeling suggests that a reduction in GSK3 ability to phosphorylate βcat is one key step in Wnt signaling activation , placing GSK3 activity in the center of ‘destruction complex inactivation’ ( Lee et al . , 2003 ) . It is intriguing to speculate about GSK3-mediated APC regulation in the context of Wnt signaling . Since GSK3 activity is inhibited by Wnt signaling , key residues in R2 and B that drive APC:Axin interaction dynamics would no longer be phosphorylated . Our data suggest this would decrease APC dynamics in the Axin complex ( since APC's Arm rpts would more strongly associate with Axin ) , perhaps reducing βcat transfer out of the destruction complex . Thus the drop in GSK3 activity upon Wnt signaling would not only downregulate the destruction complex via its main target , βcat , but through a core destruction complex component , APC , by acting via R2 and B . It will be of interest to explore further these complex interactions in both Wnt off and Wnt on states . One intriguing property of the APC2AA mutant in Drosophila is that Wnt regulation of βcat was maintained , but βcat levels were elevated in both Wnt off and Wnt on regions , presumably reflecting a less efficient destruction complex . This re-focused our attention on the much higher levels of βcat seen in APC2 APC1 mutants than occur in wildtype cells where Wg signal turns the destruction complex ‘OFF’ . These results suggest the destruction complex remains active when Wnt signal is on , but operates less effectively—this is very consistent with recent in vitro work ( Hernandez et al . , 2012 ) . A destruction complex operating at reduced levels could lead to high enough βcat levels to activate Wnt target genes but could maintain low enough levels to be quickly shut down when needed; it would also prevent the apoptosis seen in some cell types at extremely high βcat levels . The drop in GSK3 activity upon Wnt signaling would both decrease phosphorylated βcat and , based on our model , inhibit APC facilitating βcat transfer to the E3-Ligase . This speculative hypothesis would merge two recent studies proposing either reduced βcat phosphorylation ( Hernandez et al . , 2012 ) or a key regulated step involving transfer of βcat to the E3-ligase ( Li et al . , 2012b ) as key steps in Wnt regulation of the destruction complex . It will be intriguing to further probe the role of phosphorylation of R2/B in the mechanism of destruction complex action . Drosophila APC2 , human APC's Arm rpts ( 1–1012aa ) , and Axin were cloned as in Roberts et al . ( 2011 ) . In short , constructs were cloned into pECFP-N1 ( Clontech , Mountain View CA ) via Gateway ( Invitrogen , Waltham MA ) with either an N-terminal 3XFlag-tag or GFP-tag ( Roberts et al . , 2011 ) . The C-terminal-RFP vector was generated by cloning the Gateway cassette and TagRFP from pTag-RFP ( Evrogen , Russia ) into pECFP-N1 . For Phospho-APC2 constructs Serine 688 and 692 were changed using PCR stitching . Drosophila Axin fragments were N-term ( 1–54aa ) , RGS ( 55–171aa ) , GSK3 binding region ( 171–494aa ) , βcat binding region ( 494–532aa ) , PP2A binding region ( 532–666aa ) , and DIX ( 666–747aa ) . 1° antibodies: βcat ( BD Transduction , San Jose CA , 1:1000 ) , FlagM2 ( Sigma , St . Louis MO , 1:1000 , 1μg/ml IP ) , GFP ( Novus Biologicals , Littleton CO , 1:2000 ) , βcat-S33/37 ( Abcam , UK , 1:1000 ) , γ-tubulin ( Sigma , 1:5000 ) , tagRFP ( Evrogen , 1:5000 ) , hAPC1 ( Calbiochem , Billerica MA , 1:1000 ) , hAxin1 ( Cell Signaling , Danvers MA , 1:2000 ) , aPKCγ ( Santa Cruz Biotechnology , Dallas TX , 1:2000 ) . 2° antibodies: Alexa 568 and 647 ( Invitrogen , 1:1000 ) , HRP anti-mouse and anti-rabbit ( Pierce , Rockford IL , 1:50 , 000 ) . SW480 cells were cultured in L15 medium ( Corning , Tewksbury MA ) + 10% heat-inactivated FBS+1X Pen/Strep ( Gibco , Waltham MA ) at 37 °C without CO2 . Lipofectamine 2000 ( Invitrogen , Waltham MA ) was used for transfections following manufacturer's protocol . For immunostaining and IP cells were processed after 24 hr . Immunostaining was as described in Roberts et al . ( 2011 ) . In short cells were fixed 5 min in 4% formaldehyde/1XPBS , rinsed 5 min in 0 . 1% Triton-100/1XPBS , blocked with 1% normal goat serum/1XPBS , and incubated in antibody . Samples were mounted in Aquapolymount ( Polysciences , Warrington PA ) . For drug treatment 30 μM LiCl ( Sigma; dissolved in L15 ) , 2 μM BIO ( Tocris , UK ) or 25 μM MG132 ( Calbiochem; both dissolved in 99% EtOH ) were added 24 hr after transfection , and incubated 6h . Immunostained samples were imaged on a LSM Pascal microscope ( Zeiss ) and processed with the LSM image browser ( Zeiss , Germany ) . SIM microscopy was carried out on the Deltavision OMX ( GE Healthcare Life Sciences , Pittsburgh PA ) using 4% formaldehyde fixed samples mounted in Vectashield ( Vector , Burlingame CA ) following manufacturer's protocol . Images were processed using Imaris 5 . 5 , ImageJ and the LSM Image Browser . PhotoshopCS4 ( Adobe , San Jose , CA ) was used to adjust levels so that the range of signals spanned the entire output grayscale and to adjust brightness and contrast . We roughly calculated levels of expression by comparative immunoblotting , suggesting that expression levels were in the order: hAxin1-GFP > fly Axin-GFP > endogenous hAxin1 . Since fly Axin is not recognized by hAxin1 antibodies we did this in two steps , first comparing levels of tagged human Axin1 vs Drosophila Axin using antibodies against the GFP epitope tag , and then comparing the levels of the transfected human Axin1 vs the endogenous Axin1 protein in SW480 cells , using antibodies against human Axin1 ( Figure 1—figure supplement 1C ) . Because differences in expression levels took us out of the linear range of film , we diluted the more concentrated sample by a known amount . Protein bands in immunoblots of diluted samples ( Figure 1—figure supplement 1C3–4 ) were quantitated in ImageJ . Levels were normalized by ( 1 ) determining γ-tubulin levels in Figure 1—figure supplement 1C1–2 where samples were equally loaded , ( 2 ) by measuring γ-tubulin of diluted samples , and by calculating in the dilution factor . Once all samples were normalized to γ-tubulin the ratio between GFP-tagged fly and human Axin was calculated . Next , the ratio between hAxin1-GFP and endogenous hAxin1 was determined . Lastly the overexpression levels of fly GFP-Axin to endogenous hAxin1 were determined using the formula: ( ( ratio GFP-FlyAxin to hAxin1-GFP ) × ( ratio hAxin1-GFP to endogenous hAxin1 ) ) / Transfection efficiency . We then carried out a similar procedure for APC2 , comparing levels of tagged human APC1 cloned so as to mimic the truncated APC1 seen in SW480 cells vs tagged Drosophila APC2 using antibodies to the Flag epitope , and then levels of tagged truncated human APC1 vs that of the endogenous truncated APC1 protein ( Figure 1—figure supplement 1D ) . Lastly the overexpression levels of fly GFP-Axin to endogenous hAxin1 were determined using the formula ( ( ratio Flag-fly APC2 to Flag-hAPC1-1338 ) × ( ratio Flag-hAPC1-1338 to endogenous hAPC1-1338 ) ) /transfection efficiency . Z-projections of cell image stacks were generated using ImageJ . βcat fluorescent intensity: Cells were outlined , mean intensity measured , background subtracted , and βcat average intensity of a transfected cell normalized to mean of the βcat intensity of 2–3 adjacent untransfected cells . 10 cells were each measured in 3 independent experiments . Puncta colocalization of APC2ΔSAMPs with Axin in BIO/LiCl treated cells were determined by scoring for puncta formation in the APC2ΔSAMPs channel . 100 cells were scored in 3 independent experiments . APC:Axin complex size: Particle Analyzer of ImageJ was used . Background was subtracted and threshold for particles set to 200 . Cytoplasmic puncta of 10 cells were averaged . Cell images were taken with LSM Pascal ( Zeiss ) with a resolution of 5 . 7 pixel/μM . Mean number of particles per cells was calculated from size measurements . Puncta volumes were measured using Imaris Software ( Bitplane , Concord MA ) from image z-stacks acquired on the Deltavision OMX ( GE Healthcare Life Sciences ) . For comparing sequences , ClustalW2 ( EMBL , Germany ) was used for alignment . Statistical tests used the Student's t-test . FRAP was conducted using Eclipse TE2000-E microscope ( Nikon , Japan ) 24–72 hr after transfection . Drug treated samples were measured 6–48 hr after drug treatment ( 30–72 hr after transfections ) . Movies were taken at 1 frame/3 s or 1 frame/6 s for 20 min and bleaching was conducted for 8 s with 100% laser . Movies were processed using FRAP analyzer in ImageJ . Bleached area and cell were outlined , background was subtracted and the movie was processed with FRAP profiler . Values were normalized and recovery plateau and standard error were calculated by averaging 10 movies . For t1/2 values were processed in GraphPad ( La Jolla CA ) using non-linear regression ( curve fit ) -one phase decay . t1/2 of 10 movies was averaged and standard error calculated . IPs were as described in Li et al . ( 2012b ) . In short cells were lysed on ice 15 min in 150 mM NaCl , 30 mM Tris pH 7 . 5 , 1 mM EDTA , 1% Triton-X-100 , 10% glycerol , 0 . 5 mM DTT , 0 . 1 mM PMSF + proteinase/phosphatase inhibitors ( EDTA-free , Pierce ) , lysates cleared by centrifugation at 13 , 200 rpm 30 min at 4 °C , and preincubated with Sepharose beads 2h at 4 °C . 1 μg/ml antibody was added and incubated overnight at 4 °C . Beads were blocked in 5% BSA/1xPBS overnight at 4 °C , added to antibody-lysis mix , incubated 1–2 h at 4 °C , washed 5× with lysis buffer 4 °C , mixed with 2xSDS and incubated 10 min at 96 °C . Cell lysis for immunoblotting was similar . For βcat protein levels centrifugation was at 13 , 200 rpm for 30 min or 1000 rpm for 10 min at 4 °C . Drug treated cells were harvested after 6h . Proteins were run on 8% or 6% ( hAPC1 ) SDS gels and blotted to nitrocellulose . IP quantification was in ImageJ using Gel plot analyzer . IP baits were normalized to amount of IPed protein , and βcat protein levels were normalized to loading control γ-tubulin . R2/B fragments of hAPC1 ( 1355aa–1465aa ) and fly APC2 ( 632aa–733aa ) were cloned via Gateway into pdest15-GST tagged vector . Protein expression was induced via IPTG ( Apex , San Diego CA ) and protein purification was conducted using Glutathione beads ( Sigma ) . Kinase assays using human GSK3β ( Sigma#G4296 ) were conducted following the manufacturer's protocol . The GSK3 substrate peptide YRRAAVPPSPSLSRHSSPHQ ( pS ) EDEE ( based on human muscle glycogen , Signalchem , Canada ) was used as a positive control . Samples were run using Tricine SDS-PAGE ( Schagger , 2006 ) , and the gel was Coomassie stained and measured for radioactivity by exposing to film . Transgenic fly lines were generated by Best Gene ( Chino Hills , CA ) . APC2 transgenes were crossed into the APC2g10 APC1Q8 double mutant backgrounds as described previously ( McCartney et al . , 2006 ) . Maternal/zygotic double mutants for both APCs were generated using the FRT/FLP/DFS technique ( Chou and Perrimon , 1996 ) . Fly crosses and heat-shock conditions were as described in Roberts et al . ( 2011 ) . Embryonic lethality and cuticle preparations were conducted as described in Wieschaus and Nüsslein-Volhard ( 1998 ) .	An embryo starts off as a small ball of stem cells , each of which has the potential to become any type of cell in the body . Adult organs and tissues also contain small numbers of stem cells that can replace old or damaged cells . In both of these processes , stem cells need to ‘decide’ when they should start to change into a more specialized cell type , and which cell fate to choose ( e . g . , liver cell vs kidney cell ) . A signaling pathway involving Wnt proteins helps to direct many of these decisions . But if the ‘Wnt signaling pathway’ becomes activated at the wrong time , it can lead to cancer . For example , the first step in development of colon cancer is the inappropriate activation of Wnt signaling , and is most often caused by mutations in the gene that encodes a protein called APC . The APC protein is a tumor suppressor and normally inhibits Wnt signaling . However , even after over 20 years of effort , it remains largely mysterious how APC does this . APC is known to work with another protein called Axin as part of a large protein machine . This protein complex performs one of the first steps in a process that ultimately marks a key component of the Wnt signaling pathway for destruction . Pronobis et al . have now used a range of techniques to define APC's role in this so-called ‘destruction complex’ . This analysis revealed the internal structure of a complex made from APC and Axin , and showed that cable- and sheet-like assemblies of Axin were intertwined with APC cables . Further experiments then revealed how APC and Axin proteins are added into or leave these complexes , and showed that this is critical for this protein machine to work . Pronobis et al . 's data also suggest that APC plays two roles , which make the destruction complex more efficient . Firstly , it can interact with Axin via two separate interaction sites that help to assemble the destruction complex . Secondly , specific features in APC allow it to interact with a third protein ( called GSK3 ) , which can then regulate how APC interacts with Axin . One of the next challenges will be to uncover how APC helps to transfer the components of Wnt signaling to the next step of their destruction , and to clear up the role played by GSK3 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	A novel GSK3-regulated APC:Axin interaction regulates Wnt signaling by driving a catalytic cycle of efficient βcatenin destruction
Actin filaments assemble inside the nucleus in response to multiple cellular perturbations , including heat shock , protein misfolding , integrin engagement , and serum stimulation . We find that DNA damage also generates nuclear actin filaments—detectable by phalloidin and live-cell actin probes—with three characteristic morphologies: ( i ) long , nucleoplasmic filaments; ( ii ) short , nucleolus-associated filaments; and ( iii ) dense , nucleoplasmic clusters . This DNA damage-induced nuclear actin assembly requires two biologically and physically linked nucleation factors: Formin-2 and Spire-1/Spire-2 . Formin-2 accumulates in the nucleus after DNA damage , and depletion of either Formin-2 or actin's nuclear import factor , importin-9 , increases the number of DNA double-strand breaks ( DSBs ) , linking nuclear actin filaments to efficient DSB clearance . Nuclear actin filaments are also required for nuclear oxidation induced by acute genotoxic stress . Our results reveal a previously unknown role for nuclear actin filaments in DNA repair and identify the molecular mechanisms creating these nuclear filaments . Actin was first identified in muscle over seventy years ago ( Straub , 1942 ) and has been established as a component of non-muscle cells for nearly half a century ( Hatano and Oosawa , 1966 ) . Subsequent work revealed how actin filaments help organize the cytoplasm of all eukaryotic cells , supporting many fundamental biological processes , including: motility , division , phagocytosis , endocytosis , and membrane trafficking . The first reports of actin inside the nucleus of a cell appeared forty years ago ( LeStourgeon et al . , 1975 ) , and since that time , actin has been found in the nuclei of many different cell types , linked to a variety of nuclear processes ( Pederson and Aebi , 2002 ) . Recent work has identified the molecular mechanisms that control the nuclear concentration of actin , uncovering new roles for the actin-binding proteins profilin and cofilin as co-factors for actin's nucleocytoplasmic transport . The nuclear export factor exportin-6 ( XPO6 ) binds profilin-actin complexes in the nucleus—as well as a handful of other actin-binding proteins—and shuttles them into the cytoplasm ( Stüven et al . , 2003; Bohnsack et al . , 2006 ) . Conversely , nuclear import of actin is regulated by importin-9 ( IPO9 ) , which transports cofilin–actin complexes from the cytoplasm into the nucleus ( Dopie et al . , 2012 ) . In many organisms , the export factor XPO6 is not expressed in the oocytes of , and so the germinal vesicles contain a high concentration of actin ( Stüven et al . , 2003 ) . In Xenopus laevis oocytes , this germinal vesicle actin forms a filamentous mesh that protects nucleoli from gravity-induced aggregation ( Feric and Brangwynne , 2013 ) . Actin filaments associated with germinal vesicles of starfish oocytes facilitate nuclear envelope breakdown and form a contractile net that facilitates chromosome capture during mitosis ( Lénárt et al . , 2005; Mori et al . , 2014 ) . Several studies have also implicated nuclear actin filaments in oocyte transcription ( reviewed in Belin and Mullins , 2013 ) . In contrast , most somatic cells express some amount of XPO6 , and therefore , have a much lower concentration of actin in the nucleus than in the cytoplasm . Also , unlike Xenopus germinal vesicles , mammalian somatic nuclei contain relatively small amounts of filamentous actin ( Belin et al . , 2013 ) , suggesting that monomeric actin may play an important role . Monomers of actin and several actin-related proteins ( Arps ) , for example , are conserved components of chromatin-remodeling complexes ( Farrants , 2008 ) , and nuclear actin monomers inhibit the activity of the serum-responsive transcriptional co-activator MRTF ( myotonin-related transcription factor ) ( Vartiainen et al . , 2007; Mouilleron et al . , 2008 ) . Many reports have also linked actin to the regulation of RNA polymerases , although there are conflicting data on whether this activity depends on monomers or filaments ( Belin and Mullins , 2013 ) . Functions for filamentous actin in somatic cell nuclei are slowly beginning to emerge . Serum stimulation of quiescent fibroblasts ( Baarlink et al . , 2013 ) and integrin engagement in spreading cells ( Plessner et al . , 2015 ) induce transient ( <60 s ) bursts of nuclear actin polymerization , driven by the nucleation activity of formin-family proteins mDia1 and mDia2 . These short-lived filaments appear to promote activity of the transcriptional co-activator MRTF by depleting monomeric actin from the nucleus . Serum stimulation also activates the actin-severing protein MICAL-2 , which reversibly oxidizes actin monomers , rendering them incapable of inhibiting MRTF-dependent transcription ( Lundquist et al . , 2014 ) . Environmental stresses also promote actin assembly in somatic cell nuclei . Heat shock , dimethyl sulfoxide ( DMSO ) , depletion of ATP , and oxidative stress all induce formation of nuclear filament bundles that contain large amounts of cofilin ( Fukui , 1978 , Fukui and Katsumaru , 1980; Iida et al . , 1992; Pendleton et al . , 2003; Kim et al . , 2009 ) . In addition to its function as a co-factor for nuclear import , cofilin appears to play a structural role in these cofilin–actin rods , which are highly oxidized and appear to be held together by intermolecular disulfide bonds between cofilin molecules ( Pfannstiel et al . , 2001; Bernstein et al . , 2012; Zhang et al . , 2013 ) . Little is known about the physiological role of these cofilin–actin rods but they sense and perhaps regulate the reducing potential of the nucleus ( Bernstein et al . , 2012; Munsie et al . , 2012 ) . Many functions proposed for nuclear actin have been controversial , due in part to a lack of molecular tools for visualizing and perturbing actin inside the nucleus without affecting cytoplasmic actin ( Belin et al . , 2013 ) . The discovery of actin's nuclear import and export factors , along with the recent identification of some of the molecular mechanisms that create nuclear actin filaments , now enable us to make more specific perturbations of actin inside the nucleus . In addition , we and others have developed fluorescent probes that enable us to visualize actin monomers and filaments in the nuclei of live cells ( Baarlink et al . , 2013; Belin et al . , 2013; Plessner et al . , 2015 ) . Using these recently developed tools , we discovered that DNA damage induced by various genotoxic agents triggers formation of actin filaments inside the nucleus of mammalian cells . These filaments promote efficient repair of DNA double-strand breaks ( DSBs ) and are required for a DNA damage-associated burst of oxidation in the nucleus . DNA damage-induced nuclear actin structures differ in both composition and mechanism of assembly from those triggered by serum stimulation or by non-specific cell stresses . Specifically , we find that the actin regulators Formin-2 ( FMN2 ) and Spire-1/2 nucleate nuclear actin assembly in response to DNA damage . Homologs of Formin-2 and Spire-1/2 interact directly ( Quinlan et al . , 2007; Vizcarra et al . , 2011; Montaville et al . , 2014 ) and collaborate to form functional actin networks in mouse and Drosophila oocytes . Murine oocyte FMN2 and Spire are required for migration of meiotic spindles to the cortex , extrusion of polar bodies , and radial transport of vesicles ( Schuh and Ellenberg , 2008; Schuh , 2011 ) . In Drosophila , homologs of FMN2 ( Cappuccino ) and Spire collaborate to build actin networks required for oocyte polarity ( Quinlan et al . , 2005; Bor et al . , 2015 ) . We suggest that DNA damage-induced nuclear actin filaments may facilitate movement of chromatin and repair factors after DNA damage . Previously , we developed and validated a fluorescent probe , Utr230-EGFP-3×NLS ( Utr230-EN ) , capable of imaging actin filaments inside nuclei of live cells ( Belin et al . , 2013 ) . In mammalian tissue culture cells , the distribution of Utr230-EN exhibits one of three basic patterns: ( i ) accumulation on cytoplasmic stress fibers and diffuse localization throughout the nucleoplasm ( ∼53% of cells ) ; ( ii ) cytoplasmic stress fibers and punctate structures in the nucleoplasm ( ∼40% of cells ) ; and ( iii ) primarily cytoplasmic localization to stress fibers and/or aggregates at the nuclear periphery ( ∼7% of cells ) ( Figure 1—figure supplement 1 ) . Further analysis revealed that the punctate nucleoplasmic structures recognized by Utr230-EN are endogenous structures that contain short ( <0 . 5 µm ) actin filaments ( Belin et al . , 2013 ) ( Figure 1A ) . To characterize these nuclear actin filaments , we performed an immunofluorescence screen to test for co-localization with various nuclear landmarks; among the co-localization candidates , we selected was Rap1 , a marker for telomeres . We did not detect significant co-localization of nuclear actin with intact telomeres in HeLa cells , but when we uncapped the telomeres using MT-hTer-47A RNAi ( Marusic et al . , 1997; Li et al . , 2004 ) , we reproducibly observed the formation of larger nuclear actin structures , including the formation of long ( >1 µm ) filaments ( Figure 1B ) . 10 . 7554/eLife . 07735 . 003Figure 1 . DNA damage induces nuclear actin polymerization . ( A ) Localization of Utr230-EGFP-NLS ( Utr230-EN ) in unstressed HeLa cells . ( B ) Utr230-EN localization after induction of DNA damage via: 50 J/m-s UV exposure , 120-min incubation in media supplemented with 0 . 01% methyl methanosulfonate ( MMS ) , 120-min incubation in media supplemented with 50 pg/ml neocarzinostatin ( NCS ) , or 1 week post-transfection with telomere uncapping reagent MT-hTer-47A . ( C ) Percentage of cells with long ( >1 micron ) nuclear filaments after treatment with telomere uncapping reagent MT-hTer-47A ( 47A ) for 5–7 days . N = 83–90 cells . ( D ) Average number of long nuclear filaments per cell after treatment with telomere uncapping reagent 47A for 5–7 days . N = 83–90 cells . ( E ) Percentage of cells with long ( >1 micron ) nuclear filaments after incubation in 0 . 01% MMS for 120′ . N = 87–118 cells . ( F ) Average number of long nuclear filaments per cell incubation in 0 . 01% MMS for 120′ . N = 87–118 cells . ( G ) Comparison of length distributions of long nuclear filaments after treatment with either 47A for 5–7 days or 120′ in 0 . 01% MMS . Open circles indicate outliers . N = 94–95 filaments . ( H ) Utr230-EN localization after 120 minute incubation in 0 . 01% MMS including several classes of nuclear filaments , listed from top to bottom: elongated nucleoplasmic , clustered , peri- and intra-nucleolar . Nuclear filaments are indicated with yellow arrows . ( I ) Co-localization of peri-nucleolar filaments with an antibody detecting nucleolar protein fibrillarin . Asterisks indicate p-values < 10E-3 ( ) or 10E-4 ( * ) for all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 00310 . 7554/eLife . 07735 . 004Figure 1—figure supplement 1 . Distribution of Utr230-EN localization patterns in HeLa cells . The subcellular distribution of Utr230-EN ( diffuse nuclear and cytoplasmic , filamentous nuclear and cytoplasmic , or cytoplasmic only ) in untreated cells and cells treated with 0 . 01% or 0 . 05% MMS for 120′ . N = 206–311 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 004 Uncapped telomeres resemble double-strand DNA breaks , so we hypothesized that nuclear actin assembly might be triggered as part of a more general DNA damage response . To test this idea , we damaged HeLa cell DNA using a variety of genotoxic agents: incubation with 0 . 01% methyl methanesulfonate ( MMS ) ; incubation with 50 pg/ml neocarzinostatin ( NCS ) ; or exposure to 50 J/m-s ultraviolet radiation . These various treatments all triggered assembly of actin filaments in the nucleus ( Figure 1B ) . We compared nuclear actin filaments generated by telomere uncapping with those induced by genotoxic agents by counting the fraction of cells that contain nuclear filaments longer than 1 µm as well as the mean length of these long nuclear filaments in cells treated with either MT-hTer-47A for 5–7 days or 0 . 01% MMS for 2 hr . The fraction of cells with long filaments; the mean filament length; and the standard deviation of filament length were similar after the two treatments ( Figure 1C–G ) . Further analysis of nuclear filaments induced by MMS revealed the formation of three distinct classes of nuclear actin structure: ( i ) elongated nucleoplasmic filaments ( >1 µm ) ; ( ii ) amorphous , nucleoplasmic clusters; and ( iii ) nucleolus-associated filaments ( Figure 1H ) . The third class of structures includes both peri- and intra-nucleolar filaments , as judged by co-localization with the nucleolus marker , fibrillarin ( Figure 1I ) . To verify that DNA damage-induced nuclear actin filaments are endogenous structures , and not artifacts produced by expression of the Utr230-EN probe , we visualized them using other probes for filamentous actin . In HeLa cells not expressing Utr230-EN , fluorescent derivatives of phalloidin faintly but reproducibly detected all three classes of DNA damage-induced nuclear filaments ( Figure 2A ) . Detecting fluorescent phalloidin in the nucleus following DNA damage , however , presented technical challenges . High concentrations of fluorescent phalloidin bound to actin filaments in the cytoplasm swamp the much smaller signal from the nucleus . For example , the faint phalloidin fluorescence coming from the nucleus was difficult to detect by confocal microscopy , in part due to the relatively low light throughput of the pinholes . Wide-field deconvolution microscopy provided much more signal and enabled us to more easily observe the faint signal from phalloidin-bound filaments in the nucleus . 10 . 7554/eLife . 07735 . 005Figure 2 . Comparison of nuclear-targeted actin reporters for detection of DNA damage-induced nuclear actin filaments . ( A ) Faint detection of elongated nuclear actin filaments in the nucleoplasm and peri- and intra-nucleolar regions after 0 . 01% MMS treatment by phalloidin in untransfected cells . Arrows indicate locations of nuclear filaments , including elongated nucleoplasmic filaments ( top two rows ) and phalloidin staining of intra-nucleolar ( third row ) and peri-nucleolar ( bottom row ) structures . ( B ) Detection of elongated nuclear actin filaments after 0 . 01% MMS treatment by nuclear-targeted variants of three common live-cell actin reporters: Utr230-EN , Lifeact-EGFP-NLS ( Lifeact-EN ) , and F-tractin-EGFP-NLS ( F-tractin-EN ) . ( C ) Percentage of cells containing phalloidin-stainable nuclear filaments before and after MMS treatment within cell lines expressing actin reporters shown in ( B ) . N values are as shown . ( D ) Distribution of the number of phalloidin-stainable nuclear filaments per cell after MMS treatment within cell lines expressing actin reporters shown in ( B ) . Closed circles indicate outliers . ( E ) Distributions of lengths of long ( >1 micron ) phalloidin-stainable nuclear filaments per cell after MMS treatment within cell lines expressing actin reporters shown in ( B ) . N = 82–100 filaments . Open circles indicate outliers . ( F ) Scatter plot depicting the number of Utr230-EN detected nuclear filaments per cell as a function of the concentration of Utr230-EN in the nucleus following treatment with 0 . 01% MMS for 120 minute . N = 206 cells . ( G ) Scatter plot depicting the number of Lifeact-EN detected nuclear filaments per cell as a function of the concentration of Lifeact-EN in the nucleus following treatment with 0 . 01% MMS for 120′ . N = 214 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 005 We also detected DNA damage-induced nuclear actin filaments using Enhanced Green Fluorescent Protein ( EGFP ) fusions of the actin-binding peptides Lifeact and F-tractin ( Schell et al . , 2001; Riedl et al . , 2008 ) targeted to the nucleus ( Figure 2B ) . Using these probes , however , we detected filaments in a smaller fraction of 0 . 01% MMS-treated cells than in our Utr230-EN cell line . This is likely due to the fact that the higher fluorescence baseline of Lifeact-EN and F-tractin-EN in the nucleus obscures filaments ( Figure 2B ) , and this is consistent with a parallel study in which we observed much higher levels of diffuse fluorescence from Lifeact- and F-tractin-based probes in the cytoplasm compared to Utr230 ( Belin et al . , 2015 ) . To test this interpretation , we calculated the percentage of cells with phalloidin-stainable nuclear filaments in untransfected cells and cell lines expressing Utr230- , Lifeact- , or F-tractin-EN . In all cases , ∼5% of untreated cells contain phalloidin-stainable nuclear actin , whereas after induction of DNA damage by 0 . 01% or 0 . 05% MMS incubation , the percent of cells with phalloidin-stainable nuclear actin increases to ∼15% and ∼20% , respectively ( Figure 2C ) . Thus , the low frequency of cells with nuclear filaments detected by Lifeact- and F-tractin-EN does not correlate with fewer cells containing phalloidin-stainable nuclear actin . These results indicate that the DNA damage-induced nuclear filaments recognized by Utr230-EN are endogenous structures and that expression of nucleus-targeted actin probes does not significantly perturb nuclear actin assembly . When we limited our analysis to cells that contain large ( >1 µm ) nuclear actin structures , we observed—by all of our labeling methods—that the average number of actin structures per nucleus also increases significantly after DNA damage . We noticed , however , that the distribution filaments per nucleus is skewed toward both higher filament counts and longer filament lengths in cells expressing Lifeact-EN or Utr230-EN ( Figure 2D , E ) . In the Lifeact-EN cell line , high filament density in the nucleus correlates with higher Lifeact-EN expression levels , suggesting that Lifeact stabilizes filaments or enhances polymerization rate ( Figure 2G ) . The number of filaments per nucleus , however , does not correlate with Utr230-EN expression level ( Figure 2F ) , suggesting that this probe does not promote filament formation in the same way as Lifeact . It is possible that Utr230-EN expression enhances the DNA damage response or that the effect of this probe on nuclear filament stability saturates at very low concentrations , below those required for live-cell imaging . Given the poor contrast of the F-tractin-EN and Lifeact-EN probes and the concentration-dependent stabilizing effects of Lifeact-EN , we regard Utr230-EN as the best available tool for imaging nuclear actin filaments generated by DNA damage . Many DNA repair proteins have been identified by their effects on the kinetics of DSB detection and clearance . To determine whether nuclear actin plays a role in DNA repair , we compared the number of DSBs induced by MMS in cells with different concentrations of nuclear actin . Initially , we quantified DSBs by expressing a fluorescently tagged fragment of 53BP1 , a commonly used live-cell marker for DSBs ( Dimitrova et al . , 2008 ) . In our hands , however , the number of fluorescent foci generated by this 53BP1 reporter correlates directly with the expression level of the construct , making it unsuitable for quantitative studies ( Figure 3—figure supplement 1A ) . Instead , we quantified DSBs by immunofluorescence , using antibodies against either 53BP1 or histone variant pS139 H2AX . This method yielded consistent DSB counts across multiple replicate samples treated with the same dose of MMS ( Figure 3A ) . 10 . 7554/eLife . 07735 . 006Figure 3 . Nuclear actin polymerization is required for efficient double-strand break clearance . ( A ) Double-strand break ( DSB ) sites detected by 53BP1 immunofluorescence after 30′ , 60′ , 90′ , and 120′ incubations in 0 . 01% MMS . ( B ) Average number of DSB foci detected by 53BP1 immunofluorescence per cell after stable shRNA knockdown of the nuclear actin import factors , importin-9 ( IPO9 ) or cofilin , after 0 . 01% MMS incubation . N = 125–172 cells per condition . ( C ) Average number of DSB foci detected by gamma H2AX immunofluorescence per cell after stable shRNA IPO9 knockdown following 0 . 01% MMS incubation . N = 182–241 cells per condition . ( D ) Partial rescue of control DSB foci levels after IPO9 knockdown by overexpression of wild-type actin-NLS-P2A-mCherry but not non-polymerizing R62D mutant of actin-NLS-P2A-mCherry . N = 118–150 cells per condition . ( E ) Full rescue of non-Target control DSB foci levels after IPO9 knockdown by overexpression of wild-type actin-NLS-P2A-mCherry but not non-polymerizing R62D mutant of actin-NLS-P2A-mCherry . N = 262–291 cells per condition . ( F ) Comparison of the distributions of long ( >1 micron ) nuclear filaments per cell and 53BP1 foci counts . N = 206 cells . ( G ) Co-localization assays between Utr230-EN and DSBs after 120′ incubation in 0 . 01% MMS . Asterisks indicate p-values < 10E-2 ( * ) , 10E-3 ( ) , or 10E-4 ( * ) for all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 00610 . 7554/eLife . 07735 . 007Figure 3—figure supplement 1 . Knockdown of exportin-6 or IPO9 modulates filament formation after DNA damage . ( A ) Scatter plot comparing the number of 53BP1-mTagBFP2 foci per cell and 53BP1-mTagBFP2 expression level following treatment with 0 . 01% MMS for 120′ . N = 354 cells . ( B ) Localization of Utr230-EN after 0 . 01% MMS treatment following transient siRNA knockdown of exportin-6 ( XPO6 ) or IPO9 . ( C ) Quantification of elongated nuclear filament formation after increasing doses of MMS in IPO9 , XPO6 , and control siRNA knockdown lines . N = 126–150 cells . ( D ) Quantification of DSBs per cell detected by 53BP1 antibody after increasing doses of MMS in IPO9 , XPO6 , and control knockdown cell lines . N = 126–150 cells . Asterisks indicate p-values < 10E-2 ( * ) or 10E-3 ( ** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 007 To modulate the concentration of actin in the nucleus , we knocked down expression of either the nuclear import factor , IPO9 , or the nuclear export factor , XPO6 . For these experiments , we employed siRNAs previously validated in human cells ( Bohnsack et al . , 2006; Dopie et al . , 2012; Belin et al . , 2013 ) . Knocking down expression of XPO6 increased the fraction of cells with detectable nuclear actin structures as well as the average number of nuclear structures in each cell . This increase occurred in both untreated cells and cells treated with MMS . Conversely , cells from which IPO9 had been depleted lacked detectable nuclear actin structures , regardless of whether they were treated with MMS ( Figure 3—figure supplement 1B , C ) . Loss of XPO6 had no effect on the number of DSB foci generated by MMS treatment . Loss of IPO9 , however , significantly increased the number of DSBs per cell , judged using antibodies against 53BP1 ( Figure 3B , Figure 3—figure supplement 1D ) or pH2AX ( Figure 3C ) . We also observed an increase in the number of DSB foci when we depleted nuclear actin by knocking down its nuclear import co-factor , cofilin ( Figure 3B ) . To verify that IPO9 knockdown affects the number of DSB foci by decreasing the concentration of actin in the nucleus rather than by some other , off-target effect , we attempted to rescue the phenotype by expressing actin targeted to the nucleus by an IPO9-independent mechanism . In these experiments , we expressed either: an mCherry-3×NLS control construct; wild-type actin fused to mCherry-3×NLS; or a non-polymerizing R62D actin mutant fused to mCherry-3×NLS . Overexpressing wild-type actin-mCherry-NLS—but not the R62D mutant actin construct—decreased the number of DSB foci but did not completely rescue the loss of IPO9 ( Figure 3D ) . We figured that the partial rescue could be due to the fact that fluorescent actin fusions cannot be incorporated into some actin filaments , particularly those generated by formin-family nucleation factors ( Chen et al . , 2012 ) . We , therefore , modified our rescue constructs by inserting a 2A-peptide site to produce bicistronic expression of both a 3×NLS-tagged actin and an mCherry expression reporter ( Kim et al . , 2011; Huss and Lansford , 2014 ) . Overexpression of this non-fluorescent , wild-type actin-NLS , but not the R62D mutant , resulted in a complete rescue of DSB foci counts ( Figure 3E ) . These results indicate that nuclear actin polymerization is required for efficient DSB clearance and suggested that nuclear actin filaments may be generated by a formin-family actin nucleator . Finally , we asked whether the number of nuclear actin filaments correlates with the severity of DNA damage by comparing nuclear filaments and 53BP1 foci in the same cells . Surprisingly , we found no correlation between the numbers of nuclear actin filaments and DSBs ( Figure 3F ) and we observed only limited overlap between the 53BP1 and nuclear actin signals for all classes of DNA damage-induced filaments ( Figure 3G ) . More detailed analysis of our co-localization data revealed that , out of 198 filaments found in nuclei of cells treated with 0 . 01% MMS for 2 hr , only 22 ( 11 . 1% ) overlap or lie adjacent to 53BP1 foci . We cannot , however , rule out the possibility that some fraction of the observed overlap represents a functional interaction . One way to address this possibility would be to simultaneously track DSBs and nuclear actin filaments in live cells , but this would require a more reliable live-cell DSB marker . Given that expression of our fluorescent protein-tagged 53BP1 derivative is correlated with higher foci counts ( Figure 3—figure supplement 1A ) , which may include numerous spuriously detected loci , we found this tool to be unsuitable for quantitative live-cell imaging . We hypothesized that the DNA damage-induced nuclear actin structures we observe might be created by the same mechanisms that produce cofilin–actin rods in response to cell stress ( Fukui , 1978; Iida et al . , 1992 ) or nuclear actin filaments in response to serum stimulation ( Baarlink et al . , 2013 ) . The molecular mechanisms that produce cofilin–actin rods are not well understood , but as their name suggests , they contain the actin-binding protein cofilin ( Munsie et al . , 2012 ) . Due to the poor performance of commercially available antibodies in immunofluorescence , we raised our own polyclonal antibody against human cofilin in rabbit ( Covance ) capable of labeling nuclear cofilin–actin rods induced by treatment of cells with 10% DMSO . In contrast to the nuclear filaments generated by DMSO , the nuclear actin structures triggered by 0 . 01% MMS contained no detectable cofilin ( Figure 4A ) . 10 . 7554/eLife . 07735 . 008Figure 4 . DNA damage-induced nuclear actin filaments are distinct from cofilin–actin rods and mDia1/2-generated nuclear filaments . ( A ) Cofilin antibody staining in cells expressing Utr230-EN and treated with either 10% DMSO for 30′ or 0 . 01% MMS for 120′ . ( B ) Average number of nuclear actin filaments per cell after 120′ 0 . 01% MMS treatment in cells stably expressing mDia1 , mDia2 , or non-Target control hairpins . N = 115–128 cells per condition . ( C ) Distribution of nuclear actin filament classes after 120′ 0 . 01% MMS treatment in cells stably expressing mDia1 , mDia2 , or non-Target control hairpins . N = 115–128 cells per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 00810 . 7554/eLife . 07735 . 009Figure 4—figure supplement 1 . Validation of mDia1 and mDia2 knockdown by Western blot . Western blots in whole-cell lysate of knockdown lines generated using stably expressed Mission shRNA constructs ( Sigma ) . The RNAi Consortium Number ( TRCN ) for shRNA constructs is shown . Images have been enhanced for contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 009 Nuclear actin filaments generated in response to serum stimulation require the activity of formin-family actin nucleators , mDia1 and mDia2 ( Baarlink et al . , 2013 ) . To test whether mDia1 or mDia2 contribute to the production of DNA damage-induced actin filaments , we knocked down their expression using shRNAs . Compared to cells treated with a non-target hairpin , we found no effect of knocking down either mDia1 or mDia2 on the abundance or morphology of nuclear filaments produced by MMS treatment ( Figure 4B , C; Figure 4—figure supplement 1 ) . We conclude that DNA damage-induced filaments represent a biologically and biochemically distinct class of nuclear actin structures . To identify candidate regulators of DNA damage-induced nuclear actin assembly , we sifted through a list of mammalian proteins found by Matsuoka et al . ( 2007 ) to be phosphorylated in response to DNA damage . This list contains many proteins not previously associated with DNA repair , including several known actin regulators . One such protein is the actin nucleator Formin-2 ( FMN2 ) . In addition to its phosphorylation , the expression of FMN2 is upregulated in response to both ultraviolet radiation and hypoxia ( Yamada et al . , 2013a ) . To test whether FMN2 is involved in nuclear actin assembly , we used shRNAs to deplete it from cells expressing the Utr230-EN nuclear actin probe ( Figure 5—figure supplement 1 ) . When we induce DNA damage with 0 . 01% MMS , depletion of FMN2 completely inhibits nuclear actin assembly . Both the average number of elongated filaments and the distribution of filament types are nearly identical between MMS-treated , FMN2-knockdown cells and untreated controls expressing a non-target hairpin ( Figure 5A , B ) . 10 . 7554/eLife . 07735 . 010Figure 5 . Formin-2 is required for DNA damage-induced nuclear actin polymerization . ( A ) Average number of nuclear actin filaments per cell after 120′ 0 . 01% MMS treatment in cells stably expressing Formin-2 ( FMN2 ) or non-Target control hairpins . N = 117–151 cells per condition . ( B ) Distribution of nuclear actin filament classes after 120′ 0 . 01% MMS treatment in cells stably expressing FMN2 or non-Target control hairpins . N = 117–151 cells per condition . ( C ) Rescue of FMN2 hairpin in ( A ) by overexpression of hairpin-resistant mutant of mCherry-FMN2 . N = 105–127 cells per condition . ( D ) Rescue of FMN2 hairpin in ( B ) by overexpression of hairpin-resistant mutant of mCherry-FMN2 . N = 105–127 cells per condition . ( E ) Knockdown of FMN2 via stably selected hairpins increases the number of DSB foci after 0 . 01% MMS incubation . N = 238–323 cells per condition . ( F ) Rescue of control DSB foci levels after knockdown of FMN2 by overexpression of hairpin-resistant mutant of mCherry-FMN2 . N = 229–264 cells per condition . Asterisks indicate p-values < 10E-2 ( * ) or 10E-4 ( * ) for all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 01010 . 7554/eLife . 07735 . 011Figure 5—figure supplement 1 . Validation of FMN2 knockdown by Western blot . Western blots in whole-cell lysate of knockdown lines generated using stably expressed Mission shRNA constructs ( Sigma ) . The RNAi Consortium Number ( TRCN ) for shRNA constructs is shown . Images have been enhanced for contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 011 To verify the specificity of our hairpin RNA constructs , we attempted to rescue FMN2 knockdown by overexpressing an mCherry-FMN2 variant containing seven silent , hairpin-resistant mutations . Ectopic expression of this mCherry-FMN2 mutant completely restored the number and morphology of the actin structures generated by MMS treatment ( Figure 5C , D ) . As with IPO9 knockdown , we found that FMN2 knockdown produces a significant increase in DSB foci observed following MMS treatment . The increase in 53BP1 foci was also rescued by overexpression of hairpin-resistant mCherry-FMN2 ( Figure 5E , F ) . Since FMN2 homologs from other vertebrate organisms interact with Spire-family actin nucleators , we knocked down human Spire family members Spire-1 and Spire2 and determined their effect on DNA damage-induced filament assembly ( Figure 6—figure supplement 2 ) . Knocking down Spire-1 or Spire-2 individually has little effect on DNA damage-induced filament formation ( Figure 6—figure supplement 1 ) , but knocking down expression of both proteins at the same time produces the same effect as knocking down FMN2 , ablating MMS-induced nuclear filament formation ( Figure 6 ) . 10 . 7554/eLife . 07735 . 012Figure 6 . Composite knockdown of Spire-1 and Spire-2 inhibits DNA damage-induced nuclear actin assembly . ( A ) Distribution of nuclear actin filament classes after 120′ 0 . 01% MMS treatment in cells both stably expressing either Spire-1 ( SPIR1 ) or non-Target control hairpins and transiently transfected with either Spire2 ( SPIR2 ) or non-Target control siRNA . N = 111–185 cells per condition . ( B ) Average number of nuclear actin filaments per cell for knockdown conditions described in ( A ) . Asterisks indicate p-values < 10E-3 ( ) or 10E-4 ( * ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 01210 . 7554/eLife . 07735 . 013Figure 6—figure supplement 1 . Individual knockdown of either Spire-1 or Spire2 does not affect nuclear actin filament polymerization in response to DNA damage . ( A ) Average number of nuclear actin filaments per cell after 120′ 0 . 01% MMS treatment in cells stably expressing Spire-1 or non-Target control hairpins . N = 178–224 cells per condition . ( B ) Distribution of nuclear actin filament classes after 120′ 0 . 01% MMS treatment in cells stably expressing Spire-1 or non-Target control hairpins . N = 178–224 cells per condition . ( C , D ) Repeat of ( A , B ) for Spire-2 knockdown . N = 112–149 . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 01310 . 7554/eLife . 07735 . 014Figure 6—figure supplement 2 . Validation of Spire-1 and Spire-2 knockdowns by Western blot . Western blots in whole-cell lysate of knockdown lines generated using stably expressed Mission shRNA constructs ( Sigma ) or Silencer Select siRNAs ( Life Technologies ) . The RNAi Consortium Number ( TRCN ) for shRNA constructs and product ID for siRNAs are shown . Images have been enhanced for contrast . Only knockdown reagents yielding a successful knockdown ( Spire1 shRNA 219870 and Spire2 siRNA s39068 ) were used for analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 014 Previous work revealed that human FMN2 translocates from the cytoplasm into the nucleus in response to hypoxia ( Yamada et al . , 2013b ) . To determine whether DNA damage also induces nuclear translocation , we imaged subcellular distribution of mCherry-FMN2 before and after incubation with 0 . 01% and 0 . 05% MMS . We observed that this DNA-damaging agent produces a dose-dependent nuclear accumulation of FMN2 ( Figure 7A , B; Figure 7—figure supplement 1 ) , consistent with a direct role for FMN2 in the assembly of actin filaments in the nucleus following DNA damage . To better understand the nuclear accumulation of FMN2 , we searched for a potential nuclear localization sequence ( NLS ) in human FMN2 using the online bioinformatics tool , cNLS-Mapper ( Kosugi et al . , 2009 ) . We identified two putative NLS regions near the N-terminus of FMN2 , beginning at amino acids 6 and 411 of the NCBI reference sequence , accession number NP_064450 . 3 ( Figure 7C ) . To determine whether these regions are functional NLSs , we fused each to an EGFP reporter and expressed it in HeLa cells . Both sequences were sufficient to drive accumulation of EGFP in the nucleus , with stronger nuclear targeting observed for EGFP-NLS2 ( Figure 7D , E ) . Mutation of NLS2 at three positively charged residues , K414A/R415A/R416A , efficiently blocks nuclear accumulation of mCherry-FMN2 in response to DNA damage ( Figure 7F ) . 10 . 7554/eLife . 07735 . 015Figure 7 . Formin-2 accumulates in the nucleus in response to DNA damage . ( A ) Localization of transiently expressed mCherry-FMN2 in untreated cells and after 120′ incubation in 0 . 01% or 0 . 05% MMS . ( B ) Average ratio of nuclear vs cytoplasmic integrated fluorescence intensity of mCherry-FMN2 in untreated cells and after 0 . 01% and 0 . 05% MMS . N = 71–94 cells per condition . ( C ) Domain organization of human FMN2 including two putative nuclear localization sequence ( NLS ) sites identified using cNLS Mapper ( Kosugi et al . , 2009 ) . ( D ) Localization of EGFP fused to each putative NLS . ( E ) Nucleocytoplasmic ratio of EGFP fused to each putative NLS . N = 140–165 cells . ( F ) Nucleocytoplasmic ratio of the K414A/R415A/R416A NLS2 mutant mCherry-FMN2 before and after 0 . 05% MMS treatment . N = 108–115 cells . Asterisks indicate p-values < 10E-3 ( ) or 10E-4 ( * ) for all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 01510 . 7554/eLife . 07735 . 016Figure 7—figure supplement 1 . Gallery of mCherry-FMN2 localization before and after MMS treatment . Examples of mCherry-FMN2 localization in ( A ) untreated cells and cells treated with ( B ) 0 . 01% MMS for 120′ or ( C ) 0 . 05% MMS for 120′ . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 016 To determine whether nucleolus-associated actin filaments are involved in regulating a DNA damage-specific change in nucleolar morphology , we labeled nucleoli using fibrillarin and measured ( i ) the number of nucleoli per cell and ( ii ) nucleolar area after MMS treatment ( Figure 8 ) . We performed these experiments in both control cells expressing a non-target hairpin and an IPO9 shRNA knockdown line . Surprisingly , while the average number of nucleoli is constant , induction of DNA damage decreases nucleolar size by an average of ∼20% . Equivalent decreases in nucleolar area were observed in both control and IPO9 knockdown cells , however , indicating that the nucleolar area reduction is nuclear actin-independent . 10 . 7554/eLife . 07735 . 017Figure 8 . Nucleolar size decreases after DNA damage through a nuclear actin-independent pathway . ( A ) Localization by immunofluorescence of heterochromatin marker H3K9me3 and nucleolar marker fibrillarin in untreated cells stably expressing non-Target and IPO9 hairpins . ( B ) Repeat of ( A ) following 0 . 01% MMS incubation for 120′ . ( C ) Distribution of nucleolar areas ( in pixels ) before ( N = 2313 nucleoli ) and after 0 . 01% MMS incubation ( N = 2299 nucleoli ) in IPO9 knockdown cells . ( D ) Distribution of nucleolar areas ( in pixels ) before ( N = 2094 nucleoli ) and after 0 . 01% MMS incubation ( N = 2180 nucleoli ) in control cells . ( E ) Distribution of average numbers of nucleoli per cell before ( N = 406 cells ) and after 0 . 01% MMS incubation ( N = 386 cells ) in IPO9 knockdown cells . ( F ) Distribution of average numbers of nucleoli per cell before ( N = 375 cells ) and after 0 . 01% MMS incubation ( N = 411 cells ) in control cells . ( G ) Box plots comparing distributions shown in ( C , D ) . Asterisks indicate p-values < 10E-4 ( * ) . ( H ) Bar plots comparing averages from distributions shown in ( E , F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 017 In some cases , DNA damage has been shown to result in changes in the oxidative state of the nucleus , and it has been recently shown that oxidative stress can activate a subset of DNA-repair signaling cascades independently of DNA damage ( Guo et al . , 2010; Ditch and Paull , 2012 ) . Oxidative pathways also regulate previously studied cases of stress-induced nuclear actin polymerization ( Pfannstiel et al . , 2001; Bernstein et al . , 2012; Lundquist et al . , 2014 ) . To assess whether oxidative signaling is involved in regulation of nuclear actin filaments induced by MMS , we generated a construct containing the redox-sensing fluorophore roGFP2 fused with 3×NLS ( roGFP2-NLS ) ( Lohman and Remington , 2008 ) . roGFP2 is variant of Green Fluorescent Protein ( GFP ) engineered to introduce 2 cysteines to the interior of the GFP beta barrel , and it has been extensively used to measure oxidation changes in live cells ( Merksamer et al . , 2008; Al-Mehdi et al . , 2012 ) . The roGFP2 excitation spectrum contains 2 peaks , at 488 nm and at 405 nm . In the oxidized form of roGFP2 , excitation at 405 nm is increased and excitation at 488 nm is decreased , relative to the reduced form . Thus , the ratio of emission intensities following excitation of roGFP2 at these wavelengths under different conditions can be used to measure the relative oxidation state . We used roGFP2-NLS to measure nuclear oxidation levels and found that treatment with 0 . 01% MMS generates an oxidative burst in the nucleus ( Figure 9A ) . Depletion of nuclear actin by IPO9 knockdown has no effect at this MMS dosage ( Figure 9B ) . Curiously , however , when the MMS treatment was increased to 0 . 05% , IPO9 knockdown abolished nuclear oxidation; we observed the same result following knockdown of FMN2 ( Figure 9C–E ) . We measured the distribution of classes of nuclear actin filaments at MMS doses ranging from 0 . 01% to 0 . 05% , and found that increased MMS levels correlate with increases in the fraction of cells containing clustered bodies of nuclear filaments ( Figure 9F ) . These results suggest that clustered filaments may participate in regulating the nuclear redox state . 10 . 7554/eLife . 07735 . 018Figure 9 . Nuclear actin is required for regulation of nuclear oxidation after acute DNA damage . ( A ) False-colored images of roGFP2-NLS signal ( ratio of emission intensities after 488-nm and 405-nm excitation ) after 120′ in 0 . 01% MMS in non-Target control and IPO9 knockdown cell lines . ( B ) Average roGFP2-NLS signal integrated across the nuclear area in non-Target control or IPO9 hairpin lines in untreated and 0 . 01% MMS-treated cells . N = 108–144 cells per condition . ( C ) False-colored images of roGFP2-NLS signal ( ratio of emission intensities after 488-nm and 405-nm excitation ) after 120′ in 0 . 05% MMS in non-Target control and IPO9 knockdown cell lines . ( D ) Average roGFP2-NLS signal in nuclei of non-Target control or IPO9 hairpin lines in untreated and 0 . 05% MMS-treated cells . N = 106–124 cells per condition . ( E ) Average roGFP2-NLS signal in the nuclei of cells stably expressing non-Target or FMN2 shRNA in untreated cells and cells treated with 0 . 01% or 0 . 05% MMS for 120′ . N = 109–162 cells . ( F ) Distribution of nuclear actin filament classes after incubation in MMS concentrations from 0 . 01% to 0 . 05% for 120′ . N = 95–159 cells per condition . Asterisks indicate p-values < 10E-3 ( ) or 10E-4 ( * ) for all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 07735 . 018 Biochemical , cell biological , and genomic studies have revealed many ancient and important connections between actin family proteins and DNA ( Belin and Mullins , 2013 ) , but we still know little about the functions of actin inside the nucleus . For example , many chromatin-remodeling complexes contain monomeric actin or Arps as subunits , but the contribution these subunits make to chromatin remodeling remains poorly understood . In addition , cellular stresses such as ATP and nutrient deprivation , heat shock , and various chemical toxins induce the assembly of cytoplasmic actin bundles as well as cofilin–actin rods in the nucleus . Although they are thought to sense reactive oxygen species ( Bernstein et al . , 2012 ) , the precise function of nuclear cofilin–actin rods remains unclear . One of the most convincing functions proposed for nuclear actin is regulation of the actin-binding transcription factor MRTF . In this example , serum stimulation of quiescent cells induces assembly of actin filaments in both the cytoplasm and the nucleus ( Baarlink et al . , 2013 ) . Assembly of filaments in the cytoplasm drives MRTF into the nucleus , while the assembly of filaments inside the nucleus depletes monomeric actin from this compartment and relieves its inhibitory effect on MRTF ( Vartiainen et al . , 2007 ) . This mechanism relies on the nucleation activity of Diaphanous-family formins , mDia1 and mDia2 , and highlights the existence of a stable pool of polymerizable actin monomers inside the nucleus . We have uncovered another , previously unknown signaling pathway that generates actin filaments inside the nucleus , one that relies on the Cappuccino-family formin , FMN2 , and two Spire-family actin regulators , Spire-1 and Spire-2 . In response to DNA damage , these nucleation factors create a variety of actin structures in the nucleus , including long nucleoplasmic filaments , nucleolus-associated filaments , and amorphous clusters . These may represent functionally distinct species or possibly different time points in the evolution of a single structure . Regardless , as judged by the number of 53BP1 foci , nuclear actin assembly promotes clearance of double-strand DNA breaks . In contrast to the role of nuclear actin in MRTF function , however , filament formation appears to be more important than monomer depletion in the response to DNA damage . A functional role specific to filaments is supported by the fact that removing all actin ( monomer and polymer ) from the nucleus has the same effect on MMS-induced 53BP1 foci as simply inhibiting filament formation by depleting the nucleation factor , FMN2 . If monomer depletion were the primary function of DNA damage-induced filaments , then IPO9 knockdown would not have the same effect on DNA repair as filament inhibition . Little work has been done on human FMN2 , but mouse and Drosophila FMN2 homologs are relatively well studied . The nucleation activity of the Drosophila FMN2 homolog , Cappuccino , is regulated by intramolecular contacts between a C-terminal region near the actin-nucleating domain and an N-terminal sequence ( Bor et al . , 2012 ) . This autoinhibitory interaction is structurally distinct from the DID/DAD interaction that regulates activity of the Diaphanous-family formins . We have identified two nuclear-localization sites in the N-terminus of human FMN2 , which also contains two previously discovered DNA damage-induced phosphorylation sites . If human FMN2 is also autoinhibited by an interaction between N- and C-terminal regions , it is possible that DNA damage-induced phosphorylation promotes both nuclear translocation and actin nucleation . Nuclear actin filaments could potentiate the DNA damage response by several mechanisms: ( 1 ) altering mobility or organization of chromatin; ( 2 ) recruiting or delivering repair factors to sites of damage; or ( 3 ) binding and sequestering nuclear factors that would otherwise inhibit DNA repair . The third possibility is consistent with the minimal co-localization we observe between nuclear actin filaments and 53BP1 foci . Recent studies , however , have also demonstrated that DNA damage alters the mobility of DNA loci , perhaps facilitating their interaction with DNA repair proteins that are expressed at extremely low concentrations ( Dimitrova et al . , 2008; Miné-Hattab and Rothstein , 2013 ) . One attractive possibility is that myosin motors in the nucleus may drive clustering of repair components or DSB sites to facilitate homologous recombination . This would be analogous to the role of FMN2/Spire-generated actin filaments in generating contractile structures that link cargo vesicles in mouse oocytes . The DNA damage-induced recruitment of actin filaments to nucleoli is somewhat mysterious . We initially hypothesized that the nucleolar filaments may regulate some DNA damage-specific change in nucleolar morphology , analogous to the role of actin filaments in germinal vesicles of Xenopus oocytes . Although we observed a 20% decrease in average nucleolar area after MMS treatment , depleting nuclear actin by knocking down IPO9 had no measurable effect on this phenomenon . Given that nucleoli are the sites of rRNA transcription , we speculate that nuclear actin filaments may play a role in rRNA production or export , perhaps promoting stress-induced changes in translation . Clarifying the potential nucleolar functions of actin will require the identification of more nucleolus-associated actin-binding proteins . In addition to filamentous structures , DNA damage also induces formation of amorphous actin clusters in the nucleus . The number and size of these clusters increases with increasing doses of genotoxic agents , suggesting that they may be related to pre-apoptotic signaling . Consistent with this idea , we found that blocking nuclear filament assembly inhibits nuclear oxidation induced by high concentrations of MMS . Several prior studies suggested that actin acts as a sensor of cellular oxidation levels . For example , in the yeast Saccharomyces cerevisiae , cellular oxidation triggers formation of amorphous ‘actin bodies’ generated by intermolecular disulfide bonds ( Farah and Amberg , 2007 , 2011 ) . Similar sensing of nuclear oxidation has also been proposed for nuclear cofilin–actin rods , which contain disulfide-linked cofilin oligomers ( Bernstein et al . , 2012 ) . While more work will be required to determine whether DNA damage-induced nuclear filament clusters are also oxidation sensors , we demonstrate clearly that actin can alter the nuclear redox state . Given that all previously observed cases of nuclear actin assembly are connected to oxidation , we suggest that nuclear actin may be an important mediator of redox signaling . Together with other recent studies , our work reveals that multiple cellular signals can trigger assembly of a variety of actin structures in the nucleus , each with a unique morphology , function , and regulatory mechanism . One common theme , however , appears to be a link between the state of actin in the nucleus and the state of actin in the cytoplasm . For example , both serum stimulation ( Vartiainen et al . , 2007; Baarlink et al . , 2013 ) and DNA damage ( Zuchero et al . , 2012; and unpublished observations ) induce a dramatic increase in cytoplasmic actin polymerization . In both cases , this cytoplasmic actin polymerization correlates with nuclear accumulation of formin-family actin nucleators . We , therefore , speculate that changes in the actin monomer/polymer ratio in the cytoplasm may facilitate nuclear accumulation of actin regulators , coordinating the cytoplasmic and nuclear responses to stimuli . We used pEGFP-C1 or pmCherry-C1 ( Clontech , Mountain View , CA , United States ) as the host vector for all EGFP fusions , with N-terminal EGFP fusions inserted into the unique AgeI and NheI sites . Lifeact and all NLS sequences were cloned directly using annealing primers purchased from Elim Biopharmaceuticals . F-tractin and 3×NLS-peptide 2A inserts were synthesized and inserted into backbone vectors using custom cloning services from GenScript . Human actin was subcloned from cDNA purchased from GE Dharmacon ( formerly Open Biosystems ) . 53BP1 fragment was subcloned from Addgene plasmid #19835 ( Dimitrova et al . , 2008 ) and was provided by Beth Cimini . mCherry-FMN2 was provided by Sonia Rocha and includes the following N-terminal insertion in comparison to the NCBI human FMN2 reference sequence ( accession number NP_064450 . 3 ) : AGATCTCATTCGATTCGCACGGTGGAGATTAAAGTCCCCGAGATAGAGGAAACGTTTTTCGCGCCCAGGTTCAGCGAGGAGCCGCGCGGGGGCAGAGGGGGCGGCGGCGGCGGGCGGGGAGCCAGGCCCGAGCTGCGTTCTGCGCAGCCATTGGTGGGCGCCGCACTCTGCACTGAGCATGTTCGCGCCCCGCCGGCCCCTAGCCGCAGCCGCAGCCGCAGCGACGGCAGCCACGGGAGCCGCCGCGCATTATGCAAAGCGGCGGCAGATGCGAGCGGGGCCAGCCGGGCGCGCGTCGGCCTCCCCTCCCAGCGGCTCCCCCCGCCGCCGCCTGACTCTCCCGGGAGACTCCCTAGGCCCGGGATTGCACC . roGFP2 was a gift from Philip Merksamer and Ferroz Papa . Lentiviral packaging vectors were provided by John James . Mutagenesis of mCherry-FMN2 was performed using a Q5 polymerase site-directed mutagenesis ( SDM ) kit according to the manufacturer's protocol , modified to include the Q5 High GC Enhancer and an extension temperature of 75°C ( New England Biolabs , NEB ) . Primer sequences for SDM were generated using the companion NEB primer design tool . All new constructs generated for this study are being made available on Addgene . HeLa cells and HEK293T lentivirus-generating cells ( ATCC ) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% Fetal Bovine Serum ( FBS , Atlas Biologicals , Fort Collins , CO , United States ) , 2 mM L-glutamine and penicillin–streptomycin ( UCSF Cell Culture facility ) at 37°C with 5% CO2 . For transient transfection of non-knockdown constructs in HeLa cells , cells were transfected using Lipofectamine LTX ( Life Technologies ) according to the manufacturer's protocol . All transient transfections were performed 24–72 hr prior to data collection . Stable cell lines of Utr230-EGFP-NLS were generated as described in Belin et al . ( 2013 ) . For knockdown constructs , lentiviral packaging vectors pMDG . 2 and pCMV∆8 . 91 were co-transfected with shRNA vectors into HEK293T cells using GeneJuice ( EMD Millipore ) as described in James and Vale , 2012 . Lentiviral particles were harvested 2 and 3 days following transfection , sterile-filtered using 0 . 2-μm Steriflip filter units ( EMD Millipore ) and either used immediately or stored long-term at −80°C . Mock siRNA and Silencer Select siRNAs directed against human IPO9 , XPO6 , and Spire2 were purchased from Life Technologies . Transient reverse transfection was performed using Lipofectamine RNAiMAX ( Life Technologies ) according to the manufacturer's protocol . At 8–24 hr following transient transfection , the cell medium was replaced . Cells were split into flasks and/or fibronectin coverslips 3 days after transfection and were either fixed for imaging or lysed for Western blotting at 5 days following transfection . Non-Target shRNA and Mission shRNAs directed against mDia1 , mDia2 , cofilin , FMN2 , Spire-1 , and IPO9 were purchased as bacterial glycerol stocks from Sigma , and lentiviral particles were generated as described above . HeLa cells were transduced with lentiviral particles and split after 8–16 hr into media supplemented with 0 . 3 μg/ml puromycin ( Thermo Fisher Scientific ) for stable selection . All knockdown constructs were pre-validated by qRT-PCR using the SuperScript III Platinum SYBR Green One-Step qRT-PCR kit ( Life-Technologies ) and run on a Stratagene Mx3005P RT-PCR thermocycler ( UCSF Center for Advanced Technology ) . RNA was insolated using the TRIzol Plus RNA Purification Kit according to the manufacturer's instructions ( Life Technologies ) . qRT-PCR primers were designed using Primer BLAST ( NIH ) . Knockdowns verified by qRT-PCR were subsequently validated by Western blotting ( performed as described in Belin et al . , 2013 but modified to include 1× Detector Block [KPL/Sera Care Life Sciences] as the primary antibody diluent ) . Our cofilin antibody was generated in rabbit by Covance from purified human cofilin; sera from inoculated rabbits rather than purified antibody was used in immunofluorescence assays . Commercial primary antibodies used in this study follow: anti-tubulin ( Sigma , Cat . No . T29026 ) , anti-fibrillarin ( Abcam , Cat . No . ab4566 ) , H3K9me3 ( Abcam , Cat . No . ab8898 ) , anti-53BP1 ( Novus Biologicals , Cat . No . NB100-305 ) , anti-H2AX pS139 ( Millipore , Cat . No . 05–636 , Clone JBW301 ) , IPO9 ( Abcam , Cat . No . ab52605 ) , mDia2 ( ECM Biosciences , Cat . No . DP3491 ) , mDia1 ( BD Biosciences , Cat . No . 610848 ) , FMN2 ( Abnova , Cat . No . H00056776-A01 ) , Spire-1 ( Aviva Systems Biology , Cat . No . OAAB12896 ) , Spire2 ( ThermoFisher—Pierce , Cat . No . PA5-24099 ) . For telomere uncapping assays , MT-hTer-47A shRNA cell lines were was generated as described in Stohr and Blackburn ( 2008 ) and cells were fixed for imaging one week after initial infection . For global DNA damage induction via UV , cells were washed three times in calcium- and magnesium-free Phosphate-Buffered Saline ( PBS ) . The PBS was removed and cells were exposed to 50 J/m-s UV in a hybridization oven . After exposure , cells were washed once and complete medium was added . UV-treated cells were then incubated at 37°C with 5% CO2 for 6 hr and then fixed for imaging . For neocarzinostatin ( NCS; Sigma ) treatment , NCS was added to complete medium to a final concentration of 50 pg/ml . Cells were washed in PBS and NCS-supplemented media was added; cells were then incubated at 37°C with 5% CO2 for 6 hr and fixed for imaging . For methyl methanosulfonate ( MMS; Sigma ) treatment , MMS was added to complete medium at a final concentration of either 0 . 01% or 0 . 05% vol/vol . Cells were washed in PBS and MMS-supplemented media was added; cells were then incubated at 37°C with 5% CO2 for 2 hr or as indicated before fixation for imaging . Cells were passaged onto glass coverslips coated with 10 μg/ml fibronectin ( Invitrogen ) and cultured overnight to 30–60% confluence . Coverslips were fixed at room temperature for 15–30 min in 4% paraformaldehyde ( prepared from 16% solution from Electron Microscopy Sciences ) in PBS ( UCSF Cell Culture Facility ) . Cells were permeabilized in 0 . 1% Triton-X-100 ( Sigma ) in PBS for 3–5 min and blocked in 1× blocking buffer ( abcam ) /PBS at room temperature for 60 min . Primary antibody incubations were performed for 60 min at room temperature or overnight ( anti-53BP1 ) in 1× blocking buffer ( abcam ) /PBS . Cells were incubated in Alexa Fluor-labeled secondary antibodies ( Invitrogen ) in 1X blocking buffer ( abcam ) /PBS for 30 min at room temperature and mounted on slides Fluoromount G with DAPI ( Affymetrix eBioscience ) . For phalloidin staining , cells were fixed as above and stained for 15 min in 0 . 1% Triton/PBS with 100 μM Alexa Fluor 568-phalloidin ( Invitrogen ) . Images were acquired using a DeltaVision RT system ( Applied Precision ) with a Photometrics CoolSnapHQ camera using a 100× 1 . 40 NA UPlanSApo objective ( Olympus ) . Images were processed for contrast enhancement and background noise reduction using ImageJ ( National Institutes of Health ) . All image analysis was performed using custom ImageJ macros ( see Source code 1 ) . All figures and statistical analyses were generated in R . Reported p-values are the results of a two-tailed t-test .	In animals , plants , and other eukaryotic organisms , a cell's DNA is contained within a structure called the nucleus , which separates it from the rest of the interior of the cell . Filaments of a protein called actin are normally found outside the nucleus , where they help give the cell its overall shape and organize its contents . However , these filaments can sometimes form inside the nucleus in response to a sudden increase in heat or another type of stress . However , it was not clear what role these actin filaments play in the nucleus because it was difficult to distinguish them from the actin filaments that form in other parts of the cell . Researchers have recently developed new techniques to study actin filaments inside the nuclei of live cells under a microscope , using fluorescent protein tags . Here , Belin et al . —including some of the researchers involved in the previous work—used this technique to investigate whether DNA damage causes actin filaments to form in the nuclei of human cells . The experiments show that DNA damage does indeed lead to the formation of actin filaments in the nucleus . In a structure within the nucleus called the nucleolus , the actin filaments are short . However , in the rest of the nucleus , the actin forms long filaments and dense clusters . Cells that contained lower levels of actin were less able to repair their DNA than normal cells . Belin et al . also identified three proteins—called Formin-2 , Spire-1 , and Spire-2—that assemble the actin filaments in the nucleus . These proteins are also required to make actin filaments in other parts of the cell . The experiments show that the level of Formin-2 increases in the nucleus after DNA damage , and that the DNA of cells lacking this protein is more severely damaged . Belin et al . 's findings reveal a new role for actin in the repair of DNA and the next challenge is to understand the details of how this works .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	DNA damage induces nuclear actin filament assembly by Formin-2 and Spire-1/2 that promotes efficient DNA repair
Distinct microbial ecosystems have evolved to meet the challenges of indoor environments , shaping the microbial communities that interact most with modern human activities . Microbial transmission in food-processing facilities has an enormous impact on the qualities and healthfulness of foods , beneficially or detrimentally interacting with food products . To explore modes of microbial transmission and spoilage-gene frequency in a commercial food-production scenario , we profiled hop-resistance gene frequencies and bacterial and fungal communities in a brewery . We employed a Bayesian approach for predicting routes of contamination , revealing critical control points for microbial management . Physically mapping microbial populations over time illustrates patterns of dispersal and identifies potential contaminant reservoirs within this environment . Habitual exposure to beer is associated with increased abundance of spoilage genes , predicting greater contamination risk . Elucidating the genetic landscapes of indoor environments poses important practical implications for food-production systems and these concepts are translatable to other built environments . Microbial activity is an inherent feature of food-processing systems , influencing the quality and healthfulness of foods for human consumption . Like other indoor environments , the building materials , substrates , and physiochemical conditions encountered by microbes in food-processing facilities differ dramatically from the outdoor conditions to which microbial life evolved ( Kelley and Gilbert , 2013 ) . How this has impacted the adaptation and ecological assemblages of microbes in food systems is largely unexplored . While the microbial communities of other indoor environments , such as homes ( Lax et al . , 2014 ) , office buildings ( Kembel et al . , 2014 ) , and hospitals ( Bokulich et al . , 2013a ) , can influence the health of their inhabitants , food-production facilities represent a unique type of built environment wherein microbial activities are intimately tied to product-quality outcomes . Thus , the sensory and safety effects of microbial growth in food-production streams have a much broader impact , such that consumer enjoyment and health can be linked to the hygiene and processing decisions at the food facility . Furthermore , both beneficial and detrimental microbial activities are well defined in foods , making these systems a useful model for exploring microbial ecosystem dynamics beyond the scope of contaminant mitigation . Fermented foods , including beer , have the additional distinction that microbial activity is central to their production , responsible for necessary transformations as well as product spoilage ( Bokulich et al . , 2012a; Bokulich and Bamforth , 2013 ) . Most modern food-fermentation practices occur under relatively aseptic conditions , employing pure starter cultures in their production , and environmental contamination represents a prevalent threat to product integrity ( Bokulich and Bamforth , 2013 ) . Beer is typically produced through fermentation of malted barley sugars ( wort ) to alcohol by pure strains of Saccharomyces cerevisiae or Saccharomyces pastorianus , and any additional organisms—including cross-contamination from different Saccharomyces strains used in the same facility—are considered contaminants ( Bokulich and Bamforth , 2013 ) . In traditional fermentation production practices , conversely , microbial communities introduced from raw materials and processing environments are central to the fermentation process . For example , in the production of coolship ( lambic-style ) beers , a type of sour beer , no starter cultures are added to the wort; instead , boiled wort is allowed to cool overnight in a shallow , open-top vessel known as a coolship where indigenous microbiota are presumably introduced to the product , initiating fermentation ( Van Oevelen et al . , 1977; Bokulich et al . , 2012b; Spitaels et al . , 2014 ) . The unique succession of indigenous microbiota in coolship ale fermentations sets them apart chemically and sensorially from other beers ( Van Oevelen et al . , 1976; Spaepen et al . , 1978 ) , making coolship breweries a particularly interesting system for tracking microbial populations in food-processing environments . Microbial spoilage genes are also well defined in brewing environments and can be studied in situ without representing a direct threat to public health . Beer is protected from wholesale microbial contamination through its alcoholic , low-pH , and antimicrobial properties , and modern sanitary technologies and practices minimize the threat of spoilage organisms that have evolved specifically to grow in beer ( Bokulich and Bamforth , 2013 ) . Nevertheless , biofilms and other environmental reservoirs remain potential sources of microbial contamination in breweries ( Timke et al . , 2005; Storgards et al . , 2006; Timke et al . , 2008; Mamvura et al . , 2011; Matoulkova et al . , 2012 ) . As wort is boiled prior to fermentation , the primary reservoir for spoilage microorganisms in beer production is the brewery environment . Lactic acid bacteria ( LAB ) are of particular concern , as some members of this clade resist hop antimicrobial compounds , enabling growth and spoilage of beer through acidification , hazes , and off-flavors ( Suzuki et al . , 2006; Bokulich and Bamforth , 2013 ) . Hop iso-alpha-acids , the primary antimicrobial compounds in beer , kill most Gram-positive bacteria by acting as ionophores , dissipating proton-motive force across cell walls ( Simpson and Fernandez , 1992 , 1994; Simpson , 1993a ) , and via oxidative stress ( Behr and Vogel , 2010 ) . Very few resistance mechanisms have been proposed ( Suzuki et al . , 2006 ) , primarily being the multi-drug transporters horA ( Sami et al . , 1997; Sakamoto et al . , 2001 ) and horC ( Suzuki et al . , 2005; Iijima et al . , 2006 ) , the transcriptional regulator of horC , horB ( Suzuki et al . , 2005; Iijima et al . , 2006 ) , and hitA , a divalent cation transporter ( Hayashi et al . , 2001 ) that imports manganese into the cell . These genes are all located on plasmids and transmit via horizontal gene transfer ( Suzuki et al . , 2005 , 2006 ) . The frequency and transmission of these and other beer-spoilage genes within processing environments—or any other reservoir for spoilage microbes—have yet to be tested . Here , we employ mixed molecular approaches ( Bokulich and Mills , 2012b ) and a Bayesian modeling method to interrogate the seasonal sources , reservoirs , and transmission of bacteria , fungi , and beer-spoilage genes within a North American brewery over the course of one year . Given the inherent role of environmental microbiota in conducting coolship ale fermentations , this serves as a model system for studying mechanisms for microbial transfer within food-processing systems . Information on population and gene flow extends to other food-processing systems where environmental microbiota are involved positively or negatively in the production , stability , and safety of human nutrition . Short-amplicon marker-gene sequencing was employed to survey the bacterial and fungal consortia inhabiting the entire brewery environment . A total of 501 samples were collected during three seasons , representing the main processing surfaces and equipment used throughout the brewing process ( Figure 1 ) . Beta-diversity ( between sample ) comparisons provide useful assessments of the taxonomic similarity between different sites . Bray–Curtis dissimilarity of complete microbial profiles reveals that many samples cluster by processing room and substrate type regardless of season ( Figures 2–3 ) . Thus , fermenter samples cluster , associated with Bacillaceae; cellar production areas , associated with Micrococcaceae ( including the beer-spoiling genera Kocuria and Micrococcus ) ; wort , malt , and hotside ( wort-preparation ) surfaces , associated with Enterobacteriaceae , Leuconostocaceae , Candida santamariae , Pichia , and Rhodotorula; barrel-room floor samples , associated with Staphylococcaceae and Carnobacteriaceae; and beer samples , associated with Lactobacillaceae and Enterobacteriaceae . Barrels cluster , associated with Aspergillus , Eurotium , and Penicillium; coolship and barrel-room samples with Cryptococcus and Cladosporium . These taxonomic trends each demonstrate significant site associations ( Kruskal–Wallis Bonferroni-corrected p < 0 . 05 ) . S . cerevisiae was common throughout the brewery , but especially in the fermentation cellar . LAB and acetic acid bacteria were found sporadically at different sites and times , including on and near packaging equipment and fermenters inoculated with LAB ( Figure 3 ) . 10 . 7554/eLife . 04634 . 003Figure 1 . Brewery map and simplified brewing process diagrams . ( A ) Floorplan of brewery details surface sampling key and indicates separate sections of the brewery . LAC/Br fermenters indicate they were inoculated intentionally with lactic acid bacteria or Dekkera spp . , respectively , at the time of sampling . ( B ) Process diagram for conventional beer brewing , illustrating the relationship between brewing stages and sections of the brewery . Grain is milled and taken to the brewhouse ( hotside ) area where it is mashed ( steeped in hot water ) to form wort , which is lautered ( extracted from the grain by filtering and spraying with hot water ) and then boiled with hops . Boiled wort is cooled and pumped to the fermentation cellar where it is inoculated with Saccharomyces and fermented . Optionally , barrel-aged beers are transferred to barrels after fermentation . Finished beers are transferred to conditioning tanks in a separate section of the brewery where they are cooled , carbonated , and then packaged . ( C ) Process diagram for coolship beer brewing . Same as conventional , but following boiling wort is pumped to the coolship room where it is left to cool overnight , exposed to the atmosphere . The following morning , the wort is pumped to barrels in which it is fermented and aged for 1–3 years . In the Autumn samples , this occurred in the barrel room in the main brewery , but in Spring and Summer this moved to a newly built facility dedicated to sour beers . All coolship and sour beers were packaged on separate equipment in this second facility . The distinction between coolship beers and sour beers is the use of this coolship; other sour beers are produced using conventional brewing methods ( panel B ) , but are fermented with organisms other than Saccharomyces yeasts . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 00310 . 7554/eLife . 04634 . 004Figure 2 . Taxon abundance heatmaps depict genus-level relative abundance of fungi across sampling sites detected by marker-gene sequencing . The relative abundances ( RA ) of each genus ( columns ) within each sample ( rows ) are indicated by the color of the intersecting tile . Sample types are indicated by colored bars to the left of each row , classified according to the location within the brewery ( Figure 1 ) or the type of substrate ( grain , wort , hops , beer ) . Dendrograms represent Bray–Curtis dissimilarity between samples ( vertical trees ) and shared-niche similarity between taxa ( horizontal trees ) , respectively indicating taxonomic composition similarities and taxon co-occurrence patterns . Only taxa ≥0 . 05 relative abundance in at least one sample are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 00410 . 7554/eLife . 04634 . 005Figure 3 . Taxon abundance heatmaps depict family-level relative abundance of bacteria across sampling sites detected by marker-gene sequencing . The relative abundances of each genus ( columns ) within each sample ( rows ) are indicated by the color of the intersecting tile . Sample types are indicated by colored bars to the left of each row , classified according to the location within the brewery ( Figure 1 ) or the type of substrate ( grain , wort , hops , beer ) . Dendrograms represent Bray–Curtis dissimilarity between samples ( vertical trees ) and shared-niche similarity between taxa ( horizontal trees ) , respectively indicating taxonomic composition similarities and taxon co-occurrence patterns . Only taxa ≥0 . 05 relative abundance in at least one sample are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 005 To further examine this relationship , the Bayesian technique sourcetracker ( Knights et al . , 2010 ) was used to test whether raw substrates may act as sources for the microbial consortia of brewery surfaces . This tool predicts the relative proportion of contamination in sink samples ( in this case brewery surfaces ) from microbial sources ( raw ingredients and extraneous sources ) . Raw substrates ( grain , hops , yeast , beer ) and extraneous sources ( human skin , outdoor air , soil , saliva , feces , freshwater , ocean water ) from previous studies ( Caporaso et al . , 2011; Bowers et al . , 2012; Caporaso et al . , 2012; Bowers et al . , 2013 ) were tested as microbial sources . Results reveal distinct patterns of contamination across seasons ( Figure 4 ) . Grains were predicted as the largest microbial contributor to hotside areas , almost all surfaces in the coolship room , and areas of the cellar away from the main fermentation area . Hops were predicted as the major contributor to cellar fermentation areas and fermentation equipment . Yeast was predicted as the highest contributor to fermenters , conditioning tanks , and packaging equipment . Beer was predicted to be a common contaminant around fermenters and barrels within the cellar . Skin was a minor contributor to some surfaces in the cellar . Other surfaces , including most barrels and barrel-room surfaces , were most influenced by unknown sources . Outdoor air , soil , saliva , feces , freshwater , and ocean water were predicted to provide only a very low level of contamination ( <0 . 001 relative abundance ) . These results suggest that raw substrates are the main contaminant sources within the brewery , compared to extraneous sources . However , it is important to note that these predictions do not indicate a causative role for contamination . These predicted source/sink relationships could alternatively suggest that both are actually contaminated by another , untested source ( e . g , fruit flies or other vectors could transfer microbes between these and other surfaces within the brewery ) . Nevertheless , these predictions highlight potential sources of contamination , or at least shared microbial transmission patterns , between substrates and equipment/surfaces within the brewery . 10 . 7554/eLife . 04634 . 006Figure 4 . Mapping microbial contamination sources inside the brewery . Floorplans of brewery indicate the predicted relative contamination of brewery surfaces by microbial sources ( grains , hops , yeast , beer , human skin , unknown ) at each season , estimated by SourceTracker ( Knights et al . , 2010 ) . Coloration of each surface indicates the relative degree of microbial contamination from that source type ( as indicated by keys to the right of each row; units are relative abundance ) . Contamination from outdoor air , soil , feces , freshwater , ocean water , and saliva was negligible ( <0 . 001 relative abundance ) and are not shown . See Figure 1 for a floorplan key and description of surfaces . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 006 These findings suggest that substrate contact drives the microbial communities of different brewery surfaces . Consequently , raw materials ( grain , hops , yeast ) may be important vectors for any spoilage organisms encountered in the facility . This substrate-centric community structuring has also been observed in other food facilities ( Bokulich and Mills , 2013a; Bokulich et al . , 2013b , 2014 ) . However , the brewery surface niches clustered throughout time with only subtle changes across seasons , unlike wineries , which were significantly impacted by seasonal factors ( Bokulich et al . , 2013b ) . This reflects the divergent production schemes of these two foods: whereas most beers are produced year-round , wine is an inherently seasonal product . The raw materials for beer are stored and processed throughout the year in a continual production schedule , creating a stable environment on equipment surfaces that interact with these same raw substrates on a daily basis . Instead , brewery environments may function more like cheese-making environments , where facility-specific ‘house’ microbial communities form on equipment surfaces in response to idiosyncrasies in the indoor environment ( Bokulich and Mills , 2013a ) . Such a possibility would have interesting implications for sour beer breweries , and comparative studies of lambic and coolship breweries could offer insight into the brewery-specific flavor profiles displayed in these beers . By mapping surface samples to their physical location within the brewery , a spatial model emerges of microbial dispersion across brewery surfaces over time ( Figures 5–7 ) . Several populations visibly spread with time , most prominently S . cerevisiae , which progressed from dominating fermentation and packaging areas only in Autumn to being the most abundant fungus detected across the brewery in Spring and Summer . Likewise , C . santamariae displayed high abundance on hotside surfaces in Autumn but became more abundant and spread to nearby sections of the cellar in Spring and Summer ( Figure 5 ) . As building measurements were not taken during these times , the factors driving these changes cannot be assessed with these data , but warming temperatures and increasing humidity in the Spring and Summer months could be associated phenomena . Other taxa demonstrated more localized patterns of dispersion , such as Micrococcus and Kocuria , which appeared to spread the most around floors and other surfaces in cellar , barrel room , and packaging areas ( Figure 6 ) . Here , regular contact with beer runoff diluted with rinse water may support growth of these populations , which are associated with spoilage in low-alcohol beers ( Back , 1981 ) . High populations of Acetobacter and Lactobacillus were found more disparately , and specifically in areas where high volumes of wort and beer may be encountered: on conveyor belts and floors below packaging areas , hotside and cellar area sinks , and on sample ports for isolated fermenters and kegs ( Figure 6 ) . On these sites microbial communities can contact undiluted beer , for example , drips in the basin below the packaging-line belt , wort and beer collected for specific gravity ( sugar concentration ) measurements then dumped in sinks , and sampling ports on fermenters . This follows the known behavior of these bacteria , which can spoil undiluted , higher-alcohol beers under appropriate conditions ( Bokulich and Bamforth , 2013 ) . 10 . 7554/eLife . 04634 . 007Figure 5 . Spatial distribution heatmaps of fungi in brewery environments across seasons . Plots indicate relative abundance of fungal taxa detected by ITS sequence reads across brewery surfaces at different times: Autumn ( left ) , Spring ( center ) , and Summer ( right ) . See Figure 1 for a floorplan key and description of surfaces . Note that the floorplans change between seasons as some samples were only collected as specific timepoints and the wild brewing facility was built and opened during the Spring sampling time . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 00710 . 7554/eLife . 04634 . 008Figure 6 . Spatial distribution heatmaps of bacteria in brewery environments across seasons . Plots indicate relative abundance of bacterial taxa detected by 16S rRNA gene sequence reads across brewery surfaces at different times: Autumn ( left ) , Spring ( center ) , and Summer ( right ) . Scales on right represent relative abundance scale ( maximum 1 . 0 ) for each row of plots . See Figure 1 for a floorplan key and description of surfaces . Note that the floorplans change between seasons as some samples were only collected as specific timepoints and the wild brewing facility was built and opened during the Spring sampling time . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 00810 . 7554/eLife . 04634 . 009Figure 7 . Spatial distribution heatmaps of bacteria in brewery environments across seasons ( part 2 ) . Plots indicate relative abundance of bacterial taxa detected by 16S rRNA gene sequence reads across brewery surfaces at different times: Autumn ( left ) , Spring ( center ) , and Summer ( right ) . Scales on right represent relative abundance scale ( maximum 1 . 0 ) for each row of plots . See Figure 1 for a floorplan key and description of surfaces . Note that the floorplans change between seasons as some samples were only collected as specific timepoints and the wild brewing facility was built and opened during the Spring sampling time . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 009 These results illustrate the progressive dissemination of microbes in space and time within a functioning food-processing environment . Microbes not only associate with specific substrates , they exhibit patterns of dispersion within confined regions of the brewery . This yields useful insight into the transmission behavior of these organisms , and especially taxa associated with beer spoilage , through physical space . Microbial confinement within discrete zones suggests that physical barriers ( e . g . , walls ) and physiochemical conditions ( e . g . , humidity and temperature control ) can staunch the spread of some microbes . Non-production surfaces that encounter beer and/or waste streams , such as floors , sinks , and grain-handling equipment , are typically only cleaned by hose and may accumulate substrates for supporting microbial growth and survival . Aerosols and splashing occur regularly in a production brewery ( indeed , hosing down floors is one obvious cause ) , increasing the likelihood for microbial spread and colonization from these surfaces onto other surfaces , including equipment . Human and insect traffic can also increase the rate of dispersal between these surfaces ( e . g . , from one room to another ) . Physical partitions appear to inhibit this passage as observed in this brewery and , where possible , may reduce the probability for contamination . This is particularly important in packaging areas , where microbial dispersal can directly contaminate sanitary surfaces and open packages , increasing the likelihood of spoilage in finished product . Separating packaging rooms from cellar and hotside operations and using local partitions , such as blast shields , may help protect packaging surfaces and beer from such routes of contamination . Further investigations into the relationship between building materials , spatial area , indoor conditions ( e . g . , airflow , humidity , and personnel traffic ) , and the rate and extent of microbial dispersion could yield important findings for optimal brewery design from the perspective of microbial control . These findings are instructive but indeed should not be interpreted with alarm: they illustrate microbial dispersal in a functioning brewery that rarely suffers from any cases of product contamination . Different species of LAB are the principal spoilage bacteria in beer fermentations as well as important members of sour beer fermentations . However , short 16S rRNA gene amplicons are frequently inadequate to resolve reliable species-level identifications ( Bokulich and Mills , 2012a ) . Therefore , we used LAB-TRFLP ( Bokulich and Mills , 2012a ) to characterize genus- and species-level LAB community compositions in a select subset of samples . This included raw material and beer samples , as well as surfaces on which detection of Lactobacillales by marker-gene sequencing was particularly high . Results indicate that different surface and sample types exhibit distinct lactic acid bacterial patterns , corresponding to the substrates encountered at that site or contained in that sample ( Figure 8 ) . Wort samples contained a mixture of Lactobacillus delbrueckii , Lactobacillus sakei , Lactobacillus hilgardii , Leuconostoc mesenteroides , Lactococcus lactis , Streptococcus sp . , and Bacillus sp . A , most of which were only rarely detected in other fermenting and bottled beer samples . Many of these species are also rarely found in beers ( Bokulich and Bamforth , 2013 ) , but instead appear associated with grain , hence their detection in wort . Coolship and fermenting sour beers ( in this case coolship beers produced from different wort types ) were dominated by Pediococcus and/or L . lindneri , corroborating previous studies of coolship beers in this brewery ( Bokulich et al . , 2012b ) . Fermenters and barrel surfaces that contacted these fermentations near the time of sampling exhibited similar communities , though Lactobacillus brevis and Lactobacillus sp . A were more common on these surfaces than in the beers or on other surfaces . Other sour and barrel-aged beers contained unique profiles , with involvement of other Lactobacillus species only weakly detected in coolship beers . Floor and packaging area surfaces contained a more diverse mixture of LAB , but primarily the L . lindneri , L . brevis , and L . delbrueckii detected in the wort and beer samples . Interestingly , only Pediococcus was detected on grain samples , though only weak amplification could be had from these samples , suggesting low LAB populations or inhibition of PCR by grain polyphenols , possibly suppressing the detection of less abundant populations . Hop pellet samples also contained a mixture of different LAB populations , including Pediococcus , L . lindneri , and L . brevis . 10 . 7554/eLife . 04634 . 010Figure 8 . Lactic acid bacterial community composition on brewery surfaces , beers , and ingredients . LAB-TRFLP profiles of samples exhibiting high Lactobacillales relative abundance by 16S rRNA gene sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 010 These results illustrate that substrate drives the composition of LAB communities as well as whole microbial communities and highlight the risk of cross-contamination between different equipment surfaces . The detection of Lactobacillus spp . on both filler heads ( only one of which is used for sour beer packaging ) makes this observation all the more pertinent . This observation is not likely to be an exceptional case; cross-contamination between processing areas is very likely the prevailing cause of spoilage in any brewery , where microbial biofilms have been previously reported even on packaging equipment ( Timke et al . , 2005; Storgards et al . , 2006; Timke et al . , 2008; Mamvura et al . , 2011; Matoulkova et al . , 2012 ) . This observation underlines the need for constant hygiene surveillance in breweries but is hardly cause for alarm , if the exceedingly low incidence of microbial spoilage in modern brewing practice is any indication . Beer-spoiling LAB possess several mechanisms that support their growth and survival in beer ( Suzuki et al . , 2006 ) . Several hop-resistance genes are principal among these , counteracting the antimicrobial effects of iso-a-acids ( hop bittering resins ) ( Simpson , 1993b; Simpson and Fernandez , 1994 ) . Hop-resistant LAB typically contain several of these genes , including horA ( Sami et al . , 1997; Sakamoto et al . , 2001 ) , horB ( Suzuki et al . , 2005; Iijima et al . , 2006 ) , horC ( Suzuki et al . , 2005; Iijima et al . , 2006 ) , and hitA ( Hayashi et al . , 2001 ) , which are upregulated during growth in beer ( Bergsveinson et al . , 2012; Pittet et al . , 2013 ) . However , the presence of these genes within brewery environments and on brewing equipment has not been described previously . Thus , we sought to quantify the abundance of these beer-spoilage genes on brewing equipment and in beer during different times in conjunction with microbial community profiles . Results demonstrate high gene frequencies on several surfaces within the brewery ( Figure 9 ) . Sour beer samples contained the highest counts , between 2 . 0 × 104–4 . 8 × 104 copies/µl of horC , but fermenter and packaging area surfaces ( filler heads , below bottling line belt , and packaging sink ) also registered between 2 . 8–7 . 8 × 102 copies/cm2 . None of these alleles were detected on hop samples , keg samples , or one barrel bung ( stopper ) sample , though 1 . 1 × 103 total copies/cm2 were detected on a keg faucet used for attaching kegs to beer lines at the brewery . Among the genes analyzed , horC was the most abundant ( Figure 9 ) , which is interesting when considered in the context of previous work showing that presence of this gene correlates with increased hop-tolerance and beer-spoilage ability ( Fujii et al . , 2005; Iijima et al . , 2006; Bergsveinson et al . , 2012 ) and the plasmid carrying horC is the most important for supporting growth of Lactobacillus brevis in beer ( Bergsveinson et al . , 2015 ) . The preferential expression of this gene observed in these previous studies and the relative increased abundance with which it was found in this study suggests horC is an important gene for facilitating beer-spoilage and is consequently selected for in the brewery environment , particularly in areas where sour beers are produced . The purported transcriptional regulator of horC , horB , was detected at stable ratios relative to horC in all samples , supporting this putative function ( Iijima et al . , 2006 ) . The least frequently observed hop-resistance gene , hitA , is involved in manganese transport ( Hayashi et al . , 2001 ) , supporting resistance against manganese depletion by iso-a-acids ( Behr and Vogel , 2010 ) . Other studies have observed similarly low frequencies of hitA presence and expression in LAB relative to the other hop-resistance genes ( Haakensen and Ziola , 2008; Bergsveinson et al . , 2012 ) . 10 . 7554/eLife . 04634 . 011Figure 9 . Hop-resistance gene frequency on brewery surfaces and beers . ( A ) ddPCR detection of hitA , horA , horB , and horC on surfaces ( detected as copies/cm2 ) and in beers ( copies/ml ) . Bar height indicates cumulative log gene abundance; colors indicate relative gene frequencies superimposed on these bars . Two barrel bung ( stopper ) samples are depicted on the left , one has no detection . ( B ) Pearson product-moment correlation matrix between hop-resistance genes , Lactobacillales relative abundance by 16S rRNA gene sequencing , and relative abundance of the dominant lactic acid bacteria detected by LAB-TRFLP . The color and shape of correlation ellipses ( lower-left ) indicate Pearson's product-moment correlation coefficient ( r ) between intersecting variables , as depicted in the key to the right . Correlations with larger positive r values are depicted as darker blue with increasingly narrow , upward-pointing ellipses . Correlations with larger negative r values are depicted as darker red with increasingly narrow , downward-pointing ellipses . Weaker correlations are depicted as wider , lighter colored ellipses . The corresponding p values for all correlation tests are provided in the reflected intersection ( top-right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 011 The spoilage genes horA , horB , and horC all display high degrees of intercorrelation ( Pearson's r = 0 . 83–1 . 0 , p < 0 . 01 ) and significant but lesser correlation to hitA ( r = 0 . 48–0 . 64 , p ≤ 0 . 04 ) . All spoilage genes except for hitA demonstrate significant correlation with bulk detection of Lactobacillales via 16S rRNA gene sequencing ( r = 0 . 53–0 . 74 , p ≤ 0 . 03 ) , while bulk Lactobacillales and all spoilage genes but horA correlate significantly with L . lindneri detection via LAB-TRFLP ( r = 0 . 48–0 . 77 , p ≤ 0 . 04 ) . The only gene correlated with Pediococcus abundance via LAB-TRFLP was horA ( r = 0 . 57 , p = 0 . 01 ) , consistent with previous observations that horA is the primary known hop-resistance gene observed in Pediococcus spp . ( Haakensen and Ziola , 2008 ) . Interestingly , no resistance genes correlated with L . brevis , strains of which are among the most common brewery contaminants and most commonly positive for hop-resistance genes ( Haakensen and Ziola , 2008 ) . This likely reflects the strains detected in this brewery only , and L . brevis was only a minor constituent of sour beers and processing surfaces ( Figure 7 ) . These results are the first indication of hop-resistance-gene abundance within a brewery environment . As the primary reservoir for spoilage microbes in beer production , tracking spoilage genes on brewing surfaces and materials is important for understanding contamination risks arising from the environment . Sour beer contained the highest gene abundance , which is predictable given that the LAB growing in these ( coolship ) beers must be adapted to resist hop antimicrobials . Likewise , detection on barrel surfaces is also expected , given their regular and direct contact with these sour beers . Fermenters and packaging-line surfaces also exhibited fairly high levels of hop-resistance genes , with the highest detection on surfaces that contact sour beers and unsanitary surfaces ( below packaging belt , packaging-line sink ) . The only sanitary , conventional beer-making surface among these is the second filler head , which only exhibited 4 cumulative copies/cm2 of the resistance genes , compared to 2 . 1 × 102 in the first filler head , which is used for packaging sour and barrel-aged beers . Nevertheless , this second filler head is also a sanitary surface ready for packaging , and should be free of contaminants . DNA was used for ddPCR and thus these counts may represent dead cells present on the equipment surfaces; the relative proportion of viable cells is unknown . If the resistance genes detected by ddPCR do in fact represent any viable cells , this demonstrates the resilience of these bacterial populations on equipment surfaces that regularly contact ‘contaminated’ ( in this case , intentionally contaminated ) beers . These findings argue for careful separation of equipment used in conventional beer-making from that used for sour beers: equipment that contacts fermenting and finished sour beer or barrels , including pumps , hoses , and especially packaging equipment , is best dedicated to sour beer production and should not be used to handle conventional beers . No hop-resistance genes were detected on hop pellets , though LAB were present , indicating that hops are probably not a significant source of beer-spoilage bacteria . Iso-alpha-acids , the antimicrobial compounds released from hops and against which hop-resistance genes confer protection , are generated by the breakdown of humulones during boiling , and are not present in raw hops ( Steenackers et al . , 2015 ) . Undissociated humulones are significantly less inhibitory to LAB , and the antimicrobial effects of hops are dependent upon acidic conditions ( Simpson and Smith , 1992 ) . Thus , the selective pressure to acquire and maintain hop-resistance genes only exists in the presence of hopped beer , not in raw hops . Only two hop pellets were tested for hop-resistance genes , but this finding suggests that the brewery environment itself is the site of hop-resistance-gene propagation . Larger studies of hop-resistance genes on brewery surfaces and brewing materials will illuminate the role of environmental vs raw material contamination in hop-resistance-gene transfer , and sites , conditions , and mechanisms of transfer within the brewery environment . Tracking spoilage-gene flow across brewery surfaces presents a unique opportunity for understanding spoilage dynamics within food-processing systems and other built environments in general . This approach will facilitate understanding how spoilage resistance propagates within production environments and the reservoirs and vectors encouraging its spread , yielding novel insight for controlling spoilage—focused on gene transmission rather than taxonomic populations . Moving forward , this facility-surveillance model opens many questions and possibilities for mapping microbial spoilage dynamics within food-production systems . What relationship do indoor environmental factors , building design , and surface materials have with microbial transmission ? What roles do cleaning and other processes play in controlling contamination on a facility-wide scale ? What other biomarkers are associated with contamination and how does the spoilage-allele landscape behave over time and in response to these conditions ? The potential advantages of studying microbiology of breweries are not confined to food systems alone , and lessons learned here may aid our understanding of microbial communities in other built environments . For example , the hop-resistance genes include ABC multi-drug transporters similar to other antimicrobial-resistance genes ( Sakamoto et al . , 2001 ) . Studying and manipulating their transmission within breweries may aid understanding of similar gene-transfer events involved in pathogenesis in hospitals , homes , water systems , or public environments where in situ modeling of such genes is unfeasible or a potential health hazard . Breweries are a useful model for testing general theories of microbial transmission and spoilage-gene dispersal in situ for functional indoor environments and food systems , as beer spoilage is not actually detrimental to human health—merely human pleasure . Samples were collected from a single brewery located in North America . This brewery operates as a conventional brewing facility but also contains a dedicated coolship room , barrel room , and a secondary cellar building , where it produces coolship and other sour and barrel-aged beers seasonally . The coolship beers ( a type of sour beer produced without inoculation ) are produced following the classic Belgian tradition , and their microbial profiles in this brewery have been investigated previously ( Bokulich et al . , 2012b ) . The other sour beers are produced using pure-culture inocula of non-Saccharomyces yeasts and bacteria , but the term ‘sour beer’ refers to both coolship and other sour beers elsewhere in this study unless if a distinction is made between these types of beer . The secondary cellar building was built and opened during the course of the collection period . Thus , during the first sample collection time point ( Fall , 2012 ) sour beers were fermented and stored in the barrel room of the main facility . For the remaining times ( Spring and Summer , 2013 ) , all sour beers were held in this second ‘wild fermentation’ facility . A facility map and flow–through diagrams relating brewing stages to this space are presented in Figure 1 . Samples were collected in Fall 2012 , Spring 2013 , and Summer 2013 . In all , 445 surface swabs and 56 beer and ingredient samples were collected ( Figure 1 ) as described previously ( Bokulich et al . , 2013b ) . DNA was extracted using the ZR-96 Fecal DNA MiniPrep Kit ( Zymo Research , Irvine , CA ) , with bead beating in a FastPrep-24 bead beater ( MP Bio , Solon , OH ) , and stored at −20°C until further processing . Amplification and sequencing were performed as described previously for bacterial ( Bokulich , 2012b ) and fungal communities ( Bokulich and Mills , 2013b ) . The V4 domain of bacterial 16S rRNA genes was amplified using primers F515 ( 5′–NNNNNNNNGTGTGCCAGCMGCCGCGGTAA–3′ ) and R806 ( 5′–GGACTACHVGGGTWTCTAAT–3′ ) ( Caporaso et al . , 2011 ) , with a unique 8 nt barcode ( italicized poly-N section ) and 2 nt linker sequence ( bold ) at the 5′ terminus . Fungal internal transcribed spacer ( ITS ) 1 loci were amplified with primers BITS ( 5′–NNNNNNNNCTACCTGCGGARGGATCA–3′ ) and B58S3 ( 5′–GAGATCCRTTGYTRAAAGTT–3′ ) ( Bokulich and Mills , 2013b ) . Amplicons were combined into two separate pooled samples ( keeping bacterial and fungal amplicons separate ) at roughly equal amplification intensity ratios , purified using the Qiaquick spin kit ( Qiagen , Germantown , MD ) , and submitted to the UC Davis DNA Technologies Core for Illumina paired-end library preparation , cluster generation , and 250 bp paired-end sequencing on an Illumina MiSeq instrument in four separate runs ( separating bacterial and fungal libraries ) . Raw fastq files were demultiplexed , quality-filtered , and analyzed using QIIME v . 1 . 7 . 0 ( Caporaso et al . , 2010b ) . The 250-bp reads were truncated at any site of ≥3 sequential bases receiving a quality score < Q10 , and any read with <75% ( of total read length ) consecutive high-quality base calls was discarded ( Bokulich et al . , 2013c ) . Operational taxonomic units ( OTUs ) were clustered at 97% identity using QIIME's open-reference OTU-picking pipeline using UCLUST-ref ( Edgar , 2010 ) against either the Greengenes 16S rRNA gene database ( May 2013 release ) ( McDonald et al . , 2012 ) or the UNITE fungal ITS database ( Abarenkov et al . , 2010 ) , modified as described previously ( Bokulich and Mills , 2013b ) . OTUs were classified taxonomically using RDP classifier ( Wang et al . , 2007 ) for bacteria and BLAST ( Altschul et al . , 1990 ) for fungi . Any OTU comprising less than 0 . 0001% of total sequences for each run were removed ( Bokulich et al . , 2013c ) . Bacterial 16S rRNA gene sequences were aligned using PyNAST ( Caporaso et al . , 2010a ) and chimeric sequences were identified using ChimeraSlayer ( Haas et al . , 2011 ) . Sequences failing alignment or identified as chimera were removed prior to downstream analysis . OTU tables were evenly subsampled to 400 sequences per sample for all statistical tests . MANOVA with 999 permutations was used to test significant difference between sample categories based on Bray–Curtis distance using Adonis ( Anderson , 2001 ) . Kruskal–Wallis tests were used to identify significantly discriminant taxa ( with Bonferroni error correction ) between sample groups . Pearson product-moment correlation analyses were performed using R software . Environmental surveillance heatmaps were generated based on taxonomic abundance tables generated in QIIME and visualized using SitePainter 1 . 1 ( Gonzalez et al . , 2012 ) . Bacterial OTU source-sink relationships were tested using SourceTracker ( Knights et al . , 2010 ) with 1000 burn-ins , 25 restarts , and rarefaction to 100 OTUs . Source-tracking predictions used bacterial profiles of samples collected in this study , coolship beers from this brewery analyzed in a previous study ( Bokulich et al . , 2012b ) , and outdoor air ( Bowers et al . , 2012 , 2013 ) , soil , saliva , feces , human skin , freshwater , and ocean water samples ( Caporaso et al . , 2011 , 2012 ) from previously published studies as source samples . All studies were performed using bacterial V4 16S rRNA with the same F515/R806 primer pair . In order to quantify hop-resistance gene abundance on brewery surfaces , droplet digital PCR ( ddPCR ) was used to enumerate the genes horA , horB , horC , and hitA . ddPCR was performed using the QX100 Droplet Digital PCR setup and protocol ( Bio-Rad ) . ddPCR was performed in 20-µl reactions containing 3 ng/µl of DNA template , 900 nmol of each primer , 250 nM of each probe , and 1× Bio-Rad Droplet PCR Supermix . Primer and probe sequences , melting temperatures , 5′ fluorophore probe labels , and amplicon lengths for each gene target are shown in Table 1 . All probes contained a 3′ IowaBlack FQ quencher paired with different 5′ fluorescent labels ( Table 1 ) . Each 20 μl reaction was then pipetted into separate wells of a disposable eight channel droplet generation cartridge ( Bio-Rad ) and 70 µl of droplet generation oil ( Bio-Rad ) loaded into the cartridge oil wells . The cartridge was then inserted into the QX100 droplet generator ( Bio-Rad ) and each sample was portioned into droplet-sized water-in-oil emulsions . Following droplet generation , emulsion samples were transferred to a 96-well PCR plate ( Eppendorf ) . The plate was then hot-sealed with foil cover and subjected to conventional PCR in the CFX96 Touch Real-Time PCR ( Bio-Rad ) . Thermal cycling conditions consisted of an activation period for 10 min at 95°C , followed by 40 cycles of a denaturation step for 30 s at 94°C , and an annealing-extension step for 60 s at the optimal annealing temperature ( 59–59 . 6°C ) , using a ramp rate of 2 . 5°C/s for each step and a final inactivation step of 98°C for 10 min . After PCR amplification , the plate was loaded into the QX100 Droplet Digitial PCR ( Bio-Rad ) and analyzed for absolute signal quantification of each fluorescence channel in each well . Signal detection and data processing were performed using Quantasoft Software v . 1 . 3 . 2 ( Bio-Rad ) . 10 . 7554/eLife . 04634 . 012Table 1 . Hop-Resistance Gene Primers and Probes for ddPCRDOI: http://dx . doi . org/10 . 7554/eLife . 04634 . 012TargetTm5′ LabelProbeForward primerReverse primerbp†horB59FAMTCGCGGCCAAGTGATACTTATCCTGAAGTCGACACAAAATCCTGAATCAAGCCTTGATCAATCGTCAGAC88hitA59HEXACAGAATAACGGCAACCAGTGTCGCAATCCTGTTGCTTCTGATGAAATTGGCCGCTAAGAATACTTCGTAGGTGA105horA59 . 6FAMCGCCGTTCCGCTCGTCTTGATCTGCCTGGACTGGCGGATGACTATCCTGTCTCGCTCTGGCAAC104horC59 . 6HEXACCACGCCAATGCCACTAGAAGCATGGACACGGTTAATGGCACAGCGTTCGCGCCATAAAATAAGAGAGG87Tm = melt temperature ( °C ) . †Nucleotide length ( bp ) . All raw marker-gene sequencing data are publicly deposited in QIITA ( http://qiita . colorado . edu/ ) under the accession number 10105 ( http://qiita . colorado . edu/study/description/10105 ) .	Many microbes—including bacteria and fungi—can affect the food and drink we consume , for better and for worse . Some spoil food , making it less tasty or even harmful to health . However , microbes can also be important ingredients: for example , yeast ferments malted barley sugars to make the alcohol and flavor of beer . Nowadays , many beers are made under carefully controlled conditions , where the only microbes in the beer should be the strain of yeast added to the barley sugars . A more traditional ‘coolship’ method can be used to make sour beers; the barley sugars cool in an open-topped vessel and are fermented by the yeast and bacteria found naturally on the raw ingredients and in the surrounding environment . Relatively little was known about how microbes spread around and adapt to living inside buildings . Now , Bokulich et al . have used a range of molecular and statistical techniques to examine how bacteria and fungi are dispersed throughout a North American brewery that produces beer using both conventional and coolship brewing techniques . Most of the microbes found in the building originated from the raw ingredients used to make the beer , with different parts of the brewery containing different species . Over the course of a year , some species spread to new parts of the building; a statistical method predicted the sources of these microbes , and revealed some key areas and features of the brewery that affect microbial transfer . Bokulich et al . also looked at the spread of genes that enable their bacterial hosts to spoil beer , including those that protect bacteria from the antimicrobial action of the hops that flavor many beers . Lactic acid bacteria are the main cause of beer spoilage and so are usually to be avoided in breweries , but are also a normal ingredient in sour beer . In the brewery Bokulich et al . investigated , beer-spoilage and hop-resistance genes were found throughout the brewery , even in areas not used to produce sour beer . However , little beer spoilage occurred . The techniques used by Bokulich et al . to track the spread of microbes and their detrimental genes could be used in the future to understand how microbes adapt to other indoor environments . Indeed , Bokulich et al . suggest that breweries could be used as models to safely understand the factors that influence microbial movement in any food-production facility as well as other building environments .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "ecology", "microbiology", "and", "infectious", "disease" ]	2015	Mapping microbial ecosystems and spoilage-gene flow in breweries highlights patterns of contamination and resistance
Grasping requires translating object geometries into appropriate hand shapes . How the brain computes these transformations is currently unclear . We investigated three key areas of the macaque cortical grasping circuit with microelectrode arrays and found cooperative but anatomically separated visual and motor processes . The parietal area AIP operated primarily in a visual mode . Its neuronal population revealed a specialization for shape processing , even for abstract geometries , and processed object features ultimately important for grasping . Premotor area F5 acted as a hub that shared the visual coding of AIP only temporarily and switched to highly dominant motor signals towards movement planning and execution . We visualize these non-discrete premotor signals that drive the primary motor cortex M1 to reflect the movement of the grasping hand . Our results reveal visual and motor features encoded in the grasping circuit and their communication to achieve transformation for grasping . Grasping objects of different shapes and sizes appears trivial in daily life . We can distinguish between thousands of objects ( Biederman , 1987 ) and shape our hands according to their geometry in order to hold and manipulate them ( Napier , 1956; Smeets and Brenner , 1999 ) . Although such operations seem to be effortless , their underlying neuronal mechanisms are highly complex and require extensive computational resources ( Fagg and Arbib , 1998; Felleman and Van Essen , 1991 ) . The cortical grasping network needs to translate high-dimensional visual information of an object into multidimensional motor signals that control the complex biomechanics of the hand . In the primate brain , these processes are linked to the anterior intraparietal ( AIP ) , the ventral premotor ( F5 ) , and the primary motor cortex ( M1 ) ( Brochier and Umilta , 2007; Castiello , 2005; Davare et al . , 2011; Nelissen and Vanduffel , 2011 ) . Within this network , AIP provides access to the dorsal visual stream that processes vision for action ( Culham et al . , 2003; Goodale et al . , 1994 ) . In fact , neurons in AIP were shown to strongly respond to the presentation of graspable objects or 3D contours ( Murata et al . , 2000; Taira et al . , 1990; Theys et al . , 2012b ) , but could also encode specific grip types ( Baumann et al . , 2009 ) . This grasp-relevant information processed in AIP is exchanged with F5 via dense reciprocal connections ( Borra et al . , 2008; Gerbella et al . , 2011; Luppino et al . , 1999 ) . Accordingly , deactivation of AIP or F5 causes severe deficits in pre-shaping the hand while approaching an object ( Fogassi et al . , 2001; Gallese et al . , 1994 ) . In contrast to AIP , concurrent electrophysiological studies suggest that F5 is primarily encoding objects in motor terms and is storing context-specific grip type information ( Fluet et al . , 2010; Raos et al . , 2006 ) . Connections of the dorsal subdivision of F5 ( F5p ) to the spinal cord and to M1 provide further evidence for the important role of F5 for grasp movement preparation ( Borra et al . , 2010; Dum and Strick , 2005 ) . These electrophysiological and anatomical observations lead to our current understanding of the fronto-parietal network as the main circuitry for translating object attributes into motor commands ( Jeannerod et al . , 1995; Rizzolatti and Luppino , 2001 ) . In detail , it has been suggested that visual features extracted in AIP activate motor prototypes in F5 , which store hand configurations according to an object’s geometry ( Rizzolatti and Luppino , 2001 ) . However , the detailed neural mechanisms of these processes remained unclear . To create a deeper understanding of how visual information is transformed into motor commands , a precise identification and differentiation of visual and motor processes within the grasping network is required . Previous important grasping studies classified visual-dominant , visual-motor or motor-dominant neurons primarily on the phase of their activation [for AIP see Murata et al . ( 2000 ) and Sakata et al . ( 1995 ) , for F5 see Raos et al . ( 2006 ) ; Theys et al . ( 2012a ) ] , but they could not discriminate between neural coding of visual features of objects or motor features of the hand . A differentiation between visual and motor coding is challenging for multiple reasons . First , the fronto-parietal network is multimodal and can reflect sensory and motor signals simultaneously . Second , visual and motor descriptions of objects and the hand are multidimensional due to the complexity of object geometry and hand physiology . Investigations at the neural level therefore necessitate multidimensional observations from many neurons . Third , the visual and motor spaces are highly linked to each other since the form of an object often defines the shape of the grasping hand . Disassociating both neuronal representations therefore requires highly variable visual stimuli and motor responses . In this study , we took a multidimensional approach to identify and separate visual and motor processes in the grasping network of AIP , F5 , and M1 . We recorded large populations of neurons simultaneously from the entire network and compared their modulation patters to the visual attributes of highly diverse objects and the kinematic features recorded from the grasping hand . Our data revealed distinct roles of the grasping network in translating visual object attributes ( AIP ) into planning ( F5 ) and execution signals ( M1 ) and allowed visualizing the propagation of these features for grasping . Presenting 3D objects to the monkeys lead to vigorous discharge ( Baumann et al . , 2009; Murata et al . , 2000 ) of AIP-neurons ( Figure 2a–b ) . The modulated population was not only larger ( sliding ANOVA , Figure 3 ) , but also significantly faster appearing after stimulus onset than in F5 ( 49 . 7 ms and 54 . 9 ms , monkey M and Z respectively ) . Impressively , individual AIP cells were capable of differentiating object shapes at high precision ( e . g . , Figure 2a ) . To quantify this attribute , we computed a modulation depth analysis that determined the relative difference in firing rate between all pairs of conditions ( objects ) during the cue epoch ( see Materials and methods ) . The example cell of Figure 2b revealed a chequered structure caused by the shape-wise order of object conditions 00–76 ( for object id , see Figure 1b ) and a maximum modulation depth ( MD ) of 62 Hz . Statistical analysis between all conditions ( ANOVA and post-hoc Tukey-Kramer criterion , p<0 . 01; see Materials and methods ) revealed a high encoding capacity of the example neuron that could significantly separate 71% of the 946 condition pairs ( 44 conditions ) . Interestingly , the neuron decreased its MD in darkness but maintained its encoding of shape ( as also indicated in Figure 2a ) . 10 . 7554/eLife . 15278 . 006Figure 2 . Visual object processing in area AIP . ( a ) Example neuron of AIP responding to the presentation of graspable objects ( each curve represents one task condition ) . ( b ) Modulation depth plot illustrating the absolute firing rate difference in the cue epoch between all condition pairs ( conditions 00 – 76 placed on axis in ascending order ) . Warm colours: high modulation depth , cool colours: low modulation depth . ( c ) Shape-wise clustering of objects in the AIP population during the cue epoch , as demonstrated by CDA . Arrows indicate a shift in position when big horizontal cylinders ( red triangles ) were grasped from below instead from above ( black triangles ) . ( d ) Same as c , but during the grasp epoch . ( e–f ) Dendrograms illustrating the neural distance between object conditions in the simultaneously recorded AIP population in the cue and grasp epoch ( N = 62 ) . Symbols and colour code in a , c-f as in Figure 1b . Percentages in c and d describe how much variance of the data is explained by the shown components ( 1st , 2nd and 3rd ) . Note: Video 3 visualizes the N-space of AIP of an additional recording in the same animal ( Z ) . See Figure 2—figure supplement 1–2 for the averaged population results of animal Z and animal M , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 00610 . 7554/eLife . 15278 . 007Figure 2—figure supplement 1 . Visual coding for hand action in AIP in animal Z across all sessions . Neural distance between each pair of group means was measured for each recording session , averaged across sessions , and visualized as dendrograms for the ( a ) cue and ( b ) hold epoch . Symbols and colours as in Figure 1b . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 00710 . 7554/eLife . 15278 . 008Figure 2—figure supplement 2 . Visual coding for hand action in AIP in animal M across all sessions . ( a ) Firing rate plot of AIP example unit . ( b–c ) N-space of AIP during cue and hold epoch of one recording session ( simultaneously recorded , N = 90 , first three components shown ) . ( d–e ) Neural distances between each pair of group means were measured for each recording session , then averaged across all sessions and visualized as dendrograms for the cue and hold epoch . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 00810 . 7554/eLife . 15278 . 009Video 3 . Population coding in AIP . The first three canonical variables of the AIP population are shown in 3D and are animated for presentation purposes . Each symbol represents one trial . Symbols and colours as in Figure 1b . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 00910 . 7554/eLife . 15278 . 010Figure 3 . Visual processing of object shapes . ( a ) A set of six 'mixed' objects elicited different visual stimuli and different motor responses . ( b ) Percentage of tuned neurons of the AIP , F5 , and M1 population express the significant modulation with respect to the mixed objects across time ( sliding one-way ANOVA ) . ( c ) Tuned neurons ( shades of red ) were mapped to their recording location during the visual ( t = 0 . 16 s after object presentation ) and motor phase ( t = 0 . 7 s after movement onset ) . ( d ) As a contrast and to elicit pure visual responses , 'abstract' objects caused different visual stimuli but the same grip . ( e ) Similar to b , but for the abstract objects set . ( f ) Similar to c , but showing the map of tuned neurons ( shades of green ) with respect to the abstract object set . For b , e: Data is doubly aligned on cue onset and on the grasp ( go ) signal . Sliding ANOVA was computed for each session individually and averaged across all 10 recording sessions per animal . Shades represent standard error from mean ( s . e . m . ) across recording sessions . For c , f: The number of tuned neurons per channel were averaged across all recording sessions and visualized in shades of green and red for the abstract and mixed objects , respectively . Channels without any identified neurons were highlighted in light grey . Map of monkey M is mirrored along vertical axis for better comparison of both animals . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 010 To investigate this effect at the neuronal population level , we performed canonical discriminant analysis ( CDA; see Materials and methods ) , which allowed reducing the neuronal state space ( N-space ) to its most informative dimensions . Figure 2c and 2d show the first three canonical variables during the cue and grasp epoch , respectively . In them , each marker represents the neuronal state of an individual trial in the AIP population ( see Figure 1b for symbol and colour code ) . In this N-space of AIP , we found objects to be separated based on their shape . Independent of the way the objects were grasped , the neural space accurately differentiated horizontal cylinders ( black ) , vertical cylinders ( green ) , rings ( magenta ) , spheres ( orange ) , cubes ( blue ) , and bars ( black ) ( see Video 3 for an animated 3D view of a typical N-space ) . 10 . 7554/eLife . 15278 . 011Video 4 . Population coding in F5 . Joint angles and the population activity of F5 were recorded together and for visual display reduced to their most informative dimensions ( component 1 and 2 ) . The video displays the evolving hand kinematics ( top , left ) and neuronal population activity ( bottom , left ) during three subsequent grasping actions . Arrows point at these trials in the J- ( top , right ) and N-space ( bottom , right ) . The audio-track plays the spiking activity of an individual F5 neuron , which is highlighted in the raster plot in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 011 To further quantify these findings , we computed the Mahalanobis distance between all pairs of conditions in the complete N-space of AIP ( see Materials and methods ) . Hierarchical cluster analysis ( HCA ) performed on these distance measures confirmed the findings of the CDA and revealed a clear clustering according to object shape during visual presentation of the object ( Figure 2e ) that widely persisted during movement execution , although with significantly shorter neural distances ( Figure 2f ) . 100% and 91% of the objects shared their cluster with other objects of the same shape during the cue and grasp epoch , respectively . Importantly , consistent results were observed in both monkeys when performing HCA across all recording sessions ( Figure 2—figure supplement 1–2 , see Materials and methods ) . The large number of objects presented in one recording session required separating the 48 objects on different turntables ( see Figure 1 ) , often objects of similar shape . This separation created small offsets already in the fixation epoch , but at very low modulations . An extreme case is shown in Figure 2a . This might raise concerns on whether coding in AIP was really due to object shape , or rather to the object presentation order ( different turntables presented sequentially ) . However , shape-wise clustering in AIP cannot be explained by the task design for the following reasons: ( 1 ) The offsets in the fixation epoch were very small in comparison to the visual modulations observed in AIP when the objects were presented ( e . g . , see Video 5 ) . ( 2 ) The set of 'mixed' objects – presented and grasped in the same block – clustered with other objects of the same shape ( e . g . ring in mixed block clusters with other ring objects ) . Together , this demonstrates clear shape processing in AIP . The AIP population also encoded the size of objects , but differentiated this geometrical feature at clearly lower neural modulations compared to object shape . As shown in Figure 2e–f and supplements , the majority of objects were located closest to objects of similar size , but with significantly shorter neural distances in comparison to shape features . The secondary role of size was surprising , since object size has significant influence on the aperture of the grasping hand ( Jakobson and Goodale , 1991 ) . To further test the specialisation for shape processing , we tested a mixed set of objects ( Figure 3a ) , causing highly variable visual stimuli and motor responses , against an abstract object set ( Figure 3d ) that we have specifically designed to cause different visual stimuli but the same grip . In theory , a purely visual area should differentiate both sets of shapes , since they both provide different visual stimuli . In contrast , an exclusive motor area should not show modulations for the abstract objects because they require the same grip . Strikingly , the AIP population responded highly similar to the presentation ( cue epoch ) of mixed ( Figure 3b ) and abstract ( Figure 3e ) objects ( t-test , p>0 . 05; 35% vs . 35% and 20% vs . 17% comparing mixed and abstract objects in monkey Z and M , respectively ) . Thus , the equal responses to both object sets strongly supports the specialisation of AIP in processing object shapes . Importantly , AIP remained the most tuned area when the monkeys planned and grasped the abstract objects , as shown in Figure 3e . However , the number of significantly tuned cells decreased during these epochs in comparison to the responses evoked by the mixed objects ( Figure 3b ) . This reduced selectivity could indicate either motor or visual transformations that are both required for grasping: First , the abstract objects were grasped with the same hand configuration ( see Figure 3d ) . Unmodulated activity could therefore reflect the same motor affordance across the six abstract objects ( Fagg and Arbib , 1998; Rizzolatti and Luppino , 2001 ) . Second , the abstract objects have different shapes , but have the graspable handle in common that shares the same geometrical dimensions across all six objects ( see Figure 3d ) . A uniform modulation could therefore also represent visual processes that reduce the objects to its parts relevant for grasping ( i . e . the same geometries of the handle across the six abstract objects ) . Considering that the same AIP population separated the complete object set primarily on their geometrical features ( Figure 2e–f ) would suggest visual rather than motor transformations explaining uniform modulations . We found further indicators for this hypothesis when focusing on objects that caused , in contrast to the abstract objects , equal visual stimuli but different motor responses . To create such a scenario , monkeys were trained to perform power or precision grips on the same object , the handle ( condition 00 and 01 ) . Although both conditions were located most distantly in the kinematic or joint-angle ( J- ) space in both monkeys ( see Figure 4e and 5a ) , they were located closest to each other in the N-space of AIP ( see Figure 2e–f ) , clearly implying a visual representation of the handle . Statistical analysis revealed , however , that both conditions ( 00 , 01 ) slightly increased their neural distance ( Figure 2d ) towards planning and movement execution , as expressed by an average of 21% and 16% of significantly tuned AIP neurons in monkey Z and M ( ANOVA tested in grasp epoch , p<0 . 01 ) . These observations suggest a visual representation of the handle and a further differentiation of its parts that are relevant for grasping . 10 . 7554/eLife . 15278 . 012Figure 4 . Motor planning and execution in F5 . ( a ) Example neuron of F5 , responding to all 50 task conditions ( colour code as in Figure 1b ) . ( b ) Modulation depth plots ( as in Figure 2b ) in the planning and grasp epoch . ( c ) Recorded kinematics was used to drive a monkey-specific musculoskeletal model that allowed extracting 27 DOF . ( d ) A selection of DOFs is presented for three sequential grips: thumb and index finger , wrist , elbow , and shoulder . ( e ) PCA performed on the J-space during the hold epoch allowed visualizing the grip types of all conditions and trials of one recording session ( showing 1st and 2nd PCA component ) . ( f–g ) Raster plot shows the spiking activity of F5-neurons recorded from a single FMA ( F5-ventral ) . ( h ) Mean firing rates during the grasp epoch ( N-space ) were transformed with CDA to reduce and visualize the multidimensional representation of the complete F5 population ( N = 119 , simultaneously recorded ) . In d , g , example trials t1 , t2 and t3 are highlighted in yellow ( hold epoch in d , grasp epoch in g ) and marked with arrows in e , h . ( i ) Neuronal state space evolution shows the course of the task determined by the CDA . e , h and i: For visual comparison the N-space was aligned to the J-space using PCRA; Symbols and colours as in Figure 1b , symbol size corresponds to object size . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 01210 . 7554/eLife . 15278 . 013Figure 5 . Hierarchical cluster analysis of the F5 population . ( a ) Dendrogram of J-space ( 27 DOF ) . ( b–c ) Dendrogram of F5’s complete N-space during the plan and the grasp epoch . Condition numbers as in Figure 1b . A selection of grip types and their corresponding objects are illustrated . In a-c , similar motor characteristics are highlighted with coloured boxes ( see text ) . ( b , c ) is based on the complete F5 population ( N = 119 , simultaneously recorded neurons ) , in contrast to its illustration in the reduced neural space in Figure 4 . See Figure 5—figure supplement 1–2 for the averaged population results across all sessions of animal M and animal Z , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 01310 . 7554/eLife . 15278 . 014Figure 5—figure supplement 1 . F5-Motor coding in animal M across all recording sessions . Neural distances between each pair of group means were measured for each session , averaged across sessions , and visualized as dendograms for the ( a ) plan and ( b ) grasp epoch . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 01410 . 7554/eLife . 15278 . 015Figure 5—figure supplement 2 . F5-Motor coding in animal Z across all sessions . ( a ) Firing rate plot of F5 example unit . ( b ) Modulation depth plot of example unit during grasp epoch . ( c ) Reduced J-space of one session recorded with instrumented glove and ( d ) reduced N-space of F5 during the grasp epoch of the same session ( simultaneously recorded , N = 76 , first two components shown ) . ( e–f ) Neural distances between each pair of group means were measured for each session , averaged across sessions , and visualized as dendograms for the plan and grasp epoch . For c–e: Symbols and colours as in Figure 1b . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 015 Similarly , the AIP population coded all horizontal cylinders based on their shape and then differentiated the two biggest horizontal cylinders ( 55 , 56 , see Figure 2 ) , when they were grasped differently ( either from top or from below ) . Both conditions ( 55 , 56 ) required a focus on different parts of the object ( top vs . bottom edge ) as well as different hand configurations ( pronation vs . supination ) . Importantly , the neural representation of both conditions originated from the very same shape cluster in N-space ( Figure 2c , e ) that subsequently drifted further apart ( see arrows in Figure 2d , f ) . Together , these observations suggest a visual rather than motor representation in AIP . To generate grasping movements , visual attributes of objects need to be transformed into adequate motor commands before they get executed ( Jeannerod et al . , 1995; Rizzolatti and Luppino , 2001 ) . Creating a motor plan and its execution is associated with areas F5 and M1 ( Murata et al . , 1997; Raos et al . , 2006; Umilta et al . , 2007 ) . In a first analysis that compared neuronal population tuning for the mixed ( Figure 3a ) and abstract objects ( Figure 3d ) , we found evidence for a primary motor role of F5 and M1 . The F5 population was strongly activated in the motor epochs when the mixed objects were grasped differently ( up to 47% and 45% tuned neurons in monkey Z and M resp . , see Figure 3b ) , and it was strikingly unmodulated when the abstract objects were grasped similar ( Figure 3e ) . Likewise , the M1 population responded uniformly for similar grips ( Figure 3e ) , but it was modulated very strongly when different hand configurations were applied ( up to 67% and 61% in monkey Z an M resp , see Figure 3b ) . During movement planning , M1 showed no or minimal preparation activity ( Figure 3a–c ) , whereas F5 revealed a multimodal role: in the cue epoch , the tuned F5 population substantially decreased from 28% to 18% in monkey Z , and from 13% to 4% in monkey M , when comparing abstract with mixed objects . This is in strong contrast to AIP , which demonstrated equal population responses for both type of object sets . The reduced F5 modulation suggests early motor processes starting shortly after object presentation . However , 18% and 4% ( Monkey Z and M resp . ) of all F5 neurons remained their modulation when the monkeys observed the abstract geometries . Thus , at least some F5 cells coded objects in purely visual terms . It is notable , that we found significantly different contributions in motor preparation between the F5 recording sites . The visual ( Figure 3e–f ) and visuomotor modulations ( Figure 3b–c ) prior to movement primarily originated from the ventral recording array , corresponding to the F5a subdivision ( Gerbella et al . , 2011; Theys et al . , 2012a ) . In fact , 76% of the visual ( abstract objects ) and 72% of visuomotor tuned neurons ( mixed objects ) recorded from F5 were detected on the ventral site ( ANOVA p<0 . 01 , all sessions , tested in cue epoch ) , in line with previously reported enhanced decoding capabilities of planning signals from ventral F5 ( Schaffelhofer et al . , 2015a ) . In contrast , the dorsal F5 array , corresponding to F5p , mainly contributed during movement execution by a four-fold increase of its tuned population with respect to the cue epoch . The general population response of F5 was confirmed when extending our analysis to all 49 objects . Neurons were modulated by hand grasping actions and typically showed the strongest MD during motor execution . The example neuron shown in Figure 4a–b demonstrated a maximum MD of 39 . 8 Hz while grasping and allowed significantly separating 42% of all condition pairs in this epoch ( Figure 4b , right ) . Importantly , the neuron showed similar motor coding already during the preparation epoch ( r = 0 . 76 , between plan and grasp MD-maps , as shown in Figure 4b ) . Using the planning activity of the example neuron alone , about 43% of the task conditions could be separated , thereby demonstrating the important role of F5 for hand movement planning . To evaluate the relationship between neural population activity and motor actions , we recorded spiking activity ( Figure 4f–g ) together with the kinematics of the primate hand ( Figure 4c–d ) . Dimensionality reduction methods ( PCA , CDA ) were performed to express the high dimensional kinematic ( J-space ) and neural space ( N-space ) in a low dimensional fashion . Procrustes analysis ( PCRA ) was subsequently applied to align the N- to the J-space for their visual comparison ( Figure 4e , h-i ) ( see Materials and methods ) In detail , joint trajectories were recorded in 3-D space from an instrumented glove ( Schaffelhofer and Scherberger , 2012 ) and translated to joint angles utilizing a primate-specific musculoskeletal model ( Schaffelhofer et al . , 2015b ) . The mean values of a total of 27 degrees of freedom were then extracted from the hold epoch to describe the hand shapes used for grasping the objects ( J-space , Figure 4c–d ) . This provided accurate and stable , but highly variable hand configurations across the 49 tested objects . Performing PCA on this dataset allowed visualizing all correctly performed grips in a low-dimensional fashion . Thus , each marker in Figure 4e reflects one individual grip/trial . Similarly , spiking activity from a large population of neurons was recorded with the FMAs . Then , mean firing rates were extracted from epochs of interest ( e . g . , grasp ) as illustrated in Figure 4f-g ( N-space ) . On this dataset we performed CDA and PRCA to reduce and compare the multidimensional N-space and the J-space as shown in Figure 4h and Figure 4e , respectively . For an animation of Figure 4 see Video 4 . The J-space demonstrated a high variability of hand configurations across conditions and closely reflected the hand’s wrist orientation ( 1st principal component ) and hand aperture ( 2nd principal component ) . Furthermore , the reduced J-space allowed observing the most relevant kinematic observations with respect to the presented objects , as shown in Figure 4e and 5a: ( 1 ) Objects of small sizes such as the small rings ( condition ID 21 , 22 ) , spheres ( 11 , 41 , 42 ) and cubes ( 31 , 32 ) were grasped similarly ( index finger and thumb ) and thus were located close to each other in the reduced and the complete J-space . ( 2 ) Vertical cylinders and big rings ( 16 , 23–26 , 71–76 ) were enclosed with the digits and required 90° of wrist rotation . Therefore , these grips are located close to each other in J-space . ( 3 ) All abstract objects ( 91–96 ) shared a compact cluster and demonstrated their similarity in J-space . ( 4 ) Precision ( 00 ) and power grips ( 01 ) performed on the same handle required highly different hand configurations and were located distant in J-space . ( 5 ) The highest separation of hand configuration across objects of similar shape was evoked from the rings that were grasped with precision ( 21 , 22 ) or power grips ( 23–26 ) . Small rings and big rings are therefore separated in the J-space . 10 . 7554/eLife . 15278 . 016Figure 6 . Motor execution in M1 . ( a ) Example neuron of M1 in monkey M , curves show firing rates separately for all 50 task conditions ( colour code as in Figure 1b ) . ( b ) Modulation depth plots ( as in Figure 4b ) for the planning and hold epoch of the example neuron in a . ( c ) Population activity from the reduced and ( d ) the complete kinematic space ( J-space ) is compared to ( e ) the reduced and ( f ) the complete neural population space ( N-space ) of M1 during the hold epoch of the task ( N = 151 , simultaneously recorded ) . In e , N-space was aligned to J-space with PCRA; symbols and colours as in Figure 1b; symbol size corresponds to object size . Arrows t1-t3 highlight the example trials of Figure 4d , g . See Figure 6—figure supplement 1–2 for the averaged population results of animal M and animal Z , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 01610 . 7554/eLife . 15278 . 017Figure 6—figure supplement 1 . M1-Motor coding in animal M across all sessions . Neural distances between each pair of group means were measured for each session , averaged across sessions , and visualized as dendograms during the hold epoch . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 01710 . 7554/eLife . 15278 . 018Figure 6—figure supplement 2 . M1-Motor coding in animal Z across all sessions . ( a ) Firing rate plot of M1 example unit . ( b ) Modulation depth plot of example unit in hold epoch . ( c–d ) Reduced J-space of one session recorded with the instrumented glove and reduced N-space of M1 during hold epoch of the same session ( simultaneously recorded , N = 62 , first two components shown ) . ( e ) Neural distances between each pair of group means were measured for each session , averaged across all sessions , and visualized as dendograms . For c , d: Symbols and colours as in Figure 1b . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 018 Importantly , the majority of kinematic observations were also found in the N-space of F5 during motor execution ( Figure 4h ) . In contrast to AIP , conditions of different visual stimuli but equal grips were located close to each other in the neural space ( observation 1–3 ) , whereas conditions of similar visual stimuli but different motor responses caused a separation ( observation 4–5 ) . These results were further supported by the high similarity between the J- ( Figure 4e ) and N-space ( Figure 4h ) during movement execution ( for quantification see section ‘Numerical comparison’ and Figure 7 ) . The findings demonstrate clearly different coding properties with respect to AIP and reveal a primary motor role of F5 during movement execution . CDA and PRCA were further used to visualize the evolution of the F5 population during the task . Similar to observations at the single unit level ( see Figure 4a–b ) , the F5 population demonstrated first modulations and expressions of the upcoming motor actions already during motor preparation . 10 . 7554/eLife . 15278 . 019Figure 7 . Motor similarity measure . Boxplots illustrating similarity between the population coding of J-space and N-space during the hold epoch of the task , as provided by the PCRA analysis . Left: results in AIP , F5 , and M1 across all 10 recording sessions for monkey Z . Right: same for monkey M . Red horizontal lines indicate median value , boxes show lower and upper quartile of data ( 25%–75% ) , and whiskers indicate maximum and minimal values . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 019 To quantify the observations made on the reduced spaces , hierarchical cluster analysis ( HCA ) was performed with the complete population of F5 neurons ( N-space ) and joint kinematics ( J-space ) . In accordance with the low-dimensional representation , abstract forms , small objects , as well as the big rings and cylinders created individual clusters in the J-space and were located close to each other . As shown in Figure 5b , these motor characteristics were rudimentarily marked in F5 already during motor preparation ( see Figure 5b , coloured boxes ) . These clusters that emerged during motor planning persisted to a large extent during motor execution , but increased their relative distance to each other ( Figure 5c ) , in line with the higher MD of single F5 neurons during grasp execution ( Figure 4b ) . HCA performed on the averaged population response across all recording sessions ( see Materials and methods ) confirmed in both animals the motor characteristics of simultaneously recorded populations in single sessions ( Figure 5—figure supplements 1–2 ) . As discussed above , premotor preparation activity primarily originated from the ventral F5 array . Thus , the modulations observed prior to movement execution , such as observed in the CDA and hierarchical clustering , are mainly based on the ventral recording site . In contrast , both arrays significantly supported movement execution . The different modalities in both recording sites are in line with the architectonically ( Belmalih et al . , 2009 ) and connectionally ( Borra et al . , 2010; Gerbella et al . , 2011 ) distinct subdivisions F5a and F5p , which largely correspond to the ventral and dorsal recording array , respectively . Distinct connections of F5a with SII , AIP and other subdivision of F5 ( but not to M1 ) ( Gerbella et al . , 2011 ) suggest an integration of visual , motor and context specific information ( Theys et al . , 2013 ) . On the other hand , connections of F5p to the hand area of M1 and to the spinal cord ( Borra et al . , 2010 ) suggest a rather direct contribution to hand movement control . Taken together , the multimodal preparation signals , including visual and motor contribution , and the distinct motor feature coding towards planning and execution , supports the key role of F5 for visuomotor transformation . In agreement with the general population response ( Figure 3b ) , single neurons of M1 were almost exclusively modulated during movement execution and showed minimal modulations during preparatory epochs ( Umilta et al . , 2007 ) , as also indicated by the example neuron in Figure 6a–b . This neuron was capable of differentiating 52% of condition pairs when holding the object , but failed to separate any ( 0% ) of the conditions when planning the movement . ( ANOVA , Tukey-Kramer criterion , p<0 . 01 ) , thereby highlighting its major difference to F5 neurons . Although the motor relevance of the hand area of M1 has been described extensively with electrophysiological ( Schieber , 1991; Schieber and Poliakov , 1998; Spinks et al . , 2008; Umilta et al . , 2007 ) and anatomical methods ( Dum and Strick , 2005; Rathelot and Strick , 2009 ) , it has been unclear how versatile hand configurations are encoded at the population level . The high variability of objects and motor affordances in our task allowed such a description ( Figure 6 ) . Similar to the analysis performed on the F5 population , PCRA analysis compared the J-space with the N-space of M1 and revealed , highly important for the understanding of hand movement generation and as an important control , striking similarities between both representations , as demonstrated in Figure 6c , e . The similarity of J- and N-space of M1 was not only visible in the first two components , but was also quantified across all dimensions in the hierarchical cluster trees of the simultaneously recorded M1 population ( Figure 6d , f ) and when averaging the population response across all recording sessions ( Figure 6—figure supplement 1–2 for monkey M and Z , respectively ) . The large majority of conditions were assigned to the same clusters in the J- and the M1-space ( coloured boxes in Figure 6d , f , exceptions: conditions 42 , 43 ) . Furthermore , all of the motor characteristics ( 1–5 ) defined above were also observed and were even more strongly represented in M1 than in F5: again , the group of small objects ( 11 , 21–22 , 31–32 , 41–42 ) and the group of abstract forms ( 91–96 ) created strong and individual clusters , whereas small ( 21–22 ) and big rings ( 16 , 23–26 ) as well as precision ( 00 ) and power grips ( 01 ) were located distant from each other . We also quantified motor similarities between the J-space and the N-space of area AIP , F5 , and M1 across all recording sessions , similar to individual recording session analyses presented in Figures 4–6 . For this , similarity measures were performed between the J-space and the N-space of AIP , F5 and M1 using PCRA ( see Materials and methods ) . A similarity of '0' indicates a complete match between the distribution of trials ( n > 500 ) in N-space and in J-space , whereas values close to '1' represent high divergences between the clustering in both spaces . In accordance with the previous analysis , M1 demonstrated the highest similarity to the J-space ( averaged value across 10 recording sessions for monkey M: 0 . 48 , for monkey Z: 0 . 59 ) , followed by F5 ( monkey M: 0 . 61; monkey Z: 0 . 65 ) and AIP ( monkey M: 0 . 75; monkey Z: 0 . 75 ) . These values were highly consistent across recording sessions and monkeys ( Figure 7 ) . Furthermore , differences between areas were significant ( ANOVA and post-hoc Tukey-Kramer criterion , p<0 . 01 ) . Together , these results highlight the different roles of the cortical areas AIP , F5 , and M1 for the preparation and execution of grasping movements . For visualizing the communication of features between areas , we correlated the dynamic modulations on a trial-basis ( see Materials and methods ) . In this , the elicited modulation patterns of each neuronal population were correlated in a pair-wise fashion ( AIP vs . F5 , F5 vs . M1 , AIP vs . M1 ) across time ( Figure 8 , Video 5 ) . 10 . 7554/eLife . 15278 . 020Figure 8 . Temporal feature correlation between areas . The distance ( in firing rate ) between all possible trial-pairs was computed separately in the N-space of area AIP , F5 and M1 . The resulting distance maps thus represent the neuronal modulations of a population ( e . g . AIP ) for a specific time t . ( Right ) Example maps show the neural modulation patterns at key times t1 ( object presentation ) and t2 ( during hold ) . Warm colours: long neural distances , cool colours: short neural distances . ( Left ) Correlating the neuronal patterns across time ( Spearman’s r ) allowed visualizing the similarity between the areas for animal Z ( top row ) and animal M ( bottom row ) . For computation , spiking activity was aligned to the beginning of the cue and grasping onset . For an animation of the feature code correlation see Video 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 02010 . 7554/eLife . 15278 . 021Video 5 . Temporal feature correlation between areas . The neural distance ( in firing rate ) between all pairs of trials within a neuronal population provided a modulation pattern between trials ( top ) for area AIP , F5 , and M1 . Correlating these neuronal patterns for each moment in the task allowed visualizing the coding similarity between pairs of areas across time ( AIP-F5 in green , AIP-M1 in cyan , and F5-M1 in magenta ) . Feature correlations are based on all trials , whereas the presented object and corresponding grasp movement are shown for an example trial . DOI: http://dx . doi . org/10 . 7554/eLife . 15278 . 021 During object presentation , we found similar chequered patterns in area AIP in both animals ( t1 in Figure 8 ) , again confirming the coding of object shapes at the population level , but here on trial-by-trial basis . These visual feature patterns caused a significant correlation peak with area F5 in animal Z , suggesting the propagation of visual information; in animal M , however , only a minor correlation peak was observed ( Figure 8 ) . These results are consistent with the different proportion of F5 visual cells identified in animal Z and M ( Figure 3 ) , which only temporarily shared the visual coding with AIP . Interestingly , F5 and AIP demonstrated minimal similarities during movement planning , in support of the different encoding schemes of visual and motor features described at the population level ( Figures 2 , 4 and 5 ) . F5 modulations strongly increased before movement execution and were followed by M1 activity after movement onset ( see Video 5 ) , suggesting that premotor cortex drives M1 . Signals did not only follow in time but also in modulation patterns , as expressed by the impressively high correlation coefficients in both animals ( t2 in Figure 8 ) , and in agreement with similar motor coding schemes observed with the population analysis . Together , these results expand the static feature coding found with CDA by demonstrating dynamical functional coupling of AIP , F5 , and M1 during the course of the delayed grasp and hold task . We targeted the cortical grasping circuit with FMAs implanted under anatomical considerations . AIP , an end-stage area of the dorsal visual stream , receives input from parietal visual areas ( e . g . , LIP , CIP , and V6a ) as well as from the inferior temporal cortex ( e . g . , areas TEa and TEm ) ( Borra et al . , 2008; Nakamura et al . , 2001 ) . AIP further connects to pre-motor F5 via dense reciprocal projections ( Borra et al . , 2008; Gerbella et al . , 2011; Luppino et al . , 1999 ) . In agreement with these known connections , we observed strong visuomotor responses in AIP and F5 ( Figure 3 ) . Another significant connection links F5 with the hand area of M1 ( Dum and Strick , 2005; Kraskov et al . , 2011 ) . As described by Rathelot and Strick ( 2009 ) , neurons of M1 , in particular in the bank of the central sulcus , form direct connections to α-motor neurons in the spinal cord that drive the distal hand muscles . In line with these observations , the majority of M1-neurons demonstrated significant modulations during hand movement control despite highly constant reaching components ( Figure 3b ) . Together , strong grasp modulations ( Figure 3 ) and their intercommunication ( Figure 8 ) confirmed the correct positioning of the electrode arrays and the importance of all areas for visuomotor processing . While it is known that area AIP and F5 strongly respond to the presentation of objects and to the variation of their dimensions ( e . g . shape , orientation ) ( Baumann et al . , 2009; Fagg and Arbib , 1998; Murata et al . , 1997; 2000; Taira et al . , 1990 ) it was unclear whether these modulations reflect visual or motor processes . For example , responses to object presentation can either reflect object attributes or instant motor representations as we have shown here ( Figure 3 ) and as suggested previously ( Murata et al . , 1997 ) . By design , our task created associations as well as dissociations between objects and their afforded hand configurations ( Figure 4e , 5a ) in order to disentangle these normally highly linked parameters . The AIP population demonstrated a distinct visual separation of objects ( Figure 2 , 3e ) , which was largely unrelated to the observed motor characteristics ( Figure 4e , 5a ) , even when grasping in complete darkness ( Figure 7 ) . The predominant criterion for object separation was object shape , even for abstract geometries that required the same grip ( Figure 3d–f ) . These findings are in agreement with anatomical connections of AIP to the inferior temporal cortex ( Borra et al . , 2008 ) that codes perceived shapes ( Logothetis et al . , 1995; Tanaka , 1996 ) . Object size was expressed in the population but played , surprisingly , only a secondary role in AIP ( Figure 2 ) . These smaller neural distances with respect to size are remarkable , in particular since this feature is highly relevant for controlling hand aperture ( Jakobson and Goodale , 1991 ) . A possible explanation for this effect could be the higher computational effort ( more neurons ) required for processing shape in comparison to size . AIP also widely maintained its coding for visual object properties during movement execution in the dark ( Figure 2f ) , although with significantly smaller neural distances ( Figure 2e–f ) . We hypothesize that AIP serves as working memory and not only extracts , but also maintains visual object information required for motor planning and execution ( Rizzolatti and Luppino , 2001 ) . We emphasize that visual coding at the population level does not preclude motor coding of some individual cells , as suggested previously ( Murata et al . , 1997 ) . Rather , the evidence of bidirectional connections to F5 suggests that subpopulations exist in AIP that reflect feedback of motor signals . Despite AIP’s primary visual coding , we observed a distinctive coding in the N-space when the same objects were grasped differently ( e . g . , handle and cylinders in Figure 2c–d ) . These modulations could reflect either motor ( Fagg and Arbib , 1998 ) or visual transformations required for grasping . In fact , when grasping an object , both processes are required: the visual selection of ( focus on ) object parts and the selection of a corresponding hand configuration . Several indicators in our data suggest visual rather than motor transformations in AIP: First , the population separated the entire set of objects in visual terms , but differentiated same objects when grasped differently . Second , the separation of the same object when grasped differently , originated from one object shape cluster . In line with these findings , Baumann et al . ( 2009 ) demonstrated that AIP planning activity depends on the previous visual knowledge of an object ( see Figure 6 in their paper ) . Presenting an object ( the same handle as we used ) caused a strong visual activation in the AIP population that got separated when a grip type ( precision or power grip ) was instructed subsequently . In contrast , instructing the grip type before object presentation did not lead to a substantial population response and caused significant modulations only after the object was presented . These differentiation schemes support our hypothesis of visual rather than motor transformations in AIP . Grasping requires the transformation of visual descriptions of an object into adequate motor commands . F5 is densely connected to AIP ( Borra et al . , 2008; Gerbella et al . , 2011; Luppino et al . , 1999 ) and has been associated with these visuomotor processes ( Fluet et al . , 2010; Jeannerod et al . , 1995; Rizzolatti and Luppino , 2001 ) . Similar to AIP , neurons in F5 respond to the presentation of 3D objects ( Fluet et al . , 2010; Murata et al . , 1997; Raos et al . , 2006; Theys et al . , 2012a; Vargas-Irwin et al . , 2015 ) . These modulations have been discussed to reflect object or motor representations ( Fluet et al . , 2010; Murata et al . , 1997; Raos et al . , 2006; Theys et al . , 2013 ) , however without measuring corresponding hand kinematics . With many more object conditions , precise hand kinematics , and a large neuronal dataset we were able to address this question and confirm a primary motor role of the F5 population for representing the joint ( J- ) space during motor execution ( Figures 4 , 5 ) . Due to the large number of objects tested , we could demonstrate for the first time that the F5 population does not reflect stereotypical grip types ( Rizzolatti and Luppino , 2001 ) , but represents a continuum of many hand configurations ( Figure 5 ) , highly similar to M1 ( Figure 8 , Video 5 ) . Importantly , motor characteristics in the N-Space of F5 could be observed not only in motor execution epochs , but also during motor preparation ( Figure 4i , 5b ) . In contrast to AIP , the F5 population showed already in the cue epoch a reduced tuning for abstract objects requiring the same grip ( Figure 3a–b vs . Figure 3d–e ) . This suggests a rapid appearance of grip features in F5 shortly after object presentation . However , F5 also contained a small subgroup of neurons that represented pure visual information during the cue epoch ( Figure 3 ) . The vast majority of these cells have been recorded from the ventral array , i . e . , they originate from the F5a subdivision . In line , electrophysiological investigations of area F5a demonstrated selectivity for 3D shape ( Theys et al . , 2012a ) . We suggest that these neurons receive direct visual input from AIP and might be crucially involved in the activation of stored motor plans . Similar modulation patterns in AIP and F5 during object presentation suggest the communication of visual properties to premotor cortex ( Figure 8 and Video 5 ) . The presence of multimodal signals as well as the proximity of specialized sub areas for visual input and motor output suggest a central role of F5 for visuomotor transformation . In contrast to F5 , the hand area of primary motor cortex ( M1 ) showed an almost exclusive role for motor execution ( Saleh et al . , 2010; Umilta et al . , 2007 ) . We investigated , for the first time , the N-space of the hand area of M1 ( Rathelot and Strick , 2009 ) for a large repertoire of grasping actions . As expected , the N-space of M1 was closely related to the J-space ( Figure 6 ) and showed the highest motor similarity across all three areas ( Figure 7 ) . Correlating these M1 modulation patterns with F5 revealed that both areas are not only strongly interconnected ( Dum and Strick , 2005; Kraskov et al . , 2011 ) , but they also heavily share motor features for grasping ( Figure 8 ) . The high similarity of the F5 and M1 population during movement execution ( Figure 4h vs . Figure 6e , Figure 8 , and Video 5 ) is in agreement with findings of similar coding schemes between both areas in spiking ( Umilta et al . , 2007 ) and beta-band LFP activity ( Spinks et al . , 2008 ) . The earlier premotor ( vs . motor ) onset ( Figure 3b ) suggests that the movement is facilitated by F5 , whereas the closer motor similarity of M1 might reflect its advanced finger control capacity due to more direct motor connections ( Fogassi et al . , 2001; Rathelot and Strick , 2009; Schieber and Poliakov , 1998 ) . Our findings demonstrate that the cortical grasping network transforms visual object attributes into motor plans and actions . We found highly different coding schemes between the frontoparietal circuits of AIP and F5 , indicating widely separated processing of visual and motor features . These findings suggest that visuomotor transformation is achieved effectively by visual object descriptions that activate linked motor plans in the reciprocal network of AIP and F5 . Strong feature communication between F5 and M1 further suggest a common motor network that executes and refines the prepared motor plan . Two rhesus monkeys ( Macaca mulatta ) participated in this study ( animal Z: female , 7 . 0 kg body weight; animal M: male , 10 . 5 kg ) . Animal housing , care , and all experimental procedures were conducted in accordance with German and European laws governing animal welfare and were in agreement with the guidelines for the care and use of mammals in neuroscience and behavioural research ( National Research Council , 2003; see also Ethics statement ) . We developed an experimental setup that allowed us to present a large number of graspable objects to the monkeys while monitoring their behaviour , neural activity and hand kinematics . During each recording session , monkeys grasped a total of 42–48 objects of equal weight that were placed on 8 interchangeable turntables ( Figure 1a–b ) . Objects were of different shapes and sizes including rings ( diameter of 15 , 20 , 25 , 30 , 35 , and 40 mm ) , cubes ( length of 15 , 20 , 25 , 30 , 35 , and 40 mm ) , spheres ( diameter of 15 , 20 , 25 , 30 , 35 and 40 mm ) , horizontal cylinders ( diameter of 15 , 20 , 25 , 30 , 35 , 40 mm , equal length ) , vertical cylinders ( diameter: 15 , 20 , 25 , 30 , 35 , 40 mm , equal height ) , and bars ( depth of 15 , 20 , 25 , 30 , 35 , and 40 mm , equal height and width ) . Furthermore , a mixed turntable held objects of different shapes of average size . Important for this study , a special turntable was holding objects of abstract forms , which differed largely visually but required identical hand configurations for grasping ( Figure 1b ) . Both monkeys were also trained to grasp a single object , a handle , either with a precision grip or a power grip . This extended our task by two more conditions ( to a total of 50 ) that evoked similar visual , but different motor responses . In 20 recording sessions ( 10 per animal ) , each condition was repeated at least 10 times . Precision and power grips applied on the handle as well as grasping and lifting the 3D-objects were detected with photoelectric barriers ( Schaffelhofer et al . , 2015a ) . The turntable position was controlled with a step motor . Furthermore , the monkey’s eye position was monitored with an optical eye tracking system ( model AA-EL-200; ISCAN Inc . ) . All behavioural and task-relevant parameters were controlled using a custom-written control software implemented in LabVIEW Realtime ( National Instruments , Austin , TX ) . Monkeys were trained to grasp 49 objects ( Figure 1b ) in a delayed grasp-and-hold task ( Figure 1d ) . While sitting in the dark the monkeys could initiate a trial ( self-paced ) by placing their grasping hand ( left hand in monkey Z , right hand in monkey M ) onto a rest sensor that enabled a fixation LED close to the object . Looking at ( fixating ) this spot for a variable time ( fixation epoch , duration: 500–800 ms ) activated a spot light that illuminated the graspable object ( cue epoch: 700 ms ) . After the light was turned off the monkeys had to withhold movement execution ( planning epoch: 600–1000 ms ) until the fixation LED blinked for 100 ms . After this , the monkeys released the rest sensor , reached for and grasped the object ( movement epoch ) , and briefly lifted it up ( hold epoch: 500 ms ) . The monkeys had to fixate the LED throughout the task ( max . deviation: ~5 deg of visual angle ) . In trials where the handle was grasped , one of two additional LEDs was presented during the cue epoch , which indicated to perform either a precision grip ( yellow LED ) or a power grip ( green LED ) . All correctly executed trials were rewarded with a liquid reward ( juice ) and monkeys could initiate the next trial after a short delay . Error trials were immediately aborted without reward and excluded from analysis . Finger , hand , and arm kinematics of the acting hand were tracked with an instrumented glove for small primates ( Figure 1a , c ) . Seven magnetic sensor coils ( model WAVE , Northern Digital ) were placed onto the fingernails , the hand’s dorsum as well as the wrist to compute the centres of 18 individual joints in 3D space , including thumb , digits , wrist , elbow and shoulder . The method and its underlying computational model have been described previously ( Schaffelhofer and Scherberger , 2012 ) . Recorded joint trajectories were then used to drive a 3D-musculoskeletal model ( Schaffelhofer et al . , 2015b ) , which was adjusted to the specific anatomy of each monkey . The model was implemented in OpenSim ( Delp et al . , 2007 ) and allowed extracting a total of 27 DOF [see Schaffelhofer et al . ( 2015b ) for detailed list of DOF] . All extracted joint angles from the model were low-pass filtered ( Kaiser window , finite impulse response filter , passband cutoff: 5–20 Hz ) , downsampled to 50 Hz and used to describe the hand configuration in a 27-dimensional joint space ( J-space ) . Single and multiunit activity was recorded simultaneously using floating microelectrode arrays ( FMA , Microprobe Inc . , Gaithersburg , MD , USA ) . In each monkey we recorded from in total 192 channels of 6 individual arrays implanted into the cortical areas AIP , F5 , and M1 ( see Figure 1e–g ) . In each array , the lengths of the electrodes increased towards the sulcus and ranged from 1 . 5 ( 1st row ) to 7 . 1 mm ( 4th row ) . In area F5 , one array was placed in the posterior bank of the inferior arcuate sulcus approximately targeting F5a ( longer electrodes ) ( Theys et al . , 2012a ) and approaching the F5 convexity ( F5c; shorter electrodes ) . The second and more dorsally located array was positioned to target F5p . In AIP , the arrays were implanted into the end of the posterior intraparietal sulcus at the level of area PF and more dorsally at the level of area PFG . In M1 , both arrays were placed into the hand area of M1 into the anterior bank of the central sulcus at the level of the spur of the arcuate sulcus ( Rathelot and Strick , 2009 ) . See Schaffelhofer et al . ( 2015a ) for details on surgical procedures . Neural activity was recorded at full bandwidth with a sampling frequency of 24 kHz and a resolution of 16 bits ( model: RZ2 Biosignal Processor; Tucker Davis Technologies , FL , USA ) . Neural data was synchronously stored to disk together with the behavioural and kinematic data . Raw recordings were filtered offline ( bandpass cutoff: 0 . 3––7 kHz ) before spikes were detected ( threshold: 3 . 5x std ) and extracted . Spike sorting was processed in two steps: First , we applied super-paramagnetic clustering ( Quiroga et al . , 2004 ) and then revised the results by visual inspection using Offline sorter ( Plexon , TX , USA ) to detect and remove neuronal drift and artefacts . No other pre-selection was applied and single and multiunit activity have been analysed together .	In order to grasp and manipulate objects , our brains have to transform information about an object ( such as its size , shape and position ) into commands about movement that are sent to our hands . Previous work suggests that in primates ( including humans and monkeys ) , this transformation is coordinated in three key brain areas: the parietal cortex , the premotor cortex and the motor cortex . But exactly how these transformations are computed is still not clear . Schaffelhofer and Scherberger attempted to find out how this transformation happens by recording the electrical activity from different brain areas as monkeys reached out to grasp different objects . The specific brain areas studied were the anterior intraparietal ( AIP ) area of the parietal cortex , a part of the premotor cortex known as F5 , and the region of the motor cortex that controls hand movements . The exact movement made by the monkeys’ hands was also recorded . Analysing the recorded brain activity revealed that the three brain regions worked together to transform information about an object into commands for the hand , although each region also had its own specific , separate role in this process . Neurons in the AIP area of the parietal cortex mostly dealt with visual information about the object . These neurons specialized in processing information about the shape of an object , including information that was ultimately important for grasping it . In contrast , the premotor area F5 represented visual information about the object only briefly , quickly switching to representing information about the upcoming movement as it was planned and carried out . Finally , the neurons in the primary motor cortex were only active during the actual hand movement , and their activity strongly reflected the action of hand as it grasped the object . Overall , the results presented by Schaffelhofer and Scherberger suggest that grasping movements are generated from visual information about the object via AIP and F5 neurons communicating with each other . The strong links between the premotor and motor cortex also suggest that a common network related to movement executes and refines the prepared plan of movement . Further investigations are now needed to reveal how such networks process the information they receive .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Object vision to hand action in macaque parietal, premotor, and motor cortices
Like other intracellular fusion events , the homotypic fusion of yeast vacuoles requires a Rab GTPase , a large Rab effector complex , SNARE proteins which can form a 4-helical bundle , and the SNARE disassembly chaperones Sec17p and Sec18p . In addition to these proteins , specific vacuole lipids are required for efficient fusion in vivo and with the purified organelle . Reconstitution of vacuole fusion with all purified components reveals that high SNARE levels can mask the requirement for a complex mixture of vacuole lipids . At lower , more physiological SNARE levels , neutral lipids with small headgroups that tend to form non-bilayer structures ( phosphatidylethanolamine , diacylglycerol , and ergosterol ) are essential . Membranes without these three lipids can dock and complete trans-SNARE pairing but cannot rearrange their lipids for fusion . Membrane fusion on the exocytic and endocytic pathways underlies cell growth , hormone secretion , neurotransmission , and certain nutrient and pathogen uptake ( Wickner and Schekman , 2008 ) . Fusion is guided and catalyzed by families of proteins which are conserved from yeast to humans: Rab family GTPases serve as membrane-identity master switches ( Grosshans et al . , 2006 ) , binding large multisubunit effector complexes which catalyze tethering , the first step of specific membrane association . Membranes bear SNARE proteins ( Jahn and Scheller , 2006 ) , with characteristic heptad repeats with a central arginyl ( R-SNARE ) or glutamyl ( Q-SNARE ) residue ( Fasshauer et al . , 1998 ) and membrane anchor domains . SNAREs can bind ( snare ) each other , either as cis-SNARE complexes in which each SNARE is anchored to the same membrane or as trans-SNARE complexes in which SNAREs are anchored to two different apposed membranes . When the SNAREs of each membrane have been liberated from cis-SNARE complexes by the SNARE complex disassembly system of Sec18p ( NSF ) , Sec17p ( α-SNAP ) , and ATP ( Mayer et al . , 1996 ) , and the membranes have been tightly apposed through tethering , SNAREs can assemble into trans-SNARE complexes . Trans-SNARE complex assembly is a prerequisite for the bilayer rearrangements of fusion . Despite increasingly detailed knowledge of these protein catalysts , far less is known of the lipid requirements for fusion . Fusion can be reconstituted with a few recombinant SNARE proteins in liposomes of simple lipid composition ( Weber et al . , 1998; Fukuda et al . , 2000 ) such as phosphatidylcholine ( PC ) plus phosphatidylserine ( PS ) . Nevertheless , genetic screens for in vivo fusion defects and fusion assays of isolated organelles emphasize the importance of a more complex lipid composition under physiological conditions . We study membrane fusion with the vacuoles ( lysosomes ) of Saccharomyces cerevisiae ( Wickner , 2010 ) . An initial screen for altered vacuole morphology ( Wada et al . , 1992 ) , the vam phenotype , identified the vacuole-specific Rab ( now termed Ypt7p ) , the vacuole-specific SNAREs , and subunits of the Rab effector complex ( termed HOPS , for homotypic fusion and vacuole protein sorting; Seals et al . , 2000; Wurmser et al . , 2000 ) . However , later genomic screens for fragmented vacuole morphology in strains with defined nonessential gene deletions suggested that sterol and phosphoinositides were also required for fusion ( Seeley et al . , 2002 ) . With a quantitative , colorimetric assay of the fusion of purified vacuoles , biochemical studies confirmed vital roles for phosphoinositides ( Mayer et al . , 2000; Cheever et al . , 2001; Seeley et al . , 2002; Fratti et al . , 2004; Mima and Wickner , 2009; Xu and Wickner , 2010 ) , diacylglycerol ( Jun et al . , 2004 ) , and ergosterol ( Kato and Wickner , 2001; Seeley et al . , 2002 ) . It was found that each of these lipids co-localized with the Rab , Rab-effector , and SNAREs in the fusion microdomain of docked vacuoles , and that the localization of these lipids to this microdomain is interdependent with localization of the fusion proteins ( Fratti et al . , 2004 ) . Exploiting an assay of fusion of proteoliposomes consisting of vacuolar lipids , the purified prenylated Rab Ypt7p , 4 recombinant vacuolar SNAREs ( Vam3p , Vti1p , Vam7p , and Nyv1p ) , HOPS , Sec17p , Sec18p , and ATP ( Zucchi and Zick , 2011 ) , we have reexamined the roles of lipids in the fusion reaction . We find that small head-group neutral lipids that tend to form nonbilayer structures are essential for fusion at physiological SNARE concentrations . Small head-group neutral lipids are not needed for trans-SNARE pairing , but are required for the ensuing lipid rearrangements that constitute membrane fusion . Membrane fusion was measured as protected lumenal compartment mixing and concurrent lipid mixing by a modified version of our published assay ( Zucchi and Zick , 2011 ) . Reconstituted proteoliposomes ( RPLs ) were prepared with a recombinant prenylated Rab ( Ypt7p ) and with the four vacuolar SNAREs ( Nyv1p , Vam3p , Vti1p , and Vam7p , which are the vacuolar R , Qa , Qb , and Qc SNAREs; Fasshauer et al . , 1998 ) . RPLs were formed from an octylglucoside mixed-micellar solution of these proteins and vacuolar lipids during lengthy dialysis in the cold , then isolated by flotation . One set of proteoliposomes ( Figure 1A ) bears Marina Blue-linked phosphatidylethanolamine as a lipidic marker and entrapped Cy5-derivatized streptavidin as a lumenal marker , while a complementary set of RPLs bears the lipidic marker NBD-PE and lumenally entrapped biotinylated phycoerythrin . Fusion reactions are performed in the presence of a large excess of external , nonfluorescent streptavidin to bind any biotinylated R-phycoerythrin that may be released from the proteoliposomes by lysis . Upon addition of purified HOPS , Sec17p , Sec18p , and Mg2+:ATP , fusion allows the biotin-R-phycoerythrin to bind to the Cy5-streptavidin within the lumen of fused vesicles while remaining inaccessible to the external , nonfluorescent streptavidin . This is readily assayed by the ensuing Förster resonance energy transfer ( FRET ) between the Cy5 and R-phycoerythrin . Fusion is also reflected in the quenching of the fluorescence of Marina Blue-PE as it is mixed in the same bilayer with NBD-PE . The rate and extent of fusion are governed by the proteoliposomal lipid composition and by the molar ratio of SNARE proteins to lipid . Proteoliposomes of a vacuolar mixed lipid ( VML ) composition , based on the established composition of the isolated organelle ( Schneiter et al . , 1999; Zinser et al . , 1991 ) , bearing SNAREs at a 1:1000 molar ratio to lipids ( Figure 1B , left , filled circles ) or at a 1:3000 ratio ( filled squares ) undergo rapid fusion . At a 1:9000 SNARE:lipid ratio , the fusion of VML proteoliposomes ( filled triangles ) slows to the rate seen for PC/PS proteoliposomes at a 1:1000 ratio ( Figure 1B , right , open circles ) . Fusion is hardly detectable for PC/PS proteoliposomes with SNAREs at a 1:9000 molar ratio to lipids ( open triangles ) . 10 . 7554/eLife . 01879 . 003Figure 1 . Lipid composition and SNARE concentration regulate the rate of proteoliposome membrane fusion . ( A ) Membrane fusion was assayed as protected ( from external non-fluorescent streptavidin ) lumenal compartment mixing . This was measured as the FRET between biotin-phycoerythrin and Cy5-streptavidin , which had been entrapped within separate proteoliposome populations . Paired sets of proteoliposomes were prepared with either the complete vacuolar mixed lipids ( VML ) or with 70% PC/30% PS . Proteoliposomes bore Ypt7p and the 4 SNAREs , each at a 1:1000 , 1:3000 , or 1:9000 molar ratio to lipid phosphate , as described in the ‘Materials and methods’ . For each pair , half the proteoliposomes had 0 . 3% of its lipid as Marina Blue-PE and bore lumenal Cy5-streptavidin , while the complementary proteoliposomes had 1 . 5% NBD-PE and bore lumenal biotinylated-phycoerythrin . ( B ) Fusion assays were performed with Sec17p , Sec18p , HOPS , and ATP in the presence of excess nonfluorescent streptavidin , as described in the ‘Materials and methods’ . Error bars here and in subsequent figures are the standard deviations from three assays . DOI: http://dx . doi . org/10 . 7554/eLife . 01879 . 003 To place these findings in a context of the physiological concentrations of SNAREs , vacuoles were purified ( Haas , 1995 ) and analyzed for lipid phosphorus and for their bound Ypt7p , HOPS , Sec17p , Sec18p , and each of the 4 SNAREs . These proteins were from 5- to 100-fold less abundant on vacuoles as compared to proteoliposomes which were prepared with a 1:1000 SNARE:lipid molar ratio ( Table 1 ) and in which approximately half the SNAREs were shown by protease-accessibility assay ( Figure 2C ) to be exposed on the proteoliposome exterior . Thus the lower end of SNARE concentrations employed in our reconstituted reactions , while still high compared to the organelle , are closer to physiological . Only very high SNARE concentrations can partially bypass the requirement for greater lipid complexity for fusion . 10 . 7554/eLife . 01879 . 004Table 1 . Protein abundance , relative to lipids , in vacuoles or reconstituted proteoliposomes ( RPL ) fusion reactionsDOI: http://dx . doi . org/10 . 7554/eLife . 01879 . 004ProteinMolar ratio of lipid:protein in RPL reactionsMolar ratio of lipid:protein on vacuolesRatio ( RPLs/vacuoles ) of molar protein:lipid ratios in std . reactions†BJ3505DKY6218Vam7p2 × 10330 × 1046 . 5 × 1047 × 101Vam3p2 × 10311 × 10422 × 1047 × 101Vti1p2 × 10310 × 10413 × 1045 × 101Nyv1p2 × 1034 . 3 × 1048 . 1 × 1043 × 101Ypt7p4 × 1031 . 9 × 1041 . 8 × 1040 . 5 × 101Sec17p7 × 10341 × 10413 × 1043 × 101Sec18p1 × 10310 × 10413 × 10410 × 101Vps33p6 × 10317 × 10431 × 1043 × 101for SNAREs and Ypt7p , calculated , based on a 1:1000 lipid:protein ratio during reconstitution and an assumption of 50% outwardly-oriented SNAREs on proteoliposomes; for others , based on amounts of added proteins , and 0 . 74 mM lipids in standard proteoliposome reactions ( see ‘Materials and methods’ ) . †based on measured ( see ‘Materials and methods’ ) values of 2 . 17 nmol lipid per µg total vacuole protein for BJ3505 vacuoles and 1 . 00 nmol lipid per μg total vacuole protein for DKY6218 vacuoles , and standard vacuole reactions containing 3 µg protein of each vacuole in 30 µl . 10 . 7554/eLife . 01879 . 005Figure 2 . A role for neutral , non-bilayer lipids in fusion . ( A ) Fusion of proteoliposomes bearing Ypt7p and 4-SNAREs ( 1:5000 molar ratio to lipid phosphate ) , prepared with either the complete vacuole lipid mix ( squares ) or missing PE ( diamonds ) , PE and DAG ( triangles ) , or PE , DAG , and ERG ( circles ) . Fusion was assayed as the FRET between lumenal fluorescent proteins in the continuous presence of an excess of external nonfluorescent streptavidin . ( B ) Proteoliposomes were analyzed for their protein composition by SDS-PAGE and Coomassie blue staining . Lipid composition: Lane 1 , complete vacuolar lipid mix; lane 2 , PE omitted; lane 3 , PE and DAG omitted; lane 4 , PE , DAG , and ERG omitted . In each case , the percentage of PC was increased to account for the omitted lipid ( s ) . ( C ) Similar protease-accessibility of SNAREs and Rab across proteoliposomes preparations . Proteoliposomes ( 1 . 2 mM lipid ) were incubated in 15 µl RB150 with 60 mM HEPES/NaOH pH 8 . 0 for 10 min at 27°C with either no addition of protease , with 60 µg/ml of proteinase K which had been preincubated for 10 min with 1 mM PMSF prior to proteoliposome addition ( indicated by asterisk ) , with fully-active proteinase K which had not been preincubated with PMSF , or with fully active proteinase K and 1% ( wt/vol ) β-octylglucoside . After this incubation , PMSF was added to samples which had fully active proteinase K and the incubation continued for an additional 10 min . All samples were then mixed with SDS sample buffer , heated to 95°C for 5 min and subjected to SDS-PAGE . Gels were stained with Coomassie blue , and bands corresponding to Vam3p , Nyv1p , and Ypt7p quantified by scanning with a Microtek Bio-5000 scanner ( Microtek Lab , Inc . , Santa Fe Springs , CA ) and UN-SCAN-IT gel 5 . 3 software ( Silk Scientific , Orem , UT ) . For each of these three proteins , the intensity of the band from samples which never saw proteinase K was set to 100% . Dark bars correspond to VML proteoliposomes , light bars to proteoliposomes prepared without PE , DAG , and Erg . Shown is the average of two experiments +/− standard deviations . ( D ) The size distribution of various proteoliposome preparations was analyzed by dynamic light scattering with a Zetasizer nano ZS ( Malvern Instruments Inc . , Westborough , MA ) through non-invasive back-scatter at 173° . For each liposome preparation , at least four samples ( 400 µl at a lipid concentration of 20 µM ) were measured in low volume disposable sizing cuvettes at 25°C . Shown is the average diameter ( +/− standard deviation ) of independent proteoliposome preparations composed of the complete vacuolar lipid mix ( dark bars ) or without PE , ERG , DAG ( light bars ) and bearing either Nyv1 ( 1R; n = 3 ) , all four SNAREs ( 4SNARE; n = 2 ) , Vam3 and Vti1 ( 2Q; n = 1 ) , or Nyv1 , Vam3 , and Vti1 ( RQab; n = 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01879 . 005 To evaluate the role of lipids in vacuolar fusion , we started from the complete VML composition ( Mima et al . , 2008 ) with SNAREs at a 1:5000 molar ratio to lipids and sequentially removed one lipid at a time , substituting additional PC in its place . The sequential removal of PE , diacylglycerol , and ergosterol reduced the rate and extent of fusion , until fusion could no longer be detected when all three of these lipids were omitted ( Figure 2A ) . These proteoliposomes had comparable lumenal entrapment of fluorescent proteins ( data not shown ) , comparable protein composition ( Figure 2B ) , and comparable orientation of SNARE and Rab proteins , as judged by protease accessibility assays ( Figure 2C ) . Proteoliposomes vary in size from preparation to preparation , and according to their SNARE composition . Those prepared without the three non-bilayer lipids have approximately 20% smaller diameter , though this is still largely within the range seen for VML RPLs of varying SNARE composition ( Figure 2D ) . Thus , the lack of fusion signal was presumably not due to an absence of the Rab , SNAREs or lumenal probe , or to altered proteoliposome topology . Vacuoles with 3- to 9-fold elevated SNARE levels undergo lysis as well as fusion ( Starai et al . , 2007 ) , and vacuolar proteoliposomes also exhibit both behaviors ( Zucchi and Zick , 2011 ) . Lysis is inferred from the extra FRET obtained from the initially-lumenal probes when the external quencher , nonfluorescent streptavidin , is omitted ( Zucchi and Zick , 2011 ) . Proteoliposomes with the full VML lipid composition undergo fusion and lysis ( Figure 3 , squares ) , whereas RPLs lacking PE , DAG , and ERG exhibit neither fusion nor lysis ( open and filled circles ) . Thus the lack of fusion signal when PE , ERG , and DAG are absent is not due to a fusion pathway diversion into lysis . 10 . 7554/eLife . 01879 . 006Figure 3 . Omission of PE , DAG , and ERG blocks HOPS , Sec17p , and Sec18p triggered lysis as well as fusion . Proteoliposomes of complete vacuolar lipid mix or lacking PE , DAG , and ERG and with Ypt7p and the 4 vacuolar SNAREs ( 1:5000 molar ratio to lipid phosphate ) were incubated either with a large molar excess of non-fluorescent streptavidin , restricting FRET development to sealed fusion events , or without external streptavidin , yielding FRET from both fusion and from lysis ( Zucchi and Zick , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01879 . 006 The vacuole fusion pathway entails early ATP-dependent reactions that occur on separate vacuoles , termed priming , followed by tethering , which can be mediated by HOPS and Ypt7p alone ( Hickey and Wickner , 2010 ) . Tethering allows a striking enrichment of fusion proteins and lipids in a microdomain , followed by trans-SNARE complex assembly . This is followed by rearrangements of the lipid bilayers resulting in membrane fusion and the attendant mixing of lumenal compartments . To determine the stage ( s ) that require the nonbilayer-prone-lipids , paired sets of proteoliposomes were prepared with VML lipids , Ypt7p , and either the R-SNARE Nyv1p or the Q-SNAREs Vam3p ( Qa ) and Vti1p ( Qb ) . These proteoliposomes readily fuse when incubated with HOPS and Vam7p ( Qc-SNARE ) ; Sec17p and Sec18p stimulate but are not required ( Zick and Wickner , 2013 ) and were not present in these assays . A second set of proteoliposomes was prepared in parallel , but lacking PE , DAG , and ERG . There was no fusion in the absence of the three small headgroup lipids ( Figure 4A ) . After a 10 min incubation under fusion conditions , aliquots were solubilized in a RIPA buffer ( 1% Triton X-100 , 1% sodium cholate , and 0 . 1% SDS ) with affinity-purified antibody to Vam3p , then mixed with magnetic beads bearing protein A . After washing , bound proteins were eluted with hot SDS and analyzed for Nyv1p by immunoblot ( Figure 4B ) . Comparable amounts of Nyv1p had bound to Vam3p in proteoliposomes with complete vacuolar lipid mix ( Figure 4B , lane 2 ) , where rapid fusion occurred , as in proteoliposomes lacking PE , DAG , and ERG ( lane 5 ) where there was no detectable fusion ( Figure 4A ) . Nyv1p did not associate with Vam3p in trans when Vam7p was omitted ( Figure 4B , lanes 1 , 4 ) or when Vam7p was only added immediately after the RIPA buffer ( lanes 3 , 6 ) . 10 . 7554/eLife . 01879 . 007Figure 4 . Small-headgroup , nonbilayer lipids are needed for trans-SNARE docked membranes to proceed to fusion . Reconstituted proteoliposomes with either the R-SNARE or the Vam3p and Vti1p Q-SNAREs , prepared at a 1:5000 molar ratio of SNARE to lipid and either having the complete vacuolar lipid mix or without PE , ERG , or DAG were incubated in fusion reactions . Vam7p ( 0 . 5 µM ) was added where indicated , either during the fusion reaction ( lanes 2 and 5 ) or after the reaction was terminated by detergent addition ( indicated by an asterisk , lanes 3 and 6 ) . Each reaction was ( A ) assayed for lumenal content mixing and ( B ) mixed after 10 min with a 10-fold volume of a modified RIPA buffer ( 20 mM HEPES/NaOH , pH 7 . 4 , 0 . 15M NaCl , 0 . 2% bovine serum albumin ( defatted ) , 1% Triton X-100 , 1% sodium cholate , 0 . 1% sodium dodecyl sulfate , 1 mM EDTA ) with 40 µg/ml affinity-purified antibody to Vam3p and 1 µM recombinant soluble domain of Snc2p to suppress SNARE complex assembly in detergent . After addition of 10 µl of RIPA buffer-washed suspension of magnetic beads with bound protein A ( Thermo Scientific ) , samples were mixed for 1 hr at room temperature . Beads were collected by placing the tubes for 2 min onto a magnetic rack , and the unbound proteins removed . Beads were thrice washed with 1 ml of modified RIPA buffer , then proteins were eluted with SDS sample buffer at 95°C and analyzed by SDS-PAGE and immunoblot with antibodies to Nyv1p . Reactions were performed without further SNARE addition , with 0 . 5 µM Vam7p from the start of the incubation , or with the Vam7p added one minute after solubilization by RIPA buffer . The same preparations and solutions were premixed , then used in parallel for the assays of fusion and trans-assembly of SNAREs shown here . The immunoblot of one of the three trans-SNARE assays is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01879 . 007 PE , DAG , and ERG are required for fusion at SNARE:lipid molar ratios of 1:5000 , whether the proteoliposomes bear all 4 SNAREs ( Figure 2A ) or 2Q-RPLs and 1R-RPLs are incubated with added Vam7p ( Figure 4A ) . It remained possible that some other parameter of reconstitution which we could not measure , such as the trans-membrane disposition of each lipid species , might regulate fusion and be influenced by the SNARE levels . To address this possibility , we prepared RPLs bearing both the R-SNARE and QaQb ( Vam3p , Vti1p ) SNAREs , with either the complete mixture of vacuolar lipids or lacking PE , DAG , and ERG . In either case , fusion of these RQaQb RPLs was not seen without added Vam7p ( Figure 5 , diamonds ) . With ample added Vam7p ( 5 µM ) to allow maximal formation of trans-SNARE complexes , the fusion only showed a several-fold stimulation by the presence of PE , DAG , and ERG ( circles ) . However , with 5 nM added Vam7p , the small headgroup neutral lipids were still required ( squares ) . Thus the very same proteoliposomes could be used to demonstrate a strict requirement for small-headgroup lipids when the trans-SNARE complex levels were limited by the 5 nM Vam7p , while with ample Vam7p ( 5 µM ) and hence high trans-SNARE levels , this requirement was partially bypassed . Bypass thus directly reflects the formation of high levels of trans-SNARE pairs . 10 . 7554/eLife . 01879 . 008Figure 5 . The requirement for PE , ERG , and DAG is governed by the level of trans-SNARE complex . Fusion assay pairs of proteoliposomes were prepared with Ypt7p , with VML lipids or lacking PE , ERG , and DAG , as indicated , and with Nyv1p , Vam3p , and Vti1p ( the R- and Qa- and Qb-SNAREs , respectively ) at a 1:1000 molar ratio to total lipid . Fusion assays were initiated by the addition of HOPS , Sec17p , and Sec18p , as described in the ‘Materials and methods’ , as well as the indicated ( final ) concentration of Vam7p . DOI: http://dx . doi . org/10 . 7554/eLife . 01879 . 008 The vacuole fusion pathway has well-defined , sequential steps . After ATP-dependent priming , vacuoles undergo tethering . The Rab Ypt7p directly binds the oligomeric tethering complex HOPS ( Seals et al . , 2000 ) . This interaction is necessary ( Stroupe et al . , 2006 ) for tethering of vacuolar membranes , and is sufficient for tethering of proteoliposomes of a simple lipid composition such as PC/PS ( Hickey and Wickner , 2010 ) . It was therefore not surprising that comparable levels of trans-SNARE complex formed in the presence or absence of small headgroup lipids ( Figure 4B ) , but the need for these particular lipids for the ensuing fusion was unexpected . It has been suggested that fusion occurs immediately and inexorably upon trans-SNARE complex formation . However , several studies have suggested that fusion is more complex . Brunger and colleagues . ( Kyoung et al . , 2011; Diao et al . , 2012 ) have shown long kinetic separation between formation of some trans-SNARE complexes and complete membrane fusion for certain spontaneous fusion events ( for example , in the absence of calcium ) and in the presence of both synaptotagmin and complexin; upon calcium triggering , a burst of fast fusion events were then observed . During yeast vacuole fusion , lumenal compartment mixing occurs many minutes after docking is complete ( Mayer et al . , 1996 ) . Furthermore , when a single SNARE ( Vam7p ) is withheld from tethered vacuole clusters , the addition of this SNARE triggers rapid and synchronized completion of docking , as assayed by the acquisition of resistance to added antibody to Ypt7p or to Vam3p , but lumenal compartment mixing only occurs far more slowly ( Merz and Wickner , 2004 ) . Lipid ligands have been reported to block fusion after docking ( Collins and Wickner , 2007 ) . We now exploit the fully reconstituted vacuole fusion reaction to show that small-headgroup lipids that are prone to non-bilayer structures play a critical role in the progression of docked , trans-SNARE paired proteoliposomes to fusion . The essence of membrane fusion is the rearrangement of lipids from closely apposed bilayers . Models of this rearrangement , such as through hemifusion , posit intermediate states with locally crowded lipid headgroups . This headgroup crowding may constitute a substantial portion of the activation energy of the fusion process ( Chernomordik et al . , 1998; Gaudin , 2000 ) . Lipids such as PE , ergosterol , and diacylglycerol would fit into these otherwise crowded regions far better than a cylindrical lipid such as phosphatidylcholine . Upon vacuole docking , the proteins and lipids that are required for fusion become highly enriched at a ring-shaped microdomain surrounding the apposed regions of the docked membranes ( Wang et al . , 2002 , 2003; Fratti et al . , 2004 ) . Although the surface concentrations of proteins required for proteoliposome fusion are higher than those on the vacuole ( Table 1 ) , they may be comparable to those in this ring-shaped , fusion-competent vacuole microdomain . Additional studies will be required to precisely determine the physiological concentrations of these proteins and lipids in this microdomain . Previous studies of fusion , whether of vacuoles or of other model systems , have indicated a role for the three vacuolar , non-bilayer-prone lipids . Though the effects of disrupting PE biosynthesis in yeast are likely too pleiotropic to allow examination of its effects on vacuole fusion in vivo , PE is required for the fusion of mitotic Golgi membranes ( Pécheur et al . , 2002 ) . In model studies with recombinant neuronal SNAREs , PE increased the probability of docking but reduced fusion ( Domanska et al . , 2010 ) . These studies differed from those reported here in several crucial aspects: ( i ) The use of planar lipid bilayers vs highly curved proteoliposomes , ( ii ) neuronal vs vacuolar SNAREs , ( iii ) the absence of a Rab and Rab-effector docking system or of an SM protein in the neuronal reconstitution , and ( iv ) the composition and complexity of lipids in these reconstitutions . Yeast mutants defective in sterol biosynthesis have multiple small vacuoles , indicative of deficient organelle fusion ( Kato and Wickner , 2001; Seeley et al . , 2002 ) . The fusion of vacuoles purified from wild-type yeast is blocked by extraction of ergosterol by β-methylcyclodestrin , and fusion can be restored by incubation with cholesterol-loaded β-methylcyclodestrin ( Kato and Wickner , 2001 ) . Cholesterol has also been shown to be required for Semliki Forest virus fusion ( White and Helenius , 1980 ) . Diacylglycerol is also directly implicated in vacuole fusion . Whereas a wild-type yeast strain has one or a few vacuoles , deletion of PLC1 , encoding a phospholipase C which generates diacylglycerol , causes striking vacuole fragmentation ( Jun et al . , 2004 ) , a hallmark of defective fusion ( Wada et al . , 1992 ) . The cell-free fusion of purified vacuoles is blocked by recombinant C1b domain , a DAG ligand , and is stimulated by added Plc1p ( Jun et al . , 2004 ) . Both ergosterol and diacylglycerol become enriched at the fusion microdomain of docked vacuoles , and this enrichment is sensitive to fusion-protein ligands such as Sec17p and an Fab fragment of affinity-purified antibody to Vam3p ( Fratti et al . , 2004 ) . Our current studies with proteoliposomes bearing pure vacuolar proteins and lipids is thus well-grounded in in vivo genetics and morphology as well as cell-free studies of the fusion of the purified organelle . It is revealing that high concentrations of SNAREs can at least partially bypass the need for non-bilayer-prone lipids . It is currently unclear whether the sole function of trans-SNARE pairs is a close and stable tethering of apposed lipid bilayers , or whether they exert a deforming force on each bilayer to provide the energy for lipid spatial rearrangement into non-bilayer intermediates . Prenyl-anchored SNAREs can support the fusion of vacuoles ( Jun et al . , 2007 ) and of chemically-defined proteoliposomes ( Xu et al . , 2011 ) , and this has recently been extended to synaptic fusion in mice ( Zhou et al . , 2013 ) . These findings suggest that fusion does not depend on force exerted through the formation of continuous α-helices , extending from the SNARE bundle across two trans-membrane anchor domains ( McNew et al . , 1999; Li et al . , 2007 ) . How might high levels of trans-SNARE pairs be able to bypass the need for the three small headgroup lipids ? A single trans-SNARE pair may bring flat or spherical membrane bilayers into close apposition without distorting their bilayer structure , while multiple trans-SNARE pairs may draw membranes together over a wider region , creating a bend in the bilayer at the edge of the zone of apposition , as seen for docked vacuoles ( Wang et al . , 2002; Fratti et al . , 2004 ) . Such a bend may substitute for nonbilayer lipids in initiating the lipid rearrangements of fusion . Our current working model of fusion is that even single trans-SNARE pairs bring membranes into close apposition . The lipid rearrangements for fusion can be driven by the membrane bend induced by multiple trans-SNARE pairs , or can occur by the accretion of fusogenic , noncylindrical lipids when the SNARE concentrations are at low , physiological levels . The Vam7p SNARE N-domain has a Phox-homology motif , and has been shown to associate with the trans-apposed bilayer ( Xu and Wickner , 2010 ) by its affinity for PI ( 3 ) P and acidic lipids ( Lee et al . , 2006; Karunakaran and Wickner , 2013 ) and to bear apolar residues which may insert into the bilayer ( Lee et al . , 2006 ) . Similarly , neuronal SNARE-associated synaptotagmin is triggered by Ca2+ binding to insert into the lipid bilayer ( Chapman and Davis , 1998; Zhang et al . , 1998 ) and may function to bridge bilayers ( Araç et al . , 2006; Xue et al . , 2008 ) . These may also cause bilayer distortions that facilitate fusion; there has not been a direct test of whether nonbilayer-forming lipids contribute to the extraordinary rates of calcium-triggered neuronal fusion . Further tests of this model may rely on the development of assays to measure the physical strain within bilayers , and of reconstituted reactions fusing large proteoliposomes where lipid- and protein-enriched microdomains can be visualized as readily as with vacuoles . Lipids were obtained from Avanti Polar Lipids , except ergosterol was from Sigma–Aldrich ( St . Louis , MO ) , PI ( 3 ) P was from Echelon Biosciences ( Salt Lake City , UT ) , and the fluorescent lipids were from Life Technologies ( Carlsbad , CA ) . Sec18p ( Haas and Wickner , 1996 ) , Sec17p ( Schwartz and Merz , 2009 ) , Ypt7p ( Zick and Wickner , 2013 ) , HOPS ( Zick and Wickner , 2013 ) , and vacuolar SNARE proteins ( Mima et al . , 2008; Schwartz and Merz , 2009; Zucchi and Zick , 2011 ) were purified as described . Vti1p and Nyv1p were exchanged into octylglucoside buffer as described ( Zucchi and Zick , 2011 ) . Vacuolar lipids were extracted by a modification of the Bligh-Dyer method ( Bligh and Dyer , 1959 ) . Chloroform ( 100 µl ) and methanol supplemented with 0 . 1 M HCl ( 200 µl ) were added to 37 µg vacuoles , as measured by protein content ( Haas , 1995 ) , in 80 µl RB150+Mg ( 20 mM HEPES , pH 7 . 4 , 150 mM NaCl , 10% glycerol , 1 mM MgCl2 ) . This single-phase mixture was vortexed thoroughly and incubated at room temperature for 1 hr . RB150+Mg and chloroform ( 100 µl ea . ) were then added . The sample was vortexed thoroughly and centrifuged at 14000×g rpm in an Eppendorf ( Hamburg , Germany ) 5415C microcentrifuge at room temperature for 30 s . The organic layer was transferred to a 13 × 100 mm round-bottom glass tube ( 99445-13; Corning Inc . , Corning , NY ) . Chloroform ( 200 µl ) was added to the remaining aqueous layer . This sample was vortexed and centrifuged as above , and the organic layer was removed and added to the organic layer from the first extraction . RB150+Mg ( 360 µl ) and methanol-HCl ( 400 ml ) were added to the combined organic layers . This mixture was vortexed , centrifuged in a Sorvall SpeedVac SC100 ( Thermo Fisher Scientific , Waltham , MA ) at atmospheric pressure and room temperature for 30 s , and the aqueous layer was removed and discarded . Vacuole lipid levels were measured using a lipid phosphorus assay . Ammonium molybdate ( 10 µl of a 2% wt/vol solution ) was added to extracted vacuolar lipids , and to standards ( 0 , 5 , 10 25 , 50 , 75 , 100 , and 125 µl of a 1 mM NaH2PO4 solution ) . Samples were incubated at 100°C until dry ( approximately 1 hr ) . Perchloric acid ( 300 µl of a 70% vol/vol solution ) was added . Samples were incubated at 180–200°C for 1 hr with occasional vortexing , then cooled to room temperature . Ammonium molybdate ( 1 . 5 ml of a 0 . 4% wt/vol solution ) and ascorbic acid ( 225 µl of a 10% wt/vol solution ) were added . Samples were vortexed thoroughly and incubated at 100°C for 10 min , then cooled to room temperature . Absorbance at 820 nm was measured and phospholipid concentration was estimated by comparison of vacuole lipid samples to the phosphate standards . To obtain lipid concentrations for Table 1 , measured phospholipid concentrations were multiplied by 1 . 18 , to correct for a reported ergosterol:phospholipid molar ratio of 0 . 18 in vacuolar lipids ( Zinser et al . , 1991 ) . For estimation of vacuolar protein levels , 6 . 5 nmol ea . BJ3505 and DKY6218 vacuoles ( Haas , 1995 ) , here as measured by lipid content ( see previous paragraph ) rather than protein content , were analyzed by SDS-PAGE and immunoblotting for Vam7p , Vam3p , Vti1p , Nyv1p , Ypt7p , Sec17p , Sec18p , and Vps33p . Protein levels were estimated by comparison of band intensities ( measured using a ChemiDoc-It system with LabWorks version 4 . 5 . 00 . 0 software , UVP , Upland , CA ) from vacuolar samples to band intensities from standards ( 3 . 25 , 1 . 3 , 0 . 65 , 0 . 26 , 0 . 13 , 0 . 052 , and 0 . 026 pmol ea . ) of purified recombinant Vam7p , Vti1p , Nyv1p ( Mima et al . , 2008 ) , his6-tagged Vam3 cytosolic domain ( Nichols et al . , 1997 ) , Ypt7p ( Hickey et al . , 2009 ) , his6-tagged Sec17p and his6-tagged Sec18p ( Haas and Wickner , 1996 ) , and HOPS complex ( Stroupe et al . , 2009 ) . Reconstituted proteoliposomes were prepared as described ( Zick and Wickner , 2013 ) , with modifications . Chloroform solutions of lipids ( vacuolar mixed lipids; VML ) were mixed in a glass vial: 49 . 6 or 51 mol % diC18:2 PC , 15% diC18:2 PE , 1% diacylglycerol , 8% ergosterol , 2% diC18:2 PA , 18% soy PI , 4 . 4% diC18:2 PS , 1% diC16 PI ( 3 ) P and either 0 . 23% Marina Blue-PE or 1 . 5% NBD-PE ( Life Technologies ) . When small headgroup lipids were omitted , the amount of PC was adjusted to bring the sum to 100% . β-octylglucoside was added to 160 mM from a 0 . 5 M solution in methanol and samples were dried under a stream of nitrogen , then in vacuo . Samples were dissolved in a fivefold concentrate of RB150+Mg ( 0 . 1 M HEPES/NaOH , pH 7 . 4 , 0 . 75 M NaCl , 50% glycerol , 5 mM MgCl2 ) by several cycles of vortexing for 10 s , rocker mixing for 30 min , and bath sonication for 5 min , yielding mixed micellar solutions with 4 mM lipids and 50 mM detergent . Lipid micellar solutions ( 200 µl ) were mixed with a mixed micellar solution of purified Ypt7p and the indicated SNAREs ( 550 µl ) and 250 µl of either Cy5-derivatized streptavidin ( from KPL , Gaithersburg , MD; 8 µM final ) or biotinylated phycoerythrin ( Life Technologies; 4 µM final ) . Each ml of solution was added to a rinsed and knotted 6 cm segment of SpectraPor dialysis membrane , 25 kDa cutoff , 7 . 5 mm diameter ( Spectrum Labs , Rancho Dominguez , CA ) which was then knotted and dialyzed at 4°C in 250 ml of RB150+Mg ( 20 mM HEPES , pH 7 . 4 , 150 mM NaCl , 10% glycerol , 1 mM MgCl2 [Mima et al . , 2008; Zucchi and Zick , 2011] ) with 1 g of BioBeads SM-2 ( Biorad , Hercules , CA ) for at least 20 hr with continuous stirring . The isolation of proteoliposomes by flotation was as described ( Zick and Wickner , 2013 ) . After total phosphate was assayed , samples were brought to 2 mM lipid with RB150+Mg and small aliquots were frozen in liquid nitrogen and stored at −80°C . Assays of proteoliposome fusion and lysis were as described ( Zick and Wickner , 2013 ) . Adjacent wells of 384-well plates received either the mixed proteoliposomes in RB150+Mg with streptavidin or a mixture of the remaining assay components , which were added to the wells with proteoliposomes after a 10 min preincubation at 27°C . For lysis assays , duplicate wells either received streptavidin or RB150 in its place; the difference between the readings from these wells is a measure of lysis . The final assay component concentrations are: 19 . 5 mM HEPES/NaOH , pH 7 . 4 , 142 mM NaCl , 11 mM KCl , 1 . 1 mM imidazole , 9 . 8% glycerol , 3 . 3 mM sorbitol , 1 . 44 mM MgCl2 , 0 . 13 mM 2-mercaptoethanol , 1 . 1 mM potassium phosphate , 0 . 17 mM glutathione , 10 . 7 µM streptavidin , 1 . 36 mM Na2ATP , 1 . 84% bovine serum albumin ( defatted ) , 0 . 000066% Triton X-100 , 0 . 37 mM lipid from each of the 2 proteoliposome populations , 108 nM Sec17p , 0 . 55 µM Sec18p , and 0 . 12 µM HOPS .	All cells are enclosed with a membrane that is made of phospholipid molecules , and many of the structures found inside cells—such as the vacuoles in plant and fungal cells—are also enclosed with a phospholipid membrane . To form a membrane , the phospholipid molecules—which have a phosphate head and two fatty acid tails—arrange themselves in two layers , with the fatty acid tails pointing into the membrane , and the phosphate heads pointing outwards . This structure is known as a phospholipid bilayer . Vacuoles are filled with water that contains various proteins and molecules in solution , and adjust their volume to keep the concentrations of substances in the cell in balance . To do this , the vacuoles fuse with each other . This fusion process requires dramatic spatial rearrangements of the phospholipid molecules . The SNARE family of proteins plays a key role in membrane fusion . As the two membranes come together , SNARE proteins located on each membrane form a complex known as a trans-SNARE complex . This docks the vacuole in place beside another vacuole while the phospholipid molecules in the two membranes rearrange . However , much less is known about the phospholipid molecules that are involved in the fusion process . Now , Zick et al . have shown that three types of phospholipid molecules must be present for membrane fusion to be completed . These have in common that their phosphate ‘headgroups’ are small and they do not tend to form bilayers . The vacuoles can dock beside each other if these small headgroup phospholipid molecules are not present , but the bilayer lipids in the vacuole membranes cannot rearrange themselves in the absence of these particular lipids . The importance of these nonbilayer lipid molecules had not previously been established , as the majority of experiments investigating membrane fusion used concentrations of SNARE proteins that were much higher than those found physiologically . At such high concentrations , fusion can go ahead without the nonbilayer lipid molecules being present .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2014	Membranes linked by trans-SNARE complexes require lipids prone to non-bilayer structure for progression to fusion
Conjugative transfer of the integrative and conjugative element ICEclc in Pseudomonas requires development of a transfer competence state in stationary phase , which arises only in 3–5% of individual cells . The mechanisms controlling this bistable switch between non-active and transfer competent cells have long remained enigmatic . Using a variety of genetic tools and epistasis experiments in P . putida , we uncovered an ‘upstream’ cascade of three consecutive transcription factor-nodes , which controls transfer competence initiation . One of the uncovered transcription factors ( named BisR ) is representative for a new regulator family . Initiation activates a feedback loop , controlled by a second hitherto unrecognized heteromeric transcription factor named BisDC . Stochastic modelling and experimental data demonstrated the feedback loop to act as a scalable converter of unimodal ( population-wide or ‘analog’ ) input to bistable ( subpopulation-specific or ‘digital’ ) output . The feedback loop further enables prolonged production of BisDC , which ensures expression of the ‘downstream’ functions mediating ICE transfer competence in activated cells . Phylogenetic analyses showed that the ICEclc regulatory constellation with BisR and BisDC is widespread among Gamma- and Beta-proteobacteria , including various pathogenic strains , highlighting its evolutionary conservation and prime importance to control the behaviour of this wide family of conjugative elements . Biological bistability refers to the existence of two mutually exclusive stable states within a population of genetically identical individuals , leading to two distinct phenotypes or developmental programs ( Shu et al . , 2011 ) . The basis for bistability lies in a stochastic regulatory decision resulting in cells following one of two possible specific genetic programs that determine their phenotypic differentiation ( Norman et al . , 2015 ) . Bistability has been considered as a bet-hedging strategy leading to an increased fitness of the genotype by ensuring survival of one of both phenotypes depending on environmental conditions ( Veening et al . , 2008 ) . A number of bistable differentiation programs is well known in microbiology , notably competence formation and sporulation in Bacillus subtilis ( Xi et al . , 2013; Schultz et al . , 2007 ) , colicin production and persistence in Escherichia coli ( Lewis , 2007 ) , virulence development of Acinetobacter baumannii ( Chin et al . , 2018 ) , or the lysogenic/lytic switch of phage lambda ( Sepúlveda et al . , 2016; Arkin et al . , 1998 ) . Bistability may also be pervasive among many bacterial DNA conjugative systems , leading to the formation of specific conjugating donor cells at low frequency in the population ( Delavat et al . , 2017 ) . The best described case of this is the dual lifestyle of the Pseudomonas integrative and conjugative element ( ICE ) ICEclc ( Figure 1A; Minoia et al . , 2008 ) . In the majority of cells ICEclc is maintained in the integrated state , but a small proportion of cells ( 3–5% ) in stationary phase activates the ICE transfer competence program ( Minoia et al . , 2008; Delavat et al . , 2016 ) . Upon resuming growth , transfer competent ( tc ) donor cells excise and replicate the ICE ( Delavat et al . , 2019 ) , which can conjugate to a recipient cell , where the ICE can integrate ( Delavat et al . , 2016 ) . ICEclc transfer competence comprises a differentiated stable state , because initiated tc cells do not transform back to the ICE-quiescent state . Although tc cells divide a few times , their division is compromised by the ICE and eventually arrests completely ( Takano et al . , 2019; Reinhard et al . , 2013 ) . ICEs have attracted wide general interest because of the large variety of adaptive functions they can confer to their host , including resistance to multiple antibiotics ( Waldor et al . , 1996; Johnson and Grossman , 2015; Burrus et al . , 2002 ) , or metabolism of xenobiotic compounds , such as encoded by ICEclc ( Miyazaki et al . , 2015; Zamarro et al . , 2016 ) . ICEclc stands model for a ubiquitous family of genomic islands found by bacterial genome sequencing , occurring in important opportunistic pathogens such as Pseudomonas aeruginosa , Bordetella bronchiseptica , Xylella fastidiosa or Xanthomonas campestris ( Miyazaki et al . , 2015 ) . The ICEclc family of elements is characterized by a consistent ‘core’ region of some 50 kb ( Figure 1A ) , predicted to encode conjugative functions , and a highly diverse set of variable genes with adaptive benefit ( Miyazaki et al . , 2015 ) . Strong core similarities between ICEclc and the PAGI-2 family of pathogenicity islands in P . aeruginosa clinical isolates have been noted previously ( Miyazaki , 2011a; Klockgether et al . , 2007 ) . Although the existence and the fitness consequences of the ICEclc bistable transfer competence pathway have been studied in quite some detail , the regulatory basis for its activation has remained largely elusive ( Delavat et al . , 2017 ) . In terms of its genetic makeup , ICEclc seems very distinct from the well-known SXT/R391 family of ICEs ( Wozniak and Waldor , 2010 ) and from ICESt1/ICESt3 of Streptococcus thermophilus ( Carraro and Burrus , 2014 ) . These carry analogous genetic regulatory circuitry to the lambda prophage , which is characterized by a typical double-negative feedback control ( Poulin-Laprade and Burrus , 2015; Bellanger et al . , 2007 ) . Transcriptomic studies indicated that the core region of ICEclc ( Figure 1A ) is higher expressed in stationary than exponential phase cultures grown on 3-chlorobenzoate ( 3-CBA ) , and organized in at least half a dozen transcriptional units ( Gaillard et al . , 2010 ) . A group of three consecutive regulatory genes precludes ICEclc activation in exponentially growing cells , with the first gene ( mfsR ) constituting a negative autoregulatory feedback ( Figure 1B; Pradervand et al . , 2014 ) . Overexpression of the most distal of the three genes ( tciR ) , leads to a dramatic increase of the proportion of cells activating the ICEclc transfer competence program ( Pradervand et al . , 2014 ) . Despite this initial discovery , however , the nature of the regulatory network architecture leading to bistability and controlling further expression of the ICEclc genes in tc cells has remained enigmatic . The primary goal of this work was to dissect the regulatory factors and nodes underlying the activation of ICEclc transfer competence . Secondly , given that transfer competence only arises in a small proportion of cells in a population , we aimed to understand how the regulatory architecture yields and maintains ICE bistability . We essentially followed two experimental strategies and phenotypic readouts . First , known and suspected regulatory elements were seamlessly deleted from ICEclc in P . putida and complemented with inducible plasmid-cloned copies to study their epistasis in transfer of the ICE . Secondly , individual and combinations of suspected regulatory elements were expressed in a P . putida host without ICE , to study their capacity to activate the ICEclc promoters Pint and PinR , which normally only express in wild-type tc cells ( Figure 1B; Minoia et al . , 2008 ) . As readout for their activation we quantified fluorescent reporter expression from single copy chromosomally integrated transcriptional fusions , as well as the proportion of cells expressing the reporters using subpopulation statistics as previously described ( Reinhard and van der Meer , 2013 ) . On the basis of the discovered key regulators and nodes , we then developed a conceptual mathematical model to show by stochastic simulations how bistability is generated and maintained . This suggested that the ICEclc transfer competence regulatory network essentially converts a unimodal ( analog ) input signal from the ‘upstream’ regulatory branch occurring in all cells ( Figure 1B ) to a bistable ( digital ) output in a subset , and in scalable manner . We experimentally verified this scalable analog-digital conversion in a P . putida without ICEclc but with the reconstructed bistability generator . The key ICEclc bistability regulatory elements involve new previously unrecognized transcription factors , which are conserved among a wide range of Proteobacteria , illustrating their importance for the behaviour of this conglomerate of related ICEs . Previous work had implied an ICEclc-located operon of three consecutive regulatory genes ( mfsR , marR and tciR , Figure 1B ) in control of transfer competence formation ( Pradervand et al . , 2014 ) . That work had shown that mfsR codes for an autorepressor , whose deletion yielded unhindered production of the LysR-type activator TciR . As a result , the proportion of tc cells is largely increased in P . putida UWC1 bearing ICEclc-∆mfsR ( Delavat et al . , 2016; Pradervand et al . , 2014 ) . We reproduced this state of affairs here by cloning tciR under control of the IPTG-inducible Ptac promoter on a plasmid ( pMEtciR ) in P . putida UWC1-ICEclc . In absence of cloned tciR , transfer of wild-type ICEclc from succinate-grown P . putida to an ICEclc-free isogenic P . putida was below detection limit , indicating that spontaneous ICE activation under those conditions is negligible ( Figure 2A ) . In contrast , inducing tciR expression by IPTG addition triggered ICEclc transfer from succinate-grown cells up to frequencies close to those observed under wild-type growth conditions with 3-CBA ( Miyazaki and van der Meer , 2011b ) ( 10–2 transconjugant colony-forming units ( CFU ) per donor CFU , Figure 2A ) . Transfer frequencies were lower in the absence of IPTG , which indicated that leaky expression of tciR from Ptac was sufficient to trigger ICEclc transfer ( Figure 2A ) . These results confirmed the implication of TciR and thus we set out to identify its potential activation targets on ICEclc . Induction of tciR from pMEtciR in P . putida without ICEclc was insufficient to trigger eGFP production from a single-copy Pint promoter , which is a hallmark of induction of ICEclc transfer competence ( Figure 2B; Minoia et al . , 2008; Delavat et al . , 2016 ) . In contrast , in presence of ICEclc , similar induction of tciR yielded a clear increased subpopulation of activated cells ( Figure 2C ) . This suggested , therefore , that TciR does not directly activate Pint , but only through one or more other ICE-located factors . To search for such potential factors , we examined in more detail the genes in a 7 kb region at the left end of ICEclc ( close to the attL site , Figure 1A ) , where transposon mutations had previously been shown to influence Pint expression ( Sentchilo et al . , 2003 ) . In addition , three promoters had been characterized in this region ( Figure 1A; Gaillard et al . , 2010 ) , which we tested individually for potential activation by TciR ( Figure 2B ) . Promoters were fused with a promoterless egfp gene and inserted in single copy into the chromosome of P . putida UWC1 without ICEclc ( Materials and methods ) . Induction of tciR from Ptac on pMEtciR did not yield any eGFP fluorescence in P . putida UWC1 containing a single-copy PalpA- or PinR-egfp transcriptional fusion ( Figure 2B ) . In contrast , the PbisR-egfp fusion was activated upon induction of TciR compared to a vector-only control ( p=0 . 0042 , paired t-test , Figure 2B & D ) . This suggested that the link between TciR and ICEclc transfer competence proceeds through transcription activation of the promoter upstream of the gene bisR ( previously designated orf101284 ) . This transcript has previously been mapped and covers a single gene ( Gaillard et al . , 2010 ) . We renamed this gene as bisR , or bistability regulator , for its presumed implication in ICEclc bistability control ( Figure 1—figure supplement 1 , see further below ) . bisR is predicted to encode a 251-aa protein of unknown function with no detectable Pfam-domains . Further structural analysis using Phyre2 ( Kelley et al . , 2015 ) suggested three putative domains with low confidence ( between 38% and 53% , Figure 1—figure supplement 2 ) . One of these is a predicted DNA-binding domain , which hinted at the possible function of BisR as a transcriptional regulator itself . BlastP analysis showed that BisR homologs are widely distributed and well conserved among Beta- , Alpha- and Gammaproteobacteria , with homologies ranging from 43–100% amino acid identity over the ( quasi ) full sequence length ( Figure 1—figure supplement 2 ) . In order to investigate its potential regulatory function , bisR was cloned on a plasmid ( pMEbisR ) and introduced into P . putida UWC1-ICEclc . Inducing bisR by IPTG addition from Ptac triggered high rates of ICEclc transfer on succinate media ( Figure 3A ) . Deletion of bisR on ICEclc abolished its transfer , even upon overexpression of tciR , but could be restored upon ectopic expression of bisR ( Figure 3A ) . This showed that the absence of transfer was due to the lack of intact bisR , and not to a polar effect of bisR deletion on a downstream gene ( Figure 1A ) . In addition , transfer of an ICEclc deleted for tciR ( Pradervand et al . , 2014 ) could be restored by ectopic bisR expression ( Figure 3A ) . This indicated that TciR is ‘upstream’ in the regulatory cascade of BisR , and that TciR does not act anywhere else on the expression of components crucial for ICEclc transfer . IPTG induction of bisR in P . putida without ICE again did not yield activation of the single-copy Pint or PinR transcriptional reporter fusions , whereas some repression was observed on Pint itself ( Figure 3B ) . In contrast , BisR induction in P . putida UWC1 with ICEclc led to a massive activation of the same reporter constructs in virtually all cells ( Figure 3C ) , compared to a vector-only control ( Figure 2C , pME6032 ) . This suggested that BisR was an ( other ) intermediate regulator step in the complete cascade of activation of ICEclc transfer competence . Of the tested ICE–promoters within this 7 kb region , BisR induction triggered very strong expression from a single copy PalpA–egfp transcriptional fusion in all cells ( Figure 3D ) . This indicated that BisR is a transcription activator , and an intermediate regulator between TciR and further factors encoded downstream of the alpA-promoter ( Figure 1—figure supplement 1 ) . Next , we thus focused our attention on the genes downstream of the alpA-promoter . Cloning the genes from alpA all the way to inrR ( Figure 1A ) on plasmid pME6032 under control of Ptac and inducing that construct with IPTG resulted in activation of PinR–egfp and Pint–echerry expression in P . putida without ICEclc ( Figure 4A ) . Both these promoters had been silent upon activation of TciR or BisR ( Figure 2B and Figure 3B ) . This indicated that one or more regulatory factors directly controlling expression of PinR and/or Pint were encoded in this region , which we tried to identify by subcloning different gene configurations . Removing alpA from the initial construct had no measurable effect on expression of the fluorescent reporters , but replacing Ptac by the native PalpA promoter abolished all Pint reporter activation ( Figure 4B , Figure 4—figure supplement 1 ) . This suggested that PalpA is silent without activation by BisR ( see below ) and no spontaneous production of regulatory factors occurred . Removing three genes at the 3’ extremity ( i . e . , orf96323 , orf95213 and inrR ) reduced Pint–echerry reporter expression , but a fragment with a further deletion into the bisC gene was unable to activate Pint ( Figure 4B ) . Induction of inrR alone did not result in Pint activation ( Figure 4B ) . Deletion of parA and shi at the 5’ end of the fragment still enabled reporter expression from Pint , narrowing the activator factor regions down to two genes , previously named parB and orf97571 , but renamed here to bisD and bisC ( Figure 4B ) . Neither bisC or bisD alone , but only the combination of bisDC resulted in reporter expression from Pint in P . putida UWC1 without ICEclc ( Figure 4B ) , and similarly , of PinR ( Figure 4—figure supplement 1 ) . In the presence of ICEclc , inducing either bisC or bisD from a plasmid yielded a small proportion of cells expressing the Pint reporters ( Figure 4C ) . This was not the case in a P . putida carrying an ICEclc with a deletion of bisD ( Figure 4—figure supplement 2 ) , suggesting there was some sort of feedback mechanism of BisDC on itself ( see further below ) . In contrast , induction of bisDC in combination caused a majority of cells to express fluorescence from Pint in P . putida containing ICEclc ( Figure 4C ) or ICEclc-∆bisD ( Figure 4—figure supplement 2 ) . These results indicated that BisDC acts as an ensemble to activate transcription , and this pointed to bisDC as the last step in the regulatory cascade , since it was the minimum unit sufficient for activation of the Pint–promoter , which is exclusively expressed in the subpopulation of tc cells of wild-type P . putida with ICEclc ( Delavat et al . , 2016; Figure 1—figure supplement 1 ) . Induction of bisDC from plasmid pMEbisDC yielded high frequencies of ICEclc transfer from P . putida UWC1 under succinate-growth conditions ( Figure 4D ) . Expression of BisDC also induced transfer of ICEclc-variants deleted for tciR or for bisR ( Figure 4D ) . This confirmed that both tciR and bisR relay activation steps to PbisR and PalpA , respectively , but not to further downstream ICE promoters ( Figure 1—figure supplement 1 ) . Moreover , an ICEclc deleted for bisD could not be restored for transfer by overexpression of tciR or bisR , but only by complementation with bisDC ( Figure 4D ) . Interestingly , the frequency of transfer of an ICEclc lacking bisD complemented by expression of bisDC in trans was two orders of magnitude lower than that of similarly complemented wild-type ICEclc , ICEclc with tciR- or bisR-deletion ( Figure 4D ) . This was similar as the reduction in reporter expression observed in P . putida ICEclc-∆bisD complemented with pMEbisDC compared to wild-type ICEclc ( Figure 4—figure supplement 2 ) , and suggested the necessity of some ‘reinforcement’ occurring in the wild-type configuration that was lacking in the bisD deletion and could not be restored by in trans induction of plasmid-cloned bisDC . To investigate this potential ‘reinforcement’ in wild-type configuration , we revisited the potential for activation of the alpA promoter . Induction by IPTG of the plasmid-cloned fragment encompassing the gene region parA-shi-bisDC caused strong activation of reporter gene expression from PalpA in P . putida without ICEclc ( Figure 5A ) . The minimal region that still maintained PalpA induction encompassed bisDC , although much lower than with a cloned parA-shi-bisDC fragment ( Figure 5A ) . Interestingly , when the parA-shi-bisDC fragment was extended by alpA itself , reporter expression from PalpA was abolished , whereas also a fragment containing only alpA caused significant repression of the alpA promoter ( Figure 5A ) . The alpA gene is predicted to encode a 70-amino acid DNA binding protein with homology to phage regulators ( Trempy et al . , 1994; Figure 1—figure supplement 2 ) . These results would imply feedback control on activation of PalpA , since its previously mapped transcript covers the complete region from alpA to orf96323 on ICEclc , including bisDC ( Figure 1A; Gaillard et al . , 2010 ) . Although induction of BisDC was sufficient for activation of transcription from PalpA , this effectively only yielded a small subpopulation of cells with high reporter fluorescence values ( Figure 5B & C ) , in contrast to induction of the larger cloned gene region encompassing parA-shi-bisDC that activated all cells ( Figure 5B & C ) . The feedback loop , therefore , seemed to consist of a positive forward part that includes BisDC ( reinforced by an as yet unknown other mechanism ) and a modulatory repressive branch including AlpA ( Figure 1—figure supplement 1 ) . The results so far thus indicated that ICEclc transfer competence is initiated by TciR activating transcription of the promoter upstream of bisR . BisR then kickstarts expression from the alpA-promoter , leading to ( among others ) expression of BisDC . This is sufficient to induce the ‘downstream’ ICEclc transfer competence pathway ( Figure 1—figure supplement 1 ) , exemplified here by activation of the Pint and PinR promoters that become exclusively expressed in the subpopulation of transfer competent cells under wild-type conditions ( Minoia et al . , 2008 ) . In addition , BisDC reinforces transcription from the same alpA-promoter . In order to understand the importance of this regulatory architecture for generating bistability , for initiating and maintaining ( downstream ) transfer competence , we developed a conceptual mathematical model ( Figure 6A , Materials and methods , SI model ) . The model assumes the regulatory factors TciR , BisR and BisDC , typical oligomerization ( Tropel and van der Meer , 2004 ) , as well as binding of the oligomerized forms to and unbinding from their respective nodes ( i . e . , the linked promoters PbisR , PalpA and Pint ) . Binding is assumed to lead to protein synthesis and finally , protein degradation ( Figure 6A ) . We varied and explored the outcomes of different regulatory network architectures and parts , testing their effect on production of intermediary and downstream elements in stochastic simulations , with each individual simulation corresponding to events taking place in an individual cell ( Figure 6A , SI model ) . First we simulated the cellular output of BisDC in a subnetwork configuration with only BisDC activating PalpA ( i . e . , in absence of TciR or BisR , Figure 6B ) . Stochastic simulations ( n = 10 , 000 ) of this bare feedback loop with an arbitrary start of binomially distributed BisDC quantities ( mean = 8 molecules per cell , Figure 6B , INPUT ) , yielded a bimodal population with two BisDC output states after 100 time steps , one of which is zero ( black bar in histograms ) and the other with a mean positive BisDC value ( magenta ) ( Figure 6B ) . The output zero results when BisDC levels stochastically fall to 0 ( as in case of the light blue line in the panel STOCHASTIC of Figure 6B ) , since in that case there is no BisDC to stimulate its own production . Parameter variation showed that the proportion of cells with output zero from the loop is dependent on the binding and unbinding constants of BisDC to the alpA promoter , and the BisDC degradation rate ( Figure 6B , different A1 , A2 and A4-values ) . In addition , BisDC unbinding and degradation rates can influence the median BisDC output quantity in cells with positive state ( Figure 6B , case of A2 = 5 or A4 = 0 . 3 ) . This simulation thus indicated that a BisDC feedback loop can produce bimodal output , once BisDC is present . Since the feedback loop cannot start without BisDC , it is imperative to kickstart the alpA promoter by BisR ( Figure 6C ) . Simulations of a configuration that includes activation by BisR , showed how upon a single pulse of BisR , the feedback loop again leads to a bimodal population with zero and positive BisDC levels ( Figure 6C ) . Increasing the ( uniformly distributed ) mean quantity of BisR in the simulations , within a per-cell range that is typically measured for transcription factors ( Li et al . , 2014 ) , increased the proportion of cells with positive BisDC state , but did not influence their mean BisDC quantity ( Figure 6C ) . Even bimodally distributed BisR input also gave rise to bimodal BisCD output , but with a higher proportion of zero BisDC state ( Figure 6C , bimodal ) . In contrast to the BisDC loop alone , therefore , activation by BisR only influences the proportion of zero and positive BisDC states in the population , but not the mean resulting BisDC quantity in cells with positive state . In the full regulatory hierarchy of the ICE , production of BisR is controlled by TciR . Simulation of this configuration showed that bimodality already appeared at the level of BisR ( Figure 6D ) . The proportions of zero and positive states of both BisR and BisDC varied depending on the mean of uniformly distributed amounts of TciR among all cells , again sampled to within regular empiric transcription regulator quantities in individual cells ( Li et al . , 2014; Figure 6D ) . Bimodal BisDC levels are propagated by the network architecture to downstream ( ‘late’ ) promoters , as a consequence of them being under BisDC control ( Figure 6A & E ) . Importantly , simulations of an architecture without the BisDC feedback loop consistently resulted in lower protein output from BisDC–regulated promoters in activated cells than with feedback ( Figure 6E ) . This suggests two crucial functions for the ICE regulatory network: first , to convert unimodal or stochastic ( ‘analog’ ) expression of TciR and BisR among all cells to a consistent subpopulation of cells with positive ( ‘digital’ ) BisDC state , and secondly , to ensure sufficient BisDC levels to activate downstream promoters within the positive cell population ( Figure 6E ) . Through the kickstart by BisR and reinforcement by BisDC itself , bimodal expression at the alpA-promoter node can thus yield a stably expressed transfer competence pathway in a subpopulation of cells . Simulations thus predicted that the ICE regulatory network faithfully transmits and stabilizes analog input ( e . g . , a single regulatory factor uniformly or stochastically expressed at moderately low levels in all cells [Li et al . , 2014] ) to bistable output ( e . g . , a subset of cells with transfer competence and the remainder silent ) . To demonstrate this experimentally , we engineered a P . putida without ICEclc , but with a single copy chromosomally inserted IPTG-inducible bisR , a plasmid with alpA-parA-shi-bisDC under control of PalpA , and a single-copy dual Pint-echerry and PinR-egfp reporter ( Figure 7A ) . Induction from Ptac by IPTG addition yields unimodal ( analog ) production of BisR , the mean level of which can be controlled by the IPTG concentration ( Figure 7—figure supplement 1 ) . In the presence of all components of the system , IPTG induction of BisR led to bistable activation of both reporters ( Figure 7B , ABC ) . Increasing BisR induction was converted by the feedback loop into an increased proportion of fluorescent cells ( Figure 7C ) . This effectively created a scalable bimodal ( digital ) output from unimodal input , dependent on the used IPTG concentration ( Figure 7C , Figure 7—figure supplement 1 ) . The proportion of fluorescent cells was in line with predictions from stochastic simulations as a function of the relative strength of Ptac activation ( Figure 7D ) . Furthermore , in agreement with model predictions ( Figure 6C ) , the median fluorescence of activated cells remained the same at different IPTG ( and thus BisR ) concentrations ( Figure 7E ) . These results confirmed that the feedback loop architecture transforms a unimodal ( analog ) regulatory factor concentration ( BisR ) into a stabilized bimodal ( digital ) output . Pfam analysis detected a DUF2857-domain in the BisC protein , and further structural analysis using Phyre2 indicated significant similarities of BisC to FlhC ( Figure 1—figure supplement 2 ) . FlhC is a subunit of the master flagellar activator FlhDC of E . coli and Salmonella ( Claret and Hughes , 2000; Liu and Matsumura , 1994 ) . BisD carries a ParB domain , with a predicted DNA binding domain in the C-terminal portion of the protein ( Figure 1—figure supplement 2 ) . Although no FlhD domain was detected in BisD , in analogy to FhlDC we named the ICEclc activator complex BisDC , for bistability regulator subunits D and C . Database searches showed that bisDC loci are also widespread among pathogenic and environmental Gamma- and Beta-proteobacteria , and are also found in some Alphaproteobacteria ( Figure 1—figure supplement 3 ) . Phylogenetic analysis using the more distantly related sequence from Dickeya zeae MS2 as an outgroup indicated several clear clades , encompassing notably bisDC homologs within genomes of P . aeruginosa and Xanthomonas ( Figure 1—figure supplement 4 ) . Several genomes contained more than one bisDC homolog , the most extreme case being Bordetella petrii DSM12804 with up to four homologs belonging to four different clades ( Figure 1—figure supplement 4 ) . The gene synteny from bisR to inrR of ICEclc was maintained in several genomes ( Figure 1—figure supplement 3 ) , suggesting them being part of related ICEs with similar regulatory architecture . Notably , some of those are opportunistic pathogens , such as P . aeruginosa , B . petrii , B . bronchiseptica , or X . citri , and regions of high similarity to the ICEclc regulatory core extended to the well-known pathogenicity islands of the PAGI-2 ( Klockgether et al . , 2007 ) and PAGI-16 families ( Hong et al . , 2016; Figure 1—figure supplement 3 ) . Several of the ICEclc core homologs carry genes suspected in virulence ( e . g . , filamentous hemagglutinin [Sun et al . , 2016] encoded on the P . aeruginosa HS9 and Carb01-63 genomic islands ) , or implicated in acquired antibiotic resistance ( e . g , multidrug efflux pump on the A . xylosoxidans NH44784-1996 element ( Miyazaki et al . , 2015 ) , and carbapenem resistance on the PAGI-16 elements [Hong et al . , 2016] ) . This indicates the efficacy of the ICEclc type regulatory control on the dissemination of this type of mobile elements , and consequently , on the distribution and selection of adaptive gene functions they carry . ICEs operate a dual life style in their host , which controls their overall fitness as the integral of vertical descent ( i . e . , maintenance of the integrated state and replication with the host chromosome ) and horizontal transfer ( i . e . , excision from the host cell , transfer and reintegration into a new host ) ( Delavat et al . , 2017; Delavat et al . , 2016; Johnson and Grossman , 2015 ) . The decision for horizontal transfer is costly and potentially damages the host cell ( Delavat et al . , 2016; Pradervand et al . , 2014 ) , which is probably why its frequency of occurrence in most ICEs is fairly low ( <10–5 per cell in a host cell population ) ( Delavat et al . , 2017 ) . Consequently , the mechanisms that initiate and ensure ICE horizontal transfer must have been selected to operate under extremely low opportunity with high success . In other words , they have been selected to maximize faithful maintenance of transfer competence development , once this process has been triggered in a host cell . One would thus expect such mechanisms to impinge on rare , perhaps stochastic cellular events , yielding robust output despite cellular gene expression and pathway noise . ICEclc is further particular in the sense that its transfer competence is initiated in cells during stationary phase conditions ( Miyazaki et al . , 2012 ) , which restricts global transcription and activity , and may even profoundly alter the cytoplasmic state of the cell ( Parry et al . , 2014 ) . The results of our work here reveal that the basis for initiation and maintenance of ICEclc transfer competence in a minority of cells in a stationary phase population ( Reinhard et al . , 2013 ) , originates in a multinode regulatory network that further includes a positive feedback loop . Genetic dissection , epistasis experiments and expression of individual components in P . putida devoid of the ICE showed that the network consists of a number of regulatory factors , composed of MfsR , TciR , BisR and BisDC , acting sequentially on singular ( TciR , BisR ) or multiple nodes ( BisDC ) . The network has an ‘upstream’ branch controlling the initiation of transfer competence , a ‘bistability generator’ that confines the input signal , and maintains the ‘downstream’ path of transfer competence to a dedicated subpopulation of cells ( Figure 1—figure supplement 1 ) . The previously characterized mfsR-marR-tciR operon ( Pradervand et al . , 2014 ) , whose transcription is controlled through autorepression by MfsR , is probably the main break on activation of the upstream branch . This was concluded from effects of deleting mfsR , which resulted in overexpression of TciR , and massively increased and deregulated ICE transfer even in exponentially growing cells ( Pradervand et al . , 2014 ) . We showed here that TciR activates the transcription of a hitherto unrecognized transcription factor gene named bisR , but not of any further critical ICEclc promoters . Autorepression by MfsR in wild-type ICEclc results in low unimodal transcription from PmfsR ( Pradervand et al . , 2014 ) and therefore , likely , to low TciR levels in all cells . TciR appeared here as a weak activator of the bisR promoter , suggesting that only in a small proportion of cells it manages to trigger bisR transcription , as our model simulations further attested . The BisR amino acid sequence revealed only very weak homology to known functional domains , thus making it the prototype of a new family of transcriptional regulators . In contrast to TciR , BisR was a very potent activator of its target , the alpA promoter . Model simulations suggested that BisR triggers and transmits the response in a scalable manner to the bistability generator , encoded by the genes downstream of PalpA . Triggering of PalpA stimulated expression of ( among others ) two consecutive genes bisD and bisC , which code for subunits of an activator complex that weakly resembles the known regulator of flagellar synthesis FlhDC ( Claret and Hughes , 2000; Liu and Matsumura , 1994 ) . BisDC production was sufficient to activate the previously characterized bistable ICEclc promoters Pint and PinR , making it the key regulator for the ‘downstream’ branch ( Figure 1—figure supplement 1 ) . Importantly , BisDC was also part of a feedback mechanism activating transcription from PalpA , and therefore , regulates its own production . Simulations and experimental data indicated that the feedback loop acts as a scalable analog-to-digital converter , transforming any positive input received from BisR into a dedicated cell that can regenerate sufficiently high BisDC levels to activate the complete downstream transfer competence pathway . Bistable gene network architectures are characterized by the fact that expression variation is not resulting in a single mean phenotype , but can lead to two ( or more ) stable phenotypes - mostly resulting in individual cells displaying either one or the other phenotype ( Ferrell , 2012; Ferrell , 2002; Dubnau and Losick , 2006 ) . Importantly , such bistable states are an epigenetic result of the network functioning and do not involve modifications or mutations on the DNA ( Kussell and Leibler , 2005; Balázsi et al . , 2011 ) . Bistable phenotypes may endure for a particular time in individual cells and their offspring , or erode over time as a result of cell division or other mechanism , after which the ground state of the network reappears . One can thus distinguish different steps in a bistable network: ( i ) the bistability switch that is at the origin of producing the different states , ( ii ) a propagation or maintenance mechanism and ( iii ) a degradation mechanism [11] . Some of the most well characterized bistable processes in bacteria include competence formation and sporulation in Bacillus subtilis ( Dubnau and Losick , 2006 ) . Differentiation of vegetative cells into spores only takes place when nutrients become scarce or environmental conditions deteriorate ( Veening et al . , 2008; Veening et al . , 2006 ) . Sporulation is controlled by a set of feedback loops and protein phosphorylations , which culminate in levels of the key regulator SpoOA ~P being high enough to activate the sporulation genes ( Dubnau and Losick , 2006 ) . In contrast , bistable competence formation in B . subtilis is generated by feedback transcription control from the major competence regulator ComK . Stochastic variations among ComK levels in individual cells , ComK degradation and inhibition by ComS , and noise at the comK promoter determine the onset of comK transcription , which then reinforces itself because of the feedback mechanism ( Süel et al . , 2006; Maamar et al . , 2007 ) . Initiation and maintenance of the ICEclc transfer competence pathway thus resembles DNA transformation competence in B . subtilis in its architecture of an auto-feedback loop ( BisDC vs ComK ) . However , the switches leading to bistability are different , with ICEclc depending on a hierarchy of transcription factors ( MfsR , TciR and BisR ) , and transformation competence being a balance of ComK degradation and inhibition of such degradation ( Süel et al . , 2006; Maamar et al . , 2007 ) . ICEclc bistability architecture is clearly different from the well-known double negative feedback control exerted by , for example the phage lambda lysogeny/lytic phase decision in E . coli ( Arkin et al . , 1998; Bednarz et al . , 2014 ) . That switch entails essentially a balance of the counteracting transcription factors CI , CII and Cro ( Arkin et al . , 1998; Bednarz et al . , 2014 ) . Interestingly , other ICEs of the SXT/R391 family carry this typical double negative feedback loop architecture , which may therefore control their ( bistable ) activation ( Poulin-Laprade and Burrus , 2015; Bellanger et al . , 2008; Beaber and Waldor , 2004; Poulin-Laprade et al . , 2015 ) . Given the low frequencies of conjugative transfer of many different elements ( Delavat et al . , 2017 ) , bistability activation mechanisms may be much more widespread than assumed . Mathematically speaking , the ICEclc transfer competence regulatory architecture has two states , one of which is zero ( inactive ) and the other with a positive value ( activation of transfer competence ) . Stochastic modelling suggested that the feedback loop maintains positive output during a longer time period than in its absence ( although it will drop to zero at infinite time ) . Previous experimental data suggested that the tc cells indeed do not return to a silent ICEclc state , but become irreparably damaged , arrest their division ( Takano et al . , 2019 ) and wither ( Reinhard et al . , 2013 ) . However , because their number is proportionally low , there is no fitness cost on the population carrying the ICE ( Delavat et al . , 2016; Gaillard et al . , 2008 ) . The advantage of prolonged feedback output seems that constant levels of the BisCD regulator can be maintained , allowing coordinated and organized production of the components necessary for the ICEclc transfer itself . This would consist of , for example , the relaxosome complex responsible for DNA processing at the origin ( s ) of transfer , and the mating pore formation complex ( Carraro and Burrus , 2015 ) . Because Pseudomonas cells activate ICEclc transfer competence upon entry in stationary phase , the feedback loop may have a critical role to ensure faithful completion of the transfer competence pathway during this period of limiting nutrients , and to allow the ICE to excise and transfer from tc cells once new nutrients become available ( Delavat et al . , 2016 ) . Although our results were conclusive on the roles of the key regulatory factors ( MfsR , TciR , BisR , BisDC ) , there may be further auxiliary and modulary factors , and environmental cues that influence the transfer competence network . For example , we previously found that deletions in the gene inrR drastically decreased ICEclc transfer capability by 45–fold and reduced reporter gene expression from Pint ( Minoia et al . , 2008 ) . Expression of InrR alone , however , did not show any direct activation of Pint , PinR or PalpA , and InrR is thus unlikely to be a direct transcription activator protein . Our results also indicated that induction of AlpA may repress output from the PalpA promoter , and modulate the feedback loop that is initiated by BisR and maintained by BisDC . Furthermore , although induction of bisDC was sufficient to activate expression from PalpA , it was enhanced through an as yet uncharacterized mechanism involving its upstream regions . Previous results also highlighted the implication of the stationary phase sigma factor RpoS for PinR activation ( Figure 1B; Miyazaki et al . , 2012 ) , which may be more generally important for other ICEclc regulatory promoters as well . Unraveling these details in future work will be important for a full understanding of the generation and maintenance of bistability of the ICEclc family of elements , and its role in effective horizontal dissemination . Phylogenetic analyses showed the different ICEclc regulatory loci ( i . e . , bisR-alpA-bisDC-inrR ) to be widely conserved in Beta- and Gammaproteobacteria , with only few small variations in regulatory gene configurations . Most likely , these regions are part of ICEclc-like elements in these organisms , several of which have been detected previously ( Miyazaki et al . , 2015; Miyazaki , 2011a; Gaillard et al . , 2006 ) . They are further part of PAGI-2 ( Klockgether et al . , 2007 ) and PAGI-16 family genomic islands in P . aeruginosa clinical isolates ( Hong et al . , 2016 ) that have been implicated in the distribution of virulence factors and antibiotic resistance elements . The ICEclc regulatory cascade for transfer competence thus seems widely conserved , controlling horizontal dissemination of this important class of bacterial conjugative elements . Bacterial strains and plasmid constructions used in this study are shortly described in Table 1 and with more detail in Supplementary file 1 . Strains were routinely grown in Luria broth ( 10 g l–1 Tryptone , 10 g l–1 NaCl and 5 g l–1 Yeast extract , LB Miller , Sigma Aldrich ) at 30°C for P . putida and 37°C for E . coli in an orbital shaker incubator , and were preserved at –80°C in LB broth containing 15% ( v/v ) glycerol . Reporter assays and transfer experiments were carried out with cells grown in type 21C minimal media ( MM , Supplementary file 2; Gerhardt , 1981 ) supplemented with 10 mM sodium succinate or 5 mM 3-chlorobenzoate ( 3-CBA ) . Antibiotics were used at the following concentrations: ampicillin ( Ap ) , 100 µg ml–1 for E . coli and 500 µg ml–1 for P . putida; gentamicin ( Gm ) , 10 µg ml–1 for E . coli , 20 µg ml–1 for P . putida; kanamycin ( Kn ) , 50 µg ml–1; tetracycline ( Tc ) , 12 µg ml–1 for E . coli , 100 µg ml–1 or 12 . 5 µg ml–1 for P . putida grown in LB or MM , respectively . Genes were induced from Ptac by supplementing cultures with 0 . 05 mM isopropyl β-D-1-thiogalactopyranoside ( IPTG; or else at the indicated concentrations ) . Plasmid DNA was purified using the Nucleospin Plasmid kit ( Macherey-Nagel ) according to manufacturer’s instructions . All enzymes used in this study were purchased from New England Biolabs . PCR reactions were carried out with primers described in Supplementary file 3 . PCR products were purified using Nucleospin Gel and PCR Clean-up kits ( Macherey-Nagel ) according to manufacturer’s instructions . E . coli and P . putida were transformed by electroporation as described by Dower et al . , 1988 . in a Bio-Rad GenePulser Xcell apparatus set at 25 µF , 200 V and 2 . 5 kV for E . coli and 2 . 2 kV for P . putida using 2 mm gap electroporation cuvettes ( Cellprojects ) . All constructs were verified by DNA sequencing ( Eurofins ) . Different ICEclc gene configurations were cloned in P . putida with or without ICEclc , and further with different promoter-reporter fusions , using the broad host-range vector pME6032 , allowing IPTG-controlled expression from the LacIq-Ptac promoter ( Koch et al . , 2001; Table 1 ) . Genes tciR , bisR , bisC , bisDC , bisC+96323 , alpA and inrR were amplified using primer pairs as specified in Supplementary file 3 , with genomic DNA of P . putida UWC1-ICEclc as template . Amplicons were digested by EcoRI and cloned into EcoRI-digested pME6032 using T4 DNA ligase , producing after transformation the plasmids listed in Table 1 . The 6 . 4 kb ICEclc left-end fragment encompassing parA-inrR was recovered from pTCB177 ( Sentchilo et al . , 2003 ) and cloned into pME6032 ( producing pMEreg∆alpA , Supplementary file 1 ) . An alpA-parA-shi-bisD’ fragment was amplified by PCR ( Supplementary file 2 ) and cloned into pME6032 using EcoRI restriction sites ( Supplementary file 1 ) . The resulting plasmid was digested with SalI and the 4 . 8 kb fragment containing the Ptac promoter , alpA-parA-shi-bisD’ was recovered and used to replace the parA-shi-parB part of pMEreg∆alpA . This generated a cloned fragment encompassing alpA all the way to inrR ( pMEreg , Supplementary file 1 ) . Further 3’ deletions removing orf96323-inrR or bisC-inrR were generated by PstI and AfeI digestion and religation ( Supplementary file 1 ) . A DNA fragment containing PalpA , alpA , parA , shi and the 5’ part of bisD was synthesized ( ThermoFisher Scientific ) , and ligated by Quick-Fusion cloning ( Bimake ) into pMEreg∆alpA digested with PmlI and BamHI to remove the part containing lacIq , Ptac , parA , shi and bisD . This plasmid was then digested by PstI to remove orf96323-inrR and religated ( Supplementary file 1 ) . Deletions of bisR or bisD in ICEclc were constructed using the two-step seamless chromosomal gene inactivation technique as described elsewhere ( Martínez-García and de Lorenzo , 2011 ) . Activation of key ICEclc promoters was determined in strains with a single-copy chromosomal insertions to promoterless egfp or echerry genes , for most cases delivered by a suicide miniTn7 system at a fixed unique position ( Table 1 ) . In other cases , particularly in combination with other single-copy inserted gene fragments , we used miniTn5 random delivery . The promoter regions upstream of bisR or alpA were amplified in the PCR ( Supplementary file 3 ) and cloned into the promoterless egfp reporter miniTn5 delivery plasmid pBAM1 ( Martínez-García et al . , 2011 ) or into a pUC18-derived miniTn7 delivery plasmid ( Choi et al . , 2005 ) . The PinR-egfp insert was recovered from the miniTn5-based reporter system ( Minoia et al . , 2008 ) using HindIII and KpnI , and subsequently cloned into pUC18miniTn7 digested by the same enzymes . The dual miniTn5::PinR-egfp/Pint-echerry reporter has been described previously ( Minoia et al . , 2008 ) . A miniTn7::Ptac-echerry reporter was reconstructed from pZS2FUNR ( Minoia et al . , 2008 ) and the general miniTn7:Ptac suicide delivery vector ( Choi et al . , 2005; Supplementary file 2 ) . All reporter constructs were integrated in single copy into the chromosomal attBTn7 site of P . putida by using pUX-BF13 for miniTn7 , or randomly for miniTn5-based constructs ( Martínez-García et al . , 2011; Koch et al . , 2001 ) , in which case three independent clones were recovered , stored and analysed . The intactness of the inserted reporter constructs was verified by PCR amplification and sequencing . ICEclc transfer was tested with 24-h-succinate-grown donor and recipient cultures . Cells were harvested by centrifugation of 1 ml ( donor ) and 2 ml culture ( recipient , Gm-resistant P . putida UWCGC ) for 3 min at 1200 × g , washed in 1 ml of MM without carbon substrate , centrifuged again and finally resuspended in 20 µl of MM . Donor or recipient alone , and a donor-recipient mixture were deposited on 0 . 2–µm cellulose acetate filters ( Sartorius ) placed on MM succinate agar plates , and incubated at 30°C for 48 hr . The cells were recovered from the filters in 1 ml of MM and serially diluted before plating . Donors , recipients and exconjugants were selected on MM agar plates containing appropriate antibiotics and/or carbon source ( 3-CBA ) . Transfer frequencies are reported as the mean of the exconjugant colony forming units compared to that of the donor in the same assay . BisDC phylogeny was inferred from 148 aligned homolog amino acid sequences by using the Maximum Likelihood method based on the Tamura-Nei model ( Tamura and Nei , 1993 ) , eliminating positions with less than 95% site coverage . The final dataset was aligned using MEGA7 ( Kumar et al . , 2016 ) and contained a total of 2091 positions . Initial tree ( s ) for the heuristic search were obtained automatically by applying Neighbour-Joining and BioNJ algorithms to a matrix of pairwise distances estimated using the Maximum Composite Likelihood ( MCL ) approach , and then selecting the topology with superior log likelihood value . For quantification of eGFP and eCherry fluorescence in single cells , P . putida strains were cultured overnight at 30°C in LB medium . The overnight culture was diluted 200 fold in 8 ml of MM supplemented with succinate ( 10 mM ) and appropriate antibiotic ( s ) , and grown at 30°C and 180 rpm to stationary phase . 150 µl of culture were then sampled , vortexed for 30 s at max speed , after which drops of 5 µl were deposited on a regular microscope glass slide ( VWR ) coated with a thin film of 1% agarose in MM . Cells were covered with a 24 × 50 mm cover slip ( Menzel-Gläser ) and imaged immediately with a Zeiss Axioplan II microscope equipped with an EC Plan-Neofluar 100×/1 . 3 oil objective lens ( Carl Zeiss ) , and a SOLA SE light engine ( Lumencor ) . A SPOT Xplorer slow-can charge coupled device camera ( 1 . 4 Megapixels monochrome w/o IR; Diagnostic Instruments ) fixed on the microscope was used to capture images . Up to ten images at different positions were acquired using Visiview software ( Visitron systems GMbH ) , with exposures set to 40 ms ( phase contrast , PhC ) and 500 ms ( eGFP and eCherry ) . Cells were automatically segmented on image sets using procedures described previously ( Delavat et al . , 2016 ) , from which their fluorescence ( eGFP or eCherry ) was quantified . Subpopulations of tc cells were quantified using quantile-quantile-plotting as described previously ( Reinhard and van der Meer , 2013 ) . Fluorescent images for display were scaled to the same brightness in ImageJ ( Schneider et al . , 2012 ) as indicated , saved as 8-bit gray tiff-files and cropped to the display area in Adobe Photoshop ( Adobe , 2020 ) . Fluorescent reporter intensities were compared among biological triplicates . In case of mini-Tn5 insertions , this involved three clones with potentially different insertion sites , each measured individually . For mini-Tn7 inserted reporter constructs , we measured three biological replicates of a unique clone . Expression differences between mutants and a strain with the same genetic background but carrying the empty pME6032 plasmid were tested on triplicate means of individual median or 75th percentile values in a one-sided t-test ( the hypothesis being that the mutant expression is higher than the control ) . Comparison of 75th percentiles rather than median or mean is justified when populations are extremely skewed , as we previously demonstrated ( Reinhard and van der Meer , 2013 ) . Coherent simultaneous data series were tested for significance of reporter expression or transfer frequency differences in ANOVA , followed by a post-hoc Tukey test . Quantile-quantile plots were produced in MatLab ( v 2016a ) , violin -boxplots by using ggplot2 in R . ICEclc activation was simulated as a series of stochastic events in different network configurations ( as schematically depicted in Figure 6A , Supplementary file 4 ) . TciR , BisR , BisDC and protein output levels were then simulated using the Gillespie algorithm ( Gillespie , 1977; Gillespie , 1976 ) , implemented in Julia using its DifferentialEquations . jl package ( Rackauckas and Nie , 2017 ) . 10 , 000 individual simulations ( each simulation corresponding to a single ‘cell’ ) were conducted per network configuration during 100 time steps , during or after which the remaining protein levels were counted and summarized . The code for the mathematical implementation is provided in the Source code 1 .	Mobile DNA elements are pieces of genetic material that can jump from one bacterium to another , and even across species . They are often useful to their host , for example carrying genes that allow bacteria to resist antibiotics . One example of bacterial mobile DNA is the ICEclc element . Usually , ICEclc sits passively within the bacterium’s own DNA , but in a small number of cells , it takes over , hijacking its host to multiply and to get transferred to other bacteria . Cells that can pass on the elements cannot divide , and so this ability is ultimately harmful to individual bacteria . Carrying ICEclc can therefore be positive for a bacterium but passing it on is not in the cell’s best interest . On the other hand , mobile DNAs like ICEclc have evolved to be disseminated as efficiently as possible . To shed more light on this tense relationship , Carraro et al . set out to identify the molecular mechanisms ICEclc deploys to control its host . Experiments using mutant bacteria revealed that for ICEclc to successfully take over the cell , a number of proteins needed to be produced in the correct order . In particular , a protein called BisDC triggers a mechanism to make more of itself , creating a self-reinforcing ‘feedback loop’ . Mathematical simulations of the feedback loop showed that it could result in two potential outcomes for the cell . In most of the ‘virtual cells’ , ICEclc ultimately remained passive; however , in a few , ICEclc managed to take over its hosts . In this case , the feedback loop ensured that there was always enough BisDC to maintain ICEclc’s control over the cell . Further analyses suggested that this feedback mechanism is also common in many other mobile DNA elements , including some that help bacteria to resist drugs . These results are an important contribution to understand how mobile DNAs manipulate their bacterial host in order to propagate and disperse . In the future , this knowledge could help develop new strategies to combat the spread of antibiotic resistance .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "genetics", "and", "genomics" ]	2020	An analog to digital converter controls bistable transfer competence development of a widespread bacterial integrative and conjugative element
Before the outbreak that reached the Americas in 2015 , Zika virus ( ZIKV ) circulated in Asia and the Pacific: these past epidemics can be highly informative on the key parameters driving virus transmission , such as the basic reproduction number ( R0 ) . We compare two compartmental models with different mosquito representations , using surveillance and seroprevalence data for several ZIKV outbreaks in Pacific islands ( Yap , Micronesia 2007 , Tahiti and Moorea , French Polynesia 2013-2014 , New Caledonia 2014 ) . Models are estimated in a stochastic framework with recent Bayesian techniques . R0 for the Pacific ZIKV epidemics is estimated between 1 . 5 and 4 . 1 , the smallest islands displaying higher and more variable values . This relatively low range of R0 suggests that intervention strategies developed for other flaviviruses should enable as , if not more effective control of ZIKV . Our study also highlights the importance of seroprevalence data for precise quantitative analysis of pathogen propagation , to design prevention and control strategies . In May 2015 , the first local cases of Zika were recorded in Brazil and by December of the same year the number of cases had surpassed 1 . 5 million . On February 2016 , the World Health Organization declared Zika as a public health emergency of international concern ( Who , 2016 ) and in March 2016 , local transmission of Zika was recognized in 34 countries . Previously the Zika virus had circulated in Africa and Asia but only sporadic human cases had been reported . In 2007 the outbreak on Yap ( Micronesia ) was the first Zika outbreak outside Africa and Asia ( Duffy et al . , 2009 ) . Since , Zika outbreaks have been also reported in French Polynesia and in New Caledonia ( Cao-Lormeau et al . , 2014; Dupont-Rouzeyrol et al . , 2015 ) between 2013 and 2014 and subsequently , there have been cases of Zika disease in the Cook Islands , the Solomon Islands , Samoa , Vanuatu , and Easter Island ( Chile ) ( see Figure 1 in Petersen et al . [2016] ) . Zika virus ( ZIKV ) is a flavivirus , mostly transmitted via the bites of infected Aedes mosquitoes , although non-vector-borne transmission has been documented ( sexual and maternofoetal transmission , laboratory contamination , transmission through transfusion ) ( Musso and Gubler , 2016 ) . The most common clinical manifestations include rash , fever , arthralgia , and conjunctivitis ( Musso and Gubler , 2016 ) but a large proportion of infections are asymptomatic or trigger mild symptoms that can remain unnoticed . Nevertheless , the virus may be involved in many severe neurological complications , including Guillain-Barre syndrome ( Cao-Lormeau et al . , 2016 ) and microcephaly in newborns ( Schuler-Faccini et al . , 2015 ) . These complications and the impressive speed of its geographically propagation make the Zika pandemic a public health threat ( Who , 2016 ) . This reinforces the urgent need to characterize the different facets of virus transmission and to evaluate its dispersal capacity . We address this here by estimating the key parameters of ZIKV transmission , including the basic reproduction number ( R0 ) , based on previous epidemics in the Pacific islands . Defined as the average number of secondary cases caused by one typical infected individual in an entirely susceptible population , the basic reproduction number ( R0 ) is a central parameter in epidemiology used to quantify the magnitude of ongoing outbreaks and it provides insight when designing control interventions ( Diekmann et al . , 2010 ) . It is nevertheless complex to estimate ( Diekmann et al . , 2010; van den Driessche and Watmough , 2002 ) , and therefore , care must be taken when extrapolating the results obtained in a specific setting , using a specific mathematical model . In the present study , we explore the variability of R0 using two models in several settings that had Zika epidemics in different years and that vary in population size ( Yap , Micronesia 2007 , Tahiti and Moorea , French Polynesia 2013–2014 , and New Caledonia 2014 ) . These three countries were successively affected by the virus , resulting in the first significant human outbreaks and they differ in several ways , including population size and location specific features . Hence , the comparison of their parameter estimates can be highly informative on the intrinsic variability of R0 . For each setting , we compare two compartmental models using different parameters defining the mosquito populations . Both models are considered in a stochastic framework , a necessary layer of complexity given the small population size and recent Bayesian inference techniques ( Andrieu et al . , 2010 ) are used for parameter estimation . We use mathematical transmission models and data from surveillance systems and seroprevalence surveys for several ZIKV outbreaks in Pacific islands ( Yap , Micronesia 2007 ( Duffy et al . , 2009 ) , Tahiti and Moorea , French Polynesia 2013–2014 ( CHSP , 2014; Mallet et al . , 2015; Aubry et al . , 2015b ) , New Caledonia 2014 [DASS , 2014] ) to quantify the ZIKV transmission variability . Two compartmental models with vector-borne transmission are compared ( cf . Figure 1 ) . Both models use a Susceptible-Exposed-Infected-Resistant ( SEIR ) framework to describe the virus transmission in the human population , but differ in their representation of the mosquito population . Figure 1a is a schematic representation derived from Pandey et al . ( 2013 ) and formulates explicitly the mosquito population , with a Susceptible-Exposed-Infected ( SEI ) dynamic to account for the extrinsic incubation period ( time taken for viral dissemination within the mosquito ) . 10 . 7554/eLife . 19874 . 003Figure 1 . Graphical representation of compartmental models . Squared boxes and circles correspond respectively to human and vector compartments . Plain arrows represent transitions from one state to the next . Dashed arrows indicate interactions between humans and vectors . ( a ) Pandey model ( Pandey et al . , 2013 ) . HS susceptible individuals; HE infected ( not yet infectious ) individuals; HI infectious individuals; HR recovered individuals; σ is the rate at which HE-individuals move to infectious class HI; infectious individuals ( HI ) then recover at rate γ; VS susceptible vectors; VE infected ( not yet infectious ) vectors; VI infectious vectors; V constant size of total mosquito population; τ is the rate at which VE-vectors move to infectious class VI; vectors die at rate μ . ( b ) Laneri model ( Laneri et al . , 2010 ) . HS susceptible individuals; HE infected ( not yet infectious ) individuals; HI infectious individuals; HR recovered individuals; σ is the rate at which HE-individuals move to infectious class HI; infectious individuals ( HI ) then recover at rate γ; implicit vector-borne transmission is modeled with the compartments κ and λ; λ current force of infection; κ latent force of infection reflecting the exposed state for mosquitoes during the extrinsic incubation period; τ is the transition rate associated to the extrinsic incubation period . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 003 By contrast , in the second model ( Figure 1b ) based on Laneri et al . ( 2010 ) the vector is modeled implicitly: the two compartments κ and λ do not represent the mosquito population but the force of infection for vector to human transmission . This force of infection passes through two successive stages in order to include the delay associated with the extrinsic incubation period: κ stands for this latent phase of the force of infection whereas λ corresponds directly to the rate at which susceptible humans become infected . The basic reproduction number of the models ( R0 ) is calculated using the next Generation Matrix method ( Diekmann et al . , 2010 ) :R0Pandey=βHβVτγμ ( μ+τ ) R0Laneri=βγ In addition , we consider that only a fraction ρ of the total population is involved in the epidemic , due to spatial heterogeneity , immuno-resistance , or cross-immunity . For both models we define N=ρ⋅H with H the total size of the population reported by census . The dynamics of ZIKV transmission in these islands is highly influenced by several sources of uncertainties . In particular , the small population size ( less than 7000 inhabitants in Yap ) leads to high variability in transmission rates . Therefore all these models are simulated in a discrete stochastic framework ( Poisson with stochastic rates [Bretó et al . , 2009] ) , to take this phenomenon into account . Stochasticity requires specific inference techniques: thus estimations are performed with PMCMC algorithm ( particle Markov Chain Monte Carlo [Andrieu et al . , 2010] ) . Using declared Zika cases from different settings , the two stochastic models ( Figure 1 ) were fitted ( Figures 2–3 ) . These models allow us to describe the course of the observed number of cases and estimate the number of secondary cases generated , R0 . Our estimates of R0 lie between 1 . 6 ( 1 . 5–1 . 7 ) and 3 . 2 ( 2 . 4–4 . 1 ) and vary notably with respect to settings and models ( Figures 2–3 and Tables 1–2 ) . Strikingly , Yap displays consistently higher values of R0 in both models and in general , there is an inverse relationship between island size and both the value and variability of R0 . This phenomenon may be explained by the higher stochasticity and extinction probability associated with smaller populations and can also reflect the information contained in the available data . However , the two highly connected islands in French Polynesia , Tahiti and Moorea , display similar values despite their differing sizes . 10 . 7554/eLife . 19874 . 004Figure 2 . Results using the Pandey model . Posterior median number of observed Zika cases ( solid line ) , 95% credible intervals ( shaded blue area ) and data points ( black dots ) . First column: particle filter fit . Second column: Simulations from the posterior density . Third column: R0 posterior distribution . ( a ) Yap . ( b ) Moorea . ( c ) Tahiti . ( d ) New Caledonia . The estimated seroprevalences at the end of the epidemic ( with 95% credibility intervals ) are: ( a ) 73% ( CI95: 68–77 , observed 73% ) ; ( b ) 49% ( CI95: 45–53 , observed 49% ) ; ( c ) 49% ( CI95: 45–53 , observed 49% ) ; ( d ) 39% ( CI95: 8–92 ) . See Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 00410 . 7554/eLife . 19874 . 005Figure 3 . Results using the Laneri model . Posterior median number of observed Zika cases ( solid line ) , 95% credible intervals ( shaded blue area ) and data points ( black dots ) . First column: particle filter fit . Second column: Simulations from the posterior density . Third column: R0 posterior distribution . ( a ) Yap . ( b ) Moorea . ( c ) Tahiti . ( d ) New Caledonia . The estimated seroprevalences at the end of the epidemic ( with 95% credibility intervals ) are: ( a ) 72% ( CI95: 68–77 , observed 73% ) ; ( b ) 49% ( CI95: 45–53 , observed 49% ) ; c ) 49% ( CI95: 45–53 , observed 49% ) ; d ) 65% ( CI95: 24–91 ) . See Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 00510 . 7554/eLife . 19874 . 006Table 1 . Parameter estimations for the Pandey model . Posterior median ( 95% credible intervals ) . All the posterior parameter distributions are presented in Figures 6–9 . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 006Pandey modelYapMooreaTahitiNew CaledoniaPopulation sizeH689216 , 200178 , 100268 , 767Basic reproduction numberR03 . 2 ( 2 . 4–4 . 1 ) 2 . 6 ( 2 . 2–3 . 3 ) 2 . 4 ( 2 . 0–3 . 2 ) 2 . 0 ( 1 . 8–2 . 2 ) Observation rater0 . 024 ( 0 . 019-0 . 032 ) 0 . 058 ( 0 . 048-0 . 073 ) 0 . 060 ( 0 . 050-0 . 073 ) 0 . 024 ( 0 . 010-0 . 111 ) Fraction of population involvedρ74% ( 69–81 ) 50% ( 48–54 ) 50% ( 46–54 ) 40% ( 9–96 ) Initial number of infected individualsHI ( 0 ) 2 ( 1–8 ) 5 ( 0–21 ) 329 ( 16–1047 ) 37 ( 1–386 ) Infectious period in human ( days ) γ-15 . 2 ( 4 . 1–6 . 7 ) 5 . 2 ( 4 . 1–6 . 8 ) 5 . 2 ( 4 . 1–6 . 7 ) 5 . 5 ( 4 . 2–6 . 8 ) Extrinsic incubation period in mosquito ( days ) τ-110 . 6 ( 8 . 7–12 . 5 ) 10 . 5 ( 8 . 6–12 . 4 ) 10 . 5 ( 8 . 6–12 . 6 ) 10 . 7 ( 8 . 9–12 . 5 ) Mosquito lifespan ( days ) μ-115 . 6 ( 11 . 7–19 . 3 ) 15 . 3 ( 11 . 4–19 . 1 ) 15 . 1 ( 11 . 2–19 . 0 ) 15 . 4 ( 11 . 6–19 . 4 ) 10 . 7554/eLife . 19874 . 007Table 2 . Parameter estimations for the Laneri model . Posterior median ( 95% credible intervals ) . All the posterior parameter distributions are presented in Figures 10–13 . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 007Laneri modelYapMooreaTahitiNew CaledoniaPopulation sizeH689216 , 200178 , 100268 , 767Basic reproduction numberR02 . 2 ( 1 . 9–2 . 6 ) 1 . 8 ( 1 . 6–2 . 0 ) 1 . 6 ( 1 . 5–1 . 7 ) 1 . 6 ( 1 . 5–1 . 7 ) Observation rater0 . 024 ( 0 . 019–0 . 033 ) 0 . 057 ( 0 . 047–0 . 07 ) 0 . 057 ( 0 . 049–0 . 069 ) 0 . 014 ( 0 . 010–0 . 037 ) Fraction of population involvedρ73% ( 69–78 ) 51% ( 47–55 ) 54% ( 49–59 ) 71% ( 27–98 ) Initial number of infected individualsHI ( 0 ) 2 ( 1–10 ) 9 ( 1–28 ) 667 ( 22–1570 ) 82 ( 2–336 ) Infectious period in human ( days ) γ-15 . 3 ( 4 . 1–6 . 6 ) 5 . 3 ( 4 . 1–6 . 7 ) 5 . 2 ( 4 . 1–6 . 7 ) 5 . 4 ( 4 . 1–6 . 8 ) Extrinsic incubation period in mosquito ( days ) τ-110 . 7 ( 8 . 8–12 . 7 ) 10 . 6 ( 8 . 6–12 . 6 ) 10 . 5 ( 8 . 5–12 . 5 ) 10 . 8 ( 8 . 9–12 . 8 ) Regarding model variability , R0 estimates are always higher and coarser with the Pandey model than with the Laneri model ( cf . Tables 1–2 ) . The Pandey model has two additional estimated parameters ( in particular , the mosquito lifespan ) , which can explain the higher variability of the output . It is worth noting that these parameters are very sensitive ( see Materials and methods ) . The difference in R0 may also be linked to the difference in the estimated initial number of infected individuals ( HI ( 0 ) ) , which is higher in the Laneri model than in the Pandey model . Because of the high proportion of asymptomatic cases ( the ratio of asymptomatic:symptomatic is estimated to be 1:1 . 3 , V . -M Cao-Lormeau personal communication ) , it is hard to determine which scenario is more realistic , the time between introduction of the disease into the island and the first reported symptomatic case being unknown in most settings . For the durations of infectious and intrinsic incubation ( in human ) and extrinsic incubation ( in mosquito ) periods , the posterior density ressembles the informative prior ( cf . Figures 6–13 ) , indicating the models’ incapacity to identify properly these parameters without more informative data . Moreover , these parameters have a clear sensitivity ( see Materials and methods ) and precise field measures are therefore crucial for reliable model predictions . The fraction ρ of the population involved in the epidemic is well estimated when the seroprevalence is known ( in Yap and French Polynesia ) . In these cases , the proportion of the population involved is slightly greater than the seroprevalence rate , indicating a very high infection rate among involved individuals . In New Caledonia , as no information on seroprevalence was available , the fraction of population involved displays very large confidence intervals ( cf . Tables 1 and 2 ) , indicating that the model did not manage to identify properly this parameter with the available data . In this case , this parameter is highly correlated to the observation rate r ( cf Figures 17 and 21 ) : r and ρ seem unidentifiable without precise information on seroprevalence . The reproduction number R0 is a key parameter in epidemiology that characterizes the epidemic dynamics and the initial spread of the pathogen at the start of an outbreak in a susceptible population . R0 can be used to inform public health authorities on the level of risk posed by an infectious disease , vaccination strategy , and the potential effects of control interventions ( Anderson and May , 1982 ) . In the light of the potential public health crisis generated by the international propagation of ZIKV , characterization of the potential transmissibility of this pathogen is crucial for predicting epidemic size , rate of spread and efficacy of intervention . Using data from both surveillance systems and seroprevalence surveys in four different geographical settings across the Pacific ( Duffy et al . , 2009; CHSP , 2014; Mallet et al . , 2015; DASS , 2014; Aubry et al . , 2015b ) , we have estimated the basic reproductive number R0 ( see Figures 2–3 and Tables 1–2 ) . Our estimate of R0 obtained by inference based on Particle MCMC ( Andrieu et al . , 2010 ) has values in the range 1 . 6 ( 1 . 5–1 . 7 ) – 3 . 2 ( 2 . 4–4 . 1 ) . Our R0 estimates vary notably across settings . Lower and finer R0 values are found in larger islands . This phenomenon can at least in part be explained by large spatial heterogeneities and higher demographic stochasticity for islands with smaller populations , as well as the influence of stochasticity on biological and epidemiological processes linked to virus transmission . This phenomenon can also be specific to the selection of the studied islands or can reflect a highly clustered geographical pattern , the global incidence curve being the smoothed overview of a collection of more explosive small size outbreaks . However , it is notable that the two French Polynesian islands yield similar estimates of R0 despite differing population sizes . Indeed , other important factors differ among French Polynesia , New Caledonia and Yap , such as the human genetic background and their immunological history linked to the circulation of others arboviruses . Moreover , whilst both New Caledonia and French Polynesia populations were infected by the same ZIKV lineage and transmitted by the same principle vector species , Aedes aegypti , the epidemic in Yap occurred much earlier with a different ZIKV lineage ( Wang et al . , 2016 ) and vectored by a different mosquito species Aedes hensilli ( Ledermann et al . , 2014 ) . In French Polynesia , the vector Aedes polynesiensis is also present and dominates in Moorea with higher densities than in Tahiti . Finally , different vector control measures have been conducted in the three countries . To date , studies investigating Zika outbreaks in the Pacific have always estimated R0 using a deterministic framework . Using a similar version of the Pandey model in French Polynesia , Kucharski et al . ( Kucharski et al . , 2016 ) estimated R0 between 1 . 6 and 2 . 3 ( after scaling to square root for comparison ) for Tahiti and between 1 . 8 and 2 . 9 in Moorea . These estimates are slightly lower and less variable than ours . This difference can be explained firstly by the chosen priors on mosquito parameters and secondly because our model includes demographic stochasticity . Moreover , they predicted a seroprevalence rate at the end of the epidemic of 95–97% , far from the 49% measured . In Yap island , a study ( Funk et al . , 2016 ) used a very detailed deterministic mosquito model , and estimated an R0 for Zika between 2 . 9 and 8 . In this case , our lower and less variable estimates may come from the fact that our model is more parsimonious in the number of uncertain parameters , especially concerning the mosquito population . Finally , a third study ( Nishiura et al . , 2016a ) relied on another method for R0 calculation ( based on the early exponential growth rate of the epidemic ) in French Polynesia as a whole and in Yap . Again , the obtained parameters are lower than ours in French Polynesia and higher in Yap . The first estimations for South America using a similar methodology ( Nishiura et al . , 2016b; Towers et al . , 2016; Gao et al . , 2016 ) lead to similar R0 values . In all these studies a deterministic framework is used excluding the possibility of accounting for the high variability of biological and epidemiological processes exacerbated by the small size of the population . In these three studies , like in ours , it is worth noting that little insight is obtained regarding mosquito parameters . Posterior distribution mimics the chosen prior ( cf . Figures 6–13 ) . Both the simulation of the epidemics and the estimated R0 are highly sensitive to the choice of priors on mosquito parameters , for which precise field measures are rare . In the absence of sufficient data , the modeling of mosquito-borne pathogen transmission is a difficult task due to non-linearity and non-stationarity of the involved processes ( Cazelles and Hales , 2006 ) . This work has then several limitations . First , our study is limited by the completeness and quality of the data , with regard to both incidence and seroprevalence , but , above all , by the scarcity of information available on mosquitoes . Incidence data is aggregated at the island scale and cannot disentangle the effects of geography and observation noise to explain bimodal curves observed in Yap and New Caledonia . Moreover , although all data came from national surveillance systems , we had very little information about the potential degree of under-reporting , especially due to the high proportion of mildly symptomatic cases , at a time when the dangerous complications associated with the virus were unknown . Moreover , some cases might have been misdiagnosed as other flaviviruses , due to similarity in clinical manifestations or cross-reactivity in assays . Seroprevalence data were gathered from small sample sizes and were also sensitive to cross reactivity in populations non naive to dengue . In addition , they were missing in New Caledonia , which leads to strong correlation between our estimation of the observation rate and the fraction of the population involved in the epidemic . Because of the high proportion of asymptomatic or mildly symptomatic cases , the magnitude of the outbreaks is difficult to evaluate without precise seroprevalence data ( Metcalf et al . , 2016 ) or detection of mild , asymptomatic or pre-symptomatic infections ( Thompson et al . , 2016 ) . Considering vectors , no demographic data were available and this partly explains the large variability of our R0 estimations . Secondly , incidence and seroprevalence data were difficult to reconcile; the use of incidence data led to higher infection rates than those observed in the seroprevalence data . This difficulty has been overcome by considering that only a fraction of the population ( ρ ) is involved in the epidemic and then our model manages to reproduce the observed seroprevalence rate . This exposed fraction could be the result of spatial heterogeneity and high clustering of cases and transmission , as observed for dengue . The small dispersal of the mosquito and the limited population inter-mingling likely leads to considerable spatial variation in the extent of exposure to the virus and pockets of refugia in Tahiti as elsewhere ( Telle et al . , 2016 ) . For instance , previous dengue infection rates in French Polynesia display large spatial variations even within islands ( Daudens et al . , 2009 ) . Finer scale incidence and seroprevalence data would be useful to explore this . Another explanation for higher predicted than observed infection rates could be due to interaction with other flaviviruses . The Zika outbreak was concomitant with dengue outbreaks in French Polynesia ( CHSP , 2014; Mallet et al . , 2015 ) and New Caledonia ( DASS , 2014 ) . Examples of coinfection have been reported ( Dupont-Rouzeyrol et al . , 2015 ) but competition between these close pathogens may also have occurred . Finally , mathematical models with vectorial transmission may tend to estimate high attack rates , sometimes leading to a contradiction between observed incidence and observed seroprevalence . Assumptions on the proportionality between infected mosquitoes and the force of infection , as well as the density-dependence assumption in these models could be questioned . Indeed even if these assumptions are at the heart of the mathematical models of mosquito-borne pathogen transmission ( Reiner et al . , 2013; Smith et al . , 2014 ) , a recent review ( Halstead , 2008 ) and recent experimental results ( Bowman et al . , 2014; Harrington et al . , 2014 ) question these important points . On a final note , the estimates of R0 for ZIKV are similar to but generally on the lower side of estimates made for two other flaviviruses of medical importance , dengue and Yellow Fever viruses ( Favier et al . , 2006; Imai et al . , 2015; Massad et al . , 2003 ) , even though caution is needed in the comparison of studies with differing models , methods and data sources . Interventions strategies developed for dengue should thus enable as , if not more effective control of ZIKV , with the caveat that ZIKV remains principally a mosquito-borne pathogen with little epidemiological significance of the sexual transmission route . Even though further work and data are needed to support this hypothesis ( Brauer et al . , 2016 ) , two recent studies indicated that sexual transmission alone is not sufficient to sustain an epidemic ( Gao et al . , 2016; Towers et al . , 2016 ) . In conclusion , using recent stochastic modeling methods , we have been able to determine estimates of R0 for ZIKV with an unexpected relationship with population size . Further data from the current Zika epidemic in South America that is caused by the same lineage as French Polynesia lead to estimates in the same range of values ( Nishiura et al . , 2016b; Towers et al . , 2016; Gao et al . , 2016 ) . Our study highlights the importance of gathering seroprevalence data , especially for a virus that often leads to an asymptomatic outcome and it would provide a key component for precise quantitative analysis of pathogen propagation to enable improved planning and implementation of prevention and control strategies . During the 2007 outbreak that struck Yap , 108 suspected or confirmed Zika cases ( 16 per 1000 inhabitants ) were reported by reviewing medical records and conducting prospective surveillance between April 1st and July 29th 2007 ( Duffy et al . , 2009 ) . In French Polynesia , sentinel surveillance recorded more than 8700 suspected cases ( 32 per 1000 inhabitants ) across the whole territory between October 2013 and April 2014 ( CHSP , 2014; Mallet et al . , 2015 ) . In New Caledonia , the first Zika case was imported from French Polynesia on 2013 November 12th . Approximately 2500 cases ( 9 per 1000 inhabitants ) were reported through surveillance between January ( first autochtonous case ) and August 2014 ( DASS , 2014 ) . For Yap and French Polynesia , the post-epidemic seroprevalence was assessed . In Yap , a household survey was conducted after the epidemic , yielding an infection rate in the island of 73% ( Duffy et al . , 2009 ) . In French Polynesia , three seroprevalence studies were conducted . The first one took place before the Zika outbreak , and concluded that most of the population was naive for Zika virus ( Aubry et al . , 2015a ) . The second seroprevalence survey was conducted between February and March 2014 , at the end of the outbreak , and reported a seroprevalence rate around 49% ( Aubry et al . , 2015b ) . The third one concerned only schoolchildren in Tahiti and was therefore not included in the present study . Demographic data on population size were based on censuses from Yap ( Duffy et al . , 2009 ) , French Polynesia ( Insee , 2012 ) , and New Caledonia ( Insee , 2014 ) . R0 is calculated using the Next Generation Matrix approach ( NGM ) ( 19 ) . In order to analyse the influence of parameter values on the model’s outputs , a sensitivity analysis was performed , using LHS/PRCC technique ( Blower and Dowlatabadi , 1994 ) , on Tahiti example . Similar results were obtained for the other settings . Three criteria were retained as outputs for the analysis: the seroprevalence at the last time point , the intensity of the peak of the outbreak and the date of the peak . We used uniform distributions for all parameters , which are listed in Tables 6 and 7 . For model parameters , we used the same range as for the prior distribution . For initial conditions , the observation rate r and the fraction involved in the epidemic ρ , we used the 95% confidence interval obtained by PMCMC , in order to avoid unrealistic scenarios . 10 . 7554/eLife . 19874 . 011Table 6 . Sensitivity analysis in Pandey model . Tahiti island . 1000 parameter sets were sampled with latin hypercube sampling ( LHS ) , using 'lhs' R package ( Carnell , 2016 ) . On each parameter set , the model was simulated deterministically in order to explore variability in parameters without interference with variations due to stochasticity . PRCC were computed using the 'sensitivity' R package ( Pujol et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 011ParametersDistributionSeroprevalencePeak intensityPeak dateModel parametersR02Uniform[0 , 20]0 . 870 . 90−0 . 55βVUniform[0 , 10]−0 . 66−0 . 730 . 35γ-1Uniform[4 , 7]−0 . 250 . 100 . 20σ-1Uniform[2 , 7]−0 . 03−0 . 100 . 15τ-1Uniform[4 , 20]−0 . 05−0 . 070 . 06μ-1Uniform[4 , 30]−0 . 56−0 . 700 . 49Initial conditionsHI ( 0 ) Uniform[2 . 10-5 , 0 . 011]0 . 05−0 . 020 . 02VI ( 0 ) Uniform[10-4 , 0 . 034]0 . 11−0 . 00−0 . 26Fraction involved and observation modelρUniform[0 . 46 , 0 . 54]0 . 470 . 15−0 . 03rUniform[0 . 048 , 0 . 072]−0 . 040 . 030 . 0510 . 7554/eLife . 19874 . 012Table 7 . Sensitivity analysis in Laneri model . Tahiti island . 1000 parameter sets were sampled with latin hypercube sampling ( LHS ) , using 'lhs' R package ( Carnell , 2016 ) . On each parameter set , the model was simulated deterministically in order to explore variability in parameters without interference with variations due to stochasticity . PRCC were computed using the 'sensitivity' R package ( Pujol et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 012ParametersDistributionSeroprevalencePeak intensityPeak dateModel parametersR02Uniform[0 , 20]0 . 620 . 93−0 . 50γ-1Uniform[4 , 7]0 . 010 . 620 . 15σ-1Uniform[2 , 7]−0 . 03−0 . 540 . 21τ-1Uniform[4 , 20]−0 . 03−0 . 700 . 47Initial conditionsHI ( 0 ) Uniform[10-5 , 0 . 015]0 . 050 . 02−0 . 32L ( 0 ) Uniform[2 . 10-5 , 0 . 004]0 . 050 . 00−0 . 16Fraction involved and observation modelρUniform[0 . 49 , 0 . 59]0 . 800 . 340 . 02rUniform[0 . 048 , 0 . 068]−0 . 010 . 01−0 . 02 For all criteria , the key parameters in both models are transmission parameters ( R0 and βV ) . High values for R0 are positively correlated with a large seroprevalence and a high and early peak . On the contrary , high values for the parameters introducing a delay in the model , the incubation periods in human ( σ-1 ) and in mosquito ( τ-1 ) , are associated with a lower and later peak , and have no significant effect on seroprevalence . Moreover , the simulations are clearly sensitive to the other model parameters , in particular the mosquito lifespan ( μ-1 ) in Pandey model . Concerning other parameters , the initial conditions have a noticeable effect on the date of the peak only . As expected , the fraction involved in the epidemic ( ρ ) influences the magnitude of the outbreak , by calibrating the proportion of people than can be infected , but it has no significant effect on the timing of the peak . These complementary results include PMCMC results for both models in the four settings: the epidemic trajectories regarding the human compartments for infected and recovered individuals ( Figures 4 , 5 ) , the detailed posterior distributions for all parameters ( Figures 6–13 ) and correlation plots for all models ( Figures 14–21 ) . 10 . 7554/eLife . 19874 . 013Figure 4 . Infected and recovered humans evolution during the outbreak with Pandey model . Simulations from the posterior density: posterior median ( solid line ) , 95% and 50% credible intervals ( shaded blue areas ) and observed seroprevalence ( black dots ) . First column: Infected humans ( HI ) . Second column: Recovered humans ( HR ) . ( a ) Yap . ( b ) Moorea . ( c ) Tahiti . ( d ) New Caledonia . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 01310 . 7554/eLife . 19874 . 014Figure 5 . Infected and recovered humans evolution during the outbreak with Laneri model . Simulations from the posterior density: posterior median ( solid line ) , 95% and 50% credible intervals ( shaded blue areas ) and observed seroprevalence ( black dots ) . First column: Infected humans ( HI ) . Second column: Recovered humans ( HR ) . ( a ) Yap . ( b ) Moorea . ( c ) Tahiti . ( d ) New Caledonia . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 01410 . 7554/eLife . 19874 . 015Figure 6 . Posterior distributions . Pandey model , Yap island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 01510 . 7554/eLife . 19874 . 016Figure 7 . Posterior distributions . Pandey model , Moorea island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 01610 . 7554/eLife . 19874 . 017Figure 8 . Posterior distributions . Pandey model , Tahiti island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 01710 . 7554/eLife . 19874 . 018Figure 9 . Posterior distributions . Pandey model , New Caledonia . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 01810 . 7554/eLife . 19874 . 019Figure 10 . Posterior distributions . Laneri model , Yap island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 01910 . 7554/eLife . 19874 . 020Figure 11 . Posterior distributions . Laneri model , Moorea island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02010 . 7554/eLife . 19874 . 021Figure 12 . Posterior distributions . Laneri model , Tahiti island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02110 . 7554/eLife . 19874 . 022Figure 13 . Posterior distributions . Laneri model , New Caledonia . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02210 . 7554/eLife . 19874 . 023Figure 14 . Correlation plot of MCMC output . Pandey model , Yap island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02310 . 7554/eLife . 19874 . 024Figure 15 . Correlation plot of MCMC output . Pandey model , Moorea island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02410 . 7554/eLife . 19874 . 025Figure 16 . Correlation plot of MCMC output . Pandey model , Tahiti island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02510 . 7554/eLife . 19874 . 026Figure 17 . Correlation plot of MCMC output . Pandey model , New Caledonia . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02610 . 7554/eLife . 19874 . 027Figure 18 . Correlation plot of MCMC output . Laneri model , Yap island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02710 . 7554/eLife . 19874 . 028Figure 19 . Correlation plot of MCMC output . Laneri model , Moorea island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02810 . 7554/eLife . 19874 . 029Figure 20 . Correlation plot of MCMC output . Laneri model , Tahiti island . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 02910 . 7554/eLife . 19874 . 030Figure 21 . Correlation plot of MCMC output . Laneri model , New Caledonia . DOI: http://dx . doi . org/10 . 7554/eLife . 19874 . 030 The estimation tools are provided by the open source software SSM ( Dureau et al . , 2013 ) ( State Space Models , RRID:SCR_014647 ) , available at https://github . com/JDureau/ssm and https://github . com/sballesteros/ssm-predict . The codes for the implementation of each model are provided as supplementary file .	Zika virus is an infectious disease primarily transmitted between people by mosquitoes . While most people develop mild flu-like symptoms , infection during pregnancy can interfere with how the baby’s head and brain develop . Until recently , the virus had only been seen sporadically in Africa and Asia , but since 2007 , outbreaks have been recorded on several Pacific islands . In 2015 , the Zika virus reached the Americas , and within six months over 1 . 5 million cases had been reported in Brazil alone . There is an urgent need to understand how the Zika virus moves within a population in order to help policymakers , and public health professionals , plan treatment and control of outbreaks of the disease . Researchers often use predictive models to estimate how a disease will spread . A parameter commonly calculated by these models is the “basic reproductive number” , or R0 , which represents the average number of additional cases of the disease caused by one infected individual . Using models that incorporated data from Zika virus outbreaks that occurred on several Pacific islands , Champagne et al . have produced estimates of R0 that range from 1 . 5-4 . 1 . The R0 values are greater than one , indicating that infection will spread within a population , but in the same range as those obtained for dengue fever , another closely related mosquito-borne disease . This suggests that by taking appropriate measures , the spread of Zika and dengue can be controlled to similar extents . A closer look at the relationship between the population size and the predicted R0 value for each Pacific island revealed an unexpected inverse relationship: the smaller the population , the larger the value of R0 . Since other regional factors may also explain these large differences between settings , further work is needed to disentangle context-specific from disease-specific factors . In this respect , data about seroprevalence ( the number of people whose blood shows evidence of a past infection ) in different populations is crucial for precisely analyzing the spread of Zika virus .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health" ]	2016	Structure in the variability of the basic reproductive number (R0) for Zika epidemics in the Pacific islands
The internal N6-methyladenosine ( m6A ) methylation of eukaryotic nuclear RNA controls post-transcriptional gene expression , which is regulated by methyltransferases ( writers ) , demethylases ( erasers ) , and m6A-binding proteins ( readers ) in cells . The YTH domain family proteins ( YTHDF1–3 ) bind to m6A-modified cellular RNAs and affect RNA metabolism and processing . Here , we show that YTHDF1–3 proteins recognize m6A-modified HIV-1 RNA and inhibit HIV-1 infection in cell lines and primary CD4+ T-cells . We further mapped the YTHDF1–3 binding sites in HIV-1 RNA from infected cells . We found that the overexpression of YTHDF proteins in cells inhibited HIV-1 infection mainly by decreasing HIV-1 reverse transcription , while knockdown of YTHDF1–3 in cells had the opposite effects . Moreover , silencing the m6A writers decreased HIV-1 Gag protein expression in virus-producing cells , while silencing the m6A erasers increased Gag expression . Our findings suggest an important role of m6A modification of HIV-1 RNA in viral infection and HIV-1 protein synthesis . Interactions of host proteins with HIV-1 substantially modulate viral replication and pathogenesis ( Goff , 2007; Moir et al . , 2011 ) . Host proteins that interact with HIV-1 nucleic acids or viral proteins can either enhance or inhibit viral replication in cells . The secondary structure model of HIV-1 RNA and its interactions with viral proteins have been recently analyzed ( Kutluay et al . , 2014; Lavender et al . , 2015; Watts et al . , 2009; Wilkinson et al . , 2008 ) ; however , it is less clear whether host proteins can post-transcriptionally modify HIV-1 RNA , which may affect interactions between RNA and host or viral proteins , thereby affecting HIV-1 infection . N6-methyladenosine ( m6A ) is the most prevalent internal messenger RNA ( mRNA ) modification in eukaryotic organisms and plays pivotal roles in post-transcriptional regulation of gene expression ( Fu et al . , 2014; Jia et al . , 2013; Yue et al . , 2015 ) . This methylation is reversible ( Jia et al . , 2011; Zheng et al . , 2013 ) and is specifically recognized by a family of reader proteins ( Liu et al . , 2014; Wang et al . , 2015 ) . The m6A modification is widely distributed in mammalian mRNAs , enriched in the 3’ untranslated region ( UTR ) near the stop codon and also present in the 5’ UTR and long exons ( Dominissini et al . , 2012; Meyer et al . , 2012; Zhou et al . , 2015 ) . It has been known for almost 40 years that RNAs of influenza virus , adenovirus , Rous sarcoma virus , and simian virus 40 are m6A-methylated ( Beemon and Keith , 1977; Canaani et al . , 1979; Hashimoto and Green , 1976; Krug et al . , 1976 ) , although the impact of the m6A modification on viral replication remains unclear . The recent studies revealed that the m6A modification of HIV-1 RNA significantly affects viral replication and gene expression ( Kennedy et al . , 2016; Lichinchi et al . , 2016 ) . Lichinchi et al . reported that HIV-1 mRNA contains multiple m6A modifications and viral infection in a CD4+ T-cell line increases the m6A levels in both host and viral mRNAs ( Lichinchi et al . , 2016 ) , suggesting a dynamic regulation of m6A methylomes during HIV-1 infection . In contrast , Kennedy et al . only found four clusters of m6A modifications in the 3’ UTR region of the HIV-1 RNA genome that enhance viral gene expression ( Kennedy et al . , 2016 ) . The dynamic addition , removal , and recognition of m6A in cellular mRNAs and other types of nuclear RNAs are coordinately regulated by three groups of host proteins , including adenosine methyltransferases ( writers ) , m6A demethylases ( erasers ) , and m6A-selective-binding proteins ( readers ) ( Fu et al . , 2014 ) . The methyltransferase complex is composed of METTL3 , METTL14 and WTAP ( Wilms’ tumor 1-associating protein ) , which add m6A modification to nuclear RNAs ( Liu et al . , 2014; Ping et al . , 2014 ) . The RNA demethylases FTO ( fat mass and obesity-associated protein ) and AlkBH5 ( AlkB family member 5 ) remove the m6A modification of RNAs ( Fu et al . , 2014; Jia et al . , 2011 ) . Three host proteins , including YTHDF1 , 2 , and 3 ( YTHDF1–3 ) , have been identified as selective m6A-binding proteins ( readers ) in mammalian cells ( Dominissini et al . , 2012; Wang et al . , 2014 , 2015 ) . These m6A-reader proteins preferentially bind methylated mRNAs and control the stability and translation of target mRNAs ( Dominissini et al . , 2012; Liu et al . , 2014 ) . Human YTHDF1–3 proteins contain a conserved YTH RNA-binding domain that preferentially binds the m6A-containing RNAs and a P/Q/N-rich region that is associated with different RNA-protein complexes ( Fu et al . , 2014 ) . Lichinchi et al . recently reported that silencing of the m6A writers or the eraser AlkBH5 decreases or increases HIV-1 replication , respectively ( Lichinchi et al . , 2016 ) . Kennedy et al . showed that m6A modifications in the 3’ UTR region of HIV-1 RNA enhance viral gene expression by recruiting cellular YTHDF proteins ( Kennedy et al . , 2016 ) . However , neither study examined the m6A modification of HIV-1 RNA and its effect on HIV-1 replication in primary CD4+ T-cells , nor systemically analyzed the role of the m6A writers , erasers , and readers in HIV-1 replication . Here we show that HIV-1 RNA contains multiple m6A modifications enriched in the 5' and 3' UTRs and within several coding genes . We mapped the specific sites in HIV-1 RNA bound by YTHDF proteins in HIV-1-infected cells . We found that overexpression of YTHDF proteins in target cells significantly inhibited HIV-1 infection , while knockdown of these proteins in primary CD4+ T-cells enhanced HIV-1 infection . Furthermore , knockdown of the m6A writers or the erasers decreased or increased HIV-1 Gag synthesis and virion release in virus-producing cells , respectively . Our findings suggest important functions of the m6A reader , writer , and eraser proteins in modulating HIV-1 gene expression and viral infection through the m6A modification of HIV-1 RNA . To investigate the presence of m6A in HIV-1 RNA and to map the m6A modification within HIV-1 RNA , we isolated RNA samples from CD4+ Jurkat T-cells or primary CD4+ T-cells infected with replication-competent HIV-1NL4-3 , and performed immunoprecipitation ( IP ) with poly ( A ) -enriched RNA using m6A-specific antibodies , followed by high-throughput RNA sequencing ( m6A-seq ) ( Dominissini et al . , 2012 ) . We identified similar profiles of m6A peaks in HIV-1 RNA from these two cell types , which are mainly enriched in the 5' and 3' UTRs as well as the rev and gag genes of the HIV-1 genome ( Figure 1A , B ) . To confirm the m6A modification of HIV-1 RNA from virus-producing cells , we transfected HEK293T cells with a plasmid containing full-length HIV-1 proviral DNA ( pNL4-3 ) and extracted total RNA from the transfected cells . Using the same m6A-seq approach , we identified multiple m6A peaks in HIV-1 RNA , which are enriched in the 5' and 3' UTRs and within overlapped HIV-1 coding genes , such as tat , rev , env , and nef ( Figure 1—figure supplement 1 ) . These results confirm the m6A modification of HIV-1 RNA despite some differences in m6A distributions in HIV-1 infected CD4+ T-cells compared to transfected HEK293T cells . 10 . 7554/eLife . 15528 . 003Figure 1 . HIV-1 RNA contains m6A modifications and YTHDF1–3 proteins bind to m6A-modified HIV-1 RNA . ( A–B ) The distribution of m6A reads from m6A-seq mapped to HIV-1 genome ( red line ) in HIV-1 infected Jurkat cells ( A ) or primary CD4+ T-cells ( B ) . Baseline signal from the RNA-seq of input samples is shown as a black line . A schematic diagram of HIV-1NL4-3 genome is shown above . TAR , transacting response element; RRE , Rev response element . Jurkat cells ( A ) or primary CD4+ T-cells ( B ) were infected with HIV-1NL4-3 and total RNA was extracted for m6A-seq at 72 or 96 hr post-infection ( hpi ) , respectively . ( C ) YTHDF1-3 proteins bind to the HIV-1 gRNA . HeLa/CD4 cells overexpressing FLAG-tagged YTHDF1-3 proteins were infected with HIV-1NL4-3 ( MOI= 5 ) for 72 hr and used in CLIP-seq assay to identify their binding sites on HIV-1 gRNA . The distribution of mapped reads ( >16 nt ) with corresponding nucleotide positions are shown , forming peaks as putative binding positions . Asterisks mark the peak clusters overlapping with identified m6A peaks , indicating high-confident YTHDFs binding sites . Read density was normalized to the total number of mapped reads in each sample ( YTHDF1: 28438; YTHDF2: 232568; YHTDF3: 124915 ) . The data presented are representative of results from two independent experiments ( n=2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 00310 . 7554/eLife . 15528 . 004Figure 1—figure supplement 1 . HIV-1 RNA contains m6A modifications . HEK293 T cells were transfected with a proviral DNA-containing plasmid ( pNL4-3 ) . Total RNA was extracted at 48 hr post-transfection and immunoprecipitated with an m6A-specific antibody . Enriched RNA was subjected to next generation sequencing . Peaks show the relative abundance of m6A sites on the HIV-1 genome . The distribution of m6A reads from m6A-seq mapped to HIV-1 genome ( red line ) . Baseline signal from the RNA-seq of input samples is shown as a black line . A schematic diagram of HIV-1NL4-3 genome features is shown above . TAR , transacting response element; RRE , Rev response element . The data presented are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 00410 . 7554/eLife . 15528 . 005Figure 1—figure supplement 2 . Quantification of HIV-1 RNA m6A level using liquid chromatography-mass spectrometry . HIV-1 RNA ( 250 ng ) was isolated from highly purified HIV-1MN virions ( total 600 μg of p24 capsid ) and subjected to quantitative analysis of the m6A level using LC-MS/MS ( n=3 of each sample ) . The results are presented are from representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 00510 . 7554/eLife . 15528 . 006Figure 1—figure supplement 3 . Distribution of m6A in cellular RNAs and the frequency of m6A motifs in HIV-1-infected cells . ( A–B ) Pie charts show the distribution of m6A peaks in the 5′ UTR , coding DNA sequence ( CDS ) , 3′ UTR , and noncoding regions of transcripts from uninfected and HIV-1-infected Jurkat T-cells ( A ) or primary CD4+ T-cells ( B ) . The m6A peak distribution in HIV-1-specific RNAs is also shown . ( C–D ) Frequency of the RRACH motif ( C ) and the GGACU motif ( D ) within the m6A peaks in cellular RNAs from the uninfected control and HIV-1-infected cells . Data presented are the average results of duplicated samples ( n=2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 00610 . 7554/eLife . 15528 . 007Figure 1—figure supplement 4 . Gene ontology ( GO ) analysis of m6A-modified cellular genes in HIV-1 infected cells . ( A and B ) GO terms specific to virus related pathways and corresponding p values , clustered from methylated genes detected in Jurkat cells ( A ) or primary CD4+ T cells ( B ) infected with HIV-1 . ( C and D ) GO graphs showing functional clusters from genes with unique m6A peaks identified in HIV-1-infected Jurkat cells ( C ) or primary CD4+ T-cells ( D ) when compared to uninfected cells . Data presented are the average results of duplicated samples ( n=2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 007 To investigate whether HIV-1 virion RNA contains m6A , we isolated HIV-1 RNA from highly purified HIV-1 virions derived from infected CD4+ T-cells ( Rossio et al . , 1998; Wang et al . , 2008 ) , and then performed a quantitative analysis of m6A level using liquid chromatography-mass spectrometry ( Jia et al . , 2011 ) . Our data showed that m6A in HIV-1 RNA was approximately 0 . 1% of total adenosines ( Figure 1—figure supplement 2 ) . Considering 35 . 8% of HIV-1 genomic RNA ( gRNA ) ( 9173 nucleotides ) are adenosines ( van Hemert et al . , 2014 ) , our data suggest approximately 3–4 sites of the m6A modification in each copy of HIV-1 gRNA , which match the numbers of m6A peaks identified by m6A-seq ( Figure 1A , B and Figure 1—figure supplement 1 ) . Together , these results confirm that HIV-1 RNA contains m6A modifications at multiple sites within the viral genome . To examine the effect of HIV-1 infection on m6A modifications of cellular RNAs , we compared the distribution of m6A peaks in cellular RNAs from HIV-1 infected and uninfected T-cells . In Jurkat and primary CD4+ T-cells , HIV-1 infection did not significantly affect the percentages of total m6A peaks mapped to the human genome in the 5′ UTR , coding DNA sequence ( CDS ) , 3′ UTR and noncoding regions ( Figure 1—figure supplement 3A , B ) . The biological effects of the slightly altered m6A topology ( <1% ) remain to be investigated . To determine whether the preferential m6A motif usage in the host cells was altered by HIV-1 infection , we performed consensus sequence analyses within the m6A peaks to determine the preferred motifs in cellular RNAs . HIV-1 infection of Jurkat cells or primary CD4+ T cells slightly increased the frequency of the RRACH motif within the m6A peaks by 0 . 2–0 . 8% , but slightly decreased the GGACU motif frequency by 0 . 2–0 . 4% ( Figure 1—figure supplement 3C , D ) . These data suggest that the preferential usage of the RRACH motifs in m6A modification of cellular RNAs could be altered during HIV-1 infection . We also performed the GO analysis of m6A-modified cellular genes in HIV-1 infected Jurkat cells and primary CD4+ T cells and found numerous genes with known functions in viral infection-related pathways . We defined these genes as viral-specific genes and performed a separate analysis of the distribution and motif of methylation peaks on these genes ( Figure 1—figure supplement 4A , B ) . We have also performed an individual GO analysis of genes with unique methylation peaks in infected samples , the results showed that these genes enrich in functional clusters , such as metabolism , immune system process , multicellular organismal process , and development ( Figure 1—figure supplement 4C , D ) , indicating widespread impacts on host biological systems induced by HIV-1 infection . YTHDF1–3 proteins are reader proteins that specifically bind to m6A-methylated cellular RNAs ( Wang et al . , 2014 , 2015 ) . We utilized the crosslinking and immunoprecipitation ( CLIP ) assay combined with RNA-seq ( Hafner et al . , 2010; Liu et al . , 2015 ) to map the binding sites of YTHDF1–3 proteins in the HIV-1 genome in infected HeLa/CD4 cells that overexpressed individual FLAG-tagged YTHDF1–3 proteins . We identified multiple CLIP peaks of YTHDF1–3 protein-bound HIV-1 RNA , including the transactivation response element ( TAR ) in the 5’ UTR leader sequence , the env gene , the rev gene , and the 3’ UTR . Some YTHDF-binding sites ( e . g . at the 3’ and 5’ UTR and the gag gene ) in HIV-1NL4-3 infected HeLa/CD4 cells partially overlap with the identified m6A-containing regions in the HIV-1 genome in HIV-1NL4-3 infected CD4+ Jurkat T-cells or primary CD4+ T-cells ( marked by asterisks ) , indicating high-confident YTHDF1-3 binding sites . Different cell types used in these experiments might contribute to the difference of the m6A sites and YTHDFs-bound sites in HIV-1 RNA genome . Overall , these data demonstrate that YTHDF1-3 proteins bind to m6A-modified HIV-1 genomic RNA during viral infection . 10 . 7554/eLife . 15528 . 008Figure 2 . YTHDF1–3 proteins negatively regulate post-entry HIV-1 infection in HeLa cells . ( A–B ) Overexpression of YTHDF1–3 proteins in HeLa cells significantly inhibits HIV-1 infection compared to vector control cells . ( A ) Overexpression of YTHDF1–3 proteins in HeLa cells was confirmed by immunoblotting . ( B ) HeLa cells overexpressing YTHDF1–3 proteins were infected with HIV-1 Luc/VSV-G at an MOI of 0 . 5 and viral infection was measured by luciferase activity at 24 hpi . ( C ) Overexpression of YTHDF1–3 proteins inhibits HIV-1 Gag protein synthesis in infected cells . HeLa cells overexpressing individual YTHDF1–3 proteins or the vector control cells were infected by HIV-1-Luc/VSV-G at an MOI of 0 . 5 . At 24 hpi , the expression of HIV-1 Gag and YTHDF1–3 proteins ( FLAG-tagged ) was determined using immunoblotting . GAPDH was used as a loading control and mock-infected vector control cells were used as a negative control . ( D and E ) Individual knockdown of endogenous YTHDF1–3 proteins in HeLa cells significantly increases HIV-1 infection compared to vector control cells . HIV-1 infection assays were performed as described for panel B . p<0 . 05 , p<0 . 005 , and *p<0 . 0005 , compared to vector control without AZT treatment . All results are shown as mean ± SD ( n=3 ) and data presented are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 008 Because HIV-1 RNAs are present in the cytoplasm and the nucleus of infected cells at different stages of viral lifecycle ( Goff , 2007 ) , we hypothesized that YTHDF1–3 proteins may directly interact with methylated HIV-1 RNAs , thereby affecting the metabolism and/or processing of the viral RNA . To examine the roles of YTHDF1–3 proteins in post-entry HIV-1 infection , we either overexpressed or knocked down the individual YTHDF proteins in human cell lines , and examined the effect on HIV-1 infection using a single-cycle , luciferase reporter HIV-1 pseudotyped with vesicular stomatitis virus G protein ( VSV-G ) to overcome the requirement of HIV-1 receptors during viral entry ( Wang et al . , 2007 ) . Compared to vector control cells , at 24 hr post-infection ( hpi ) , overexpression of individual FLAG-tagged YTHDF1–3 proteins in HeLa cells ( Figure 2A ) significantly inhibited HIV-1 infection by approximately 10-fold ( Figure 2B , p<0 . 0005 ) and drastically reduced the synthesis of full-length viral Gag protein ( Pr55 ) ( Figure 2C ) . In contrast , stable knockdown of individual , endogenous YTHDF1–3 proteins in HeLa cells ( Figure 2D ) significantly increased HIV-1 infection by four- to 14-fold ( p<0 . 05 ) relative to control cells ( Figure 2E ) . Overexpression or knockdown of YTHDF1–3 proteins in HeLa cells did not affect cell proliferation ( data not shown ) . To confirm these observations in CD4+ T-cells , we generated Jurkat cell lines with knockdown of individual , endogenous YTHDF1–3 proteins ( Figure 3A ) and did not observe a significant change in proliferation of the knockdown cells relative to parental or vector-control Jurkat cells ( Figure 3B ) . The partial knockdown of YTHDF1 or YTHDF3 in Jurkat cells increased HIV-1 infection by three- to four-fold ( p<0 . 005 ) , while YTHDF2 knockdown slightly increased viral infection ( Figure 3A and C ) at 24 hpi . Furthermore , knockdown of individual , endogenous YTHDF1–3 proteins in activated primary CD4+ T-cells from healthy donors enhanced HIV-1 infection by approximately two-fold ( p<0 . 005 ) ( Figure 3D and E ) , confirming the effects observed in cell lines despite a lesser extent . The treatment of cells with the HIV-1 reverse transcriptase inhibitor azidothymidine ( AZT ) was used as a negative control to show the expected HIV-1 inhibition ( Figures 2B , E , 3C and E ) . Overall , these data suggest that overexpression of YTHDF1–3 proteins significantly inhibits HIV-1 infection , while knockdown of these proteins efficiently promotes HIV-1 gene expression . Thus , endogenous YTHDF1–3 proteins in CD4+ T-cells act as negative regulators to inhibit post-entry HIV-1 infection . 10 . 7554/eLife . 15528 . 009Figure 3 . YTHDF1–3 proteins negatively regulate post-entry HIV-1 infection in CD4+ T-cells . ( A ) Individual knockdown of endogenous YTHDF1–3 proteins in Jurkat CD4+ T cells was confirmed by immunoblotting . ( B ) Knockdown of YTHDF1–3 proteins does not affect proliferation of Jurkat cells . Jurkat cells ( 2 × 104 ) were seeded and cultured for 3 days . At the times indicated , cell proliferation was measured using the MTS assay . ( C ) Knockdown of YTHDF1–3 proteins significantly increases HIV-1 infection compared to vector control cells . ( D ) Individual knockdown of YTHDF1–3 proteins in activated primary CD4+ T-cells from a healthy donor . ( E ) Knockdown of YTHDF1–3 proteins significantly increases HIV-1 infection compared to vector control cells . ( A and D ) GAPDH was used as a loading control . ( C and E ) The vector controls without AZT were set as 100% . The reverse transcriptase inhibitor AZT treated cells were used as positive control for productive HIV-1 infection . p<0 . 05 , p<0 . 005 , and p<0 . 0005 , compared to vector control without AZT treatment . All results are shown as mean ± SD ( n=3 ) and data presented are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 009 To investigate the mechanisms of HIV-1 inhibition by YTHDF1–3 proteins , we assessed the stage of HIV-1 life cycle affected by the YTHDF1–3 proteins . We first measured levels of HIV-1 late reverse transcription ( RT ) products in infected cells , which represent the levels of the full-length viral cDNA ( St Gelais et al . , 2015 ) . Overexpression of each of the YTHDF1–3 proteins in HeLa cells significantly reduced the level of HIV-1 late RT products by four- to ten-fold ( p<0 . 005 ) compared to the vector control cells at 24 hpi ( Figure 4A ) , suggesting that the inhibition of viral reverse transcription contributes to the impairment of post-entry HIV-1 infection . In contrast , the knockdown of individual YTHDF1–3 proteins in HeLa cells elevated the levels of HIV-1 late RT products by two- to three-fold compared to vector control cells ( Figure 4B ) . Furthermore , the level of HIV-1 2-LTR circles in infected HeLa cells , a surrogate marker of nuclear import of viral cDNA ( Dong et al . , 2007 ) , was also significantly reduced over 10-fold ( p<0 . 05 ) by overexpression of YTHDF1–3 proteins ( Figure 4C ) , corresponding to the reduced late RT products observed in this experiment . Using our established Jurkat cell lines with stable knockdown of individual YTHDF1–3 proteins ( Figure 3A ) , we found that the levels of HIV-1 late RT products were significantly increased in YTHDF1 down-regulated cells by 2 . 7-fold ( p<0 . 05 ) compared to control cells , while the knockdown of YTHDF2 or YTHDF3 only increased late RT products by 20–30% ( Figure 4D ) . As a negative control , AZT-treated cells showed inhibition of HIV-1 post-entry infection as expected ( Figure 4A–D ) . 10 . 7554/eLife . 15528 . 010Figure 4 . YTHDF1–3 proteins inhibit post-entry HIV-1 infection by blocking viral reverse transcription . HeLa cells over-expressing or knocking-down ( shRNA ) individual YTHDF1–3 proteins were infected with HIV-1-Luc/VSV-G at an MOI of 0 . 5 . ( A , B and D ) Genomic DNA was isolated from the cells 24 hr post-infection and HIV-1 late reverse transcription ( RT ) products were quantified by qPCR . ( C ) YTHDF family proteins reduce the formation of HIV-1 2-LTR circles in infected HeLa cells . At 24 hr post-infection , DNA was isolated from the cells and the 2-LTR circles were analyzed by qPCR and normalized to GAPDH levels . AZT treated vector control cells were used as a negative control for HIV-1 inhibition . p<0 . 05 compared to the vector control without AZT treatment . All results are shown as mean ± SD ( n=3 ) and data presented are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 01010 . 7554/eLife . 15528 . 011Figure 4—figure supplement 1 . YTHDF1–3 proteins negatively regulate HIV-1 gag mRNA expression . Specific shRNAs or scrambled shRNA vector-treated cells were infected with HIV-1 Luc/VSV-G at an MOI of 0 . 5 . Total RNA was isolated from the cells 24 hr post-infection and HIV-1 gag mRNA levels were quantified using qRT-PCR . ( A and B ) HIV-1 gag mRNA levels in the infected HeLa cells with overexpression ( A ) or knockdown ( B , shRNA ) of YTHDF1–3 proteins . ( C ) HIV-1 gag mRNA levels in the HIV-1 infected Jurkat cells after knockdown of YTHDF1–3 proteins . AZT treated vector control cells were used as a negative control of HIV-1 infection ( A–C ) . The vector controls without AZT were set as 100% . p<0 . 05 , p<0 . 005 , and p<0 . 0005 , compared to vector control without AZT treatment . All results are shown as mean ± SD ( n=3 ) and data presented are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 011 The effects on HIV-1 late reverse transcription mediated by YTHDF1–3 would lead to altered viral gene expression . To examine the impacts of the YTHDF proteins on viral gene expression , we quantified HIV-1 gag mRNA in infected cells at 24 hpi . The HIV-1 gag mRNA level in HeLa cells with overexpression of individual YTHDF1–3 showed a four-fold reduction ( p<0 . 0005 ) compared to vector control cells ( Figure 4—figure supplement 1A ) . In contrast , the knockdown of individual YTHDF1–3 in HeLa cells increased the level of gag mRNA by eight- to 12-fold ( p<0 . 05 ) compared to vector control cells ( Figure 4—figure supplement 1B ) . The knockdown of YTHDF3 in Jurkat cells significantly increased the level of HIV-1 gag mRNA by two-fold ( p<0 . 05 ) compared to control cells , while the knockdown of YTHDF1 or YTHDF2 did not have a significant effect ( Figure 4—figure supplement 1C ) . The different effects of YTHDF1–3 silencing on gag mRNA expression in HeLa and Jurkat cells might result from the difference in the knockdown efficiency in these cells ( Figures 2D and 3A ) . These data suggest that YTHDF1–3 proteins could negatively regulate HIV-1 mRNA transcription , in addition to inhibiting viral reverse transcription in HIV-1 infected cells . We hypothesize that YTHDF1–3 proteins could inhibit the reverse transcription of HIV-1 gRNA through directly binding to the gRNA . To test this hypothesis , we used a single-cycle , VSV-G-pseudotyped HIV-1 to infect HeLa cells overexpressing individual YTHDF1–3 proteins or vector control cells , immunoprecipitated YTHDF proteins from the infected cells at 3 hpi , and then measured HIV-1 gRNA levels in the IP samples . The presence of YTHDF proteins was confirmed in the input and IP samples ( Figure 5A ) . The quantification of the HIV-1 gRNA by qRT-PCR revealed a strong and specific association ( p<0 . 005 ) of HIV-1 gRNA with YTHDF proteins in HIV-1-infected YTHDF1–3-expressing cells compared to control cells ( Figure 5B ) . To examine the impact of YTHDF1–3 on HIV-1 gag RNA kinetics , we quantified HIV-1 gag RNA levels in YTHDF1–3-expressing HeLa cells and vector control cells over a time course of 6–24 hpi . The relative levels of gag RNA in HIV-1 infected cells were normalized to that of the vector control cells at 6 hpi . In the control cells , compared to 6 hpi ( set as 100% ) , the level of gag RNA was reduced to 40% at 12 hpi and then increased to 80% at 24 hpi ( Figure 5C ) , suggesting degradation of HIV-1 gRNA at 12 hpi during the reverse transcription and then increased gag mRNA at 24 hpi during viral gene transcription . In contrast , the levels of gag RNA in YTHDF1–3-expressing cells were reduced to 40% at 12 hpi and to 13–25% at 24 hpi ( p<0 . 0005 ) compared to that of the vector control cells at 6 hpi ( Figure 5C ) . These data suggest that YTHDF1–3 proteins can degrade HIV-1 gag RNA in infected cells , thereby leading to inhibition of HIV-1 reverse transcription . 10 . 7554/eLife . 15528 . 012Figure 5 . YTHDF1–3 proteins bind to HIV-1 gRNA in infected cells . ( A ) Immunoblotting of YTHDF1–3 proteins in the input and immunoprecipitation ( IP ) samples from HIV-1-Luc/VSV-G infected HeLa cells . FLAG antibodies were used to immunoprecipitate FLAG-tagged YTHDF1–3 proteins overexpressed in HeLa cells after HIV-1 infection . A short and long exposure of the immunoblot is shown . ( B ) HIV-1 gRNA is bound by YTHDF1–3 proteins expressed in HeLa cells . HeLa cells stably overexpressing FLAG-tagged YTHDF1–3 proteins or empty vector control cells ( Ctrl ) were infected with HIV-1-Luc/VSV-G at an MOI of 5 for 3 hr . Cell lysates were immunoprecipitated with anti-FLAG , RNA was extracted and HIV-1 gag RNA levels were measured . p<0 . 005 compared to the vector control cells . ( C ) YTHDF1–3 affect HIV-1 gag RNA kinetics . HIV-1 gag RNA levels in YTHDF1–3-expressing HeLa cells were quantified by qRT-PCR . The relative levels of gag RNA in infected cells were normalized to that of the vector control cells at 6 hr post-infection ( hpi ) . **p<0 . 0005 , compared to the control cells at 6 hpi ( set as 100% ) . All results are shown as mean ± SD ( n=3 ) and data presented are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 012 We examined the role of the m6A writers in HIV-1 protein expression and viral release in virus-producing cells . We knocked down endogenous METTL3 , METTL14 , or both , in HEK293T cells using siRNA , and then transfected the cells with an HIV-1 proviral DNA plasmid ( pNL4-3 ) . We determined the levels of HIV-1 Gag protein expression in the cells and the capsid p24 protein released in the supernatants . Interestingly , we found that partial knockdown of METTL3 , METTL14 , or both , inhibited HIV-1 Gag expression in the cells by 60–70% ( Figure 6A ) , and reduced the levels of HIV-1 p24 release by 30–50% compared to control cells ( Figure 6B ) . These results suggest that the m6A writers are required for efficient HIV-1 protein synthesis , and that m6A modification of HIV-1 RNA could facilitate translation of viral proteins . 10 . 7554/eLife . 15528 . 013Figure 6 . The m6A writers and erasers affect HIV-1 Gag expression in virus-producing cells . ( A and B ) Individual or combined knockdown of endogenous METTL3 and METTL14 inhibits HIV-1 Gag protein expression . HEK293T cells were transfected with indicated siRNA , and then with an HIV-1 proviral DNA plasmid ( pNL4-3 ) . Cells and supernatants were collected for analyses at 36 hr post-transfection . ( A ) Expression of METTL3 , METTL14 and HIV-1 Gag proteins in the transfected HEK293T cells was detected by immunoblotting . ( C and D ) Knockdown of endogenous AlkBH5 , FTO , or both promotes HIV-1 Gag protein expression . HEK293T cells were transfected with indicated siRNA , and then with pNL4-3 . Cells and supernatants were collected at 36 hr post-transfection . ( C ) Expression of AlkBH5 , FTO and HIV-1 Gag proteins in the cells was detected by immunoblotting . ( A and C ) GAPDH was used as a loading control . Relative levels of Gag expression were normalized to GAPDH levels . ( B and D ) HIV-1 capsid p24 levels in supernatants were measured by ELISA . The relative levels ( % ) are also shown . p<0 . 05 compared to the siRNA control . The results are shown as mean ± SD ( n=3 ) and data presented are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 013 We next examined the role of the m6A erasers in HIV-1 protein expression and viral release in virus-producing cells . We knocked down endogenous AlkBH5 , FTO , or both , in HEK293T cells using siRNA , and then transfected the cells with the pNL4-3 plasmid . We determined the levels of HIV-1 Gag protein expression in the cells and the capsid p24 protein released in the supernatants . Interestingly , we found that partial knockdown of FTO significantly promoted HIV-1 Gag synthesis in the cells by 2 . 5- to 6 . 5-fold ( Figure 6C ) , and increased the levels of HIV-1 p24 release by two- to three-fold compared to control cells ( Figure 6D ) . Thus , the m6A modification of HIV-1 RNA can enhance HIV-1 protein synthesis . Our results are in agreement with a recent report ( Lichinchi et al . , 2016 ) showing that silencing of the m6A writers ( METTL3 and METTL14 ) or the eraser AlkBH5 decreases or increases HIV-1 p24 expression in the infected MT4 cells , respectively . The m6A modification of cellular mRNAs is coordinately regulated by the writers , erasers , and readers to control the metabolism and processing of methylated RNA ( Fu et al . , 2014 ) . We found that HIV-1 RNA is m6A-methylated in infected cells , and that binding of YTHDF1–3 proteins to m6A-methylated HIV-1 RNA inhibits viral reverse transcription and gene expression . In contrast , partial knockdown of the m6A writers decreased HIV-1 Gag synthesis and viral release , while partial knockdown of FTO had the opposite effects , suggesting that m6A modification of HIV-1 RNA could enhance HIV-1 protein synthesis and viral release . Based on our results , we propose a working model suggesting that YTHDF proteins inhibit post-entry HIV-1 infection by blocking viral reverse transcription and mRNA expression , while the m6A modification of HIV-1 RNA can promote viral protein translation ( Figure 7 ) . 10 . 7554/eLife . 15528 . 014Figure 7 . Proposed mechanisms and dynamics of m6A modification of HIV-1 RNA in regulating viral infection in cells . In the nucleus , the m6A writers ( METTL3 and METTL14 ) add the m6A marker to HIV-1 genomic RNA ( gRNA ) or mRNA , and the m6A erasers ( FTO and AlkBH5 ) remove the m6A modifications of HIV-1 RNA . The m6A modification of HIV-1 RNA can promote viral protein translation in cells . In contrast , cytoplasmic m6A readers ( YTHDF1–3 ) bind to m6A-modified HIV-1 gRNA , which can result in inhibition of HIV-1 reverse transcription ( RT ) , viral mRNA expression , and thereby HIV-1 infection in cells . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 014 It is possible that YTHDF protein-mediated inhibition of HIV-1 infection can result from indirect effects on cellular RNA stability or gene expression , rather than direct inhibition of HIV-1 replication . We noticed a differential level of the effect on HIV-1 infection in different cell types by manipulating the individual YTHDF1–3 proteins . It is possible that the different effects are due to distinct cellular functions and mechanisms of YTHDF proteins ( Fu et al . , 2014 ) . Recent studies indicated that YTHDF1 is responsible for translation promotion , and that YTHDF2 is responsible for mRNA decay , while the function YTHDF3 is unclear , but likely to aid in the temporal-spatial transport and delivery of mRNA ( Wang and He , 2014; Wang et al . , 2014 , 2015 ) . Indeed , we observed a unique high peak within the rev gene of HIV-1 RNA bound by YTHDF1 ( Figure 1C ) , while YTHDF1 and YTHDF2 appear to have similar inhibitory effects on HIV-1 infection . It is possible that YTHDF1 and YTHDF2 may interact with different host proteins that directly or indirectly affect HIV-1 replication and lead to similar effects on viral inhibition . Furthermore , these three YTHDF proteins may have functional redundancy , and individual knockdown of one YTHDF protein may result in a modest effect because the other two could compensate the function . A recent study suggests that the dynamic m6A modification of cellular mRNA is a result of stress-induced nuclear localization and upregulation of YTHDF2 ( Zhou et al . , 2015 ) . It is conceivable that HIV-1 infection of cells may induce the changes of cellular localization and expression levels of YTHDF proteins , thereby affecting HIV-1 RNA replication and viral infection . The mechanisms by which m6A modification of HIV-1 RNA regulates viral infection remain to be elucidated . Lichinchi et al . recently showed that m6A modification of a conserved adenosine ( A7883 ) in the stem loop II region of HIV-1 Rev response element ( RRE ) RNA increased binding of HIV-1 Rev protein to the RRE and facilitated nuclear export of RNA , thereby enhancing HIV-1 replication ( Lichinchi et al . , 2016 ) . However , other m6A sites in the HIV-1 genome ( such as in the gag , pol , tat , and rev genes ) might also be critical for viral replication . Future studies using targeted mutations of the identified m6A sites in HIV-1 genome may lead to a loss of the effects on HIV-1 infection by YTHDF1–3 knockdown or overexpression . During the revision of this manuscript , Kennedy et al . reported that m6A modification of HIV-1 mRNAs enhances viral replication and gene expression ( Kennedy et al . , 2016 ) . Our data of the inhibitory effects on HIV-1 infection by YTHDF1–3 proteins are different from the results recently reported by Kennedy et al . ( Kennedy et al . , 2016 ) . They showed that overexpression of YTHDF1–3 proteins in HEK293T cells enhanced HIV-1 mRNA and protein expression in a single-cycle infection , and that overexpression or knockout of YTHDF2 in CEM-SS T-cells increased or decreased HIV-1 replication and protein expression , respectively ( Kennedy et al . , 2016 ) . The mapping of m6A sites in HIV-1 RNA between the previous data ( Kennedy et al . , 2016; Lichinchi et al . , 2016 ) and the results presented in this manuscript are also different . Different approaches , cell lines and other reagents used in these studies may contribute to distinct results obtained , which remain to be clarified in the future . However , we report here consistent m6A mapping results using Jurkat T-cells and primary CD4+ T-cells and systematic evaluation of the roles of the m6A writers , readers , and erasers in HIV-1 infection . Chemical modifications of viral RNAs such as m6A may protect viral genomes or mRNA from recognition by cellular innate immunity proteins ( Fu et al . , 2014 ) . The m6A modification of HIV-1 RNA could serve as an immune evasion strategy , wherein the virus may escape detection by innate immunity against infection . In response to HIV-1 infection , the host may evolve and utilize the YTHDF1–3 proteins to bind the m6A-modified viral RNA and inhibit its reverse transcription and subsequent viral mRNA expression . Our findings can stimulate further studies on the precise role of m6A in HIV-1 RNA replication and the mechanisms of m6A regulation pathways that impact HIV-1 infection in cells . Because m6A-modified RNA has also been found in other viruses ( Beemon and Keith , 1977; Canaani et al . , 1979; Hashimoto and Green , 1976; Krug et al . , 1976 ) , this modification pathway may represent a novel and conserved target for antiviral development . Human healthy primary CD4+ T-cells were isolated from healthy blood donors’ buffy coats ( purchased from American Red Cross Blood Service , Columbus , OH ) using anti-human CD4-coated magnetic particles according to the manufacturer’s instructions ( BD Biosciences , San Jose , CA ) as described ( St Gelais et al . , 2014 , 2015 ) . Isolated CD4+ T cells were maintained in complete RPMI media containing interleukin-2 ( 20 U/ml , PeproTech , Rocky Hill , NJ ) and activated with phytohemagglutinin A ( PHA , Sigma-Aldrich , St . Louis , MO ) as described ( St Gelais et al . , 2014 ) . HEK293T cells , Jurkat cells , and the HIV-1 indicator cell line GHOST/X4/R5 were cultured as described ( St Gelais et al . , 2014 , 2015 ) . HeLa cells overexpressing the empty vector ( pPB-CAG ) , YTHDF1 , YTHDF2 , or YTHDF3 were maintained in complete DMEM containing 2 µg/ml of puromycin . All parental cell lines were obtained from the American Type Culture Collection ( ATCC , Manassas , VA ) and the identity of the cell lines has been authenticated using short tandem repeat profiling or genotyping methods as described ( Wu et al . , 2004 ) . All the cell lines were tested negative for mycoplasma contamination using a PCR-based universal mycoplasma detection kit ( ATCC ) . Cell proliferation of HeLa cells or Jurkat cells with YTHDF1–3 overexpression or knockdown respectively were determined by the MTS assay ( Promega , Madison , WI ) as described ( St Gelais et al . , 2015 ) . Cells ( 2 × 104 ) were plated in triplicate in a 96-well plate and cultured for 3 days and the absorbance was read at 490 nm at the indicated times . HIV-1 proviral DNA vector pNL4-3 , pNL-Luc-E−R+ containing a firefly luciferase reporter gene and the empty vector control were described ( de Silva et al . , 2012 ) . The pPB-CAG plasmid vector was used to overexpress the YTHDF1–3 proteins in HeLa cells . pLenti vectors carrying specific YTHDF1–3 shRNAs ( Table 1 ) were used to knockdown of YTHDF1–3 proteins in different cell types as described ( St Gelais et al . , 2015 ) . Jurkat cells transduced with lentivirus containing shRNAs specific for YTHDF1 , YTHDF2 , and YTHDF3 were maintained in puromycin ( 3 µg/ml ) containing complete RPMI media . AlkBH5 , FTO , METTL3 and METTL14 gene expression in HEK293T cells was silenced by two rounds of siRNA transfection using specific siRNA ( Qiagen , Valencia , CA , sequences listed in Table 2 ) transfected with the Lipofectamin RNAiMax reagent ( Invitrogen , Waltham , MA ) according to the manufacturer protocol ( reversible siRNA transfection method ) . Briefly , HEK293T cells ( 1 . 5 × 105 ) were transfected with specific siRNA or a non-specific control ( 80 nM ) . At 24 hr post-transfection , media were replaced and the second round of siRNA transfection was performed using the same siRNA concentration ( 80 nM ) . The pNL4-3 construct ( 0 . 5 μg ) was transfected into the cells ( 1 . 5 × 105 ) 6 hr after of the second round transfection and cells were harvested for immunoblotting 36 hr after the proviral DNA transfection . 10 . 7554/eLife . 15528 . 015Table 1 . The shRNA sequences used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 015shRNASequences ( 5’-3’ ) Non-specific ( vector ) controlCCGGCAACAAGATGAAGAGCACCAACTCGAGTTGGTGCTCTTCATCTTGTTGTTTTTYTHDF1CCGGCCCGAAAGAGTTTGAGTGGAACTCGAGTTCCACTCAAACTCTTTCGGGTTTTTGYTHDF2CCGGCGGTCCATTAATAACTATAACCTCGAGGTTATAGTTATTAATGGACCGTTTTTGYTHDF3CCGGGATAAGTGGAAGGGCAAATTTCTCGAGAAATTTGCCCTTCCACTTATCTTTTTG10 . 7554/eLife . 15528 . 016Table 2 . The siRNA sequences used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 016siRNASequences ( 5’-3’ ) METTL35’-CTGCAAGTATGTTCACTATGA-3’ 5’-AGGAGCCAGCCAAGAAATCAA-3’METTL145’-TGGTGCCGTGTTAAATAGCAA-3’ 5’-AAGGATGAGTTAATAGCTAAA-3’FTO5’-AAATAGCCGCTGCTTGTGAGA-3’AlkBH55’-AAACAAGTACTTCTTCGGCGA-3’ Single-cycle , luciferase reporter HIV-1 stock ( HIV-Luc/VSV-G ) was generated by calcium phosphate co-transfection of HEK 293T cells with the pNL-Luc-E–R+ and pVSV-G as described ( St Gelais et al . , 2014 ) . The infectious units of virus stocks were evaluated by limiting dilution on GHOST/X4/R5 cells as described ( Wang et al . , 2007 ) . HIV-1 infection assays using luciferase reporter viruses were performed using a multiplicity of infection ( MOI ) of 0 . 5 as described ( de Silva et al . , 2012 ) . Cell lysates were obtained 24 hpi and analyzed for luciferase activity using a commercially available kit ( Promega ) according to the manufacturer’s instructions . Total cell protein was quantified using a bicinchoninic acid assay ( BCA; Pierce , Waltham , MA ) and all luciferase results were normalized to total protein amounts . HIV-1 capsid p24 levels in supernatants were measured by an enzyme-linked immunosorbent assay ( ELISA ) using anti-p24-coated plates ( AIDS and Cancer Virus Program , NCI-Frederick , MD ) as described ( Wang et al . , 2007 ) . Jurkat cells were infected with replication-competent HIV-1NL4-3 at an MOI of 0 . 5 as described ( Wang et al . , 2007 ) . At 72 hpi , cells were washed 3 times and harvested for total RNA extraction using RNAeasy kit ( Qiagen ) . PHA-activated primary CD4+ T cells were infected with HIV-1NL4-3 ( 40 ng p24 equivalent HIV-1 per 106 cells ) and cells were harvested at 96 hpi for total RNA extraction using RNAeasy kit ( Qiagen ) . High-throughput sequencing of HIV-1 methylome was carried out using m6A-seq ( Dominissini et al . , 2012 ) and followed the protocol published previously ( Dominissini et al . , 2012 ) . In brief , total RNA containing HIV-1 RNA was extracted from the cells and purified by poly ( dT ) beads . Purified polyadenylated RNA was mixed with 2 . 5 μg of affinity purified anti-m6A polyclonal antibody ( 202003; Synaptic Systems , Goettingen , Germany ) in IPP buffer ( 150 mM NaCl , 0 . 1% NP-40 , 10 mM Tris-HCl , pH 7 . 4 ) and incubated for 2 hr at 4°C . RNA was used for library generation with the small RNA sequencing kit ( NEB , Ipswich , MA ) . Sequencing was carried out on Illumina HiSeq 2000 according to the manufacturer’s instructions . HIV-1 gRNA ( 250 μg ) was isolated from highly purified HIV-1MN virions ( total p24 capsid 600 μg ) ( Rossio et al . , 1998; Wang et al . , 2008 ) using an RNeasy Mini kit ( Invitrogen ) , and subjected to quantitative analysis of m6A level using LC-MS/MS as described ( Jia et al . , 2011 ) . Quantitative PCR ( qPCR ) was performed to assess the relative levels of HIV-1 late reverse transcription ( RT ) products and 2-LTR circles as described ( de Silva et al . , 2012; Dong et al . , 2007 ) . Reverse transcription PCR ( RT-PCR ) was used to measure HIV-1 gag mRNA as described ( Dong et al . , 2007 ) . To amplify HIV-1 late RT products in cells transduced with pLenti vectors , a different set of primers ( LW59 and LW60 ) were used as described ( St Gelais et al . , 2015 ) . Sequences of PCR primers and probes are listed in Table 3 . All HIV-1 stocks used for PCR assays were treated with DNaseI ( 40 U/ml; Ambion , Waltham , MA ) prior to infections to avoid plasmid DNA contamination . 10 . 7554/eLife . 15528 . 017Table 3 . Sequences of PCR primers and probes used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 15528 . 017PrimersSequences ( 5’-3’ ) HIV-1 gag forwardCTAGAACGATTCGCAGTTAATCCTHIV-1 gag reverseCTATCCTTTGATGCACACAATAGAGUnspliced GAPDH forwardGGGAAGCTCAAGGGAGATAAAATTCUnspliced GAPDH reverseGTAGTTGAGGTCAATGAAGGGGTCSpliced GAPDH forwardGGAAGGTGAAGGTCGGAGTCAACGGSpliced GAPDH reverseCTGTTGTCATACTTCTCATGGTTCACMH531 forward ( for HIV-1 late reverse transcription ( RT ) products ) TGTGTGCCCGTCTGTTGTGTBB reverse ( for late RT products ) GGATTAACTGCGAATCGTTCHIV-1 late RT product probeTCGACGCAGGACTCGGCTTGCT2-LTR probeAAGTAGTGTGTGCCCGTCTGTTGTGTGACTC2-LTR forwardGCCTGGGAGCTCTCTGGCTAA2-LTR reverseGCCTTGTGTGTGGTAGATCCALW59 ( forward , alternative for late RT detection in shRNA vector-transduced cells ) GACATAGCAGGAACTACTAGTACCCLW60 ( reverse , alternative for late RT detection in shRNA vector-transduced cells ) GGTCCTTGTCTTATGTCCAGAATGC The antibodies used in this study were: anti-GAPDH ( clone 4G5 , AbD serotec , Atlanta , GA ) , anti-FLAG ( F-3165 , Sigma-Aldrich ) , anti-FTO ( ab124892 , Abcam , Cambridge , MA ) , anti- AlkBH5 ( HPA007196 , Sigma-Aldrich ) , anti-METTL3 ( 15073-1-AP , Proteintech Group , Rosemont , IL ) , anti-METTL14 ( HPA038002 , Sigma-Aldrich ) , anti-YTHDF1 ( ab99080; Abcam ) , anti-YTHDF2 ( ABE542 , EMD Millipore , Billerica , MA ) , anti-YTHDF3 ( ab103328; Abcam ) , and anti-HIV-1 Gag ( clone #24–2 , the NIH AIDS Reagent Program ) . Cells were harvested and lysed in cell lysis buffer ( Cell Signalling , Beverly , MA ) supplemented with protease inhibitor cocktails ( Sigma-Aldrich ) . Immunoblotting was performed as described ( St Gelais et al . , 2015 ) . Detection of GAPDH ( glyceraldehyde-3-phosphate dehydrogenase ) expression was used as a loading control . HeLa cells expressing pPB-CAG vector or YTHDF1–3 ( 2 . 5 × 106 cells ) were seeded in a 60 mm-diameter culture plate . Cells were infected with HIV-Luc/VSV-G at an MOI of 5 for 3 hr . Cells were UV-cross-linked , lysed in cell lysis buffer ( Sigma-Aldrich ) . The cells were incubated in lysis buffer for 10 min on ice with frequent mixing and were sonicated to ensure maximum lysis . The lysed cell suspension was centrifuged for 5 min at 9 , 300 ×g at 4°C . The supernatant was transferred to fresh tubes and equal amount of cell lysates were mixed with anti-FLAG-conjugated protein G beads and rotated for 2 hr at 4°C . After the incubation , beads were washed 3 times with cell lysis buffer . Co-immunoprecipitated RNA was isolated from the immunoprecipitates using Trizol ( Invitrogen ) , and RNeasy columns ( Qiagen ) with an on-column DNase I treatment ( Qiagen ) and eluted with RNase-free water . Equal volumes of RNA were used as a template for first-strand cDNA synthesis , according to the manufacturer’s instructions . We followed a previously reported protocol of the PAR ( photoactivatable ribonucleoside-enhanced ) -CLIP assay ( Hafner et al . , 2010 ) with the following modifications . As HIV-1 infection was inhibited by the addition of 4-thiouridine ( data not shown ) , we omitted that step and performed crosslinking directly . Briefly , HeLa/CD4 cells stably expressing individual YTHDF1-3 proteins were seeded in thirty 15-cm diameter plates one day before HIV-1NL4-3 infection . At day 2 , the cells were infected with HIV-1NL4-3 at a multiplicity of infection ( MOI ) of 5 and cells were washed to remove cell-free viruses . At 72 hr post-infection , the cells were washed once with 10 mL ice-cold PBS . Uncovered cell plates were placed on a tray with ice and irradiated with 0 . 15 J/cm2 of 254 nm UV light three times in a Stratalinker 2400 ( Stratagene , Santa Clara , CA ) . Cells were scraped off in PBS and transferred to centrifugation tubes and collected by centrifugation at 500 × g for 5 min at 4°C . The cell pellets were lysed in 3 volumes of 1% NP40 lysis buffer and incubated on ice for 10 min . The cell lysates were cleared by centrifugation at 13 , 000 × g for 15 min at 4°C . Cleared cell lysates were incubated with RNase T1 to a final concentration of 0 . 2 U/μl , at 22°C for 15 min , and immediately put on ice for 5 min to quench . Samples were then centrifuged at 13 , 000 × g for 10 min at 4°C and the supernatant was taken . Anti-FLAG M2 magnetic beads ( Sigma M8823 , 20 μl slurry/15-cm diameter plate ) were washed in IP buffer ( 50 mM HEPES ( pH 7 . 5 ) , 0 . 3 M KCl , 0 . 05% NP40 ) for 5 times and resuspended in cleared cell lysates and incubated at 4°C overhead rotator for 2 hr . After 2 hr , beads were washed with 1 mL IP buffer for 3 times and beads were changed into a new tube for the final wash and resuspended into 100 μl IP buffer . RNAse T1 was added to the beads at a final concentration of 10 U/μl , incubated at 22°C for 6 min , immediately put on ice for 5 min to quench . Beads were washed with 1 mL high salt buffer ( 50 mM Tris ( pH 7 . 5 ) , 500 mM KCl , 0 . 05% NP40 ) five times , then with 1 mL T4 polynucleotide kinase ( PNK ) buffer containing 50 mM Tris ( pH 7 . 5 ) , 10 mM MgCl2 , 50 mM NaCl ( without DTT ) two times , finally resuspended into 100 μl PNK reaction mix ( 95 μl commercial PNK buffer and 5 μl T4 PNK ( Promega ) and incubated at 37°C for 15 min . Then 10 μl PNK and 1 . 1 μl 10 mM ATP ( to a final concentration of 100 μM ) were added to the mixture and kept at 37°C for another 20 min , followed by washing with 1 mL PNK buffer ( without DTT ) five times and with 1 mL high salt buffer five times . The bound RNA fragments were eluted from the beads by proteinase K digestion twice at 55 °C for 20 and 10 min , respectively . The eluate was further purified using RNA Clean and Concentrator kit ( Zymo Research ) . RNA was used for library generation with NEBNext Small RNA Library Prep kit ( NEB ) . Sequencing was carried out on Illumina HiSeq 4000 according to the manufacturer’s instructions . For bioinformatics analysis , after removing the adapter sequence , only reads that are longer than 16 bp were kept . The reads were mapped to the reference genomes ( both human hg38 and HIV-1NL4-3 ( GenBank: M19921 . 2 ) using Bowtie2 . Reads uniquely mapped to HIV-1 genome ( not to human genome ) were used in the subsequent analysis . HeLa cells over-expressing individual YTHDF1–3 proteins or the pPB-CAG vector were infected with HIV-1-Luc/VSV-G ( MOI of 0 . 5 ) . Cells were collected at 6 , 12 and 24 hpi . Total RNA was isolated from the cells using RNeasy columns ( Qiagen ) with on-column DNase I treatment ( Qiagen ) and eluted with RNase-free water . Quantitative RT-PCR was used to measure HIV-1 gag RNA levels as described ( Dong et al . , 2007 ) . Sequences of PCR primers are listed in Table 3 . All HIV-1 stocks used for infection were treated with DNaseI ( 40 U/ml; Ambion ) prior to infections to avoid plasmid DNA contamination . GO analysis was performed using the GO Enrichment Analysis tool from the Gene Ontology Consortium ( Gene Ontology , 2015 ) . GO graphs were plotted using the Web server REVIGO ( Supek et al . , 2011 ) . Data were analyzed using Mann-Whitney’s U test or one-way ANOVA test with Prism software and statistical significance was defined as p<0 . 05 . Data accession: all the raw data and processed files have been deposited in the Gene Expression Omnibus ( http://www . ncbi . nlm . nih . gov/geo ) and accessible under GSE85724 .	HIV infection is a global health challenge . The antiviral drugs that are currently available limit the ability of the virus to multiply in infected individuals , but they rarely eliminate the virus entirely . A better understanding of how HIV behaves in the cell would help researchers to find a cure for persistent HIV infection . When HIV enters an immune cell , its genetic material – in the form of molecules of ribonucleic acid ( RNA ) – is used as a template to make molecules of DNA . This viral DNA can integrate into the host cell’s DNA , where it is used as a template to make more viral RNA molecules , which are then used to make viral proteins . Some of the viral RNAs are also packaged into new virus particles . In cells , RNA molecules are often subject to a chemical modification called adenosine methylation , which regulates how that RNA is translated into proteins . Specific enzymes add molecules called methyl tags to particular locations on the RNA , while other enzymes remove them . A family of proteins called YTHDF1–3 recognize and bind to these methyl tags on the RNA and influence how much protein is produced from the target RNA . There is evidence to suggest that the cell can add methyl tags to HIV RNA . However , the extent to which this happens , and what effects this modification has on HIV replication and viral protein production , are not clear . Tirumuru et al . addressed these questions by analyzing how changing the levels of YTHDF1–3 proteins and the enzymes that add or remove methyl tags in human cells affected HIV infection . The experiments show that YTHDF1–3 proteins inhibited HIV infection in immune cells called T-lymphocytes by recognizing HIV RNA that had been methylated , mainly by targeting the step where the viral RNA is copied into DNA . Altering the levels of the enzymes that add or remove methyl tags in the cells can change the amount of methyl tags attached to RNA molecules , which alters the amount of HIV protein produced . For example , when more RNA molecules had methyl tags , the cells produced more HIV proteins . These findings suggest that adenosine methylation plays an important role in regulating the ability of HIV to thrive and multiply in T-lymphocytes , which are an important target for HIV . Since the RNAs of other human viruses , such as influenza virus , can also be modified by adenosine methylation , drugs that target this pathway could have the potential to be used to fight a variety of viral illnesses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2016	N6-methyladenosine of HIV-1 RNA regulates viral infection and HIV-1 Gag protein expression
Many studies are uncovering functional roles for long noncoding RNAs ( lncRNAs ) , yet few have been tested for in vivo relevance through genetic ablation in animal models . To investigate the functional relevance of lncRNAs in various physiological conditions , we have developed a collection of 18 lncRNA knockout strains in which the locus is maintained transcriptionally active . Initial characterization revealed peri- and postnatal lethal phenotypes in three mutant strains ( Fendrr , Peril , and Mdgt ) , the latter two exhibiting incomplete penetrance and growth defects in survivors . We also report growth defects for two additional mutant strains ( linc–Brn1b and linc–Pint ) . Further analysis revealed defects in lung , gastrointestinal tract , and heart in Fendrr−/− neonates , whereas linc–Brn1b−/− mutants displayed distinct abnormalities in the generation of upper layer II–IV neurons in the neocortex . This study demonstrates that lncRNAs play critical roles in vivo and provides a framework and impetus for future larger-scale functional investigation into the roles of lncRNA molecules . Mammalian genomes encode thousands of long noncoding RNAs ( lncRNAs ) , which are emerging as key regulators of cellular processes ( Rinn and Chang , 2012; Mercer and Mattick , 2013 ) . Gain- and loss-of-function approaches in cell-based in vitro systems have been useful in uncovering important roles for lncRNAs , such as modulating chromatin states , maintaining cellular identity ( i . e . pluripotency ) and regulating cell cycle and translation ( Tsai et al . , 2010; Guttman et al . , 2011; Wang et al . , 2011; Yoon et al . , 2012 ) . Genome-wide association and expression profiling studies in humans have also identified correlations between lncRNA mutation , misregulation and disease states ( Visel et al . , 2010; Cabili et al . , 2011; Brunner et al . , 2012 ) . Yet , direct in vivo , genetic evidence of the functional significance of lncRNAs as a class of transcripts remains elusive . Functional studies in knockout mouse models have provided compelling evidence for the requirement and sufficiency of particular transcripts for organ development and function . However , small and large-scale efforts have focused primarily on protein coding genes , leaving long noncoding RNAs vastly understudied ( Edwards et al . , 2011; White et al . , 2013 ) . Of the few lncRNA mutant mice generated to date , most involved previously studied and classic examples of lncRNAs ( e . g . , Xist , H19 , Kcn11ot1 , Malat1 ) ( Anguera et al . , 2011; Gomez et al . , 2013; Gordon et al . , 2010; Moseley et al . , 2006; Nakagawa et al . , 2011; Ripoche et al . , 1997; Zhang et al . , 2012 ) , leaving the more recent large-scale RNA-Seq-derived catalogs of lincRNAs largely unexplored . The progress made in these early studies to initiate functional characterization of lncRNAs in vivo now warrants a larger investigation of their biological importance to development and disease . To address this directly , and to begin to explore the roles of lncRNAs in vivo , we have established a novel cohort of lncRNA knockout mice . We focused on a subgroup of lncRNAs called long intergenic noncoding RNAs ( lincRNAs ) , such that genetic deletion would not overlap known protein coding genes or other known gene annotations . We have implemented a generalized and logical lincRNA candidate selection process that leverages a collection of cell-based functional assays , RNA-sequencing data and computational analyses . This approach led us to identify 18 lincRNAs for targeted deletion in mouse . Initial characterization of these new knockout strains demonstrated key functional roles in viability , development of the cerebral cortex and other developmental processes . In this study , we describe the observed phenotypes for five strains within this collection . Collectively , these data provide evidence that lincRNAs play central roles in mammalian development and physiology . To more globally understand the physiological significance of lincRNAs , we combined a step-wise lincRNA selection pipeline with a genetic approach to engineer a cohort of lincRNA knockout mouse strains for functional analysis . We integrated several computational and experimental data sets to select bona fide lincRNA candidates across three lincRNA catalogs ( Guttman et al . , 2009; UCSC and Refseq ) . First , all transcripts with identifiable Pfam domains or those overlapping known non-lncRNA annotations ( e . g . , annotated protein coding genes , microRNAs , tRNAs and pseudogenes ) were excluded . Second , we excluded any remaining transcripts with conserved protein coding potential , including the potential to produce small peptides , by performing stringent codon-substitution frequency ( CSF ) analysis ( Lin et al . , 2011 ) . We previously demonstrated that this algorithm is capable of discriminating known functional small peptides down to 11 amino acids ( Guttman and Rinn , 2012; Guttman et al . , 2013 ) . Using selective criteria , we restricted candidate lincRNAs to those with CSF scores <−200 ( ranging from −205 to −14 , 771 , Figure 1A ) . Then , we examined existing ribosome profiling datasets to quantitate the ability of these transcripts to engage the ribosome ( Figure 1—figure supplement 1 ) . None of the tested candidates were shown to have clear translation efficiency or codon bias according to proposed standards ( Guttman et al . , 2013 ) . Finally , through genome lift-over of our mouse lincRNAs to the human genome ( build hg19 ) , we examined mass spectrometry data by Gascoigne et al . ( 2012 ) to discard transcripts with mapped peptides . We kept transcripts with only two or less mass spectrometry tags , consistent with low levels of background ribosome association observed by Guttman et al . ( 2013 ) ( Supplementary file 1B ) . Thus , based on an exhaustive analysis of existing annotations , standard evolutionary and ribosome profiling metrics , as well as mass spectrometry data , these candidates do not appear to contain protein coding potential . 10 . 7554/eLife . 01749 . 003Figure 1 . Properties of the 18 lincRNA candidates and Mendelian inheritance . ( A ) List of the 18 lincRNA candidates for targeted deletion in mouse and overview of criteria used for their selection . ( B ) Heatmap of log10 FPKM+1 expression levels of the 18 lincRNAs in a panel of adult mouse tissues and cell lines via RNA-Seq . ( C ) Guilt-by-association ( GBA ) analysis for 17/18 lincRNA candidates . Individual tiles represent significant ( p<1 . 0 × 10−6 ) gene sets from the CP Reactome collection at MSigDB . Tiles are filled based on the Z-score of the Pearson correlation values between a given lincRNA and the genes within the gene set across a compendium of RNA-Seq samples . ( D ) Mendelian inheritance of the 18 lincRNA mutant alleles from the progeny of heterozygote intercrosses . Numbers of observed and expected ( in parenthesis ) wild-type ( +/+ ) , heterozygote ( +/− ) and homozygote ( −/− ) mice are indicated . Mice were genotyped at weaning age . The p value is based on X2 test . † The Spasm gene is located on the X chromosome . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 00310 . 7554/eLife . 01749 . 004Figure 1—figure supplement 1 . Ribosome release score of lncRNAs . The ribosome release score ( RRS ) measures evidence of ribosome disassociation at the stop codon of a putative coding region . A putative coding region is defined as the region between a start codon and then next in-frame stop codon . Its putative 3′-UTR is defined as the region beginning immediately after the stop codon and ending at the next start codon in any frame . The RRS score is defined as the total number of ribosome-associated reads overlapping the putative coding region , divided by the number of ribosome-associated reads contained within the 3′-UTR . This ratio is then normalized by the same ratio for total RNA reads . ( A ) The ORF with maximum RRS score for each lincRNA is shown . ( B ) Maximum RRS scores of lincRNAs are compared to the distribution of RRS scores for known coding genes . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 00410 . 7554/eLife . 01749 . 005Figure 1—figure supplement 2 . lincRNA candidates knockout targeting strategy . Genomic locus and targeted deletion scheme for the 18 lincRNA candidates . Briefly , each lincRNA gene was replaced by a ß-galactosidase ( lacZ ) reporter cassette containing a KOZAK-ATG sequence , polyadenylation signal , and a LoxP-flanked neomycin ( neo ) resistance gene driven by the human ubiquitin C promoter ( hUb1 ) ( mammalian cells ) and EM7 promoter ( for gap repair cloning selection in bacteria ) . Arrows indicate location of the primer sets used for genotyping . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 005 From this stringent list of lincRNAs , we selected 18 candidates by further considering additional criteria: ( 1 ) Presence of an observed orthologous human lincRNA with syntenic conservation as determined by TransMap ( Zhu et al . , 2007 ) ; ( 2 ) Presence of canonical chromatin features of actively transcribed genes such as histone H3 lysine 4 trimethylation ( H3K4me3 ) at the promoter and histone H3 lysine 36 trimethylation throughout the gene body ( Guttman et al . , 2009 , 2010 ) ; ( 3 ) Local ‘enhancer-like’ signatures based on published P300 , RNA polymerase II , histone lysine 4 monomethylation ( H3K4me1 ) and histone lysine 27 acetylation ( H3K27ac ) chromatin states ( Shen et al . , 2012 ) . We selected 6/18 candidates lncRNA loci that have putative ‘enhancer’ modifications , thus allowing to investigate the newfound roles of lncRNAs in ‘enhancer’ activity ( Ørom et al . , 2010; Wang et al . , 2011 ) . Collectively , this step-wise selection process resulted in 18 high-quality candidate lincRNAs with diverse features , for targeted deletion and phenotypic characterization ( Figure 1A , Supplementary file 1A for coordinates and nomenclature details ) . We first examined the gene-expression patterns of the candidate lincRNAs using RNA-sequencing of various adult tissues and cell types . Several lincRNAs presented more restricted patterns of expression , suggesting strong tissue specificity , for example Celr in embryonic stem ( ES ) -derived neural stem cells ( NSCs ) , linc–Enc1 in mouse ES cells , Manr and linc–Cox2 in lung , whereas a select few showed more ubiquitous expression across tissues ( linc–Pint , Spasm , linc–Ppara , Tug1 ) ( Figure 1B ) . Most of the candidate lincRNAs ( 12/18 ) were expressed in the adult brain or in ES-derived NSCs ( Figure 1B ) . Since most candidate lincRNAs in this screen have unknown biological roles , we leveraged our RNA-sequencing data using guilt-by-association ( GBA ) analysis to generate hypotheses on functional significance by comparing the expression of each lincRNAs to protein coding genes of known function ( ‘Materials and methods’ ) ( Figure 1C ) . GBA predicted lincRNA activities across a wide range of pathways and biological processes , ranging from regulation of the cell cycle and chromosome organization and maintenance , to neuronal differentiation and immune response ( Figure 1C ) . Overall , these analyses suggested diverse potential roles for our candidate lincRNAs . To study the function of the selected 18 lincRNAs in vivo and to resolve , at the cellular level , their expression pattern in different organs , we generated knockout mouse strains for each candidate by replacing the lincRNA gene with a lacZ reporter cassette ( ‘Materials and methods’ , Figure 1—figure supplement 2 for details ) ( Valenzuela et al . , 2003 ) . These new strains , which are the first lincRNA knockout models to incorporate a reporter , more than double the number of models available for investigation and constitute an important resource that will be used to better understand the functional contribution of lincRNAs to mammalian biology . To assess the requirement for each lincRNA in embryonic development and viability , we examined the progeny from heterozygote intercrosses for all 18 strains . Genotyping of weanlings ( 21 days old ) revealed normal Mendelian segregation of mutant alleles in 15 of the 18 strains ( Figure 1D ) . For the three remaining strains Peril , Mdgt and Fendrr , the progeny of heterozygote intercrosses contained much lower numbers of homozygote mutants than expected . Only 13 Peril−/− mice ( of an expected 32 ) , and 6 Mdgt−/− mice ( of an expected 17 ) were found at weaning age , indicating that deletion of Peril and Mdgt leads to reduced viability with >50% and 65% penetrance , respectively ( Figure 1D ) . Closer examination of Mdgt pups revealed that homozygous mutants died within 2 weeks after birth . For Fendrr , no homozygous mutants were found at weaning age ( Figure 1D ) , indicating that the lethal phenotype for this strain is fully penetrant . Thus , 3 out of the 18 ( 17% ) lincRNA knockout strains generated have a lethal phenotype , confirming that specific lincRNA genes are required for viability . To determine the onset of Peril−/− mice lethality , we monitored survival at both early and late stages of embryonic development . Since normal Mendelian ratios were observed up to E18 . 5 ( Figure 2A ) , and pups appeared macroscopically normal at birth ( Figure 2B ) , we monitored newborns to see if lethality occurred perinatally . We observed that 50% ( 5/10 ) of Peril−/− pups died within 2–20 days after birth ( Figure 2A ) . A similar percentage ( 52% , 11/21 ) of newborn deaths occurred in the progeny of intercrosses from surviving Peril−/− mice ( Figure 2A , lower panel ) . These results confirmed the reduced viability of Peril mutants ( ∼50% penetrance ) with pups dying early after birth . 10 . 7554/eLife . 01749 . 006Figure 2 . Deletion of Peril leads to reduced viability . ( A ) Genotyping results from heterozygote intercrosses ( Upper panel ) at different developmental stages ( pups dying within days after birth ) and homozygote intercrosses ( Lower panel ) at birth . The p value is based on X2 test . ( B ) Newborn ( P0 ) Peril−/− mutants and wild-type littermates . ( C ) Peril genomic locus and targeting scheme . ( D ) RNA-Seq expression profile for Peril across a panel of mouse tissues and cell types . ( E ) Single-molecule FISH targeting Peril in wild type mouse embryonic stem cells ( mES ) . Nuclei were stained with DAPI . ( F ) Whole mount and coronal section lacZ stainings reporting expression of Peril in the brain and spinal cord of a heterozygote E14 . 5 embryo . LGE/CGE , Lateral and Caudal Ganglionic Eminence; P , Pons; CP , Choroid Plexus; SC , Spinal Cord; D , dorsal; V , ventral; R , rostral; C , caudal . Scale bars = 1 mm , whole brains; 500 µm , sections . ( G ) Relative expression levels of Peril as assessed by RT-PCR and expression estimates ( FPKM ) of the neighboring genes Sox2 and Sox2ot as assessed by RNA-Seq in E18 . 5 brain of homozygote mutant and wild-type littermates ( n = 2 each ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 00610 . 7554/eLife . 01749 . 007Figure 2—figure supplement 1 . Peril E18 . 5 brain differential RNA-Seq analysis . ( A ) lacZ staining reporting expression of Peril in brain and spinal cord of a heterozygote E18 . 5 embryo . SC , Spinal Cord . Scale bars = 1 mm whole brains , 500 µm sections . Scale bars = 1 mm whole brains , 500 µm sections . SC , spinal cord . ( B ) GSEA of Peril−/− vs wild-type E18 . 5 total brain RNA-Seq . Each tile is a significant ( q<0 . 001; Mann-Whitney , BH ) gene set from the Reactome collection at mSigDB , based on the Peril−/−/wild-type ranking of test-statistic values from a Cuffdiff2 differential analysis . Tiles are shaded based on the z-score of the test-statistic for genes within the given gene set , relative to all genes for a given condition to show direction of expression relative to wild-type . ( C ) Heatmap of significant ( q<0 . 05 , CuffDiff2 ) differentially expressed genes ( log10 FPKM+1 ) in Peril−/− vs wild-type E18 . 5 total brain by RNA-Seq . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 007 Peril is a seven exon transcript derived from a 18 . 2 Kb genomic locus located ∼110 Kb downstream of the pluripotency factor Sox2 ( Figure 2C ) . Using mouse ES cell ( mES ) cDNA , we were able to clone two distinct isoforms ( 1 , 831 bp and 631 bp ) ( Figure 2C ) . RNAseq expression profiling from a panel of mouse tissues and cell lines showed that Peril was highly enriched in mouse ES cells , but also expressed at lower levels in adult brain and testes ( Figure 2D ) . The Peril transcript is predominantly nuclear as revealed by RNA fluorescence in situ hybridization ( FISH ) in mouse ES cells ( Figure 2E ) . Using the knocked-in lacZ reporter in heterozygote embryos , tight temporal and spatial regulation of Peril was found in the brain and spinal cord of E14 . 5 and E18 . 5 embryos ( Figure 2F , Figure 2—figure supplement 1A ) . To identify putative pathways affected by deletion of Peril , we performed RNA-sequencing on E18 . 5 brains from Peril−/− and wild-type littermates ( n = 2 ) . Analysis of Peril expression levels confirmed the deletion of Peril ( Figure 2G ) . Expression of the pluripotency factor Sox2 and its overlapping noncoding RNA Sox2ot were not significantly affected in the knockout brains ( Figure 2G ) . Differentially regulated genes and gene set enrichment analysis ( GSEA ) revealed downregulation of genes involved in cell cycle regulation , energy metabolism , immune response , and mRNA and protein processing in Peril−/− brains relative to wild-type controls ( Figure 2—figure supplement 1B and C ) . This indicates that the transcriptional programs within the brain are in fact affected by deletion of Peril . Yet , it remains unclear if these changes underlie the observed lethality . Fendrr is a 2 , 380 bp transcript consisting of six exons . It is transcribed from a bidirectional promoter shared with the protein coding gene Foxf1a , located 1 , 354 bp from its transcriptional start site ( Figure 3A ) . An orthologous human FENDRR ( LINC–FOXF1 ) transcript expressed from a syntenic region was identified within our catalog of human lincRNAs ( Cabili et al . , 2011 ) . We previously demonstrated that this lincRNA is predominantly nuclear and physically associates with the PRC2 Polycomb complex ( Khalil et al . , 2009 ) . 10 . 7554/eLife . 01749 . 008Figure 3 . Fendrr−/− pups have multiple defects in lung , heart and gastrointestinal tract . ( A ) Fendrr locus and targeting strategy . Arrows indicate location of the primers used for genotyping . ( B ) Genotyping results from heterozygote intercrosses at embryonic stages E14 . 5 , E18 . 5 and at birth ( P0 ) . The p value is based on X2 test . ( ) All newborns died within 24 hr after birth . ( C ) Fendrr−/− E18 . 5 embryos and wild-type littermates . ( D ) RNA-Seq expression profile for Fendrr across a panel of mouse tissues and cell types . ( E and F ) lacZ reporter stained organs and sections showing expression of Fendrr in specific regions of the lung ( Lu ) , trachea ( Tr ) and esophagus ( Es ) , but not in heart ( H ) in E14 . 5 and E18 . 5 embryo ( E ) and in the gut and stomach ( St ) ( F ) . Sm , smooth muscle; Ep , Epithelia; Me , Mesenchyme; Ly , Lymphoid aggregates . Scale bars = 1 mm whole organ , 200 µm sections . ( G ) Number of E18 . 5 embryos successfully breathing after surgical delivery . ( H ) Size difference of Fendrr−/− lungs at E14 . 5 compared to wild-type littermates ( n = 3 each ) . ( I–K ) Representative hematoxylin and eosin ( H&E ) stained sections showing unstructured vessels ( arrow ) in E14 . 5 Fendrr mutant lungs compared to wild type littermates ( n = 3 ) ( I , upper panels ) , alveolar defects at E18 . 5 ( I , lower panel ) , thinner mesenchymal layer of the mucosa and external smooth muscle layer of the oesophagus ( J ) and ventricular septal defects in the heart ( K ) of Fendrr−/− E18 . 5 embryos compared to wild type ( n = 3 ) . Scale bars= 500 µm , 100 µm for esophagus . ( L ) RNA-Seq expression levels of Fendrr and the neighboring coding gene Foxf1a in E18 . 5 lung of homozygote mutant and wild-type littermates ( n = 2 each ) . ( M ) GSEA of Fendrr−/− vs wild-type E18 . 5 lung and total brain RNA . Each tile is a significant ( q<0 . 001; Mann-Whitney , BH ) gene set from the Reactome collection at mSigDB , based on the Fendrr−/−/wild-type ranking of test-statistic values from a Cuffdiff2 differential analysis . Tiles are shaded based on the z-score of the test-statistic for genes within the given gene set , relative to all genes for a given condition to show direction of expression relative to wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 00810 . 7554/eLife . 01749 . 009Figure 3—figure supplement 1 . Fendrr−/− embryos don’t have an omphalocele . Wild type and Fendrr mutant E14 . 5 embryos were harvested and examined for the presence of an omphalocele . Numbers of embryos analyzed for each genotype are indicated . Tail and limbs were removed . Scale bar = 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 00910 . 7554/eLife . 01749 . 010Figure 3—figure supplement 2 . Fendrr , linc–Brn1b , and Peril do not act as cis-enhancer elements . Scatterplot showing Cuffdiff2 test-statistic ( Knockout/wild type ) for each gene ±1 Mb from the start site of the lincRNA . Genes with significant differential expression ( q≤0 . 05 ) are highlighted in red . In each case , there is no significant enrichment for differentially expressed genes within the ±1 Mb window relative to the background distribution of differentially genes , as determined from random sampling of windows for each comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 010 To determine the onset of Fendrr lethality , we monitored the survival of embryos at both early and late stages of embryonic development , as well as that of newborn pups . Normal Mendelian ratios were found at both E14 . 5 and E18 . 5 ( Figure 3B ) , with embryos appearing macroscopically normal prior to birth ( Figure 3C ) , suggesting that the lethality most likely occurred after birth . Importantly , we observed 8 Fendrr−/− mutant newborns ( P0 ) ( Figure 3B ) , all of which died within 24 hr , indicating perinatal lethality . During the course of our study , another group generated a Fendrr loss-of-function mouse model by replacing the first exon with multiple transcriptional termination sequences ( Grote et al . , 2013 ) . In contrast to the results presented here , Grote et al . observed lethality at E13 . 75 due to heart and body wall ( omphalocele ) defects . When analyzing E14 . 5 embryos , we found no resorbed embryos or omphalocele in our Fendrr homozygous mutants ( Figure 3—figure supplement 1 ) . Although , both studies used similar genetic background strains , a possible explanation for this discrepancy may be found in the distinct targeting strategy used to remove the Fendrr gene . Regardless , both studies confirm that loss of Fendrr is lethal in mice . Grote et al . observed Fendrr expression to be restricted to nascent lateral plate mesoderm and did not detect it in any other tissue . Using RNA-Seq expression profiling from adult mouse tissues and cell lines , we however found that Fendrr is expressed at high levels in the adult lung , and lower levels are detectable in colon , liver , spleen and brain ( Figure 3D ) . Analysis of the knocked-in lacZ reporter in E14 . 5 and E18 . 5 embryos confirmed expression of Fendrr in these tissues as well as in the trachea and all along the gastrointestinal tract ( Figure 3E , F ) . Interestingly , in the developing respiratory and digestive tracts at E14 . 5 and E18 . 5 , expression of Fendrr is restricted to the pulmonary mesenchyme surrounding the bronchiolar epithelial cells ( Figure 3E , right panels ) ; in mesenchymal cells of the developing mucosa; the muscularis externa of the gut and in lymphoid aggregations within the gut’s mucosa ( Figure 3F , bottom panel ) , all of which are derived from the lateral plate mesoderm . Perinatal lethality is often associated with respiratory failure . Since the highest expression levels of Fendrr are found in the lungs , we evaluated initiation of breathing in surgically delivered E18 . 5 embryos . After cleaning of their airways , all Fendrr−/− embryos analyzed either failed to breathe or gasped and stopped breathing within 5 hr ( n = 7 Fendrr−/− , Figure 3G ) . In contrast , respiration initiated normally and was maintained for all but one of the heterozygote and wild-type embryos ( n = 15 Fendrr+/− and n = 8 wild type ) . Fendrr−/− lungs at the pseudoglandular stage ( E14 . 5 ) were hypoplastic compared to wild type ( Figure 3H ) , and histological evaluation of the lungs revealed a decrease in the number and organization of pulmonary arteries , and a general failure of vasculogenesis within the lungs of the Fendrr−/− mutants compared to wild type ( n = 3 Fendrr−/− and n = 3 wild type; Figure 3I , upper panels ) . At E18 . 5 , Fendrr−/− lungs appeared to have fewer but larger alveoli ( n = 3 Fendrr−/− and n = 3 wild type; Figure 3I , lower panels ) . Together , these results suggest that respiratory failure observed at birth in Fendrr−/− mice could be due to a lung maturation and vascularization defect . We also observed expression of Fendrr in the esophagus and gut . We observed thinning of the mesenchymal layer of the mucosa and external smooth muscle layers in the esophagus at E18 . 5 ( n = 3 Fendrr−/− and n = 3 wild type; Figure 3J ) . Although we did not observe Fendrr expression in the heart at E14 . 5 , E18 . 5 , or postnatally , we did observe intraventricular septal heart defects prior to birth ( E18 . 5 ) ( n = 3 Fendrr−/− and n = 3 wild type; Figure 3K ) . In accordance with Fendrr having a previously described role in lateral plate mesoderm ( Grote et al . , 2013 ) , our results suggest a more general role for Fendrr in regulating the proper differentiation of mesenchyme-derived tissues across several organ systems . We next investigated if neighboring gene expression is perturbed by the deletion of Fendrr . We harvested lungs from E14 . 5 Fendrr−/− embryos and wild-type littermates ( n = 2 ) and performed differential RNA-Seq analyses . A loss of Fendrr expression in knockout relative to wild-type lungs confirmed deletion of Fendrr ( Figure 3L ) . No significant change in the expression of the Foxf1a protein coding gene was observed in the Fendrr−/− lung . Furthermore , genes within ±1 Mb of the Fendrr locus were not significantly differentially expressed any more than background levels of local enrichment ( Figure 3—figure supplement 2A , p<0 . 087; bootstrapped from random 2 Mb genomic intervals ) , suggesting that the Fendrr gene does not act as a local enhancer . GSEA identified gene sets involved in muscle differentiation and contraction as the most significant sets misregulated in Fendrr−/− lungs compared to wild type ( Figure 3M ) . This agrees with our identification of defects in the lung vasculature of the Fendrr−/− mice . Further studies will be needed to understand how specific changes in gene-expression patterns upon deletion of Fendrr contribute to the observed defects and perinatal lethality . To determine if deletion of our candidate lincRNAs affects normal development and growth postnatally , each strain was examined for gross morphological abnormalities and body weight ( BW ) measurements were taken . Homozygote mutant mice with heterozygote and wild-type littermate controls were compared over a 7–10 week period . Mdgt−/− pups displayed a severe growth retardation phenotype , which may contribute to their lethality ( Figure 4C , D ) . A week after birth , Mdgt−/− pups were already significantly smaller than heterozygote and wild-type littermates ( Figure 4C ) , with females 60% smaller ( n = 5 , p<0 . 00001 ) than wild-type littermates ( n = 10 ) and males 32% smaller ( n = 3 , p<0 . 05 ) , suggesting a sex bias in this phenotype . In Mdgt−/− survivors , this retarded growth persisted up to 10 weeks in females , which were still 37% smaller than wild types ( Figure 4D , p<0 . 0006 ) . The defect , although milder , also persisted in male mutants up to at least 8 weeks . 10 . 7554/eLife . 01749 . 011Figure 4 . Mdgt−/− surviving mice have growth defects . ( A ) Mdgt genomic locus and targeting scheme . ( B ) RNA-Seq expression profile for Mdgt across a panel of mouse tissues and cell types . ( C ) Representative example showing the reduced size of Mdgt−/− pups compared to wild type 7 days after birth ( P7 ) . ( D ) Body weight ( g ) measurements over a 10 weeks period show growth retardation in both female and male Mdgt−/− mice compared to wild type and Mdgt+/− littermates ( Females: n = 5 Mdgt−/− , n = 15 Mdgt+/− and n = 10 wild types; Males: n = 3 Mdgt−/− , n = 10 Mdgt+/− and n = 10 wild types ) . Significant p values at each time point are indicated ( * ) . ( E ) Whole mount lacZ stainings reporting expression of Mdgt in adult tissues of heterozygote mutant mice . Scale bars = 1 mm , testis , thymus; 2 mm , brain , stomach , colon . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 01110 . 7554/eLife . 01749 . 012Figure 4—figure supplement 1 . Growth retardation phenotype in lincRNA knockout strains . Body weight ( g ) measurements of wild type , lincRNA heterozygote and homozygote mutants for ( A ) linc–Pint , ( B ) linc–Brn1b and ( C ) Peril were taken at the indicated postnatal time points . Animals used for measurements ( n ) were derived from at least 2–3 litters . Paired Student's t test was used to assess statistical significance ( p ) in mean values . Scale bar = 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 012 Mdgt is a 443 bp transcript consisting of three exons transcribed from a bidirectional promoter shared with the homeobox gene Hoxd1 . The first exon for Mdgt is located only 84 bp away from the Hoxd1 transcription start site . Another noncoding transcript , AK144266 , is contained within an intron of Mdgt and appears to be expressed from the opposite strand ( Figure 4A ) . In our targeting strategy , the entire Mdgt locus was deleted , including the AK144266 transcript ( Figure 4A ) . Expression estimates obtained from RNA-Seq revealed that Mdgt is highly expressed in testes , and more moderately in brain , kidney , colon and skeletal muscles ( Figure 4B ) . lacZ staining confirmed expression in those tissues and also revealed expression in stomach and thymus ( Figure 4E ) . In addition to Mdgt , stunted growth defects were also observed in three other strains ( Figure 4—figure supplement 1 ) . 3 weeks after birth , linc–Pint−/− mice were noticeably smaller than wild-type littermates ( Figure 4—figure supplement 1A ) . By 7 weeks of age , females had 22% reduced body weight ( n = 5 , p<0 . 001 ) compared to wild type ( n = 4 ) , whereas males were 29% smaller ( n = 3 linc–Pint−/−; n = 3 wild type , p<0 . 05 ) . A decrease in body weight was also observed by three weeks of age in linc–Brn1b−/− ( 29% , p<0 . 013 , n = 3 ) and Peril−/− mice ( 35% , p<0 . 0003 , n = 4 ) when compared to wild-type littermates ( n = 3 ) ( Figure 4—figure supplement 1B , C ) . Together , these results indicate that specific lincRNAs can be important for normal body weight and growth . In order to determine whether loss of our candidate lincRNAs could lead to other developmental defects , we examined additional mutant strains that did not show perinatal lethality . Building on prior evidence that lincRNAs are abundantly expressed and spatio-temporally regulated within the brain during development and adulthood ( Mercer et al . , 2008; Qureshi et al . , 2010; Cabili et al . , 2011; Ramos et al . , 2013 ) , we concentrated on the candidate lincRNAs that were expressed in the brain or in neural progenitors . Twelve of the lincRNAs for which we have established knockout strains exhibited such expression profile ( Figure 1B ) . To focus on lincRNA candidates of potential functional relevance in neuronal development , we used syntenic orthology ( TransMap [Zhu et al . , 2007] ) and RNA-sequencing to select those with identifiable human orthologs , and whose expression was regulated during in vitro neural differentiation . Briefly , transcripts expressed along a time course of ES-derived human neural stem cells ( NSC ) differentiation ( Goff et al . , 2009 ) ( Figure 5A ) were assembled and aggregated with an existing compendium of RNA-Seq data , using our previously described lincRNA discovery pipeline ( Cabili et al . , 2011 ) . The resulting lincRNA catalog contained 24 , 737 distinct , high-quality transcript reconstructions corresponding to 14 , 259 human lincRNA genes . We observed 769 lincRNA genes with significant differential expression ( q<0 . 01; Cuffdiff2 ) between any two adjacent time points during NSC differentiation . With a false discovery rate of 1% ( Figure 5—figure supplement 1 ) , 302 of these were significantly induced relative to day 0 ( Figure 5—figure supplement 1 ) . This approach revealed that 7 lincRNAs from our mouse knockout strains have human orthologs that are dynamically induced during in vitro human neuronal differentiation ( Figure 5B ) . Interestingly , two of these , linc–Brn1a and linc–Brn1b , were almost exclusively expressed in NSCs as determined by RNA-Seq ( Figure 1B ) . These lincRNAs reside in the genomic region of Brn1 ( Pou3f3 ) , a well-studied transcription factor involved in cortical development ( McEvilly , 2002; Sugitani et al . , 2002; Dominguez et al . , 2012 ) . We thus focused on this locus , starting with linc–Brn1b , as it is not transcribed from a bidirectional promoter and therefore deletion does not disrupt the Brn1 promoter . 10 . 7554/eLife . 01749 . 013Figure 5 . linc–Brn1b is spatio-temporally regulated during corticogenesis . ( A ) Schematic overview of the in vitro human neural stem cell differentiation protocol . RNA was collected at the indicated time points and sequenced to identify significantly differentially expressed lincRNA human orthologs . ( B ) Heatmap of log2 ratios to undifferentiated ( Day 0 ) human neural stem cells for 7 of the 20 lincRNAs selected for deletion with a significant ( q<0 . 01; Cuffdiff2 ) increase in expression during differentiation . ( C ) RNA-Seq expression profile for linc–Brn1b across a panel of mouse tissues and cell types . ( D ) linc–Brn1b genomic locus and targeting strategy . ( E ) qRT–PCR confirmation of the genotype for both heterozygotes ( +/− ) and homozygous null ( −/− ) mutants . ( F ) Single-molecule RNA FISH targeting linc–Brn1b in wild-type E14 . 5 cortical neurospheres . ( G ) lacZ staining shows expression profile of linc–Brn1b at different embryonic ( E13 . 5 , E15 . 5 and E18 . 5 ) and early postnatal stages ( P7 ) in linc–Brn1b+/− telencephalon . lacZ expression in neural progenitors of both ventral telencephalon ( lateral ganglionic eminence , LGE and medial ganglionic eminence , MGE ) ( E13 . 5 ) and dorsal telencephalon ( ventricular zone , VZ and subventricular zone , SVZ ) ( E15 . 5 ) is detected . Restricted expression in the upper cortical layers is observed in both E18 . 5 and P7 cortex . ( H ) Whole mount lacZ staining of P7 linc–Brn1b+/− brain shows distinct linc–Brn1b expression in primary somatosensory ( S1 ) and visual ( V1 ) cortical areas . ( I ) β-galactosidase immunofluorescence on coronal sections of P7 linc–Brn1b−/− cortex shows linc–Brn1b expression in layer IV of the somatosensory area ( white boxes ) , specifically within the barrel structures ( white arrows ) , as determined by co-staining with the upper layer markers SATB2 and CUX1 . Abbreviations: LV , lateral ventricle; LGE , lateral ganglionic eminence; MGE , medial ganglionic eminence; CGE , caudal ganglionic eminence; CP , cortical plate; SVZ , subventricular zone; VZ , ventricular zone; Str , striatum; M1 , primary motor area; S1 , primary somatosensory area; cc , corpus callosum; V1 , primary visual cortex; A1 , primary auditory cortex; F/M , frontal motor cortex . Scale bars: 500 µm ( G ) , ( I , upper panels ) ; 100 µm ( I , lower panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 01310 . 7554/eLife . 01749 . 014Figure 5—figure supplement 1 . Differential expression of protein coding genes , lncRNAs and neuronal markers during human ES-derived neuronal differentiation time course . ( A ) Heatmap of Cuffdiff2 estimated expression values ( FPKM ) expressed as log2 fold-change to Day 0 for 5 , 100 significant protein-coding genes and 769 significant lncRNAs ( q<0 . 001; Cuffdiff2 ) from a human H1-derived NSC differentiation timecourse . ( B ) Individual expression plots for key neural stem cell markers , neuronal markers , and non-neuronal markers confirm the differentiated state of the human H1-derived NSC into predominantly neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 014 Using RNA-Seq data , we observed linc–Brn1b to be predominantly expressed in ES-derived NSCs with additional moderate expression in the adult brain and kidney ( Figure 5C ) . The linc–Brn1b gene is a ∼3 Kb transcript derived from a 6 . 8 Kb genomic locus approximately 10 Kb downstream of the Brn1 protein coding gene . To generate the linc–Brn1b knockout mice , we targeted the entire linc–Brn1b locus ( Figure 5D ) . Complete ablation was confirmed by qRT–PCR ( Figure 5E ) using adult brain cDNA as template . RNA-FISH in mouse E14 . 5 neural progenitor cells ( NPCs ) isolated from the cerebral cortex demonstrated that the linc–Brn1b transcript is predominantly nuclear with moderate cytoplasmic expression ( Figure 5F ) . While identified as a lincRNA expressed in adult brain , the spatio-temporal distribution of linc–Brn1b during brain development is not known . Therefore , we used lacZ expression from the linc–Brn1b locus in heterozygote mutants to define its expression in vivo . We found that linc–Brn1b was expressed within neural progenitors of both the ventral and dorsal telencephalon , as early as E13 . 5 ( Figure 5G ) . Detailed characterization of expression in the germinal zones of the dorsal telencephalon showed that by E15 . 5 , it was strongly expressed in progenitors of both the ventricular zone ( VZ ) and the subventricular zone ( SVZ ) of the developing cerebral cortex , and by E18 . 5 showed restricted expression in the developing upper cortical layers . In addition , whole-mount lacZ staining at P7 , showed a specific areal distribution for linc–Brn1b within both the primary somatosensory cortex and the primary visual cortex ( Figure 5H ) . Within the somatosensory cortex , linc–Brn1b was expressed in the barrel structures of the posteromedial barrel subfield ( PMBSF ) ( Figure 5I ) ; a highly organized region of cortical projection neurons that receives afferent connections from the thalamus and is responsible for the coordination of sensory inputs from the rodent vibrissae ( Petersen , 2007 ) . Collectively , the timing and location of linc–Brn1b expression within the developing cortex suggests a potential role for linc–Brn1b in area-specific development of distinct classes of projection neurons . To investigate the consequences of genetic deletion of linc–Brn1b during development of the telencephalon , we performed RNA-Seq on E13 . 5 and E15 . 5 whole telencephalons and P7 whole brains ( n = 2 ) of wild type and linc–Brn1b−/− mice . At all time points , RNA-Seq analyses identified a statistically significant reduction ( ∼50% ) in expression of the neighboring Brn1 protein coding gene in linc–Brn1b−/− brains relative to wild type ( E15 . 5 shown in Figure 6A , Figure 6—figure supplement 1A , B ) . We observed a similar decrease in BRN1 protein expression between wild type and linc–Brn1b−/− E14 . 5 cortical-derived neurospheres and E15 . 5 whole cortex samples ( Figure 6B ) . In contrast , upon knockout of linc–Brn1b , we observed a significant increase in expression of linc–Brn1a ( p<0 . 01; Cuffdiff2 ) , which shares a bidirectional promoter with Brn1 ( Figure 6A ) . This suggests opposing regulatory effects on the neighboring lincRNA and protein coding genes upon deletion of linc–Brn1b . 10 . 7554/eLife . 01749 . 015Figure 6 . linc–Brn1b−/− mice demonstrate defects in proliferation of IPCs . ( A ) RNA-Seq expression estimates from E15 . 5 wild type and linc–Brn1b−/− telencephalon for linc–Brn1b , the protein coding gene Brn1 , and neighboring genes linc–Brn1a and 2900092D14Rik . ( B ) Western blots of wild type and linc–Brn1b−/− E14 . 5 cortical neurospheres ( NS ) , and E15 . 5 cortices . ( C ) GSEA of linc–Brn1b−/− vs wild type in E13 . 5 and E15 . 5 telencephalon and P7 whole brain ( as described in Figure 3 ) . ( D and E ) Immunofluorescence staining for the mitotic marker phosphorylated histone H3 ( pH3 ) in E15 . 5 coronal sections of cortex ( D ) and pH3+ cell counts of apical progenitors ( APC ) and intermediate progenitors ( IPC ) ( E ) . ( F–H ) Immunofluorescence staining for TBR2 ( F ) , and in situ hybridization for Cux2 ( G ) and Svet1 ( H ) in wild type and linc–Brn1b−/− E15 . 5 cortex show that mutant mice have decreased expression of SVZ intermediate progenitor markers . Scale bars = 500 µm ( D–H ) . p<0 . 01 , p<0 . 001; Student’s t test . LGE , lateral ganglionic eminence; LV , lateral ventricule; ctx , cortex; CP , cortical plate; VZ , ventricular zone; SVZ , subventricular zone; IZ , intermediate zone . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 01510 . 7554/eLife . 01749 . 016Figure 6—figure supplement 1 . Brn1 transcript is reduced in linc–Brn1b−/− E13 . 5 telencephalon and P7 brain . RNA-Seq expression estimates ( FPKM ) for the Brn1 protein coding gene in E13 . 5 telencephalon ( A ) and P7 brain ( B ) harvested from wild type and linc-Brn1b-/- littermates ( n=2 ) . ( C ) The Brn1 paralog Brn2 has no significant changes in expression in the linc-Brn1b-/- mice in any condition , suggesting no contribution from this gene to the observed phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 016 Gene set enrichment analysis ( GSEA ) revealed a significant decrease in genes associated with cellular growth and proliferation in linc–Brn1b−/− relative to wild type at all three time points sampled and an increase in gene sets positively correlated with neuronal maturation ( Figure 6C ) . These results are consistent with a predicted role for linc–Brn1b derived from our initial GBA analyses ( Figure 1C ) and suggest a function for linc–Brn1b in the differentiation of neural progenitors and neurogenesis in the telencephalon . To investigate potential abnormalities of linc–Brn1b−/− mutants , we concentrated on the dorsal telencephalon ( the developing cerebral cortex ) . First , to quantify the proliferation of cortical progenitors , we evaluated the expression of the mitotic marker phosphorylated histone H3 ( pH3 ) in E15 . 5 wild type and linc–Brn1b−/− cortices ( Figure 6D ) . Quantification of pH3+ cells revealed a significant and specific reduction in the relative percentage of pH3+ intermediate progenitor cells within the SVZ ( 36 . 88% decrease; p<3 . 93e-05; Student’s t test ) in linc–Brn1b mutant vs control cortices ( Figure 6E ) . The observed abnormalities were specific to intermediate progenitors , as there was no statistically significant change in the proliferation rate of apical progenitors within the VZ ( Figure 6E ) . In agreement with a reduction in intermediate progenitors of the cortical SVZ we found decreased expression of the markers TBR2 ( Figure 6F ) , Cux2 ( Figure 6G ) , and Svet1 ( Figure 6H ) ( Tarabykin et al . , 2001; Zimmer , 2004; Molyneaux et al . , 2007; Arnold et al . , 2008; Cubelos et al . , 2008; Sessa et al . , 2008 ) in linc–Brn1b−/− mutants . Together , the data indicate that linc–Brn1b knockout mice exhibit a decrease in proliferation of cortical progenitors in vivo , and that this defect is restricted to a specific subpopulation of progenitors in the SVZ of the developing cortex . To investigate whether the observed decrease in SVZ progenitors resulted in an overall reduction in cortical thickness , we measured linc–Brn1b−/− mutants and wild-type littermates at P7 . The distance between the pia and the white matter was measured at matched medio-lateral and rostro-caudal locations ( n = 17 wild-type sections and n = 10 linc–Brn1b−/− sections ) . We reproducibly observed a significant 6 . 24% decrease in the total thickness of the neocortex in the mutants , relative to wild type ( 1317 . 31 µm ± 20 . 03 in linc–Brn1b−/− vs 1404 . 90 µm ± 15 . 07 in wild type; p<0 . 002 , Student’s t test ) ( data not shown ) . To define whether generation of projection neuron subtypes was affected in the absence of linc–Brn1b , cortices from linc–Brn1b−/− and wild-type littermates were collected at P7 and immunostained for markers of different projection neuron classes: CUX1 , a marker of upper layer II/III and IV callosal projection neurons ( Zimmer , 2004 ) ( Figure 7A ) , CTIP2 , a marker of layer V subcerebral projection neurons ( Arlotta et al . , 2005 ) ( Figure 7B ) , and TLE4 , a marker of layer VI corticothalamic projection neurons ( Koop et al . , 1996; Chen , 2005; Molyneaux et al . , 2005 ) ( Figure 7C ) . Measurements of the thickness of each cortical layer showed a distinct reduction in the upper layer II/III–IV ( 11 . 62% reduction in absolute thickness in linc–Brn1b−/− vs wild type; Figure 7A , p<0 . 0004; Student’s t test ) , which was also detectable by histological Nissl staining ( Figure 7D ) . In contrast , no significant change in the absolute thickness of either layer V ( Figure 7B , p<0 . 18; Student’s t test ) or VI ( Figure 7C , p<0 . 54; Student’s t test ) was detected . 10 . 7554/eLife . 01749 . 017Figure 7 . Abnormal cortical lamination and disruption of the barrel cortex in linc–Brn1b−/− mice . ( A ) Immunofluoresence staining and quantification on coronal sections of P7 wild type and linc–Brn1b−/− cortices for upper layer II–IV marker CUX1 show a significant reduction in absolute layer thickness and total number of CUX1+ projection neurons in linc–Brn1b−/− mice . ( B and C ) Immunofluorescence staining and quantification on coronal sections of P7 wild type and linc–Brn1b−/− cortices for layer V marker CTIP2 ( B ) , and for layer VI marker TLE4 ( C ) show not significant change in the total thickness of layer V and VI , but significant increase in the number of CTIP2+ subcerebral projection neurons and TLE4+ corticothalamic neurons in linc–Brn1b−/− mice . ( D ) Nissl staining of coronal sections of P8 wild type and linc–Brn1b−/− cortices shows overall cortical reduction and specific decrease in upper layer thickness in linc–Brn1b−/− mice . ( E ) Rorβ in situ hybridization on coronal sections of P7 wild type and linc–Brn1b−/− primary somatosensory cortex . ( F ) Cytochrome oxidase c activity on sections spanning barrel cortex shows reduction in the anteriolateral barrel subfield ( ALBSF ) in linc–Brn1b−/− mice . The loss of barrels in the ALBSF and their general disorganization are confirmed by immunofluorescence staining for vGLUT2 ( G ) and 5-HTT ( H ) . The full arrows indicate individual barrels and arrowheads point to corresponding barrels that are absent in the linc–Brn1b−/− brains . p<0 . 05 , p<0 . 01 , *p<0 . 001; Student’s t test . Scale bars = 500 µm ( A–E ) and ( G–H ) , 100 µm ( F ) . ctx , cortex; cc , corpus callosum; str , striatum; PMBSF and ALBSF , posteriomedial and anteriolateral barrel subfield . DOI: http://dx . doi . org/10 . 7554/eLife . 01749 . 017 In order to determine whether the number of specific classes of neurons within layers was affected in the mutant cortex , we quantified the percentage of CUX1+ , CTIP2+ and TLE4+ neurons . In agreement with a reduction in the thickness of layer II/III–IV , we observed that the number of CUX1+ callosal projection neurons was decreased by 12 . 97% in the mutant compared to wild type ( Figure 7A , mean = 1137 ± 69 CUX1+ cells in linc–Brn1b−/− vs mean = 1307 ± 52 in wild type; p<0 . 03; Student’s t test ) . Conversely , despite the unchanged thickness of layers V and VI , the number of CTIP2+ subcerebral projection neurons ( Figure 7B , mean = 271 ± 28 in linc–Brn1b−/− vs mean = 199 ± 14 in wild type; p<0 . 016; Student’s t test ) and TLE4+ corticothalamic projection neurons ( Figure 7C , mean = 1432 ± 37 in linc–Brn1b−/− vs mean = 1176 ± 31 in wild type; p<5 . 9 × 10−5; Student’s t test ) neurons were increased by 36 . 39% and 21 . 77% in the linc–Brn1b−/− mice , respectively . These data suggest the possibility that a subset of upper layer progenitors abnormally generate deep layer neurons in place of CUX1+ upper layer neurons in the linc–Brn1b−/− mutant . The selective reduction in the generation of upper layer cortical neurons is interesting , considering that these neurons are principally derived from the expansion of intermediate progenitors , which are reduced in linc–Brn1b−/− mice . Taken together , these results indicate that linc–Brn1b is required for proper generation of cortical neurons , in particular projection neurons of layer II/III–IV . Whole mount analysis of linc−Brn1b areal distribution showed high levels of expression in primary somatosensory cortex , an area that in rodents receives thalamocortical afferents relaying sensory information from the mustacial vibrissae ( Petersen , 2007 ) . Given the expression of linc–Brn1b within this region , and the requirement for proper specification of upper layer neurons , we investigated whether linc–Brn1b is required for the proper development of the somatosensory cortex and organization of the barrel structures . In situ hybridization against the barrel cortex marker RAR-related orphin beta ( Rorβ ) ( Jabaudon et al . , 2012 ) , on matched coronal sections of P7 wild type and knockout mouse cortices demonstrated a reduction in the total size of the barrel cortex in the linc–Brn1b−/− mice ( Figure 7E ) with a more pronounced loss of Rorβ+ neurons at the medial edge . Histochemical staining of cytochrome oxidase C activity in the somatosensory cortex was examined and showed a distinct disruption of the individual barrel structures , particularly within the anteriolateral barrel subfield ( ALBSF ) ( Figure 7F ) . A reduction in overall area and number of barrels in linc–Brn1b mutants was also observed in the highly organized posteriomedial barrel subfield ( PMBSF ) . These findings , already consistent with the Rorß in situ hybridization data , were confirmed in P7 coronal sections immunostained with vGLUT2 ( Figure 7G ) and 5-HTT ( Figure 7H ) , both of which specifically label the barrel structures . Analysis of both of these markers corroborates a specific impairment of the barrels within the ALBSF , and a general disorganization of individual barrel structures in the linc–Brn1b−/− mice . Taken together , these results demonstrate the requirement of linc−Brn1b for the proper development of different classes of projection neurons within the cerebral cortex , and suggest that the loss of linc−Brn1b may potentially have broader implications for state-dependent cortical sensory processing . In the post genomic era , thousands of long noncoding RNAs have been discovered as transcribed units in mammalian genomes . However , what fraction of these new transcripts have general functional significance in vivo is debated . While several studies have indicated a role for lincRNAs in diverse biological processes ( Ponting et al . , 2009; Rinn and Chang , 2012; Mercer and Mattick , 2013; Ulitsky and Bartel , 2013 ) , it has been suggested that most transcripts could represent nonfunctional transcriptional by-products ( Struhl , 2007; Kowalczyk et al . , 2012 ) . Early critical studies of knockout strains ( e . g . , Xist and Tsix ) did find lncRNAs implicated in X inactivation to be required for life . Yet , of the relatively few lncRNA mouse models derived since , many have displayed subtle defects or no phenotype ( Ripoche et al . , 1997; Gordon et al . , 2010; Anguera et al . , 2011; Nakagawa et al . , 2011; Zhang et al . , 2012 ) . Other strategies using RNAi and xenografts to assess the function of lncRNAs in vivo have revealed interesting roles in development and tumor growth ( Ulitsky et al . , 2011; Wang et al . , 2011; Yang et al . , 2013 ) . Together with difficulties in finding a phenotype in mouse models such as Malat1 , Neat1 ( Nakagawa et al . , 2011; Zhang et al . , 2012 ) , these findings have led some to suspect that acute inactivation of lncRNAs leads to stronger phenotypes than constitutive deletions , where compensatory events may occur . Therefore , our study , grounded in genetic deletions , demonstrates the important physiological insights that can be gleaned by constitutive lincRNA knockouts . By leveraging our large-scale RNA-sequencing and genomics studies in combination with an integrated and multifaceted candidate selection pipeline , we were successful in finding physiologically essential lincRNAs . Through the initial characterization of the 18 lincRNAs knockout strains generated here , we found three that exhibit peri- or post-natal lethality and two additional ones with distinct developmental defects . Future studies will describe in detail other promising phenotypes . A few recent studies have demonstrated that lincRNAs regulate neighboring protein coding genes and thereby may function as cis-enhancers or cis-regulatory elements ( Ørom et al . , 2010; Wang et al . , 2011 ) . Here , we were able to examine several lincRNAs and the regulation of the neighboring protein coding genes in a genetically defined context . We find one instance where the deletion of a lincRNA , Fendrr , phenocopies the neighboring protein coding gene despite a non-significant effect on Foxf1 gene expression . Interestingly , chromatin predictions of enhancers ( Shen et al . , 2012 ) suggest this lincRNA may have enhancer-like properties ( H3K4me1 and Pol II ) . However , our genetic analysis is not consistent with the notion of Fendrr as a chromatin signature-defined ‘enhancer’ . Specifically , Grote et al observed a similar lethality phenotype in Fendrr mutants , yet their strategy ( insertion of a strong transcription stop sequence ) retained almost the entire DNA segment of Fendrr . This suggests that Fendrr cannot be acting simply as a DNA enhancer , although we cannot rule out the passive act of transcription as being functional . In our model , both the endogenous promoter and first exon remain . Moreover , the introduced lacZ reporter is actively transcribed as well , thus ruling out the passive act of transcription for Fendrr activity . Finally , since this lincRNA does not appear to act as a local enhancer and does not encode a small peptide based on PhyloCSF and Ribosome profiling analyses , we believe Fendrr to be a physiologically relevant and functional RNA molecule . Interestingly , Fendrr mutant mice exhibit many features of Foxf1a protein coding gene heterozygous mutants ( Mahlapuu et al . , 2001 ) . Since we do not observe a significant decrease in Foxf1a expression in Fendrr mutants , this raises the possibility that Fendrr could act downstream of Foxf1a . The FOXF1A protein plays an important role in the development of the lung and the gastrointestinal tract ( Mahlapuu et al . , 2001 ) . Similarly , we found that Fendrr is expressed in the mesenchyme of these tissues and that homozygote mutant mice display lung and gastrointestinal tract defects most likely leading to perinatal lethality . Several cases of genetic deletions encompassing the neighboring protein coding gene FOXF1 , and also FENDRR have been observed in neonates with the lethal lung disorder ‘alveolar capillary dysplasia with misalignment of pulmonary veins’ ( ACD/MPV ) ( Szafranski et al . , 2013 ) . This supports our hypothesis that lincRNAs can be required for normal organ development and misregulated in pathological states , which could translate to human disease . Deletion of both Peril and Mdgt also results in viability defects . Mdgt is transcribed from a bidirectional promoter shared with Hoxd1 , which is located only 84 bp away from Mdgt . Although we do not exclude that Hoxd1 expression could be affected by our Mdgt deletion strategy , the phenotype of homozygous Hoxd1 mutants , which are viable and fertile , is in sharp contrast with Mdgt−/− mutants ( Guo et al . , 2010 ) . Thus , deletion of Mdgt does not phenocopy Hoxd1 , indicating a distinct function for this lincRNA . Interestingly , mice with a deletion of the adjacent Hoxd3 protein coding gene display reduced viability similar to Mdgt ( Condie and Capecchi , 1993 ) . However , contrary to Mdgt−/− , the surviving Hoxd3−/− mice do not appear to have a growth defect , again suggesting a distinct function for Mdgt . Neither the Peril nor Mdgt locus appears to be enriched for the signature enhancer modifications H3K4me1 , H3K27ac in any public mouse ENCODE datasets . In the case of Peril , no genes were found significantly differentially expressed in knockout vs wild-type brains within ±1 Mb of Peril ( Figure 3—figure supplement 2C; p<1 . 0 , bootstrapped from 1000 random genomic intervals ) . Therefore , cis effects , as suggested by other enhancer-associated lincRNAs like Hottip ( Wang et al . , 2011 ) , are unlikely . Collectively , these results suggest that lethality of the three lincRNAs observed in this study are likely not due to cis or enhancer-like RNA regulatory effects . In addition to viability phenotypes , we also describe several developmental abnormalities such as body size and cortical defects in the linc–Brn1b mutant strain . Based on currently available public data , the linc–Brn1b locus does not appear to have enrichment for chromatin marks characteristic of enhancers , but rather the canonical H3K4me3/H3K36me3 signature of active transcription in the brain . In this case , however , deletion of linc–Brn1b did affect the expression of the neighboring protein coding gene . linc–Brn1b resides near Brn1 ( Pou3f3 ) , a key transcription factor that shares redundant roles with the closely related paralog Brn2 ( Pou3f2 ) in upper layer cortical development . Interestingly , deletion of linc–Brn1b results in a significant reduction in the neighboring BRN1 protein ( Figure 6 ) . However , in contrast to linc–Brn1b mutants , deletion of Brn1 alone does not lead to defects in cortical lamination . Only when both Brn1 and Brn2 are deleted ( Brn1/2 double mutants ) is a reduction in layers II–IV neurons observed ( McEvilly , 2002; Sugitani et al . , 2002; Dominguez et al . , 2012 ) . Despite the differential regulation of the adjacent Brn1 gene , linc–Brn1b does not appear to act as a general cis-enhancer . The region has five other genes within ±1 Mb that are differentially expressed . However , this is not significant given the dramatic changes to the transcriptome in the linc–Brn1b−/− E13 . 5 telencephalon ( Figure 3—figure supplement 2B; p<0 . 225 , bootstrapped from 1000 random genomic intervals ) . We do not observe a significant difference in expression of the paralogous Brn2 protein coding gene in any linc–Brn1b−/− developmental stage analyzed when compared to wild type ( Figure 6—figure supplement 1C ) , Thus , despite an observed decrease in Brn1 expression in linc–Brn1b−/− , deletion of linc–Brn1b does not phenocopy deletion of the neighboring protein coding gene Brn1 . In fact , ablation of linc–Brn1b results in a stronger phenotype than that observed for the neighboring protein coding gene , suggesting additional roles for this lincRNA . In addition , we observe expression of linc–Brn1b in regions of the brain with no known expression of Brn1 . Together , these observations suggest that lincRNAs adjacent to important developmental regulators may act on upstream and/or distinct pathways in addition to already reported cis-enhancer-like mechanisms ( Ørom et al . , 2010; Wang et al . , 2011 ) . Further leveraging our genetically defined deletion we gleaned additional insights into lincRNA biology and transcriptional regulation in vivo . It has been noted that lincRNAs have a propensity to be transcribed from bidirectional promoters ( Cabili et al . , 2011; Pauli et al . , 2012; Ulitsky and Bartel , 2013 ) . New studies have also suggested that many lincRNAs transcribed from bidirectional promoters are unstable and likely non-functional transcripts ( Almada et al . , 2013; Sigova et al . , 2013 ) . Interestingly , Fendrr is transcribed from a bidirectional promoter . Here , we observe that deletion of Fendrr is lethal despite leaving an intact promoter , expression of the first exon and not interfering with Pol II dynamics at this locus . Similarly , RNA-Seq analysis of linc–Brn1b knockout revealed a strong and significant upregulation of linc–Brn1a , a lincRNA transcribed from a bidirectional promoter shared with Brn1 . This incongruous activity on adjacent genes ( decrease in Brn1 protein coding gene and increase in linc–Brn1a transcribed from a bidirectional promoter with Brn1 ) observed in the linc–Brn1b−/− brain is suggestive of a functional role for the dynamically regulated linc–Brn1a , rather than transcriptional noise from bidirectional transcription from the protein coding gene promoter . Together , these findings confirm that not all lincRNAs transcribed from a bidirectional promoter are irrelevant transcriptional by-products , but rather suggests that some fraction of these transcripts play critical functional roles during development . Our framework for lincRNA candidate selection for genetic analysis , based on RNA-sequencing catalogs and genomic studies , has led us to unexplored roles for lincRNAs in brain development . The brain ( and the CNS more broadly ) constitutes one of the most complex and fast evolving organs in the body . Here , it is likely that complex regulatory mechanisms of lincRNAs have evolved to build the layered control of gene expression necessary to generate the unparalleled cellular diversity and complex function of this organ . linc–Brn1b represents an example of a lincRNA that has an effect on development of the cerebral cortex . Beyond linc–Brn1b , many more lincRNAs from our screen have similar restricted patterns within progenitors in the VZ and SVZ of the telencephalon , possibly suggesting roles in neurogenesis . Others have highly cell-specific and dynamic in vivo expression patterns in distinct regions of the brain ( Mercer et al . , 2008 ) . Thus , it is likely that more extensive work will reveal additional lincRNA mutants strains with additional brain and/or behavioral defects yet unexplored here . While rigorous behavioral studies will , in the future , determine whether loss of a lincRNA may result in specific behavioral abnormalities , we have made promising initial observations on two mutant strains . One mutant strain ( Spasm ) has tremors and a propensity to develop spastic movements upon handling , while a second strain ( linc–p21 ) displays clasping of hind limbs when lifted by the tail ( Sauvageau et al . , unpublished ) . As a whole , our 18 lincRNA knockout mouse models have revealed important aspects of lincRNA biology and constitute a useful resource for many future studies on the roles of lincRNAs in mammalian development , physiology and behavior . lincRNA knockout mice were generated by replacing the selected lincRNA gene with a lacZ cassette . Briefly , targeting constructs were constructed using VelociGene technology as described previously ( Valenzuela et al . , 2003 ) . The VelociGene Allele Identification Numbers are shown in Supplementary file 1A . Linearized targeting constructs , generated by gap repair cloning containing mouse lincRNA upstream and downstream homology arms flanking a KOZAK-ATG-lacZ-pA-LoxP-hUb1-EM7-neo ( superscript R ) -pA-LoxP cassette , were electroporated into VGF1 hybrid mouse embryonic stem ( ES ) cells , derived from a 129S6S v/Ev female to a C57BL/6N male mating . Mouse ES cells carrying a heterozygous deletion of the lincRNA gene were identified by loss-of-function allele screening with 2 Taqman qPCR assays ( Supplementary file 1F ) . Simultaneous replacement of the lincRNA gene with the lacZ cassette was confirmed by gain-of-allele Taqman assays against the lacZ and neomycin resistance cassette ( Supplementary file 1F ) . Probes were labeled with 6-carboxy-fluorescein ( FAM ) on their 5′ ends and BHQ-1 on their 3′ ends . Targeted ES clones were introduced into an 8-cell stage mouse embryo using the VelociMouse method ( Poueymirou et al . , 2006 ) . Mice were backcrossed once with C57BL/6J . Mutant mice were identified by genotyping for loss of lincRNA allele and gain of lacZ cassette . Toe clips , embryos or yolk sac were digested for 30 min at 95°C in 100 μl of 25 mM Sodium Hydroxide and 0 . 2 mM EDTA . Tissue digestion was neutralized by adding 100 μl of 40 mM Tris-HCl . PCR reactions using 4 μl of digested tissue with 10 mM lacZ specific and lincRNA gene specific primer pairs ( Supplementary file 1C for sequences ) were then performed and run on a 2% agarose gel . PCR conditions were as follows: 5 min at 95°C followed by 35 cycles of 30 s at 95°C , 45 s at 60°C , 30 s at 72°C and a final step at 72°C for 2 min . Mice were housed under controlled pathogen-free conditions ( Harvard University’s Biological Research Infrastructure ) and experiments were approved by the Harvard University Committee on the Use of Animals in Research and Teaching . Viability of the 18 lincRNA mutant strains was determined at postnatal day 21 by genotyping the progeny of heterozygous intercrosses ( Supplementary file 1C for genotyping primer sequences ) . In the case of lethal strains , the developmental stage at which lethality occurs was determined by genotyping of embryos at E14 . 5 and E18 . 5 and newborns . Respiratory function ( Fendrr mutant strain ) was evaluated in surgically delivered E18 . 5 embryos from heterozygous intercrosses ( Eggan et al . , 2001 ) . After cleaning of the airways , pups were placed on a 37°C warm pad and observed for sign of breathing . Total RNA from embryonic and postnatal mouse tissues , neural stem cells , and neurospheres was isolated using TRIzol ( Life Technology , Carlsbad , CA ) /chloroform extraction followed by spin-column purification ( RNeasy mini kit , Qiagen , Venlo , Netherlands ) according to the manufacturer instructions . RNA concentration and purity were determined using a Nanodrop ( Thermo Fisher , Waltham , MA ) . RNA integrity was assessed on a Bioanalyzer ( Agilent , Santa Clara , CA ) using the RNA 6000 RNA chip . High-quality RNA samples ( RNA Integrity Number ≥8 ) were used for library preparation . mRNA-seq libraries were constructed using the TruSeq RNA Sample Preparation Kit ( Illumina , San Diego , CA ) as previously described ( Trapnell et al . , 2012a; Sun et al . , 2013 ) . 500 ng total RNA was used as input for the TruSeq libraries from mouse tissues , and 200 ng for the libraries from neural stem cells and neurospheres . Prior to sequencing , libraries were run on a Bioanalyzer DNA7500 chip to assess purity , fragment size , and concentration . Libraries free of adapter dimers and with a peak region area ( 220–500 bp ) ≥80% of the total area were sequenced . Individually barcoded samples were pooled and sequenced on the Illumina HiSeq 2000 platform . Paired-end 101 bp reads were aligned to the mouse ( mm9 ) reference genome assembly and , for the human neuronal differentiation time course also to the human ( hg19 ) assembly , using Tophat2 ( Trapnell et al . , 2009 ) with default options and assembled into transcripts with Cufflinks ( Trapnell et al . , 2012a , 2012b ) . Aligned reads and assembled transcriptome catalog were used as input for Cuffdiff2 ( Trapnell et al . , 2012a ) to determine expression levels ( FPKM , Fragments Per Kilobase per Million mapped reads ) and differential expression between conditions using default options . CummeRbund v2 . 1 ( http://compbio . mit . edu/cummeRbund/ ) was then used to process , index , and visualize the output of the Cuffdiff2 analyses . Gene set enrichment analysis ( GSEA ) and Guilt-by-Association analysis ( GBA ) were performed to predict the effect of gene expression changes on biological processes . Detailed description of GSEA and GBA analysis , RNA isolation , and libraries preparation are provided in the Extended Experimental Procedures . Cis-enhancer activity was tested by determining the number of genes with differential expression in a particular Knockout vs wild type contrast within ±1 Mb window of the targeted lincRNA . 1000 random genomic intervals of the same size were obtained and interrogated in kind to determine how often the same number of differentially expressed ( DE ) genes could be identified . The ratio of intervals with DE genes >= the number of DE genes in the target-flanking window to the number of iterations , provided a bootstrapped p value and false discovery rate estimate . GSEA was performed to predict the effect of the significant gene expression changes on biological processes in the knockout mice . For a given comparison of KO vs WT differential expression , all genes were rank-ordered by their Cuffdiff2 test-statistics and mouse gene identifiers were mapped to human HUGO gene names . Genesets from the Reactome collection at mSigDB were obtained and for each gene set , the relative enrichment or depletion within our ranked list was determined via Mann-Whitney U-test . p values were corrected for multiple tests using the Benjamini-Hochberg method . Genesets with q<0 . 001 were selected and presented as a heatmap with color mapped to the Z-score of the Cuffdiff2 test statics for genes in the specific gene set relative to all genes . Redundant genesets were aggregated into higher-level biological processes . Predictive GBA analysis for 17/18 tested lincRNAs was conducted as follows: Pearson correlation values of FPKM expression profiles were calculated for each lincRNA to all protein coding genes across a compendium of RNA-Seq samples ( combination of in-house samples and samples from [Merkin et al . , 2012] ) . Protein coding genes were then rank-ordered and subjected to the gene set enrichment analysis described above . Significant genesets for a given lincRNA represent the most likely pathways/biological processes for which this lincRNA may play a role . Expression of the knocked-in lacZ reporter gene was assessed in heterozygous mice . Embryos ( from E13 . 5 to E18 . 5 ) were fixed in 4% paraformaldehyde ( PFA ) in phosphate buffered saline ( PBS ) overnight at 4°C prior to dissection of the brain , lung and respiratory tract , digestive tract , heart , and other organs . P7 brain , from linc–Brn1b mutant strain , were dissected from pups transcardially perfused with 4% paraformaldehyde ( PFA ) , and fixed overnight at 4°C . The fixed tissues were rinsed three times at room temperature in PBS , 2 mM MgCl2 , 0 . 01% deoxycholic acid , 0 . 02% NP-40 . X-gal staining was performed by incubating the tissues for up to 16 hr at 37°C in the same buffer supplemented with 5 mM potassium ferricyanide , 5 mM potassium ferrocyanide and 1 mg/ml X-gal . Staining reaction was stopped by washing three times in PBS at room temperature , followed by 2 hr post-fixation in 4% PFA at 4°C . Stained whole organs and sagittal brain sections were imaged using a Leica M216FA stereomicroscope ( Leica Microsystems , Buffalo Grove , IL ) equipped with a DFC300 FX digital imaging camera ( Watson et al . , 2008 ) . Histology was performed at the Rodent Histopathology Service of the Dana Farber/Harvard Cancer Center Pathology Research Core . Embryos were harvested , fixed in Bouin’s solution and embebbed in paraffin . Microtome sections were stained with hematoxilin and/or eosin for histological analysis . Embryonic brains , dissected in cold PBS and fixed in 4% PFA/PBS overnight at 4°C , and P7 brains dissected from pups transcardially perfused with 4% PFA and post-fixed as described above , were processed for Nissl staining and immunofluorescence as previously described ( Englund et al . , 2005; Molyneaux et al . , 2005 ) . Nissl-stained and immunostained sections were imaged using a Nikon 90i fluorescence microscope equipped with a Retiga Exi camera ( Q-IMAGING , Surrey , Canada ) and acquired with Volocity image analysis software v4 . 0 . 1 ( Perkin Elmer , Waltham , MA ) . For quantification of overall cortical thickness , cortical layers and number of CUX1+ , CTIP2+ and TLE4+ cells within the primary somatosensory cortical area , anatomically matched sections were processed ( n = 3 linc–Brn1b−/−; n = 3 wild-type , at P7 ) . Boxes of 300 pixels in width and spanning the thickness of the cortex were superimposed at matched locations on each section , and the overall cortical thickness was measured as the distance from the pia to the white matter in each box , using ImageJ . Specific layer thicknesses were measured at the midpoint of the matched-location 300 pixel images for each of the TLE4+ , CTIP2+ and SATB2+ immunofluorescence stainings using ImageJ . Layer VI thickness was measured as the distance between the dorsal edge of the TLE4+ region and the white matter . Layer V thickness was determined by the span of the CTIP2+ region , and layer II–IV thickness were measured as the SATB2+ region between the dorsal edge of the CTIP2+ stain and the pia . In each case results were expressed as mean ± SEM . Cell counts of the specific neuronal subpopulations were obtained using the ITCN plugin for ImageJ and results were expressed as mean ± SEM . A priori criteria were defined for analysis . Statistical analysis was performed using R unpaired Student’s t test assuming equal variance was used for the pairwise comparisons . Primary antibodies and dilutions were as follows: anti-ßgal ( CGAL-45; 1:500; Immunology Consultants Laboratory , Portland OR ) , anti-SATB2 ( ab51502; 1:50; Abcam , Cambridge , UK ) , anti-CUX1 ( sc-13024; 1:100; Santa Cruz Biotechnology , Dallas , TX ) , anti-pH3 ( 06-570; 1:500; Millipore , Billerica , MA ) , anti-TBR2 ( 1:2000; Gift from Robert F Hevner Lab ) , anti-CTIP2 ( ab18465; 1:100; Abcam ) , anti-TLE4 ( sc-9125; 1:100; Santa Cruz ) , anti-vGLUT2 ( MAB5504; 1:50; Millipore ) , anti-5-HTT ( PC177L; 1:1000; Millipore ) , anti-TUJ1 ( mms-435P; 1:1000; Covance , Princeton , NJ ) . Secondary antibodies were from the Molecular Probes ( Eugene , OR ) Alexa series and were used at 1:750 dilution . Immunostained sections were imaged using a Nikon 90i fluorescence microscope equipped with a Retiga Exi camera ( Q-IMAGING ) and acquired with Volocity image analysis software v4 . 0 . 1 ( Improvision ) . linc–Brn1b−/− and wild-type littermates at P7 were anesthetized , perfused with 4% PFA , decapitated , and the brain rapidly removed . The brains were post-fixed in 4% PFA for 3 hr at 4°C . Cortices containing the barrel field were dissected and flattened as described ( Welker and Woolsey , 1974 ) , post-fixed in 4% PFA for 12–14 hr at 4°C and sectioned by using a vibratome ( 80 μm ) . Sections were incubated in phosphate buffer containing 0 . 5 mg/ml DAB , 0 . 18 mg cytochrome C , and 40 mg/ml sucrose for 3–5 hr at 37°C , rinsed , and mounted in Fluoromount-G ( SouthernBiotech , Birmingham , AL ) ( Wong-Riley , 1979; Land and Simons , 1985 ) . Nonradioactive in situ hybridization was performed on 40 μm vibratome sections mounted on superfrost slides ( Fisher Scientific , Waltham , MA ) as previously described ( Berger and Hediger , 2001; Arlotta et al . , 2005 ) . The probe for Svet1 was a gift from M Studer . Probes for RorB ( nt 1573-2087 of NM_146095 ) and Cux2 ( nt 1069-11694 of NM_007804 ) transcripts were generated by PCR from mouse brain cDNA and subcloned in pCRII-TOPO ( Life Technologies , Carlsbad , CA ) . Antisense riboprobes were generated by in vitro transcription using SP6 or T7 polymerase ( Roche Applied Science , Penzberg , Germany ) as previously described . Sense probes were used as negative controls . Single molecule FISH was performedas described by Raj et al . ( 2008 ) . Briefly , oligonucleotide probes targeting and tiling Peril ( 48 probes ) and linc–Brn1b ( 20 probes ) were conjugated to Quasar 570 fluorophores and HPLC purified ( Biosearch Technologies , Petaluma , CA ) . A list of the Peril and linc–Brn1b probes ( sequence , position , and %CG content ) is provided in Supplementary file 1E . Dissociated E14 . 5 cortical neurospheres or mouse ES cells were fixed in 2% formaldehyde for 10 min , washed twice with PBS , and permeabilized with 70% ethanol . The cells were then seeded onto previously gelatinized two-chamber cover glasses . Prior to hybridization , the cells were rehydrated in wash buffer containing 10% formamide and 2 × SSC for 5 min . Probes ( 0 . 5 ng/μl final concentration ) were hybridized in 10% dextran sulfate , 10% formamide , and 2 × SSC at 37°C overnight . After hybridization , cells were washed twice with wash buffer at 37°C for 30 min ( with DAPI added to the second wash for nuclear staining ) , and twice with 2 × SSC . After the SSC wash , the cells were equilibrated in anti-fade buffer ( 2 × SSC , 0 . 4% glucose , 10 mM Tris pH 8 . 0 ) for 3–5 min . Cells were mounted in 100 μl anti-fade buffer supplemented with 1 μl of glucose oxidase ( G2133-10KU; Sigma-Aldrich , St . Louis , MO ) and 1 μl of catalase ( C3515-10 MG; Sigma-Aldrich ) and immediately imaged with a LSM 700 Inverted Confocal microscope ( Zeiss , Jena , Germany ) . 25 Z-stacks were taken per field , using DAPI and laser 639 for excitation . H1 human neural stem cells were prepared as described previously ( Arnold et al . , 2008; Goff et al . , 2009 ) and grown at 37°C , 5% CO2 on 1:4 diluted Matrigel-coated wells in neural proliferation medium ( NPM; 50% DMEM/F12 Glutamax , 50% Neurobasal medium , 0 . 5X N2 , 0 . 5X B27 without vitamin A , 20 ng/ml FGF [Life Technologies] ) . For differentiation , cells were plated at a density of 106 cells per well in a 6-well plate and allowed to proliferate for one day in the NPM medium . Neural induction was then initiated by withdrawal of FGF and addition of BDNF by switching the medium to neural differentiation medium ( NDM; 100% Neurobasal medium , 1X B27 without vitamin A [Life Technologies] , 10 ng/ml BDNF [Peprotech , Rocky Hill , NJ] . ) Differentiating cultures were maintained by refreshing NDM every other day until collection . Samples of these cultures were collected at days 0 , 1 , 2 , and 4 . Remaining cells ( those designated for collection at days 5 , 11 , and 18 ) were replated at day 4 at a density 106 cells per well of a Poly-D-lysine/laminin-coated 6-well plate . Cells were harvested with Accutase ( Stem Cell Technologies , Vancouver , Canada ) and RNA collected as described above . linc–Brn1b−/− and WT E14 . 5-derived neurospheres ( passage 3 ) and E15 . 5 cortices were lysed in RIPA buffer ( 1% NP-40 , 1% Na-deoxycholate , 0 . 2% SDS , 50 mM Tris 7 . 4 , 500 mM NaCl ) containing protease inhibitor cocktail ( Roche Applied Science ) . Proteins were resolved by SDS-PAGE and electroblotted . Blots were probed sequentially for BRN1 ( anti-BRN1 , sc-6028-R , Santa Cruz ) and GAPDH ( anti-GAPDH , sc-365062-R , Santa Cruz ) to control for protein loading . Immunoreactive bands were detected by enhanced chemiluminescence ( SuperSignal , Thermo Fisher Scientific , Waltham , MA ) , and visualized with a Gel Doc ( Biorad ) . All primary RNA-Seq data are available at the Gene Expression Omnibus ( GSE49581 ) .	The mammalian genome is comprised of DNA sequences that contain the templates for proteins , and other DNA sequences that do not code for proteins . The coding DNA sequences are transcribed to make messenger RNA molecules , which are then translated to make proteins . Researchers have known for many years that some of the noncoding DNA sequences are also transcribed to make other types of RNA molecules , such as transfer and ribosomal RNA . However , the true breadth and diversity of the roles played by these other RNA molecules have only recently begun to be fully appreciated . Mammalian genomes contain thousands of noncoding DNA sequences that are transcribed . Recent in vitro studies suggest that the resulting long noncoding RNA molecules can act as regulators of transcription , translation , and cell cycle . In vitro studies also suggest that these long noncoding RNA molecules may play a role in mammalian development and disease . Yet few in vivo studies have been performed to support or confirm such hypotheses . Now Sauvageau et al . have developed several lines of knockout mice to investigate a subset of noncoding RNA molecules known as long intergenic noncoding RNAs ( lincRNAs ) . These experiments reveal that lincRNAs have a strong influence on the overall viability of mice , and also on a number of developmental processes , including the development of lungs and the cerebral cortex . Given that the vast majority of the human genome is transcribed , the mouse models developed by Sauvageau et al . represent an important step in determining the physiological relevance , on a genetic level , of the noncoding portion of the genome in vivo .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology" ]	2013	Multiple knockout mouse models reveal lincRNAs are required for life and brain development
Testosterone products are prescribed to males for a variety of possible health benefits , but causal effects are unclear . Evidence from randomized trials are difficult to obtain , particularly regarding effects on long-term or rare outcomes . Mendelian randomization analyses were performed to infer phenome-wide effects of free testosterone on 461 outcomes in 161 , 268 males from the UK Biobank study . Lifelong increased free testosterone had beneficial effects on increased bone mineral density , and decreased body fat; adverse effects on decreased HDL , and increased risks of prostate cancer , androgenic alopecia , spinal stenosis , and hypertension; and context-dependent effects on increased hematocrit and decreased C-reactive protein . No benefit was observed for type 2 diabetes , cardiovascular or cognitive outcomes . Mendelian randomization suggests benefits of long-term increased testosterone should be considered against adverse effects , notably increased prostate cancer and hypertension . Well-powered randomized trials are needed to conclusively address risks and benefits of testosterone treatment on these outcomes . In developed countries , rising rates of both serum testosterone level testing and therapy initiation have been observed among older male patients ( Handelsman , 2013; Layton et al . , 2014 ) . In the USA alone , it is estimated 1 . 5–1 . 7% of males are prescribed testosterone ( Baillargeon et al . , 2018; Jasuja et al . , 2017 ) . Randomized clinical trials ( RCT ) have attempted to elucidate the benefits and risks of testosterone treatment ( Bhasin et al . , 2018a; Gagliano-Jucá and Basaria , 2019 ) . These studies identified short-term beneficial effects on bone mineral density ( BMD ) , sexual function , body fat and muscle mass , and anaemia; potential adverse effects on venous thrombosis and coronary artery plaque; and no effects on cognitive function , fatigue , or hemoglobin A1c ( HbA1c ) ( Bhasin et al . , 2018a; Gagliano-Jucá and Basaria , 2019; Mohler et al . , 2018; Snyder et al . , 2018 ) . However , given the logistic and financial challenges involved in a well-powered RCT with appropriate follow-up , there is unlikely to be satisfactory evidence regarding long-term effects and risks of adverse outcomes , such as myocardial infarction ( MI ) , stroke and cancer ( Gagliano-Jucá and Basaria , 2019 ) . Given the rates of testosterone prescription , efforts to resolve the causal effects of testosterone on health outcomes have important public health implications ( Bhasin et al . , 2018a ) . Mendelian randomization ( MR ) is a technique for causal inference that leverages the random allocation of genetic variants to infer the unconfounded relationship between an exposure and outcome . Similar to the random assignment of participants to experimental groups in a RCT , genetic variants are randomly allocated at meiosis ( Davies et al . , 2018 ) . For instance , if individuals genetically randomized to produce higher testosterone develop different rates of cardiovascular disease ( CVD ) , then MR analysis supports a causal effect of testosterone on risk of CVD ( Figure 1—figure supplement 1 ) . Notably , this technique has previously replicated RCT findings , among others demonstrating causal roles for LDL cholesterol and dysglycemia on CVD risk ( Holmes et al . , 2015; Ross et al . , 2015 ) . Earlier MR studies investigating the effects of testosterone have demonstrated harmful effects on lipid levels but inconsistent effects on CVD , and they were limited by the small number of genetic variants ( Schooling et al . , 2018; Zhao et al . , 2014 ) . A recent MR study using the UK Biobank identified a large number of genetic variants associated with testosterone and found evidence for harmful effects on several types of cancers but sex-specific effects on type 2 diabetes ( T2D ) ( Ruth et al . , 2020 ) . This study highlighted the importance of performing sex-specific analyses for testosterone , but it was focused on glycemic and oncologic traits ( Ruth et al . , 2020 ) . Therefore , we sought to expand the scope of prior studies by performing a comprehensive scan of the effects of free testosterone on human disease in males . We hypothesized that MR and genetic risk score ( GRS ) analyses would enable estimation of the causal effects of longstanding exposure to high levels of free testosterone on health outcomes in males . We first conducted a genome-wide association study ( GWAS ) for calculated free testosterone ( CFT ) in male participants of the UK Biobank ( n = 161 , 268 ) cohort to identify genetic determinants of free testosterone levels . Then , using MR , we investigated the causal effects of lifelong genetically-elevated free testosterone levels on a priori health outcomes previously investigated in RCTs of testosterone treatment , encompassing: expected clinical benefits ( physical activity , strength , fat-free body mass , body fat , BMD , dementia , depression ) and potential adverse effects ( androgenic alopecia , heematocrit , T2D , prostate cancer , benign prostate hyperplasia , blood pressure , CVD , heart failure , ischemic stroke ) ( Figure 1; Bhasin et al . , 2018a; Gagliano-Jucá and Basaria , 2019; Mohler et al . , 2018; Snyder et al . , 2018 ) . Finally , we used GRS to investigate the associations of lifelong genetically-elevated free testosterone levels on 439 health outcomes , encompassing diseases ( n = 415 ) and biomarkers of health ( n = 24 ) ( Figure 1 ) . To calculate free testosterone levels , 187 , 524 males in the white , British subset of the UK Biobank cohort were excluded if they had missing levels of total testosterone , SHBG and albumin , or self-reported taking androgen medications . After these exclusions , the study population consisted of 161 , 268 males with an average CFT of 0 . 210 nmol/L ( Supplementary file 1 - Table 1 and Figure 1—figure supplement 2 ) . There were 13 , 338 genetic variants associated with CFT that reached genome-wide significance ( p<5×10−8 ) . After removing genetic variants associated with natural-log-transformed SHBG , there were 7048 genetic variants that comprised 93 independent signals carried forward for subsequent genetic analyses ( Supplementary file 1 - Table 2 and Figure 1—figure supplement 3 ) . Overall , chip-based heritability of CFT was estimated at 15% ( 95% CI = 14 to 16 ) , while these 93 independent genetic variants associated with CFT explained 3 . 7% of the total variance of CFT levels in males from the UK Biobank . In males from the UK Biobank , sample size for the quantitative risk factors ranged from 30 , 439 to 156 , 403 , while number of cases for dichotomous outcomes ranged from 1003 to 70 , 283 ( Table 1 ) . After adjusting for the 22 outcomes tested , one-sample MR analysis using IVW regression identified significant effects of CFT on hematocrit percentage , body fat-free percentage , body fat percentage , heel BMD , androgenic alopecia , and prostate cancer ( Table 1 ) . Each 0 . 1 nmol/L higher CFT had beneficial effects on increased heel BMD ( 0 . 40 SD; 95% CI = 0 . 25 to 0 . 54; p=1 . 10×10−7 ) , increased body fat-free percentage ( 1 . 91%; 95% CI = 1 . 48 to 2 . 35; p=9 . 06×10−18 ) , and decreased body fat percentage ( −1 . 88%; 95% CI = −2 . 31 to −1 . 45; p=1 . 65×10−17 ) , but deleterious effects on increased hematocrit percentage ( 1 . 37%; 95% CI = 1 . 12 to 1 . 62; p=1 . 03×10−27 ) , risk of prostate cancer ( OR = 1 . 51; 95% CI = 1 . 21 to 1 . 88; p=2 . 1×10−4 ) , and risk of androgenic alopecia ( OR = 1 . 49; 95% CI = 1 . 19 to 1 . 86; p=5 . 28×10−4 ) ( Figure 3—figure supplements 1–6 ) . Leave-one-out analyses did not identify any outlying individual genetic variants responsible for the observed effects on any significant outcomes . Sensitivity analyses were performed to detect violations of MR assumptions . Egger regression did not detect evidence of directional pleiotropy for any outcomes ( pintercept <0 . 05 ) ( Supplementary file 1 - Table 3 ) . Results using MR-RAPS were consistent with IVW regression method for all significant outcomes ( Supplementary file 1 – Table 4 ) . However , MR-PRESSO detected evidence of pleiotropic variants for hematocrit percentage , body fat-free percentage , body fat percentage , heel BMD , androgenic alopecia , whole body fat-free mass , hemoglobin A1C , glucose , handgrip strength , systolic blood pressure , diastolic blood pressure , T2D , and benign prostate hyperplasia ( Supplementary file 1 - Table 5 ) . However , removal of pleiotropic variants made no changes to the significance or interpretation of earlier results using IVW regression ( Supplementary file 1 - Table 5 ) . To discover novel effects of free testosterone , we tested for the association of a GRS for testosterone with 415 diseases and 24 biomarkers in the same subpopulation of unrelated males from the UK Biobank . Sample size for biomarkers ranged from 118 , 783 for lipoprotein ( a ) to 149 , 940 for total cholesterol , while number of cases for diseases ranged from 876 for ‘localized superficial swelling , mass , or lump’ to 40 , 960 for ‘hypertension’ ( Figure 2—source data 1 ) . After adjusting for the 439 outcomes tested , each 0 . 1 nmol/L increase in genetically-predicted CFT was significantly associated with beneficial effects on lowered C-reactive protein ( β = −0 . 085 SD; 95% CI = −0 . 119 to −0 . 052; p=6 . 15×10−7 ) but adverse effects on increased creatinine ( β = 0 . 113 SD; 95% CI = 0 . 079 to 0 . 146; p=4 . 78×10−11 ) , lowered apolipoprotein A ( β = −0 . 018 g/L; 95% CI = −0 . 026 to −0 . 01; p=1 . 55×10−5 ) , lowered HDL ( β = −0 . 074 SD; 95% CI = −0 . 109 to −0 . 039; p=3 . 62×10−5 ) , and increased risks of hypertension ( OR = 1 . 17; 95% CI = 1 . 08 to 1 . 26; p=2 . 83×10−5 ) , and spinal stenosis ( OR = 2 . 03; 95% CI = 1 . 51 to 2 . 75; p=3 . 82×10−6 ) ( Table 2 and Figure 2 ) . As confirmation , we demonstrated the GRS was indeed not associated with natural log-transformed natural log-transformed SHBG levels in males ( p=0 . 12 ) . For all statistically significant outcomes , associations were directionally consistent after removing participants taking blood pressure medication ( Supplementary file 1 - Table 6 ) or cholesterol-lowering medication ( Supplementary file 1 - Table 7 ) . Further sensitivity analyses were performed by repeating the one-sample MR analysis using 52 genetic variants associated with total testosterone in males from the UK Biobank ( Supplementary file 1 - Table 8 ) . For all statistically significant outcomes , effects observed using total testosterone genetic variants were directionally consistent with CFT , and results for all outcomes are presented in Supplementary file 1 - Tables 9 and 10 . Finally , most effect estimates for genetically-predicted testosterone in this study were comparable in magnitude to effect sizes reported in RCTs except bone mineral density ( Figure 3 ) . We herein perform MR and GRS analyses of CFT to identify effects of endogenous free testosterone in males on 461 health outcomes . All effects are reported in terms of 0 . 1 nmol/L of CFT to approximate expected effect sizes after initiation of testosterone treatment ( Bhasin et al . , 2018b ) . Among 22 a priori outcomes with suspected effects based on RCTs of testosterone treatment , MR analyses demonstrated that each 0 . 1 nmol/L increase in CFT was associated with adverse effects on increased risk of prostate cancer , risk of androgenic alopecia , and hematocrit percentage , but beneficial effects on increased heel BMD , increased body fat-free percentage and decreased body fat percentage . Findings on body composition , hematocrit , and BMD are consistent with short-term effects in randomized trials of testosterone treatment ( Bhasin et al . , 2018a ) . Although testosterone treatment has not been conclusively shown to increase risk of prostate cancer and androgenic alopecia in RCTs , androgen suppression therapies , such as of 5α-reductase inhibitors , are used as treatment for androgenic alopecia and prostate cancer ( Adil and Godwin , 2017; Andriole et al . , 2010 ) . The increased risk of prostate cancer replicates effects of testosterone observed in a previous MR analysis using independent data from the PRACTICAL consortium , and further supports the role of testosterone in development of these outcomes . As the leading cause of cancer among men , the predicted 1 . 5-fold increased risk as a result of changes in testosterone observed after initiation of testosterone treatment warrants further investigation in clinical trials and greater scrutiny in at-risk patient populations ( American Cancer Society , 2019; Bhasin et al . , 2018b ) . Furthermore , these results cast doubt on cardiovascular , cognitive , or metabolic benefit for increased testosterone , as we do not find evidence of a beneficial effect of CFT on hard endpoints , such as dementia , MI , stroke , fractures , or T2D ( Aukrust et al . , 2009 ) . Most of the estimates from MR analyses were comparable with effect sizes from RCTs ( Figure 3 ) . There was only significant heterogeneity between the effects on BMD for MR and RCT , but it is difficult to make direct comparisons due to variable change in testosterone levels after administration of testosterone in each RCT , different methods and anatomical sites of BMD estimation , and differences between short-term effects in RCTs relative to lifelong effects in MR . Among the remaining outcomes without well-established effects from RCTs , we identified evidence of novel associations between an increased GRS for CFT with adverse effects on creatinine , HDL , apolipoprotein A , hypertension , and spinal stenosis , but beneficial effects on C-reactive protein . Higher genetically-predicted free testosterone was associated with increased creatinine ( β = 0 . 113 SD; 95% CI = 0 . 079 to 0 . 146; p=4 . 78×10−11 ) . Mechanistically , effects of testosterone on renal function are unclear , but this effect may be mediated through the known effect of testosterone on increased muscle mass which is tightly related to serum creatinine ( Carrero et al . , 2009; Filler et al . , 2016; Schutte et al . , 1981 ) . HDL cholesterol ( β = −0 . 074 SD; 95% CI = −0 . 109 to −0 . 039; p=3 . 62×10−5 ) and its main protein component , apolipoprotein A ( β = −0 . 018 g/L; 95% CI = −0 . 026 to −0 . 01; p=1 . 55×10−5 ) , were both decreased with higher genetically-predicted free testosterone . Likewise , the Testosterone Trials found male participants over 65 years of age randomized to testosterone experienced mildly lowered HDL cholesterol levels after 12 months ( Mohler et al . , 2018; Snyder et al . , 2018 ) . Higher free testosterone was associated with decreased C-reactive protein ( CRP ) ( β = −0 . 085 SD; 95% CI = −0 . 119 to −0 . 052; p=6 . 15×10−7 ) . Although the Testosterone Trials did not find any change in CRP in its testosterone arm , testosterone is widely-believed to have suppressive effects on the immune system which may extend to markers of inflammation such as CRP ( Trigunaite et al . , 2015 ) . Furthermore , despite no effect on SBP or DBP , our analyses suggest 0 . 1 mol/L higher free testosterone is associated with increased risk of hypertension ( OR = 1 . 17; 95% CI = 1 . 08 to 1 . 27; p=1 . 05×10−4 ) . Given the multifactorial nature of this disease , the apparent discrepancy between blood pressure and hypertension may be explained by an effect on other risk factors that develop into hypertension . Moreover , both human and animal studies suggest a role of testosterone on hypertension . A randomized controlled trial found testosterone administration increased levels of NT-proBNP , and studies of both transgender men and anabolic steroid users have found testosterone increased arterial stiffness and blood pressure ( Bachmann et al . , 2019; Hartgens and Kuipers , 2004; Velho et al . , 2017 ) . Meanwhile , animal models have shown testosterone may aggravate hypertension and exacerbate increased production of reactive oxygen species specifically in hypertensive but not normotensive rat vascular endothelial tissue ( Chignalia et al . , 2012; Reckelhoff et al . , 1998 ) . Testosterone is widely-believed to have anti-inflammatory and osteogenic effects , but our analyses showed an association with higher risk of spinal stenosis ( OR = 2 . 03; 95% CI = 1 . 51 to 2 . 75; p=3 . 82×10−6 ) . However , the literature shows some evidence that higher testosterone is associated with greater loss of cartilage in healthy older males , and evidence from mouse models suggest testosterone has a sex-specific role in worsening osteoarthritis , a common risk factor for spinal stenosis ( Hanna , 2005; Hl et al . , 2007 ) . In comparison to previous MR studies , our results broaden the scope of the existing literature by comprehensively assessing the effects of testosterone on 461 health outcomes including hard endpoints and intermediate biomarkers . Moreover , a key strength of this study was the stringent attempt to control for pleiotropic effects of SHBG on free testosterone by conservatively removing any genetic variants in the GRS that were associated with SHBG ( p<0 . 05 ) . The apparent difference between protective effects of testosterone observed in a previous MR analysis of testosterone and lack of protective effect in our study might be a result of less stringent control for pleiotropic effects of SHBG in the previous study . Given studies have identified associations between SHBG and risk of T2D independent of testosterone and a direct role of SHBG in mediating signalling on target cells , insufficient controls for SHBG may lead to residual pleiotropic effects ( Lakshman et al . , 2010; Rosner et al . , 2010; Vikan et al . , 2010 ) . Other reasons may include genetic variants explaining less variation in testosterone levels in our study , fewer cases of T2D leading to inadequate statistical power to detect weaker effects in our study , or other differences between the populations of the UK Biobank in our study and DIAGRAM consortium used by Ruth et al . , 2020 . There are several limitations of this study . First , an assumption of the MR analysis is that the effect of the genetic variant on the outcome occurs only through free testosterone levels , such that there are no pleiotropic effects through other proteins or mechanisms ( Davies et al . , 2018 ) . This concern was minimized by the use of multiple genetic variants , which limited the likelihood of a common alternative pathway confounding our observation . Moreover , we performed several sensitivity analyses and excluded genetic variants associated with SHBG levels , which is a potential source of pleiotropy through its effects on other hormones . Although a stringent p-value threshold was selected for genetic variants , the winner’s curse phenomenon may still bias genetic effect sizes due to the same sample being used to select genetic variants and estimate effect sizes on testosterone . Additionally , one-sample MR may be susceptible to bias towards the confounded estimate if the genetic variants are ‘weak instruments’ , which can occur if the genetic variants don’t explain enough of the variance in free testosterone levels ( Davies et al . , 2018 ) . To address this concern , we confirmed the selected genetic variants were strong instruments using a common threshold in MR literature ( F-statistic >10 ) ( Davies et al . , 2018 ) . Next , the UK Biobank is generally healthier and higher socioeconomic status than the general population , so there are insufficient cases to detect effects on certain rarer outcomes , such as Alzheimer’s disease , and inadequate power to identify weaker effects of free testosterone on common outcomes . Relatedly , an inherent limitation for outcomes ascertained using linked electronic medical records is a lack of adjudication and consistent application of codes in clinical practice . In the UK Biobank , CFT levels were below the reference ranges for young healthy individuals , which may be attributable to the older age of the cohort and inherent inaccuracy of immunoassays at lower levels of total testosterone . Total testosterone levels are similarly low relative to reference ranges and comparable to previous studies in the UK Biobank ( Peila et al . , 2020; Petermann-Rocha et al . , 2020 ) . Additional sources of variability introduced into the total testosterone measurements include differences in fasting times , diets , and time of day at which blood was drawn from participants . Nevertheless , genetic variants associated with testosterone consistently replicated known effects of testosterone on established outcomes , such as body fat , body fat-free mass , and hematocrit ( Table 1 ) . Furthermore , although the free hormone hypothesis is still debated by experts , we found largely consistent effects on outcomes using genetically-predicted free testosterone and total testosterone ( Handelsman , 2017 ) . The only significant outcomes from MR analyses with free testosterone that showed no significant effect with total testosterone across all MR methods were HDL ( p=0 . 55 ) and apolipoprotein A ( p=0 . 45 ) . Finally , these results represent lifelong effects of endogenous free testosterone and may not necessarily reflect effects of exogenous testosterone treatment , which can vary in duration , age of initiation , and dosage . Taken altogether , the decision to initiate long-term testosterone use warrants careful consideration of benefits and risk . Beneficial effects on body composition , sexual function , hematocrit , and BMD should be weighed against detrimental effects on androgenic alopecia , prostate cancer , hypertension and spinal stenosis , and no detectable beneficial effects on other major clinical endpoints . Ultimately , well-designed and appropriately powered RCTs , such as the ongoing TRAVERSE trials ( clinicaltrials . gov , NCT03518034 ) , are necessary to conclusively address questions of safety and effectiveness of testosterone treatment . However , as demonstrated in this study , genetically-informed analyses can be powerful tools to aid health professionals in prioritizing allocation of limited resources towards investigating the most pressing questions . The UK Biobank is a large-scale longitudinal cohort study that recruited over 500 , 000 people between the ages of 37–73 across the United Kingdom from 2006 to 2010 ( Sudlow et al . , 2015 ) ( RRID:SCR_012815 ) . UK Biobank received ethical approval from the North West Multi-Centre Research Ethics Committee ( REC reference: 11/NW/0382 ) . This research was conducted using the UK Biobank under Application Number 15255 . For this study , UK Biobank participants were included if white British ancestry , and no self-reported androgen medication at recruitment based on field ID 20003 . In the UK Biobank , total testosterone and sex hormone-binding globulin ( SHBG ) were measured on a Beckman Coulter Unicel DXI 800 using a one-step competitive analysis and two-step sandwich immunoassay , respectively . Analytical range for the immunoassays of total testosterone and SHBG were 0 . 35 to 55 . 52 and 0 . 33 to ( 226-242 ) nmol/L , respectively . For total testosterone , within-laboratory CV for high , medium , and low concentration quality control samples were 4 . 15 , 3 . 66 , and 8 . 34% . For SHBG , within-laboratory CV for high , medium , and low concentration quality control samples were 5 . 22 , 5 . 25 , and 5 . 67% . For each blood sample drawn at recruitment , testosterone , SHBG , and albumin were each measured only once . Testosterone and SHBG measurements were flagged if they fell outside the manufacturer’s observed reportable range , or samples reported high levels of bilirubin , hemoglobin or lipids/turbidity that might interfere with the assay . Testosterone measurements were flagged if levels of total protein ( <55 or>85 g/L ) or triglycerides ( >20 mmol/L ) could interfere with the assay measurements . To monitor assay consistency , all samples were run with internal quality control samples between batches and operations used external quality assurance schemes against the ISO 17025:2005 standard . Individual-level genetic data was available for 488 , 317 participants that consented to blood collection and genotyping . Genotyping was performed with the Applied Biosystems UK Biobank Lung Exome Variant Evaluation ( UK BiLEVE ) and UK Biobank Axiom arrays ( Affymetrix Research Services Laboratory , Santa Clara , California , USA ) . Description of quality control has been previously described in detail ( Bycroft et al . , 2017 ) . Genetic variants located in the human leukocyte antigen gene complex were excluded due to extensive pleiotropic effects . For genome-wide association testing , samples were restricted to a subset of 161 , 268 males with white British ancestry , no androgen medication ( n = 2 , 137 ) , and no missing values of testosterone , SHBG , or albumin at recruitment . Free testosterone at recruitment was calculated using the Vermeulen equation ( Vermeulen et al . , 1999 ) . CFT levels were winsorized such that outlying values greater or less than four standard deviations ( SD ) away from the mean in males were set to 4 SD . This study was restricted to genetic variants from ‘v3’ release of the UK Biobank data including those present in the Haplotype Reference Consortium and 1000 Genomes panels with imputation imputation quality greater than 0 . 7 , no deviation from Hardy-Weinberg equilibrium ( p>1×10−10 ) and minor allele frequency greater than 1% ( McCarthy et al . , 2016 ) . To allow for genetic relatedness between participants , linear mixed models in BOLT-LMM were used to test for associations of genetic variants ( Loh et al . , 2015 ) . The model was adjusted for age , age2 , chip type , assessment center , and the first 20 genetic principal components . Genetic variants near the SHBG gene may alter binding affinity for testosterone thereby violating assumptions of the Vermeulen equation , or risk having pleiotropic effects through binding of other sex hormones ( Ohlsson et al . , 2011 ) . Therefore , any genetic variants associated with CFT reaching genome-wide significance ( p≤5×10−8 ) were excluded if associated with natural log-transformed SHBG levels at a stringent threshold ( p<0 . 05 ) in the same subset of the UK Biobank ( Figure 1—figure supplement 4 ) . To arrive at an independent set of genetic variants , variants associated with CFT but not SHBG were pruned based on linkage disequilibrium ( LD ) at a threshold of r2 <0 . 01 using Europeans from 1000 Genomes phase three as reference panel ( Abecasis et al . , 2012 ) ( RRID:SCR_006828 ) . Genomic inflation factor ( λ ) was 1 . 2 and calculated as the ratio of the median test statistic from the GWAS relative to the expected median test statistic under a null model ( Figure 1—figure supplement 5 ) . To distinguish between an inflated λ due to population stratification or polygenic inheritance of the trait , the intercept of an LD score regression line was determined to be 1 . 03 indicating the observed inflation could be attributed to polygenicity rather than uncontrolled population stratification . LD score regression was performed and intercept was calculated with LDSC software ( Bulik-Sullivan et al . , 2015 ) using 1000 Genomes Europeans phase three data as the LD reference panel ( Abecasis et al . , 2012 ) . For MR analyses , 22 health outcomes were selected a priori based on relevance with known or suspected effects of testosterone treatment and categorized based on expected beneficial or adverse effects from RCT data . Outcomes with expected beneficial effects were fractures at any site , heel BMD , body fat percentage , body fat-free percentage , dementia , depression , handgrip strength , and physical activity level measured by wrist-worn accelerometer . Outcomes with potential adverse effects were stroke , androgenic alopecia , benign prostate hyperplasia ( BPH ) , blood pressure , glucose , hematocrit percentage , hemoglobin A1c , heart failure , prostate cancer , MI , type 2 diabetes ( T2D ) , and venous thromboembolism . Depression was coded using a ‘broad’ definition as previously described , which included self-reported depressive symptoms with associated impairment , or having sought help for ‘nerves , anxiety , tensions or depression’ ( Howard et al . , 2018 ) . Androgenic alopecia was defined based on participants’ responses to the question , ‘Which of the following best describes your hair/balding pattern ? ’ ( field ID 2395 ) . Available options were four pictures of hair patterns ( Supplementary file 1 – Figure 1 ) . Individuals with pattern 3 or four were cases , pattern 1 and 2 were controls , and ‘do not know’ or ‘prefer not to answer’ responses were excluded . Physical activity was assessed using the overall acceleration average from wrist-worn accelerometer devices over the course of approximately 7 days . Following UK Biobank recommendations , individuals were excluded from the analysis based on poorly calibrated data ( field ID: 90016 ) or having worn the device for insufficient time to get a stable measure of physical activity ( field ID: 90015 ) ( Doherty et al . , 2017 ) . Blood pressure measures were coded as the average of two automated measurements of blood pressure taken a few moments apart by a registered nurse using an Omron 705 IT electronic blood pressure monitor . Body fat percentage and whole body fat-free mass were estimated based on impedance measurements from a Tanita BC418MA body composition analyser . Heel BMD was estimated as a T-score based on quantitative ultrasound index through the calcaneus relative to that expected in someone of the same sex . Handgrip strength was calculated as the average of right and left hands measured using a Jamar J00105 hydraulic hand dynamometer . hemoglobin A1C was measured using high performance liquid chromatography analysis on a Bio-Rad VARIANT II Turbo . Glucose was measured using hexokinase analysis on a Beckman Coulter AU5800 . Hematocrit percentage was measured using a Coulter LH750 and calculated as the relative volume of packed erythrocytes to whole blood , computed by the formula: redbloodcells∗meancorpuscularvolume10 . Detailed descriptions of all 22 outcomes are shown in Supplementary file 1 – Table 11 . For hypothesis-free GRS analyses , we included 24 blood biomarkers measured at recruitment and 415 diseases derived from linked electronic medical records ( Supplementary file 1 - Table 12; Brion et al . , 2013; Denny et al . , 2013; Wu et al . , 2019 ) . Disease outcomes were defined using the previously published ‘PheCode’ scheme to aggregate ICD-10 codes from hospital episodes ( field ID 41270 ) , death registry ( field ID 40001 and 40002 ) , and cancer registry ( field ID 40006 ) records ( Denny et al . , 2013; Wu et al . , 2019 ) . Given the small number of cases for many disease outcomes , any outcomes with detectable odds ratios less than 0 . 5 or greater than 2 per 0 . 1 nmol/L at 80% power were excluded ( ncases < 871 ) based on approximate changes in response to testosterone supplementation ( Bhasin et al . , 2018b; Brion et al . , 2013; Traustadóttir et al . , 2018 ) . After these exclusions , there were 415 diseases that remained for subsequent analyses in this study . Furthermore , all blood biomarkers measured by the UK Biobank at recruitment were included except estradiol and rheumatoid factor , which were complicated by majority missing values below the limit of detection of the assay ( nbiomarkers = 24 ) . Detailed descriptions of all 439 outcomes ( 415 diseases and 24 biomarkers ) are shown in Supplementary file 1 – Table 12 . In a subset of unrelated males with White British ancestry , the association of all independent genetic variants associated with CFT were determined for each of the 22 a priori outcomes using additive genetic models in BGENIE v1 . 2 and adjusted for the same covariates as the model for CFT ( Bycroft et al . , 2017 ) . For each of the 22 outcomes , one-sample MR analysis was used to combine the effect of each independent genetic variant on CFT with its effect on the outcome using the inverse variance-weighted ( IVW ) method ( Burgess et al . , 2016 ) . Effect estimates were reported per 0 . 1 nmol/L increase in CFT levels based on approximate changes in response to testosterone treatment ( Bhasin et al . , 2018b ) . For dichotomous outcomes , odds ratios were approximated as previously described ( Adams et al . , 2018 ) by converting linear effect estimates from BGENIE to log-odds scale using: log⁡OR= k ( 1-k ) , where k is the proportion of cases for the given outcome . Given the polygenic nature of testosterone and potential for pleiotropy , for outcomes with statistically significant effects using the IVW method , standard sensitivity analyses were conducted to correct for pleiotropic effects , such as MR-Egger , MR-RAPS , and MR-PRESSO ( Bowden et al . , 2015; Verbanck et al . , 2018 ) . To investigate and correct for directional pleiotropy on each outcome , we performed Egger regression . For outcomes with y-intercept of the regression line significantly different from 0 ( p<0 . 05 ) , there was evidence of directional pleiotropy and the causal estimate from MR Egger was reported to attempt to control for pleiotropic effects ( Bowden et al . , 2015 ) . As a sensitivity analysis robust to idiosyncratic pleiotropy and weak instrument bias , MR-RAPS ( Robust Adjusted Profile Score ) was conducted using overdispersion and Tukey’s loss function ( Zhao et al . , 2018 ) . To detect and correct for potential bias from invalid variants with pleiotropic effects , we performed the MR-PRESSO ( Mendelian Randomization Pleiotropy RESidual Sum and Outlier ) test with 10 , 000 simulations ( Verbanck et al . , 2018 ) . The global test p-value evaluated whether there was any overall horizontal pleiotropy among all genetic variants . For outcomes with significant p-values ( p<0 . 05 ) , outlying genetic variants with predicted pleiotropic effects were removed and MR analysis repeated to correct for horizontal pleiotropy . The distortion test evaluated whether removal of the pleiotropic variants resulted in a significantly different causal estimate ( p<0 . 05 ) . Leave-one-out analysis was performed such that the IVW MR analysis was repeated after each genetic variant was excluded to identify effects on an outcome that are driven by a single outlying genetic variant . Furthermore , the set of genetic variants used in MR analysis were assessed for ‘weak instrument bias’ , which can result in biased estimates if genetic variants don’t explain enough variance in exposure ( e . g . , CFT ) levels ( Pierce et al . , 2011 ) . Lastly , as a sensitivity analysis , all MR and GRS analyses were repeated using genetic variants associated with total testosterone . Finally , for significant outcomes , we compared estimated effect sizes from this MR study with reported effect sizes from random controlled trials of testosterone therapy , where possible , in Figure 3 ( Cui et al . , 2014; Fernández-Balsells et al . , 2010; Ng Tang Fui et al . , 2016; Zhang et al . , 2020 ) . In consideration of ‘weak instrument bias’ , the F-statistic was 66 for the genetic variants associated with CFT , which was considered a strong instrument based on the recommended threshold of greater than 10 ( Davies et al . , 2018 ) . MR-PRESSO was performed using the MR-PRESSO package and all other MR analyses were implemented using the TwoSampleMR package ( Hemani et al . , 2018; Verbanck et al . , 2018 ) ( RRID:SCR_019010 ) . A genetically-predicted value of CFT was determined for each individual by constructing weighted GRS in the unrelated White British subset of UK Biobank males ( n = 157 , 252 ) . Weighted GRS were calculated by multiplying the effect of each CFT-associated genetic variant by the number of effect-corresponding alleles and summing this value for each individual . The GRS was tested for association with outcomes using logistic or linear regression models for case-control or quantitative outcomes , respectively , and adjusted for the same covariates as the GWAS for CFT . Effect estimates were reported per 0 . 1 nmol/L increase in CFT levels based on approximate changes in response to testosterone treatment ( Bhasin et al . , 2018b ) . As sensitivity analyses , we repeated GRS analyses after excluding males that self-reported taking blood pressure ( n = 38 , 676 ) or cholesterol medication ( n = 35 , 737 ) at recruitment based on field ID 6177 . As a set of sensitivity checks , we repeated all GWAS , MR , and GRS analyses using total testosterone . In the White British subset of the UK Biobank , there were 175 , 421 males with total testosterone measured with an average 11 . 9 nmol/L ( Figure 1—figure supplement 6 ) . In this population , a genome-wide association study was conducted for total testosterone as described herein for CFT . After removing genetic variants associated with natural-log-transformed SHBG and LD pruning for independent SNPs ( r2 <0 . 01 ) , there were 52 independent genetic variants associated ( p<5×10−8 ) with total testosterone in males from the UK Biobank ( Supplementary file 1 – Table 8 ) . All statistical analyses were performed under R version 3 . 6 . 0 , unless otherwise specified ( RRID:SCR_001905 ) . A two-sided p-value less than 5 × 10−8 for GWAS , 2 . 27 × 10−3 ( 0 . 05/22 outcomes ) for a priori MR analyses , and 1 . 14 × 10−4 ( 0 . 05/439 outcomes ) for hypothesis-free GRS analyses was considered statistically significant .	Men experience a gradual decline in their testosterone levels as they grow older . However , the effects of testosterone and the consequences of supplementation on the human body have been unclear . Scientists use so-called randomized controlled trials to establish cause-and-effect and to reduce bias . In these experiments , participants are randomly assigned to a either a treatment group ( that receives the intervention being tested ) or a control group ( that either receives an alternative intervention , a dummy or placebo , or no intervention at all ) . Randomization ensures that both groups are balanced , and any resulting differences can be attributed to the treatment . However , randomized controlled trials are time-consuming and expensive , so trials of testosterone have had relatively small numbers of participants and short follow-up periods . This makes it difficult to draw conclusions about any potential effects of testosterone administration on less common diseases in men . Now , Paré et al . investigated the effects of naturally produced testosterone using Mendelian randomization , which mimics randomized trials by exploiting the fact that parents randomly pass on their unique genetic variants to their children at conception . This random assignment of genetic variants leads to its informal namesake , “nature’s clinical trial” , and provides the ability to study cause-and-effect for any genetically determined factors , such as testosterone levels . Paré et al . studied the long-term effects of testosterone on 22 diseases previously explored in randomized controlled trials , and hundreds of other traits and diseases that have not been investigated in any randomized controlled trials yet . The Mendelian randomization analysis made it possible to examine the effects of lifelong naturally elevated testosterone levels on 469 traits and diseases . Paré et al . found that testosterone increased the density of bone mineral and decreased body fat . However , it also increased the risks of prostate cancer , high blood pressure , baldness and a condition affecting the spine . It also increased the number of red blood cells and decreased a marker of inflammation , which may be beneficial or detrimental depending on the context . This shows that genetic analyses can be powerful methods to prioritize the allocation of limited resources towards investigating the most pressing clinical questions . The results of this study may help inform physicians and patients about the effects of long-term testosterone use . Ultimately , large randomized controlled trials are needed to conclusively address the cause-and-effect on these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "genetics", "and", "genomics" ]	2020	Effects of lifelong testosterone exposure on health and disease using Mendelian randomization
Many plants attract and reward pollinators with floral scents and nectar , respectively , but these traits can also incur fitness costs as they also attract herbivores . This dilemma , common to most flowering plants , could be solved by not producing nectar and/or scent , thereby cheating pollinators . Both nectar and scent are highly variable in native populations of coyote tobacco , Nicotiana attenuata , with some producing no nectar at all , uncorrelated with the tobacco's main floral attractant , benzylacetone . By silencing benzylacetone biosynthesis and nectar production in all combinations by RNAi , we experimentally uncouple these floral rewards/attractrants and measure their costs/benefits in the plant's native habitat and experimental tents . Both scent and nectar increase outcrossing rates for three , separately tested , pollinators and both traits increase oviposition by a hawkmoth herbivore , with nectar being more influential than scent . These results underscore that it makes little sense to study floral traits as if they only mediated pollination services . Plants attract pollinators to move pollen from one flower to another to ensure outcrossing and frequently pay for these pollination services by rewarding floral visitors with nectar ( Kevan and Baker , 1983; Schemske and Bradshaw , 1999; Raguso and Willis , 2005 ) . Even though it is clear that both floral scent and nectar provide fitness benefits for plants , rewardless flowers have evolved in all major groups of angiosperms ( Renner , 2006 ) . Orchids , in particular , have frequently evolved deceptive pollination systems , in which flowers attract pollinators by mimicking mating partners or oviposition sites without offering rewards ( Schiestl , 2005 ) . But rewardless , nectar-free , flowers are commonly found within species that normally provide nectar , and this is surprising , as the occasional nectar-free flowers would have a disadvantage if visitors have the sensory abilities to avoid rewardless flowers ( Karban , 2015 ) . Early on , theorists ( Bell , 1986 ) recognized that most flowers hide their rewards , for example , deep in the corolla tube , which thwarts the easy visual evaluation of a flower's standing nectar volume and developed an ESS model for the proportion of cheating flowers and discriminating visitors that would be evolutionarily stable . Researchers have since uncovered evidence consistent with the predictions of the model . Gilbert et al . ( 1991 ) found that nectar secretion was highly variable within plants of a population and suggested that floral visitors could distinguish between low and high nectar secreting plants . Recent research suggests that hawkmoth pollinators can use humidity as a proxy for the presence of nectar ( von Arx et al . , 2012 ) . To examine the importance of nectar for pollination services and to study the fitness advantages of nectar-cheating plants , researchers have used native varieties with reduced nectar accumulation , introgression lines ( Brandenburg et al . , 2012a , 2012b ) , artificial flowers ( e . g . , Ishii et al . , 2008 ) , or conducted direct manipulations of nectar quantities by adding artificial nectar to flowers ( e . g . , Mitchell and Waser , 1992; Jersáková et al . , 2008 ) . Ishii et al . ( 2008 ) found that pollinators avoided inflorescences with greater numbers of empty flowers . Smithson ( 2002 ) added nectar to rewardless orchids , and while the addition changed bee behavior , it did not influence plant fitness . Brandenburg et al . ( 2012a ) found that nectar-deficient Petunia lines produced fewer seeds than did nectar-replete control plants , because Manduca sexta moths reduced their probing times in low-nectar plants , which in turn , reduced pollen transfer and thus seed set . While rewards keep pollinators moving pollen from one plant to another , other cues , such as floral scent , provide honest signals that advertise the occurrence of the rewards ( Wright and Schiestl , 2009 ) . Floral scent is known to play a central role in attracting insect pollinators to flowers ( Galen and Newport , 1988; Jürgens et al . , 2002; Klahre et al . , 2011; Byers et al . , 2014; Riffell et al . , 2014 ) . The effect of floral scent on the pollination success of single pollinator species has been studied mainly with scent augmentations and additions to existing scent bouquets ( e . g . , Majetic et al . , 2009; Shuttleworth and Johnson , 2010 ) . Shuttleworth and Johnson ( 2010 ) , for example , showed that single sulphur compounds are responsible for the shift between wasp and fly pollination in Eucomis ( Hyacinthaceae ) . Byers et al . ( 2014 ) found altered bumblebee visitation rates in response to single volatile compounds which were added to the scent bouquet of Mimulus species . In most studies , only one pollinator species was investigated at a time , frequently in very specialized model systems , often the sexually deceptive pollination systems of orchids ( Schiestl , 2005; Schiestl and Schlüter , 2009 ) . Several studies investigated fitness outcomes of these manipulations . Majetic et al . ( 2009 ) for example , found a positive effect on both pollinator visitation and seed production in Hesperis matronalis by augmenting inflorescences with scent extracts . Kessler et al . ( 2008 ) genetically manipulated the biosynthesis of the most abundant floral volatile , benzylacetone ( BA ) , in the flowers of the wild tobacco ( Nicotiana attenuata ) to demonstrate that BA emission is required to maximize both maternal and paternal fitness in the field when the entire native community of floral visitors had access to the flowers . Few studies have investigated the influence of scent and nectar on the entire pollinating community simultaneously . Floral traits are shaped not only by interactions with mutualists , but also with antagonist , such as herbivores , florivores , seed feeders , or nectar robbers ( Strauss and Whittall , 2006; Andrews et al . , 2007; Kessler et al . , 2013 ) . Both floral scent and nectar are known to influence antagonists and again manipulative experiments have been essential in illuminating these interactions . Floral scents allow florivores to locate their host plants ( Theis , 2006; Theis and Adler , 2012 ) . Several studies have shown that individual compounds in floral scent bouquets specifically deter florivores and nectar robbers ( Galen et al . , 2011; Junker et al . , 2011; Kessler et al . , 2013 ) and thus complex floral scents can both attract pollinators and deter antagonists . Nectar quantity has also been shown to influence herbivory , as inferred by adding nectar or sugar solutions to flowers to increase the standing volume of nectar . In both Datura wrightii ( Adler and Bronstein , 2004 ) , as well as N . attenuata ( Kessler , 2012 ) nectar addition increased oviposition by M . sexta , a pollinator as an adult as well as a devastating herbivore as a caterpillar for both plant species . The inference from both studies is that plants could reduce their herbivore load by producing only small amounts or no nectar at all . As suggested by the early ESS modeling efforts ( Bell , 1986 ) , a few nectar-free plants dispersed in a large population of nectar-producing plants could realize a fitness benefit from reduced herbivore loads , if at low frequencies pollinators cannot learn their locations ( Gumbert and Kunze , 2001 ) or otherwise identify their lack of rewards ( Smithson and MacNair , 1997; Smithson , 2009 ) . A few studies investigated the influence of floral traits for both pollinators and herbivores at the same time . Theis et al . ( 2014 ) , for example , found that floral sesquiterpenoids from Cucurbitaceae were the best predictors of flower preference for both the specialist pollinator squash bees and the specialist herbivore cucumber beetle . Schiestl et al . ( 2014 ) showed that the herbivory-induced reduction of floral volatiles reduced the attractiveness of flowers to pollinators . These studies however were focusing on floral signals and did not consider the function of floral rewards , and so while the ESS explanation for rewardless flowers being maintained by frequency dependent selection is well established ( Bell , 1986 ) , we still lack studies that evaluate the relative selective pressures of mutualistic and antagonistic floral visitors on floral reward provisioning strategies . In this study , we rigorously investigate the consequences of reducing floral scent and nectar independently and simultaneously by RNAi not only for mutualistic interactions but also for oviposition rate , which is an excellent predictor of future herbivory . We measured outcrossing rates afforded by the entire natural pollinator community in N . attenuata's native habitat , by three specific pollinators ( hummingbirds and two hawkmoth species ) , and oviposition by M . sexta . Different approaches were used in the field and laboratory to study pollination and herbivory . At our field site at the Lytle Ranch Preserve in SW Utah , we conducted common garden experiments using plants silenced by RNAi in the production of floral scent ( CHAL; Kessler et al . , 2008 ) , floral nectar ( SWEET9; Lin et al . , 2014 ) , or both ( CHALxSWEET9 ) , in comparison to empty vector-transformed control plants ( EV ) . The RNAi constructs silenced only the targeted pathways and were otherwise isogenic , and the transformed plants were morphologically indistinguishable from EV transformed and wild-type ( WT ) plants . This approach allowed for the study of the entire panoply of interactions simultaneously , including different floral visitors , and herbivores to gain an unbiased picture of the interactions of individual players in the complex web of interactions that occur when flowers advertise for pollinator services . To examine the natural genetic variability in floral advertisement and rewards for pollinator services in native populations of N . attenuata plants , seeds from plants from 13 native populations , collected between 1993 and 2009 within a 200 km radius of our field station in the SW USA ( Utah , Arizona ) , were used . The variance in standing nectar volume was assessed by selecting 52 plants in the glasshouse . The average standing nectar volume was 3 . 3 μL and varied between single genotypes between 1 . 3 μL and 5 . 7 μL per flower ( Figure 1 ) . In addition to this fourfold difference in nectar volume , we also assessed the variability of BA emission from individual flowers from the same sample of plant populations . The average emission was 5 . 0 ng BA per flower per night , varying between a minimum emission of 0 . 5 ng and a maximum of 33 . 9 ng , equivalent to a 70-fold difference in emission rates ( Figure 1 ) . Nectar sugar concentrations were in contrast , rather constant . Sugar concentration in newly opened flowers varied between 14 . 6 and 22 . 8% , with an average of 18 . 2% . 10 . 7554/eLife . 07641 . 003Figure 1 . Characterization of 52 plants from different natural populations . Mean standing nectar volume ( n = 6 to 8 flowers/plant ) , nectar sugar concentration ( n = 6 to 8 flowers/plant ) , and benzylacetone ( BA ) level ( n = 1 flower/plant ) . Nectar volume and sugar concentration were measured from newly opened flowers at the end of the nectar production period ( 5–6 am ) . BA level was measured over the night ( 8 pm–6 am ) and calculated from peak areas normalized using tetralin as an internal standard . Inset: mean ( +SE ) standing nectar volume in newly opened flowers at the end of the nectar production period ( 5–6 am ) in three native phenotypes ( Ut-WT , 83 , and 84; n = 7–9 ) found within a separate sample of 424 native Nicotiana attenuata plants that were screened for the presence of nectar . Asterisks indicate significant differences , as informed by a nonparametric Mann–Whitney U-test p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 07641 . 003 In a second screening , 424 individual plants were screened for the presence or absence of nectar and scent . Seeds for this experiment were collected from 75 natural populations in Utah , Arizona , and California between 1990 and 2009 . Of these , two plants produced no nectar at all . These lines ( 83 and 84 ) were inbred for two generations and plants of the T2 generations of both lines also did not produce any nectar in any flowers at all stages of development ( Figure 1 ) , demonstrating that the lack of nectar production was heritable . All flowers from all 424 plants in this analysis produced BA in some measureable quantity . We found no evidence for a linear correlation between BA emission and nectar volume ( y = 0 . 1875x + 1 . 02 , R2 = 0 . 01 ) , BA emission and nectar sugar concentration ( y = 0 . 1233x + 3 . 89 , R2 = 0 . 01 ) , or nectar volume and nectar sugar ( y = −1 . 4471x + 22 . 99 , R2 = 0 . 41 ) . While we cannot exclude that genotypes completely lacking floral scent occur in native populations , our results demonstrate that native populations harbor nectar-free ‘cheating’ plants at low frequencies . BA emission however does not occur over the entire night , but is commonly released in a sharp peak in the first part of the night ( Kessler et al . , 2010 ) , depending on the genotype and the environment . Hence , there are times in the night were one plant is emitting BA , while a neighboring plant is not in genetically diverse populations . Since nectar and scent production were uncorrelated in natural populations , we used RNAi to completely uncouple nectar and scent production in all combinations in an isogenetic background to rigorously examine their costs and benefits . To investigate the consequences of a lack in nectar , floral scent , and both on pollination and herbivory , EV , SWEET9 , CHAL , and CHALxSWEET9 plants were randomly planted 4 m apart in a field plot in the SW USA ( Utah ) in the plant's native habitat ( Figure 2A ) . Since N . attenuata is a self-compatible plant , anthers were removed from five flowers per plant , and all additional flowers were removed . Hence , seeds of treated flowers were produced entirely from pollinator-mediated outcrossing with neighboring WT plants , which were planted between the rows of transformed plants and were allowed to flower normally . To evaluate the overall pollinator attractiveness of the lines , we allowed floral visitors free access to treatment plants for the flowers' lifetime ( 3 days ) . All three lines lacking either floral scent , nectar or both , produced significantly fewer seeds in comparison to similarly antherectomized EV plants ( Figure 2B; Friedman signed rank test χ2 = 10 . 22 , df = 3 , n = 16–20 , p = 0 . 017 ) . CHAL plants produced only 22 . 9% ( p < 0 . 001 ) , Sweet9 9 . 7% ( p < 0 . 001 ) , and CHALxSWEET9 11 . 6% ( p < 0 . 001 ) of the seeds produced by EV from outcrossed pollen . The most frequent floral visitors of N . attenuata at night in the field at the time of the experiment were M . sexta , Manduca quinquemaculata , and Hyles lineata hawkmoths ( Sphingidae ) , and Archilochus alexandri hummingbirds ( Trochilidae ) during the day time . In order to evaluate each species individually , we covered experimental plants at night with mesh cones ( Kessler et al . , 2010 ) to only allow hummingbirds to access the flowers . While CHAL ( 75 . 6%; Friedman signed rank test χ2 = 7 . 66 , df = 2 , n = 24 , p = 0 . 02; p = 0 . 835 ) plants produced almost as many seeds as EV plant , SWEET9 plants produced considerably fewer seeds than did EV plants ( 13 . 1%; p = 0 . 074 ) and significantly fewer seeds than did CHAL plants ( 17 . 4%; p = 0 . 005; Figure 2C ) . 10 . 7554/eLife . 07641 . 004Figure 2 . Benefits and cost of floral scent and nectar , as revealed from pollination and oviposition experiments in the field and tent . Means +SEM of either seed production of antherectomized flowers or oviposition by Manduca sexta on transformed plants silenced in the production of floral scent ( CHAL ) , floral nectar ( SWEET9 ) , or both ( CHALxSWEET9 ) in comparison to empty vector control plants ( EV ) . ( A ) Field plot in N . attenuata's native habitat at the Lytle ranch preserve in Santa Clara , Utah , USA . ( B ) Seeds sired from pollen transferred by the native community of floral visitors over the complete life span of a flower ( 3 days ) . ( C ) Exclusive pollination by Archilochus alexandri hummingbirds in the field during a 12 hr day . ( D ) Tent set-up used for single species pollinations in Isserstedt , Germany . Seeds sired from pollination of single M . sexta ( E ) or Hyles lineata ( F ) individuals during a 9-hr night . Eggs oviposited per transformed line ( n = 10 ) on different days by the native community of M . sexta in the field ( G ) or by single individuals in the tent ( H ) . Letters indicate significant differences inferred by a Friedman signed rank test p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 07641 . 00410 . 7554/eLife . 07641 . 005Figure 2—figure supplement 1 . Nectar accumulation in flowers of transformed lines in the field . Mean ( +SE ) standing nectar volume in newly opened flowers at the end of the nectar production period ( 5–6 am ) in the field . Different letters indicate significant differences , as informed by a Friedman signed rank test p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 07641 . 00510 . 7554/eLife . 07641 . 006Figure 2—figure supplement 2 . Analysis of flower volatiles . Benzylacetone ( BA ) level from flowers in the field during the first night of opening ( bars represent mean +SE amount of BA trapped per flower , n = 8 ) . Different letters indicate significant differences , as informed by Friedman signed rank test p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 07641 . 00610 . 7554/eLife . 07641 . 007Figure 2—figure supplement 3 . Principal components analysis ( PCA ) of leaf volatiles . Leaf volatiles were collected overnight from un-attacked leaves from seven individual plants per genotype . 94 volatiles were detected and analyzed by PCA , and principal components ( PCs ) 1 and 2 of the transgenic lines were plotted against each other ( dashed circles represent 95% confidence intervals ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07641 . 007 Of the three hawkmoth species occurring in the field , we examined two ( M . sexta and H . lineata ) in a tent set-up . To simulate field conditions for the moths , we used a mesh tent large enough for natural moth pollination behavior ( 24 m × 8 m; Figure 2D ) and examined the consequences of pollination services from single hawkmoths of a given species at a time , something that was not possible to do in the field . To avoid inbreeding or artificial larval diet effects , only moths that had been reared from field-collected eggs and maintained on a plant diet were used . We used a different moth on each experimental day . M . sexta strongly distinguished between EV and all other lines ( Figure 2E; Friedman signed rank test χ2 = 29 . 90 , df = 3 , n = 30 , p < 0 . 001 ) . Although SWEET9 lines produced only 44 . 6% of the seeds produced by EV plants ( p = 0 . 006 ) , the avoidance of plants which produce no floral scent had an even stronger effect on seed set . CHAL plants produced only 17 . 4% of the seeds of EV ( p < 0 . 001 ) , just as CHALxSWEET9 plants which produced only 5 . 2% of the seeds produced by EV plants ( p < 0 . 001 ) . Interestingly , H . lineata , a second naturally occurring hawkmoth species , showed a different response to that of M . sexta ( Figure 2F; Friedman signed rank test χ2 = 12 . 35 , df = 3 , n = 35 , p = 0 . 006 ) . Both CHAL ( 96 . 6%; p = 0 . 853 ) and SWEET9 ( 111 . 69%; p = 0 . 578 ) plants produced a number of seeds equivalent to those of EV plants . Only when the flowers lacked both floral traits ( CHALxSWEET9 ) was seed set mediated by H . lineata pollinator activity significantly reduced ( 21 . 3%; p = 0 . 003 ) . Our results highlight the value of using both floral scent attraction as well as a reward in order to cope with an unpredictable community of pollinators . While floral scent is an essential floral cue for M . sexta , it is clearly not for H . lineata . While nectar is essential for pollination by A . alexandri , the lack of nectar does not reduce the pollination services provided by H . lineata visitations . Though more generalist pollinators such as H . lineata seem to pollinate even when scent or nectar is missing , they strongly reject plants which are unable to produce both floral traits . However , for the more specialized pollinator , M . sexta , both floral scent attraction and nectar reward are important to maximize outcrossing . Although all lines lacking a certain floral trait showed reduced seed set , plants unable to produce nectar still produced more than twice as many seeds than plants unable to produce floral scent . Again , we found a stronger negative effect in plants unable to produce both traits . In contrast , A . alexandri , the only daytime visitor , rejected plants which did not produce nectar , but surprisingly was also less attracted to plants unable to produce floral scent , although visiting flowers at times when BA is not emitted . One explanation for this could be the accumulation of floral scent in nectar ( Raguso , 2004 ) , which could influence the taste of the nectar for these pollinators ( Kessler and Baldwin , 2007 ) . Another hypothesis could be that hummingbirds also use olfaction and are able to use nectar scent to locate a proper nectar source . Birds , long suspected to use scent in homing and navigation ( Keeton , 1974 ) have recently been suggested to use scent for prey localization ( Mantyla et al . , 2008 ) . Hummingbirds , in contrast to hawkmoths , have smaller territories and pollinate flowers only in a small area close to their nests ( Norton et al . , 1982 ) . This allows them to learn cues and plant positions which are correlated with the presence of food sources ( Ackerman et al . , 1994; Campbell et al . , 1997; Hurly and Healy , 2002 ) or even with the quality of the food ( Pérez et al . , 2011 ) . In contrast to hummingbirds , large hawkmoths like M . sexta cover much larger distances each night and rely more on direct perception rather than memory in visiting their food sources ( Raguso and Willis , 2003 , 2005 ) , as they most likely never visit the same plant twice in nature , although moths are also known to learn when exploring new food sources ( Riffell et al . , 2013 ) . The inferences we can draw about learning in moths from our results are limited . The pollination assays were conducted with naive moths that lacked experience with rewardless and scentless N . attenuata plants . Experienced H . lineata might have behaved differently , as has been shown in orchid pollinating bumblebees which were able to learn to avoid deceptive orchids ( Internicola and Harder , 2012 ) . Our experiments did not directly measure the behavior of the floral visitors and data on the variance in seed set resulting from pollinator visitations , limits the inferences we can draw about pollinator behavior . While there was no evidence that pollinators can sense the presence of nectar , we cannot reject this hypothesis ( von Arx et al . , 2012 ) . For future research , it would be interesting to investigate how geitonogamy ( Fisogni et al . , 2011 ) would be influenced in rewardless plants . We measured only the pollen flow from one plant to another and not within a given inflorescence . The behaviors of the three tested pollinator species appear to be quite different , and these differences have the potential to influence pollen transfer within an inflorescence and thereby decrease the chance of receiving outcrossed pollen . H . lineata , M . sexta , and M . quinquemaculata , for example , show different patterns of movement in both the field and the tent . While H . lineata moves sequentially from one flower to the next—visiting all flowers within one inflorescence ( even visiting the same flower multiple times ) —and from one plant to the next plant , both Manduca species visit only a few flowers within an inflorescence before moving on and skipping as many as 10 intervening plants ( personal observations ) . N . attenuata is capable of producing the full set of seeds of comparable mass and number from self-pollination as from outcrossing ( Baldwin et al . , 1997 ) . Hence , the plant can compensate for the lack of pollinators by selfing , and the function of producing floral nectar and scent is to ensure outcrossing to maintain genetic variability . Given that N . attenuata seeds can spend hundreds of years in the seed bank before germinating ( Preston and Baldwin , 1999 ) , outcrossing will likely be associated with longevity in the seedbank . In other systems outcrossing is thought to increase herbivore resistance in future generations ( Bello-Bedoy and Núnez-Farfán , 2011 ) . How does the production of floral scent or nectar influence M . sexta herbivory ? To answer this question , the number of flowers was reduced to the same number on all plants on a given day and oviposited eggs were counted the following morning . In the field , all lines lacking a certain floral trait had reduced loads of eggs ( Figure 2G; Friedman signed rank test χ2 = 11 . 15 , df = 3 , n = 4 , p = 0 . 0109 ) . On CHAL plants , only 43 . 1% ( p = 0 . 046 ) of the eggs found on EV plants were oviposited . The effect was even stronger in plants unable to produce nectar ( SWEET9 , 10 . 8%; CHALxSWEET9 , 6 . 5% , p = 0 . 046 ) . Comparable results were obtained in a similar experiment , conducted in the tent with single moths , ( Figure 2H; χ2 = 13 . 73 , df = 3 , n = 11 , p = 0 . 0033 ) . CHAL plants received only 81 . 9% ( p = 0 . 096 ) of the eggs of EV plants , while SWEET9 received 52 . 5% ( p = 0 . 058 ) and CHALxSWEET9 received 48 . 4% ( p = 0 . 007 ) of eggs compared with EV . To control for any other leaf-based influences on oviposition , we conducted control experiments with plants without flowers , by using non-flowering plants or plants from which flowers had been removed before the experiment . In the non-flowering stage , no differences in oviposition were found between the different lines ( Friedman signed rank test χ2 = 2 . 81 , df = 3 , n = 6 , p = 0 . 4214 ) . In the field , oviposition rates ( over eight experimental days ) were too low for statistical analyses , when plants did not flower . Although , plants unable to produce floral scent had a significantly lower chance of receiving an egg from M . sexta , nectar seems to play a larger role in host choice . In contrast to the pollination results , we found no elevated effect in the CHALxSWEET9 cross compared to the single silenced line , which is consistent with the strong effect of nectar alone . The fact that plants lacking floral scent have reduced ovipositions as well as reduced seed set is probably due to the difficulties hawkmoths have in locating scentless N . attenuata flowers ( Kessler et al . , 2008 ) . It is intriguing that it is the reward more than the attractant which influences the female hawkmoth's oviposition decisions , while the attractant plays a similar role as the reward when the same species is acting as a pollinator . Of course our conclusions are context dependent , depending on the frequency of low-nectar or low-scented plants in a native population . Clearly , there needs to be an attractive component ( e . g . , floral scent ) coupled with the nectar reward to allow moths to associate food sources with certain floral signals ( Daly and Smith , 2000 ) or oviposition sites . Similarly , moths would learn to avoid floral advertisements if they are not associated with a reward ( Haber , 1984; Smithson , 2001 ) . In an evolutionary context , our findings suggest that for an interaction with M . sexta , N . attenuata plants should minimize their nectar accumulation while maximizing their floral scent emission in order to maximize fitness . Indeed N . attenuata produces tiny amounts of nectar in comparison to other sympatric plant species that compete for M . sexta's pollination services ( Raguso et al . , 2003; Riffell et al . , 2008 ) , and thus seems to have optimized the attractant–reward proportions to maximize its fitness when dealing with this pollinator which is also an herbivore , albeit at a different life stage . D . wrightii , for example , a plant which shares M . sexta as both pollinator and herbivore , is a perennial plant that accumulates 20 times more nectar , produces a large amount of foliage that allows it to better compensate for losses to herbivores , and recruits fewer pollinating species than does N . attenuata ( Kessler , 2012 ) . Pollination success of N . attenuata is ensured by recruiting several pollinator species , which may influence the evolution of floral scent and nectar reward . In the absence of M . sexta , species such as hummingbirds which only pollinate plants capable of producing nectar , and probably will also only continue to pollinate if a sufficient quantity of nectar is available , are important pollen vectors . H . lineata also provides adequate pollination services , but without apparent evolutionary pressure on these N . attenuata floral traits , as this species seems to require neither an attractive signal nor a reward , at least if these plants are surrounded by other nectar and scent producing plants . Moreover H . lineata is also not an herbivore on N . attenuata during its larval stage . These two sphingid species show dramatic differences in their responses to both floral traits . Only M . sexta seems to rely on the cues provided by the flower and distinguishes between honest and cheating plants in a mixed N . attenuata population . For H . lineata , either floral scent or nectar is sufficient for the moth to provide full pollination services . In bee pollinated Brassica rapa plants , the honesty of a floral signal—that a certain scent would consistently lead to a reward—plays a key role in the plants attractiveness to pollinators ( Knauer and Schiestl , 2015 ) , something which seems not to be the case for N . attenuata and its pollinators , as we found no association between BA emission and nectar volume , or nectar sugar concentration . Our data are consistent with a trade-off between rewarding pollinators and avoiding herbivores in the evolution of floral nectar in N . attenuata , but why is BA emission so variable ? Since antagonists also use floral scent to locate host plants , increasing floral scent emissions would not only increased M . sexta pollination but could also incur costs by possibly increasing florivory or nectar robbing . Carpenter bees , for example , tend to rob more N . attenuata flowers on plants that emit BA than from plants that do not ( Kessler et al . , 2008 ) and experimentally enhancing BA emissions increased rates of browsing ( Baldwin et al . , 1997 ) . In this study , all pollinators visited N . attenuata at different times between dusk and dawn and it is an interesting question how this temporal order of nectar removal influences the pollination services they provide for the plants . A . alexandri visits N . attenuata approximately 2 h before dusk until dusk and again in the morning after dawn for 2 h . H . lineata visits from 1 h before dusk until 1 h after dawn , and finally both Manduca species pollinate and oviposit after dusk until dawn , with periods of enhanced activity depending on moonlight , temperature , and other conditions . We had expected that these different visiting times would blur the differences in the seed set amongst the lines differing in scent and nectar production when plants were exposed to the entire community of floral visitors at the field station , but the results were just the opposite . The differences between EV plants and SWEET9 as well as CHAL plants were maximized when plants were exposed to the entire community of pollinators at once , suggesting that the behavioral responses of each different pollinator to a given floral trait reinforce each other in ways that can only be understood through more detailed behavioral observations . N . attenuata flowers secrete nectar slowly over the night until approximately 4 am ( Kessler , 2012 ) , theoretically keeping a low but constant amount of nectar over the entire night , even if floral visitors removed the majority of nectar in the dusk or in the first half of the night . Unfortunately , we do not know if nectar secretion patterns change after flowers have been visited and nectar has been removed , but from nectar measurements , we know that flowers visited by M . sexta ( which invariably removes the standing volume of nectar ) again contain a residual amount of nectar in the morning . Similar questions remain about the costs associated with the production of nectar . One would expect that nectaring hummingbirds or white-lined sphinx moths in the dusk could influence M . sexta oviposition by removing all the nectar in the 2 h before M . sexta becomes active . Clearly , there remains a large gap between our understanding of how particular flower traits influence the behavior of the flower visitors and the pollination services that they provide . Our model system appears particular in that pollinators are at the same time herbivores , but this ecological dilemma is not uncommon for many plant species ( Irwin , 2010 ) , and the lessons we draw from our study are likely applicable for other plant systems with differently structured interactions with insects . For example , in the Silene latifolia–Hadena bicruris nursery pollination system , male moths provide a fitness benefit for the plant , while interactions with females come with the substantial costs of seed-feeding caterpillars ( Labouche and Bernasconi , 2009 ) . The highly specialized Yucca-Tegiticula pollination/seed predation system would be another example . Plants rarely interact with just one partner and usually face multiple selection pressures from mutualists and antagonists at the same time , which may shape the different floral traits of a plant . In most cases , there will be less and more effective pollinators ( Conner et al . , 1995; Barthelmess et al . , 2006; Matsuki et al . , 2008; Miller et al . , 2014 ) , as well as herbivores which are attracted by floral signals such as scent or nectar ( Adler and Bronstein , 2004; Theis , 2006 ) . Even purely mutualistic interactions like those known from bee pollination systems , probably evolved in the context of avoiding herbivory and excluding less efficient pollinators at the same time . Pollinator networks are known to be highly flexible and change frequently over time within a season ( Olesen et al . , 2008 ) , as well as between years ( Olesen et al . , 2011 ) and this flexibility likely explains why plants recruit additional pollinator species rather than specialize if they rely on outcrossing . Most pollinator networks are nested ( Pawar , 2014 ) and this nestedness is thought to stabilize mutualistic networks ( Rohr et al . , 2014 ) , reducing interspecific competition and enhancing the number of coexisting species ( Bastolla et al . , 2009 ) , with the end result of increasing the coexistence of several pollinating species on a plant . The high temporal plasticity in species composition coupled with the low variation in network structure properties in pollination networks makes such a scenario likely ( Petanidou et al . , 2008 ) . Flowers face a multidimensional challenge . They have to ensure outcrossing by an unpredictable number of pollinating species and individuals , all of which have different preferences and behaviors , and at the same time need to keep herbivores at bay . Here , we demonstrate that nectar-free , pollinator cheating , varieties of N . attenuata occur in nature , and that producing no nectar can substantially reduce ovipositions by M . sexta , but that this comes with a reduction in pollination services . While cheating on the nectar rewards works for one pollinator species , it does not for others . In contrast , cheating with regards to the attractant floral scent has little influence on M . sexta oviposition , but dramatically reduces outcrossing by M . sexta . All three examined pollinators show different responses to nectar and scent and thus the interactions of multiple species are likely responsible for fine-tuning the evolution of both nectar and floral scent , as both H . lineata , as well as A . alexandri are important pollen vectors in the absence of M . sexta . Both traits in combination are required to maximize maternal fitness in nature particularly in ensuring M . sexta visitations , which we hypothesize delivers the greatest diversity of pollen genotypes compared to H . lineata and A . alexandri . Hence , the close association between N . attenuata and M . sexta may result from M . sexta's ability to move pollen over longer distances than the other pollinators , and thus provide superior outcrossing services among N . attenuata populations which are frequently isolated by the large distances that frequently occur between fires , and hence N . attenuata populations . The long time between fires at any particular location may place a premium on outcrossing rates , as heterozygosity may increase herbivore resistance in future generations ( Mescher et al . , 2009; Bello-Bedoy and Nunez-Farfan , 2011 ) , as well as longevity in the seedbank . We conclude that herbivores , as well as pollinators shape the evolution of floral traits and that it makes little sense to study floral traits as if they only mediate pollination services . Wild-type ( WT ) N . attenuata plants , obtained from seeds collected from a native population in 1988 at the DI Ranch ( Santa Clara , UT ) and subsequently inbred for 22 or 30 generations , were transformed with Agrobacterium tumefaciens ( strain LBA 4404 ) ( Krügel et al . , 2002 ) containing the construct pRESC5CHAL to silence N . attenuata chalcone synthase ( Kessler et al . , 2008 ) , or the construct pSOL8SWEET9 to silence N . attenuata sweet9 ( Lin et al . , 2014 ) . The vector construction and transformation procedures have been previously described ( Krügel et al . , 2002 ) . For both CHAL and SWEET9 , we choose one line which harbored only one single insertion of the construct and which had completely normal growth and the strongest reductions in either floral volatile or nectar production . Both lines have been fully described previously: CHAL A-07-283-5 ( Kessler et al . , 2008 ) , SWEET9 A-10-198-4 ( Lin et al . , 2014 ) . To create a hemizygous cross between CHAL and SWEET9 , anthers of CHAL flowers of the T2 generation were removed and flowers hand pollinated with pollen of a SWEET9 ( T2 generation ) plant . The resulting T3 crossed seeds were used in all field and tent experiments together with the T3 generations of CHAL and SWEET9 plants . As a control , we used EV line A-04-266-3 transformed with pSOL3NC ( Bubner et al . , 2006 ) , which is known to be completely comparable to wild-type plants ( Schwachtje et al . , 2008 ) . Seed germination was performed in both glasshouse and field as described by Krügel et al . ( 2002 ) . For tent or glasshouse experiments , plants were grown individually in 1 ( glasshouse ) or 2 L pots ( tent ) in the glasshouse at 26–28°C under 16 h of light supplied by Philips SON-T Agro 400 ( Philips , Germany , http://www . lighting . philips . co . uk/ ) sodium lights until used for experiments in glasshouse or tent . The tent ( Amiran , Kenya , www . amirankenya . com ) has a dimension of 8 × 24 m and is about 4 m high . With a distance of 2 m among plants , 60 plants can fit into the tent . The sides and the front the tent is covered with mesh , which allows for air exchange . In the tent , plants were positioned 2 m apart from each other in rows , with three different transformed plants in one row , followed by a row of three WT plants , which served as pollen donors . In total , five plants per transgenic line were positioned in the tent per night for pollination experiments or 10 plants per transgenic line were used for oviposition trials . In the field , seedlings were transferred into previously hydrated 50-mm peat pellets ( Jiffy 703 , Always Grows , Sandusky , OH , http://www . alwaysgrows . com/ ) 14 days after germination and were gradually adapted to the environmental conditions of high sun and low relative humidity of the Great Basin Desert habitat over 14 days by keeping the seedlings in the shade . Adapted size-matched seedlings were transplanted into the field plot at the Lytle Ranch Preserve ( Santa Clara ) . Seedlings were watered every other day until roots were established . The four different transformed genotypes were randomly planted in rows with 4 m between each other and with one pair of WT plants between two transformed plants , which served as pollen donors in pollination experiments . Seeds of the transformed N . attenuata lines were imported under US Department of Agriculture Animal and Plant Health Inspection Service ( APHIS ) notification numbers 07-341-101 , 10-349-101m , and 11-350-102m and the field experiments were performed under notification numbers 06-242-03r , 10-349-102r , and 13-350-101r . To characterize the phenotypes of all transformed lines , nectar accumulation ( Figure 2—figure supplement 1 ) , nectar sugar concentration , floral scent emission ( Figure 2—figure supplement 2 ) , as well as leaf volatile emissions ( Figure 2—figure supplement 3 ) were measured in the field . While there were no differences in leaf volatile emission and nectar sugar concentration among the four lines , the lack of nectar in Nasweet9-silenced lines ( SWEET9 and CHALxSWEET9 ) , as well as the lack of floral BA emission in Nachal1-silenced lines ( CHAL and CHALxSWEET9 ) were confirmed . No phenological differences , such as plant shape , size , or flower size among these four lines , would suggest a lack of fitness benefit for plants producing no nectar or no scent . In the case of the hemizygous cross between CHAL and SWEET9 ( CHALxSWEET9 ) , we observed that occasionally plants started to secrete nectar in late developmental stages . This phenomenon is likely caused by methylation events which occur in aging hemizygous lines ( Weinhold et al . , 2013 ) . In all experiments , we therefore measured nectar production daily and excluded plants which had lost their initial no-nectar phenotype . All flowers were removed daily before opening . On experimental days , flowers were allowed to open in the evening and inflorescences were covered with plastic bags ( Plastibrand , Wertheim , Germany ) to exclude pollinators and to retard nectar evaporation . Nectar volume was measured with 25 μL glass capillaries ( BLAUBRAND , Wertheim , Germany ) on the following morning between 5 and 6 am , at the time of the maximal nectar accumulation ( Kessler , 2012 ) . To quantify nectar accumulations , the entire corolla was removed with mild pressure from the flower . The entire nectar remains in the corolla tube and can be squeezed from the tube into a glass capillary ( Rothe et al . , 2013 ) . Nectar in the capillary was quantified with a ruler . To quantify nectar sugar concentration , we used a portable refractometer ( Optech , Sliedrecht , the Netherlands ) with a range from 0 to 32% and a resolution of 0 . 2% . Nectar of 4–7 flowers per plant was measured to assess the variability in the different native phenotypes ( single plants ) , and two flowers per plant were measured in 10 plants of each of our transformed lines , as well as the nectar-free natural phenotypes . For statistics , the average nectar volume or nectar sugar concentration of all measured flowers of a plant was used . Plant volatiles were collected on silicone tubing ( ST ) and analyzed as described in Kallenbach et al . ( 2014 ) . Briefly , polyethylene terephthalate ( PET ) containers were used to enclose leaves ( 300 mL sampling volume ) and flowers ( 30 mL sampling volume ) for headspace sampling . Leaf volatiles were collected in the glasshouse overnight using a similar , fully expanded , mature , non-senescent stem leaf from each plant ( n = 7 , per genotype ) . Flower volatiles from EV , CHAL , SWEET9 , and CHALxSWEET9 ( Figure 2—figure supplement 2 ) were collected on ST as described ( Kallenbach et al . , 2014 ) in the field by trapping flowers overnight ( 8 pm–6 am ) on the first night of opening . Flower volatiles from native populations ( Figure 1 ) were collected using Super-Q traps and analyzed as described in Wu et al . ( 2008 ) . Empty trapping containers distributed among plants were used as background controls . BA levels were calculated using external calibrations . Volatile analysis was performed on a TD-20 thermal desorption unit ( Shimadzu , Germany , www . shimadzu . de ) connected to a quadrupole GC-MS-QP2010Ultra ( Shimadzu ) . Individual STs were placed in 89 mm glass thermo desorption ( TD ) tubes ( Supelco , Germany , sigmaaldrich . com ) and desorbed under a stream of nitrogen at 60 ml min−1 for 8 min at 200°C . Desorbed volatiles were cryo-focused at −20°C onto a Tenax adsorbent trap in front of the column . After desorption , the Tenax trap was heated to 230°C within 10 s , and analytes were injected with a 1:20 split ratio onto a ZB-Wax-plus column ( 30 m long , 0 . 25 mm i . d . , 0 . 25 μm film thickness; Phenomenex , Germany , www . phenomenex . com ) with He as the carrier gas at a constant linear velocity of 40 cm s−1 . The TD-GC interface was maintained at 230°C . Two different Gas Chromatograph ( GC ) oven gradients for the profiling of leaf and flower headspaces were used . For leaf samples , the oven was held at 40°C for 5 min , then ramped to 185°C at 5 . 0°C min−1 , and finally to 230°C at 30°C min−1 , where it was held for 0 . 5 min . For analysis of flower samples , the oven was held at 60°C for 1 min , then ramped to 150°C at 30°C min−1 , then to 200°C at 10°C min−1 , and finally to 230°C at 30°C min−1 and held for 1 min . Electron impact ( EI ) spectra were recorded at 70 eV in scan mode from 33 to 400 m/z using a scan speed of 2000 Da s−1 . The transfer line was maintained at 240°C and the ion source at 220°C . Data processing and export were performed using the Shimadzu GCMS solutions software ( v2 . 72 ) . Flower volatiles were analyzed against our reference standards library . For leaf volatiles analysis , raw data files were converted to netCDF format and processed using the R ( http://www . r-project . org ) packages XCMS ( Tautenhahn et al . , 2008 ) and CAMERA ( http://www . bioconductor . org/biocLite . R ) as described ( Gaquerel et al . , 2010 ) . After removal of contaminants and known artifact peaks , 94 compounds were detected , and the Metaboanalyst software ( 5 , 6 ) was used to perform principal components analysis ( PCA ) . As N . attenuata is a self-compatible plant , anthers of flowers were removed in the morning between 5 and 7 am from experimental flowers , just before anthesis ( Kessler et al . , 2008 ) and 12 hr before flowers open and are accessible for pollinators . Five flowers per plant were antherectomized . All other floral buds which were about to open within the next 2 days were removed . WT plants located between the experimental plants of the different genotypes were allowed to flower and served as pollen donors . This procedure was used in both tent and field . In a control experiment in the tent , all four lines used in the experiments were equally able to produce seeds from hand-pollinated flowers ( one-way ANOVA: F3 , 36 = 1 . 19 , p = 0 . 33; n = 10 ) . EV plants produced on average 233 ± 23 seeds per flower , CHAL 242 ± 22 , SWEET9 209 ± 21 , and CHALxSWEET9 265 ± 19 . We have no data available for the longevity of seeds of the four lines in the seedbank , which would allow predictions about fitness benefits for future generations of these lines . As such data require years to collect , it is beyond the scope of this study . In a 2011 field experiment , the response of hummingbirds was tested without including the CHALxSWEET9 cross , as these seeds were not available at this time . To exclude night-time pollinators and thus to evaluate only hummingbird pollination , plants were covered with mesh cones ( Kessler et al . , 2010 ) between 8 pm and 6 am . By visual observation , it was ensured that only A . alexandri hummingbirds were visiting the flowers during the times when flowers were accessible to visitors . Bees were not present on the field plot at the time of hummingbird experiments on 2 days in 2011 . Experiments planned to investigate the pollination services of the entire community of floral visitors in 2012 and 2013 failed due to fungal or bacterial diseases which killed plants before they started to flower . In 2014 , we conducted an experiment , in which we allowed the entire community ( day- and night-pollinators ) of floral visitors to access treated plants on four experimental days . All plants in this experiment were accessible for floral visitors over the entire lifespan of the atherectomized flowers ( 3 days ) . All these experiments have not been influenced by nectar robbing from carpenter bees ( Xylocopa spp . ) . On all experimental days , we monitored their presence , and experiments were not conducted if carpenter bees occurred on the plot , which happened in the later parts of the 2011 and 2014 seasons , when pollination experiments had to be stopped . In the tent , plants were treated similarly to those of the field experiments , and one single male M . sexta or H . lineata was released in the tent to interact with the experimental plants overnight . Only moths raised on N . attenuata or N . tabacum ( M . sexta ) or Fuchsia sp . ( H . lineata ) from field-collected eggs were used for experiments in the tent . In the following morning , plants were moved from the tent into the glasshouse until capsules had matured ( approximately 14 days ) . On each experimental day , new plants and a new moth were used . In total seven M . sexta and seven H . lineata moths were tested in these experiments . The same plants used for pollination experiments were used for M . sexta oviposition in the field . Flower number of all plants was reduced to five newly opening flowers every afternoon so as to standardize the apparency of each plant's inflorescence to the moths . Each morning , eggs were counted and removed from the plants . The collected eggs , as well as eggs collected from D . wrightii plants at this time of the season were used to rear moths for experiments in the tent . All caterpillars which hatched from these eggs proved to be M . sexta , which suggests that M . sexta accounts for all the oviposition data in our experiments , and not M . quinquemaculata , a closely related species which shares host plants with M . sexta in both the larval and the adult stages . Of the nine experimental days , M . sexta moths oviposited on only four of these days . In the tent experiments , unlike in the pollination trials , no WT pollen donor plants were necessary , which allowed for a larger number of experimental plants in the tent at a time . 10 plants per transformed line were randomly arranged in the tent with a distance of 2 m between plants . Plants were exchanged and rearranged daily , and a freshly mated female M . sexta moth was released into the tent on each day . For all pollination assays , experiments were conducted over several days . In all cases , the day did not influence the line effect significantly as revealed by a one-way ANOVA p > 0 . 05 . Therefore , each plant was used as a replicate and the average seed number produced from the five flowers on a plant was used for statistical analyses . For oviposition data , days were used as replicates and average values of the different lines were compared . Significant differences for all pollination , oviposition , as well as phenotyping data ( nectar volume and floral volatile emissions ) data were calculated using a Friedman signed rank test which was performed in R2 . 11 . 1 ( Crawley , 2007; R Development Core Team , 2007 ) . Control hand-pollinations were compared using a Fisher's PLSD post-hoc test following a one-way ANOVA , which were performed using StatView program version 5 . 0 ( SAS Institute Inc . , NC , USA; www . statview . com ) .	Flowering plants have evolved a number of different approaches to reproduction . Some use their own pollen and self-fertilize , while others use pollen from other nearby plants . This fertilization by other plants is called ‘outcrossing’ and introduces new genetic variation into each generation , which is extremely important for the evolutionary process . Some flowering plants rely on animals to help with pollination , attracting visitors with floral scents and rewarding the visitors with sugar-rich nectar . But scent and nectar also attract herbivores that damage the plants . This causes a dilemma for flowering plants , which has led some to evolve to not produce scent or to offer no nectar while masquerading as a plant that does . Previous studies into the costs and benefits of such strategies have looked at the effects of either floral scent or nectar , but no-one has uncoupled the effects of these two traits on both pollination and herbivore attack . Kessler et al . have addressed this issue in wild tobacco plants , which can both self-fertilize and outcross , and which produce varying amounts of scent and nectar . The experiments were conducted under mesh tents and in field trials in the plant's natural habitat: the Great Basin Desert in Utah . Kessler et al . used a gene-silencing technique called ‘RNA interference’ to inhibit the production of scent or nectar , either separately or together . When grown in field trials , under conditions that prevent self-fertilization , these tobacco plants are frequently visited by a hummingbird and three species of hawkmoth . All four of these animals pollinate the tobacco plants , but one of the moths also lays eggs that hatch into caterpillars , which damage the plant . Kessler et al . monitored the effects that the loss of scent , nectar or both had on visits by each pollinator and on outcrossing . These experiments revealed that scent is essential to attract one hawkmoth species but not for another ( called Hyles lineata ) . Furthermore , while , the hummingbird needs nectar , the H . lineata moth does not; but this moth won't visit flowers that lack both scent and nectar . The experiments also show that , for the moth that lays its eggs on the tobacco plants , both scent and nectar increase pollination and egg laying , but nectar has a stronger effect . Thus reducing nectar , as this tobacco plant does in the wild , is one strategy that can be used to reduce herbivore attack by caterpillars . Together , these findings highlight that it is important to study both herbivores and pollinators when attempting to understand the evolution of floral traits .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "ecology", "short", "report", "plant", "biology" ]	2015	How scent and nectar influence floral antagonists and mutualists
After the European colonization of the Americas , there was a dramatic population collapse of the Indigenous inhabitants caused in part by the introduction of new pathogens . Although there is much speculation on the etiology of the Colonial epidemics , direct evidence for the presence of specific viruses during the Colonial era is lacking . To uncover the diversity of viral pathogens during this period , we designed an enrichment assay targeting ancient DNA ( aDNA ) from viruses of clinical importance and applied it to DNA extracts from individuals found in a Colonial hospital and a Colonial chapel ( 16th–18th century ) where records suggest that victims of epidemics were buried during important outbreaks in Mexico City . This allowed us to reconstruct three ancient human parvovirus B19 genomes and one ancient human hepatitis B virus genome from distinct individuals . The viral genomes are similar to African strains , consistent with the inferred morphological and genetic African ancestry of the hosts as well as with the isotopic analysis of the human remains , suggesting an origin on the African continent . This study provides direct molecular evidence of ancient viruses being transported to the Americas during the transatlantic slave trade and their subsequent introduction to New Spain . Altogether , our observations enrich the discussion about the etiology of infectious diseases during the Colonial period in Mexico . European colonization in the Americas resulted in a frequent genetic exchange mainly between Native American populations , Europeans , and Africans ( Aguirre-Beltrán , 2005; Rotimi et al . , 2016; Salas et al . , 2004 ) . Along with human migrations , numerous new species were introduced to the Americas including bacterial and viral pathogens , which played a major role in the dramatic population collapse that afflicted the immunologically naïve Indigenous inhabitants ( Acuña-Soto et al . , 2004; Lindo et al . , 2016 ) . Among these pathogens , viral diseases , such as smallpox , measles , and mumps , have been proposed to be responsible for many of the devastating epidemics during the Colonial period ( Acuña-Soto et al . , 2004 ) . Remarkably , the pathogen ( s ) responsible for the deadliest epidemics reported in New Spain ( the Spanish viceroyalty that corresponds to Mexico , Central America , and the current US southwest states ) remains unknown and is thought to have caused millions of deaths during the 16th century ( Acuña-Soto et al . , 2004 ) . Indigenous populations were drastically affected by these mysterious epidemics , generically referred to as Cocoliztli ( ‘pest’ in Nahuatl ) , followed by Africans and to a lesser extent European people ( Acuña-Soto et al . , 2004; Malvido and Viesca , 1982; Somolinos d’Árdois , 1982 ) . Accounts of the 1576 Cocoliztli epidemic were described in autopsy reports of victims treated at the ‘Hospital Real de San José de los Naturales’ ( HSJN ) ( Malvido and Viesca , 1982; Wesp , 2017 ) , the first hospital in Mexico dedicated specifically to treat the Indigenous population ( Malvido and Viesca , 1982; Wesp , 2017; Figure 1a and b ) . The study of ancient viral genomes has revealed important insights into the evolution of specific viral families ( Barquera et al . , 2020; Duggan et al . , 2016; Düx et al . , 2020; Kahila Bar-Gal et al . , 2012; Krause-Kyora et al . , 2018; Mühlemann et al . , 2018a; Mühlemann et al . , 2018a; Mühlemann et al . , 2018b; Neukamm et al . , 2020; Pajer et al . , 2017; Patterson Ross et al . , 2018; Xiao et al . , 2013 ) , as well as their interaction with human populations ( Spyrou et al . , 2019 ) . To explore the presence of viral pathogens in circulation during epidemic periods in New Spain , we leveraged the vast historical and archeological information available for the Colonial HSJN . These include the skeletal remains of over 600 individuals recovered from mass burials associated with the hospital’s architectural remnants ( Figure 1b ) . Many of these remains were retrieved from burial contexts suggestive of an urgent and simultaneous disposal of the bodies , as in the case of an epidemic ( Meza , 2013; Wesp , 2017 ) . Prior bioarcheological research has shown that the remains of a number of individuals in the HSJN collection displayed dental modifications and/or morphological indicators typical of African ancestry ( Meza , 2013 ) , consistent with historical and archeological research that documents the presence of a large number of both free and enslaved Africans and their descendants in Colonial Mexico ( Aguirre-Beltrán , 2005 ) . Indeed , a recent paleogenomics study reported a sub-Saharan African origin of three individuals from this collection ( Barquera et al . , 2020 ) . Here we describe the recovery and characterization of viral pathogens that circulated in New Spain during the Colonial period , using ancient DNA ( aDNA ) techniques ( Figure 1—figure supplement 1 ) . For this work , we sampled skeletal human remains recovered from the HSJN where archeological context suggest victims of epidemics were buried ( Meza , 2013 ) and from ‘La Concepcion’ chapel , one of the first catholic conversion centers in New Spain ( Moreno-Cabrera et al . , 2015; Figure 1a ) . We report the reconstruction of ancient hepatitis B virus ( HBV ) and human parvovirus B19 ( B19V ) genomes recovered from these remains , providing a direct molecular evidence of human viral pathogens of African origin being introduced to New Spain during the transatlantic slave trade . We sampled the skeletal remains from two archeological sites , a Colonial Hospital and a Colonial chapel in Mexico City ( Figure 1a and b ) . For the HSJN , 21 dental samples ( premolar and molar teeth ) were selected based on previous morphometric analyses and dental modifications that suggested an African ancestry ( Hernández-Lopez and Negrete , 2012; Karam-Tapia , 2012; Meza , 2013; Ruíz-Albarrán , 2012 ) . The African presence in the Indigenous Hospital might reflect an urgent response to an epidemic outbreak since hospitals treated patients regardless of the origin of the affected individuals during serious public health crises ( Meza , 2013 ) . Dental samples of five additional individuals were selected ( based on their conservation state ) from ‘La Concepción’ chapel ( COY ) , which is located 10 km south of the HSJN in Coyoacán , a Pre-Hispanic Indigenous neighborhood that became the first Spanish settlement in Mexico City after the fall of Tenochtitlan ( Moreno-Cabrera et al . , 2015 ) . Following strict aDNA protocols , we processed these dental samples to isolate aDNA for next-generation sequencing ( NGS ) ( Figure 1—figure supplement 1 , Materials and methods ) . Tooth roots ( which are vascularized ) can be a good source of pathogen DNA ( Key et al . , 2017 ) , especially in the case of viruses that are widespread in the bloodstream during systemic infection . Accordingly , a number of previous studies have successfully recovered ancient viral DNA from tooth roots ( Barquera et al . , 2020; Krause-Kyora et al . , 2018; Mühlemann et al . , 2020; Mühlemann et al . , 2018a; Mühlemann et al . , 2018b ) . Metagenomic analyses with MALT ( Herbig et al . , 2018 ) ( Materials and methods ) on the NGS data using the Viral NCBI RefSeq database as a reference ( Pruitt et al . , 2007 ) revealed 17 samples containing at least one normalized hit to viral DNA ( abundances were normalized to the smallest library since each sample had different number of reads ) ( Materials and methods ) , particularly similar to Hepadnaviridae , Herpesviridae , Parvoviridae , and Poxviridae viral families ( Figure 1c , Figure 1—figure supplement 2 , Materials and methods ) . These viral hits revealed the potential to recover ancient viral genomes from these samples . We selected 12 samples for further screening ( Figure 1c , Figure 1—figure supplement 3 ) based on the DNA concentration of the NGS library and the quality of the hits to a clinically important virus ( HBV , B19V , papillomavirus , smallpox ) . To isolate and enrich the viral DNA fraction in the sequencing libraries , biotinylated single-stranded RNA probes designed to capture sequences from diverse human viral pathogens were synthesized ( Supplementary file 1A ) . The selection of the viruses included in the capture design considered the following criteria: ( 1 ) DNA viruses previously retrieved from archeological human remains ( i . e . , hepatitis B virus , human parvovirus B19 , variola virus ) , ( 2 ) representative viruses from families capable of integrating into the human genome ( i . e . , Herpesviridae , Papillomaviridae , Polyomaviridae , Circoviridae ) , or ( 3 ) RNA viruses with a DNA intermediate ( i . e . , Retroviridae ) . Given the size constraints of the probe kit , only a couple of genes were selected from some viral families ( Materials and methods , Supplementary file 1A ) . Additionally , a virus-negative aDNA library , which showed no hits to any viral family included in the capture assay ( except for a frequent Poxviridae-like region identified as an Alu repeat; Tithi et al . , 2018 ) was captured and sequenced as a negative control ( HSJN177 ) to estimate the efficiency of our capture assay . Only one post-capture library had an ~100-fold increase of Hepadnaviridae-like hits ( HBV ) , while three more libraries had an ~50–200-fold increase of Parvoviridae-like hits ( B19V ) ( Figure 1c , Supplementary file 1B ) , compared to their corresponding pre-capture libraries ( Materials and methods ) . In contrast , the captured negative control ( HSJN177 ) presented a negligible enrichment of these viral hits ( Figure 1c , Supplementary file 1B ) . Independently , a metagenomic analysis using Kraken2 ( Wood et al . , 2019 ) and Pavian ( Breitwieser and Salzberg , 2020 ) was performed on the non-human ( unmapped ) reads as part of a different study ( Bravo-Lopez et al . , 2020 ) . Our samples presented bacterial constituents of human oral and soil microbiota at different proportions between the samples ( Figure 1—figure supplements 4–7 ) . Although no lethal bacterial pathogen was retrieved , some ancient dental pathogens ( Tannerella forsythia ) were reconstructed and described in more detail by Bravo-Lopez et al . , 2020 ( Figure 1—figure supplements 4–7 ) . We verified the authenticity of the reads mapped to HBV ( BWA ) or B19V ( BWA/blastn ) in the enriched libraries ( Materials and methods ) by querying the reads against the non-redundant ( nr ) NCBI database using megaBLAST ( Altschul et al . , 1990 ) . This step was performed to avoid including in the genome assembly reads that were mapped by BWA or blastn as HBV or B19V , but with a similar identity to a different taxon in the nr database ( and absent in DS1; Materials and methods ) . Therefore , we only retained reads for which the top hit was to either B19V or HBV ( Supplementary file 1C ) . To confirm the ancient origin of these viral reads , we evaluated the misincorporation damage patterns using the program mapDamage 2 . 0 ( Jónsson et al . , 2013 ) , which revealed an accumulation of C to T mutations towards their 5′ terminal site with an almost symmetrical G to A pattern on the 3′ end ( Figure 2a , Figure 2—figure supplement 1 ) , as expected for aDNA ( Briggs et al . , 2007 ) . Three ancient B19V genomes were reconstructed ( Figure 2b , Supplementary file 1C ) with sequence coverages between 92 . 37% and 99 . 1% , and average depths of 2 . 98–15 . 36× along their single-stranded DNA ( ssDNA ) coding region , which excludes the double-stranded DNA ( dsDNA ) hairpin regions at each end of the genome ( Luo and Qiu , 2015 ) . These dsDNA inverse terminal repeats ( ITRs ) displayed considerably higher depth values ( <218× ) compared to the coding region consistent with the better postmortem preservation of dsDNA compared to ssDNA ( Lindahl , 1993; Figure 2b ) . In addition , we reconstructed one ancient HBV genome ( Figure 2c , Supplementary file 1C ) at 30 . 8× average depth and with a sequence coverage of 89 . 9% , including its ssDNA region at a reduced depth ( <10× ) . The reconstructed ancient HBV genome shows a 6 nucleotide ( nt ) insertion in the core gene , which is characteristic of the genotype A ( Kramvis , 2014 ) . Further phylogenetic analyses ( Materials and methods ) revealed that the Colonial HBV genome clustered with modern sequences corresponding to sub-genotype A4 ( previously named A6 ) ( Pourkarim et al . , 2014; Figure 3a , Figure 3—figure supplement 1 ) . The genotype A ( HBV/GtA ) has a broad diversity in Africa reflecting its long history in this continent ( Kostaki et al . , 2018; Kramvis , 2014 ) , while the sub-genotype A4 has been recovered uniquely from African individuals in Belgium ( Pourkarim et al . , 2010 ) and has never been found in the Americas . Regarding the three Colonial B19V genomes from individuals HSJN240 , COYC4 , and HSJNC81 ( C81A ) , these were phylogenetically closer to modern B19V sequences belonging to genotype 3 ( Figure 3b , Figure 3—figure supplements 2–3 ) . This B19V genotype is divided into two sub-genotypes: 3a , which is mostly found in Africa , and 3b , which is proposed to have spread outside Africa recently ( Hübschen et al . , 2009 ) . The viral sequences from the individuals HSJN240 and COYC4 are similar to sub-genotype 3b genomes sampled from immigrants ( Morocco , Egypt , and Turkey ) in Germany ( Schneider et al . , 2008; Figure 3b , Figure 3—figure supplements 2–3 ) while the sequence of the individual HSJNC81 is more similar to a divergent sub-genotype 3a strain ( Figure 3b , Figure 3—figure supplements 2–3 ) retrieved from a child with severe anemia born in France ( Nguyen et al . , 1999 ) . These observations support the African origin of the reconstructed Colonial viral genomes . In order to use our reconstructed viral genomes as molecular fossils to recalibrate each virus phylogeny and perform evolutionary inferences , we first needed to estimate if the phylogenetic relationships among B19V or HBV genomes had a temporal structure ( i . e . , sufficient genetic change between sampling times to reconstruct a statistical relationship between genetic divergence and time ) ( Rambaut et al . , 2016 ) . In the context of viruses , temporal structure is canonically tested with a root-to-tip distance and date randomization analyses ( see Firth et al . , 2010; Rieux and Balloux , 2016 ) . Similarly to previous studies ( Krause-Kyora et al . , 2018; Patterson Ross et al . , 2018 ) , we found little or no temporal structure for this HBV phylogeny containing all genotypes ( R2 = 0 . 1351; correlation coefficient = 0 . 3676 ) ( Figure 3—figure supplement 5a-c ) . The complex evolution of HBV may not be prone to an appropriate genetic dating since multiple inter-genotype recombination and cross-species transmission ( Human-Ape ) events ( Krause-Kyora et al . , 2018 ) occurred throughout its evolution . Since the entire genotype A has been identified as a recombinant genotype before ( Mühlemann et al . , 2018a ) , we analyzed it independently and identified a stronger temporal signal within this genotype ( R2 = 0 . 722; correlation coefficient = 0 . 8498 ) ( Figure 3—figure supplement 5d-f ) . In the case of B19V , we identified a temporal structure when including all three genotypes ( R2 = 0 . 3837; correlation coefficient = 0 . 6194 ) ( Figure 3—figure supplement 6a-c ) , in agreement with previous studies ( Mühlemann et al . , 2018b ) . Furthermore , we corroborated this temporal structure was not an artifact by a set of tip-dated randomized analyses ( Rieux and Balloux , 2016 ) , where none of the 95% highest posterior density ( HPD ) intervals of the clock rate overlapped with the correctly dated dataset ( Figure 3—figure supplement 7 ) . Given its strong temporal structure , we then performed a dated coalescent phylogenetic analysis for B19V ( Supplementary file 1D ) . We inferred a median substitution rate for B19V of 1 . 03 × 10–5 ( 95% HPD: 8 . 66 × 10–6–1 . 21 × 10–5 ) s/s/y under a strict clock and a constant population prior , and a substitution rate of 2 . 62 × 10–5 ( 95% HPD: 1 . 50 × 10–5–3 . 98 × 10–5 ) s/s/y under a relaxed log normal clock and a constant population prior . The divergence times from the most recent common ancestor of genotypes 1 , 2 , and 3 under a strict clock were 7 . 19 ( 95% HPD: 6 . 98–7 . 46 ) , 2 . 11 ( 95% HPD: 1 . 83–2 . 51 ) , and 3 . 64 ( 95% HPD: 3 . 04–4 . 33 ) ka , respectively . The inferred substitution rates and divergence times from the most recent common ancestor for genotypes 1 and 2 were similar to previous estimations ( Mühlemann et al . , 2018b ) that included much older sequences , while the divergence of genotype 3 was subtly older since no other ancient genotype 3 had been reported previously . Next , we used the shotgun data generated to determine the mitochondrial haplogroup of the hosts , as well as their autosomal genetic ancestry using the 1000 Genomes Project ( 1000 Genomes Project Consortium , 2015 ) as a reference panel ( Figure 4a , Supplementary file 2A ) . The nuclear genetic ancestry analysis showed that all three HSJN individuals , from which the reconstructed viral genomes were isolated , fall within African genetic variation in a principal component analysis ( PCA ) plot ( Figure 4a ) , while their mitochondrial aDNA belong to the L haplogroup , which has high frequency in African populations ( Supplementary file 2A , Figure 2—figure supplement 2 ) . Additionally , we performed 87Sr/86Sr isotopic analysis on two of the HSJN individuals using tooth enamel as well as phalanx ( HSJN240 ) or parietal bone ( HSJNC81 ) to provide insights on the places of birth ( adult enamel ) and where the last years of life were spent ( phalanx/parietal ) . The 87Sr/86Sr ratios measured on the enamel of the individual HSJNC81 ( 0 . 71098 ) and HSJN240 ( 0 . 71109 ) are similar to average 87Sr/86Sr ratios found in soils and rocks from West Africa ( average of 0 . 71044 , Figure 4—figure supplement 1 , Supplementary file 2 ) , as well as to 87Sr/86Sr ratios described in first-generation Africans in the Americas ( Barquera et al . , 2020; Bastos et al . , 2016; Fricke et al . , 2020; Price et al . , 2012; Schroeder et al . , 2009 ) . In contrast , the 87Sr/86Sr ratios on the parietal and phalange bones from the HSJNC81 ( 0 . 70672 ) and HSJN240 ( 0 . 70755 ) show lower values similar to those observed in the Trans Mexican Volcanic Belt where the Mexico City Valley is located ( 0 . 70420–0 . 70550 , Figure 4—figure supplement 1 , Supplementary file 2 ) . Moreover , radiocarbon dating of HSJN240 ( 1442–1608 CE , years calibrated for 1σ ) and HSJN194 ( 1472–1625 CE , years calibrated for 1σ ) ( Supplementary file 2A , Figure 4—figure supplement 2 ) indicates that these individuals arrived during the first decades of the Colonial period , when the number of enslaved individuals arriving from Africa was particularly high ( Aguirre-Beltrán , 2005 ) . Strikingly , Colonial individual COYC4 , who was also infected with an African B19V strain , clusters with present-day Mexicans and Peruvians from the 1000 Genomes Project ( Figure 4a ) . An ADMIXTURE ( Alexander and Lange , 2011 ) analysis with these data confirmed a Native American genetic component ( Figure 4b ) , as expected for an indigenous individual . The B19V ancient genome from the individual COYC4 is the first genotype 3 genome obtained from a non-African individual and suggests that following the introduction from Africa , the virus ( B19V ) spread and infected people of different ancestries during the Colonial period . In this study , we reconstructed one HBV and three B19V ancient genomes from four different individuals using NGS , metagenomics , and in-solution targeted enrichment methods ( Figure 2b , c , Figure 1—figure supplement 1 ) . Several lines of evidence support the ancient nature of these viral sequences , in contrast to environmental contamination or a capture artifact . First , our negative control was not enriched for B19V or HBV hits in our capture sequencing ( Figure 1c ) . For those samples that showed an enrichment in viral sequences after capture , the reads covered the reference genomes almost in their entirety and displayed deamination patterns at the terminal ends of the reads , as expected for aDNA ( Figure 2a ) . Moreover , it is important to notice that B19V and HBV are blood-borne human pathogens that are not present in soil or the environment , and that DNA from these viruses had never been extracted before in the aDNA facilities used for this study . The recovery of aDNA from B19V , which has a ssDNA genome ( with dsDNA terminal repeats ) , in previous studies ( Mühlemann et al . , 2018b ) as well as in our samples is noteworthy considering the NGS libraries were constructed using dsDNA as a template . Therefore , we would not expect to recover the ssDNA from B19V with this library building method . However , it is known that dsDNA intermediates form during the B19V replication cycle ( Ganaie and Qiu , 2018 ) , and that throughout the viral infection the replicating genomes are present in both the ssDNA and dsDNA forms . The sequences we retrieved must therefore correspond to the cell-free dsDNA replication intermediates . This is coherent with the peculiar coverage pattern on the B19V genome , where the dsDNA hairpins at its terminal sites and are highly covered , reflecting a better stability of these regions over time ( Figure 2b ) . Similarly , the partially circular dsDNA genome from HBV was poorly covered at the ssDNA region ( Figure 2c ) , which also goes through a dsDNA phase during replication , a similar coverage is reported in three previous ancient HBV genomes ( Krause-Kyora et al . , 2018 ) . Although HBV and B19V are also capable of integrating into the human host genome ( Yuen et al . , 2018; Janovitz et al . , 2017 ) , the uneven read coverage observed for all reconstructed viruses ( higher coverage in dsDNA regions ) suggests that these sequences do not correspond to integration events . If the B19V or HBV reads we recovered derived from integrated sequences in the human genome , we would expect an even coverage along the reference viral genome , which is not the case . Further analyses would be needed to determine if the aDNA retrieved in this and other studies comes from systemic circulating virions or from systemic cell-free DNA intermediates ( Cheng et al . , 2019 ) produced after viral replication in the bone marrow or liver for B19V and HBV , respectively ( Broliden et al . , 2006; Yuen et al . , 2018 ) . The ancient B19V genomes were assigned to genotype 3 . This genotype is most prevalent in West Africa ( Ghana: 100% , n = 11; Burkina Faso: 100% , n = 5 ) ( Candotti et al . , 2004; Hübschen et al . , 2009; Rinckel et al . , 2009 ) and a potential African origin has been suggested ( Candotti et al . , 2004 ) . It has also been sporadically found outside of Africa ( Jain et al . , 2015 ) , ( Candotti et al . , 2004; Rinckel et al . , 2009 ) in Brazil ( 50% , n = 12 ) ( Freitas et al . , 2008; Sanabani et al . , 2006 ) , India ( 15 . 4% , n = 13 ) ( Jain et al . , 2015 ) , France ( 11 . 4% , n = 79 ) ( Nguyen et al . , 1999; Servant et al . , 2002 ) , and the USA ( 0 . 85% , n = 117 ) ( Rinckel et al . , 2009 ) as well as in immigrants from Morocco , Egypt , and Turkey in Germany ( 6 . 7% , n = 59 ) ( Schneider et al . , 2008 ) . Two other genotypes , 1 and 2 , exist for this virus . Genotype 1 is the most common and is found worldwide , while the almost extinct genotype 2 is mainly found in elderly people from Northern Europe ( Pyöriä et al . , 2017 ) . Ancient genomes from genotypes 1 and 2 have been recovered from Eurasian samples , including a genotype 2 B19V genome from a 10th-century Viking burial in Greenland ( Mühlemann et al . , 2018b ) . 87Sr/86Sr isotopes on individuals from this burial revealed that they were immigrants from Iceland ( Mühlemann et al . , 2018b ) , suggesting an introduction of the genotype 2 to North America during Viking explorations of Greenland . While serological evidence indicates that B19V currently circulates in Mexico , only the presence of genotype 1 has been formally described using molecular analyses ( Valencia Pacheco et al . , 2017 ) . Taken together , our results are consistent with an introduction of the genotype 3 to New Spain as a consequence of the transatlantic slave trade imposed by European colonization . This hypothesis is supported by the 87Sr/86Sr isotopic analysis , which suggests that the individuals from the HSJN with B19V ( HSJN240 , HSJNC81 ) were born in West Africa and spent their last years of life in New Spain ( Figure 4—figure supplement 1 ) . Furthermore , the radiocarbon analysis for individuals HSJN240 and HSJN194 ( Figure 4—figure supplement 2 ) support this notion as they correspond to the Early Colonial period , during which the number of enslaved Africans arriving was higher compared to later periods ( Aguirre-Beltrán , 2005 ) . Remarkably , a B19V genome belonging to the genotype 3 was recovered from an individual ( COYC4 ) with 100% Indigenous ancestry ( Figure 4b ) . COY4 was excavated in an independent archeological site 10 km south of the HSJN ( Figure 1a ) , supporting the notion that viral transmissions between African individuals and Native Americans occurred during the Colonial period in Mexico City . The HBV genotype A is highly diverse in Africa , reflecting its long evolutionary history , and likely originated somewhere between Africa , the Middle East , and Central Asia ( Kostaki et al . , 2018 ) . The introduction of the genotype A from Africa to the Americas has been proposed based on phylogenetic analysis of modern strains from Brazil ( Freitas et al . , 2008; Kostaki et al . , 2018 ) and Mexico ( Roman et al . , 2010 ) , and more precisely of the sub-genotype A1 using sequences from Martinique ( Brichler et al . , 2013 ) , Venezuela ( Quintero et al . , 2002 ) , Haiti ( Andernach et al . , 2009 ) , and Colombia ( Alvarado-Mora et al . , 2012 ) . Recently , a similar introduction pattern was proposed for the quasi genotype A3 based on an ancient HBV genome recovered from an ancient African individual sampled in Mexico ( Barquera et al . , 2020 ) . The origin of the sub-genotype A4 is controversial since the apparent African origin is based on modern sequences recovered from African immigrants living in Europe ( Pourkarim et al . , 2010 ) . The Colonial ancient HBV genome reconstructed in our work represents the first ancient A4 linked to the transatlantic slave trade ( Figure 3a , Figure 3—figure supplement 1 ) , and the only report of this sub-genotype in the Americas , further supporting its African origin . The introduction of pathogens from Africa to the Americas has been proposed for other human-infecting viruses such as smallpox ( Mandujano-Sánchez et al . , 1982; Somolinos d’Árdois , 1982 ) , based on historical records; or yellow fever virus ( Bryant et al . , 2007 ) , HTLMV-1 ( Gadelha et al . , 2014 ) , hepatitis C virus ( genotype 2 ) ( Markov et al . , 2009 ) , and human herpes simplex virus ( Forni et al . , 2020 ) based on phylogenetic analysis of modern strains from Afro-descendant or admixed human populations . Although we cannot assert where exactly the African-born individuals in this study contracted B19V or HBV ( Africa , America , or the Middle Passage ) nor if the cause of their deaths can be attributed to such infections , the identification of ancient B19V and HBV in contexts associated with Colonial epidemics in Mexico City is still relevant in light of their paleopathological marks and the clinical information available for the closest sequences in the phylogenetic analyses . Notably , individual HSJNC81 displayed cribra orbitalia in the eye sockets and porotic hyperostosis on the cranial vault ( Figure 4—figure supplement 3 ) . The reconstructed ancient B19V genome from this individual is closest to the V9 strain , which was isolated from an infant with severe anemia and G6PD deficiency ( Nguyen et al . , 1999; GenBank: AJ249437; Figure 3b ) . The HSJN skeletal collection has a notably higher rate of cribra orbitalia and porotic hyperostosis compared to other Colonial archeological sites , marks that were proposed to be caused by an unknown infectious disease ( Castillo-Chavez , 2000 ) . These skeletal indicators are caused by irregular hematopoiesis in the bone marrow and are typically associated with genetic anemias such as thalassemia and sickle cell anemia ( Angel , 1966 ) , as well as to nutritional stress or parasitic infections ( Walker et al . , 2009 ) . It is acknowledged that B19V infection can cause severe or even fatal anemia due to the low level of hemoglobin in individuals with other blood disorders , such as thalassemia , sickle cell anemia , malaria , or iron deficiency ( Broliden et al . , 2006; Heegaard and Brown , 2002 ) . Therefore , since B19V infects precursors of the erythroid lineage ( Broliden et al . , 2006 ) , it is possible that the morphological changes found in HSJNC81 might be the result of a severe anemia caused or enhanced by a B19V infection . With our data we cannot discard the simultaneous presence of a genetic disease since the loci for thalassemia , sickle cell anemia , and G6PD deficiency were not covered with our human-mapped NGS data . Nevertheless , the identification of ancient B19V in a Colonial context is noteworthy considering several recent reports reveal that measles-like cases were actually attributable to B19V ( De Los Ángeles Ribas et al . , 2019; Rezaei et al . , 2016 ) or rubella ( Anderson et al . , 1985; Davidkin et al . , 1998; De Los Ángeles Ribas et al . , 2019; Rezaei et al . , 2016 ) , which produce a similar kind of rash and fever . Therefore , it is possible that B19V might have been responsible for some of the numerous cases attributed to measles that were described in early 16th-century Mexico ( Acuña-Soto et al . , 2004; Mandujano-Sánchez et al . , 1982; Wesp , 2017 ) , in particular historical records that document the treatment of an outbreak of measles at the HSJN in 1531 CE ( Meza , 2013 ) . Our study , however , does not reject the notorious role that measles played during the Colonial outbreaks ( as it is strongly supported by historical records ) , but provides evidence of the presence of B19V during the Colonial period in Mexico City to facilitate discussions about the paradigmatic etiology of the supposed measles epidemics reported in historical records ( Malvido and Viesca , 1982; Mandujano-Sánchez et al . , 1982; Somolinos d’Árdois , 1982 ) . This hypothesis requires additional comprehensive studies aimed at characterizing the presence of measles and rubella viruses from ancient remains , a task that currently poses difficult technical challenges given that RNA is known to degrade rapidly . In fact , most ancient viral RNA genomes have been recovered only from formalin-fixed tissue ( Düx et al . , 2020; Xiao et al . , 2013 ) . Furthermore , historical records of the autopsies of the victims of the 1576 CE Cocoliztli epidemic treated at the HSJN describe the presence of enlarged hard liver and jaundice ( Acuña-Soto et al . , 2002; Acuña-Soto et al . , 2004; Malvido and Viesca , 1982; Marr and Kiracofe , 2000; Somolinos d’Árdois , 1982 ) as well as a black spleen and lungs and heart with yellow liquid and black blood ( Acuña-Soto et al . , 2000; Malvido and Viesca , 1982; Somolinos d’Árdois , 1982 ) . This is noteworthy given that both HBV and B19V viruses proliferate in the liver and are associated with hepatitis and jaundice ( Broliden et al . , 2006; Yuen et al . , 2018 ) . The radiocarbon dating of individuals HSJN194 ( HBV ) and HSJN240 ( B19V ) suggests that these individuals died between 1472–1625 CE and 1442–1608 CE ( years calibrated for 1σ ) , respectively ( Figure 4—figure supplement 2 ) , which overlaps with the period of time when the hepatitis symptoms were reported in the autopsies after the 1576 Cocoliztli epidemic at the HSJN ( Acuña-Soto et al . , 2004; Marr and Kiracofe , 2000; Somolinos d’Árdois , 1982 ) . However , additional analyses are needed before being able to establish a link between these viruses and the wide array of symptoms described for Cocoliztli . Currently , technological limitations prevent the direct identification of ancient RNA viruses in bone or dental remains . However , future studies , with larger sample sizes from different contexts associated with the outbreak , should explore a wider range of pathogens previously suggested as potential causative agents , like arthropod-borne pathogens ( malaria , yellow fever virus , and dengue virus ) ( Marr and Kiracofe , 2000 ) or hemorrhagic fever RNA viruses ( Acuña-Soto et al . , 2004 ) . Furthermore , it is important to acknowledge that both viruses have also been previously identified in aDNA datasets not necessarily associated with disease or epidemic contexts ( Kahila Bar-Gal et al . , 2012; Krause-Kyora et al . , 2018; Mühlemann et al . , 2018a; Patterson Ross et al . , 2018 ) . Additionally , our data is not sufficient to elucidate the age when the individuals acquired the viruses or if it is related to their cause of death . In the case of HSJN194 , we cannot establish if he acquired HBV vertically or horizontally , nor if this individual presented an acute or chronic infection . Finally , although our data does not allow us to associate these viruses to a specific epidemic outbreak , the identification of HBV and B19V in Post-Contact remains opens up new opportunities for investigating their presence in similar contexts and expand our knowledge on their evolution and potential link to disease in Colonial Mexico . This type of research is particularly relevant when considering previous hypotheses favoring the synergistic action of different types of pathogens in these devastating Colonial epidemics ( Somolinos d’Árdois , 1982 ) . It is important to emphasize that our findings should be interpreted with careful consideration of the historical and social context of the transatlantic slave trade . This cruel episode in history involved the forced displacement of millions of individuals to the Americas ( ca . 250 , 000 to New Spain; Aguirre-Beltrán , 2005 ) under inhumane , unsanitary , and overcrowded conditions that , with no doubt , favored the spread of infectious diseases ( Mandujano-Sánchez et al . , 1982 ) . Therefore , the introduction of these and other pathogens from Africa to the Americas should be attributed to the brutal and harsh conditions of the Middle passage that enslaved Africans were subjected to by traders and colonizers , and not to the African peoples themselves . Moreover , the adverse life conditions for enslaved Africans and Native Americans , especially during the first decades after colonization , surely favored the spread of diseases and emergence of epidemics ( Mandujano-Sánchez et al . , 1982 ) . Integrative and multidisciplinary approaches are thus needed to understand this phenomenon in full . In summary , our study provides direct aDNA evidence of HBV and B19V introduced to the Americas from Africa during the transatlantic slave trade . The isolation and characterization of these ancient HBV and B19V genomes represent an important contribution to the ancient viral genomes reported in the Americas ( Barquera et al . , 2020; Duggan et al . , 2020; Schrick et al . , 2017 ) . Our results expand our knowledge on the viral agents that were in circulation during Colonial epidemics like Cocoliztli , some of which resulted in the catastrophic collapse of the immunologically naïve Indigenous population . Although we cannot assign a direct causality link between HBV and B19V and Cocoliztli , our findings confirm that these potentially harmful viruses were indeed circulating in individuals found in archeological contexts associated with this epidemic outbreak . Further analyses from different sites and samples will help understand the possible role of these and other pathogens in Colonial epidemics , as well as the full spectrum of pathogens that were introduced to the Americas during European colonization . Dental samples ( premolars and molars ) were obtained from 21 individuals from the skeletal collection associated with the HSJN and were selected based on morphological features indicating a possible African origin ( Hernández-Lopez and Negrete , 2012; Karam-Tapia , 2012; Meza , 2013; Ruíz-Albarrán , 2012 ) . Five additional samples were taken from ‘La Concepción’ chapel , based on their conservation state . Permits 401 . 1 S . 3-2018/1373 and 401 . 1 S . 3-2020/1310 to carry out this sampling and aDNA analyses were obtained from the Archeology Council of the National Institute of Anthropology and History ( INAH ) for the Hospital San Jose de los Naturales and ‘La Concepción’ chapel , respectively . Two of the individuals from whom the ancient viral genomes were retrieved ( HSJN194 and HSJN240 ) are mostly complete articulated skeletons and one individual ( HSJNC81 ) is an isolated cranium recovered during the early excavation stage and does not have any associated postcranial elements . The archeologists suggest that all of the individuals were deposited during an infectious disease epidemic in a mass burial context ( Figure 1b; Cabrera-Torres and García-Martínez , 1998 ) . Bone samples were transported to a dedicated ancient DNA clean-room laboratory at the International Laboratory for Human Genome Research ( LIIGH-UNAM , Querétaro , Mexico ) , where DNA extraction and NGS library construction was performed under the guidelines on contamination control for aDNA studies ( Warinner et al . , 2017 ) . Teeth were carefully cleaned with NaClO ( 70% ) and ethanol ( 70% ) superficially and later exposed to UV light for 1 . 5 min . The tooth root was sectioned from the crown and fragmented by mechanical pressure . Previously reported aDNA extraction protocols were used on approximately 200 mg of tooth root powder obtained from the HSJN and COY samples ( Dabney et al . , 2013; Rohland and Hofreiter , 2007 ) . A blank extraction control per batch was used to identify the presence of environmental and cross-sample contamination . dsDNA indexed ( 6 bp ) sequencing libraries were constructed using 30 µl of the DNA extract , as previously reported ( Meyer and Kircher , 2010 ) . Pooled libraries were sequenced on an Illumina NextSeq550 at the ‘Laboratorio Nacional de Genómica para la Biodiversidad’ ( LANGEBIO , Irapuato , Mexico ) , with a mid-output 2 × 75 format ( paired-end ) . The reads obtained ( R1 and R2 ) were merged ( >11 bp overlap ) and trimmed with AdapterRemoval 1 . 5 . 4 ( Schubert et al . , 2016 ) . Overlapping reads ( >30 bp in length , quality filter >30 ) were kept and mapped to the human genome ( hg19 ) using BWA 0 . 7 . 13 ( aln Algorithm ) ( Li and Durbin , 2009 ) . Mapped reads were used for further human analysis ( genetic ancestry and mitochondrial haplogroup determination ) , whereas unmapped reads were used for metagenomic analysis and viral genome reconstruction . The Viral RefSeq database was downloaded from the NCBI ftp server on February 2018; this included 7530 viral genomes , including human pathogens . MALT 0 . 4 . 0 ( Herbig et al . , 2018 ) software was used to taxonomically classify the reads using the viral genomes database as a reference . The viral database was formatted automatically with malt-build once , and non-human ( unmapped ) reads were aligned with malt-run using the blastn and SemiGlobal mode with an 85 minimal percent identity ( --minPercentIdentity ) and e-values of 0 . 001 ( --e ) . The RMA files were used for the normalization of the viral abundances based on the library with the smallest number of reads ( default , ( count of class/total count of sample ) * count of smallest sample ) and compared to all the samples from the same archeological site with MEGAN 6 . 8 . 0 ( Huson et al . , 2016 ) . Independently , unmapped reads ( non-human ) were taxonomically classified with Kraken2 ( Wood et al . , 2019 ) using a reference database composed of NCBI RefSeq bacterial , archaea , and viral genomes ( downloaded on November 3 , 2017 ) . The Kraken2 output was transformed to a BIOM-format table using Kraken-biom ( https://github . com/smdabdoub/kraken-biom; Dabdoub et al . , 2018 ) and then visualized with Pavian ( Breitwieser and Salzberg , 2020 ) . Detailed description of the results can be found in Bravo-Lopez et al . , 2020 . Twenty-nine viruses were included in the design of biotinylated probes ( Supplementary file 1A ) , including viral genomes previously recovered from archeological remains like B19V , B19V-V9 , and HBV ( consensus genomes ) , selected VARV genes , as well as clinically important viral families that are able to integrate into the human genome , have dsDNA genomes , or dsDNA intermediates . The HBV majority consensus genome ( >50% conservation per site ) was constructed using an alignment of modern references ( A–H genotype ) and a well-covered ( >5× coverage ) ancient genotype ( Mühlemann et al . , 2018a; LT992459 ) . Thirty VARV genes were chosen for a consensus sequence construction based on three categories; replication ( J6R , A24R , A29L , E4L , A50R , A5R , D7R , H4L , E9L ) , structural ( A27L , A25 , D8L , H3L , L1R , A33R , B5R , A16L ) , and immune host regulation ( B18R , A46R , B15R , K7R , N1L , M2L , E3L , H1L , B8R , D9R , D10 , K3L ) , and were obtained from all the available VARV genomes including three ancient genomes ( NCBI GenBank 2019 Duggan et al . , 2016; Pajer et al . , 2017 ) . The selected genes were aligned in AliView ( MUSCLE algorithm Edgar , 2004; Larsson , 2014 ) to generate a majority consensus for every gene . The generated consensus sequences targeted <20% of the VARV whole genome . For the Herpesviridae family , a total of 19 genes were selected , 6 from herpes simplex virus 1 ( UL30 , UL31 , UL19 , UL27 , US6 , UL10 ) , 6 from human cytomegalovirus ( UL54 , UL53 , UL86 , UL115 , UL75 , UL83 ) , and 7 from Epstein–Bar virus ( ORF9 , ORF69 , ORF25 , ORF47 , ORF8 , vIRF2 , K5 ) . GenBank IDs are shown in Supplementary file 1A . Selected genes from VARV and Herpesviridae were defined as 40 bp or 60 bp upstream the start codon , and downstream the stop codon , respectively , in order to ensure a uniform coverage of the entire coding region in case of a positive sample . The resulting design comprised 19 , 147 ssRNA 80 nt probes targeting , with a 20 nt interspaced distance , the whole or partial informative regions of eight viral families ( Poxviridae , Hepadnaviridae , Parvoviridae , Herpesviridae , Retroviridae , Papillomaviridae , Polyomaviridae , Circoviridae ) . To avoid a simultaneous false-positive DNA enrichment , low-complexity regions and human-like ( hg38 ) sequences were removed ( in silico ) . The customized kit was produced by Arbor Biosciences ( Ann Arbor , MI , USA ) . Capture-enrichment was performed on 30–90 ng ( depending the availability ) of the indexed libraries to pull-down viral aDNA using 60°C during 48 hr for hybridization , based on the manufacturer’s protocol ( version 4 ) . Libraries were amplified with 18–20 cycles ( Phusion U Hot Start DNA Polymerase by Thermo Fischer Scientific ) using primers for the adaptors of each post-capture library . PCR products were purified with SPRISelect Magnetic Beads ( Beckman Coulter ) and quantified with a Bioanalyzer 2100 ( Agilent ) . Amplified libraries were then pooled in different concentrations and deep sequenced on an Illumina NextSeq550 ( 2 × 75 middle output ) yielding >1 × 106 non-human reads ( Supplementary file 1C ) . In order to saturate the target viral genome , one or two non-consecutive rounds of capture were performed for HBV and B19V , respectively . Reads generated from each enriched library were analyzed exactly as the shotgun ( not-enriched ) libraries . Normalized abundances between shotgun and captured libraries were compared in MEGAN 6 . 8 . 0 ( Huson et al . , 2016 ) to evaluate the efficiency and specificity of the enrichment assay . HBV_DS2 and B19V_DS2 were aligned independently in Aliview ( Larsson , 2014; MUSCLE algorithm; Edgar , 2004 ) and curated manually to have the same lengths . The alignments were evaluated in jModelTest 2 . 1 . 10 ( Darriba et al . , 2012 ) using a corrected Akaike information criterion ( AICc ) and Bayesian information criterion ( BICc ) tests that supported with 100% confidence the evolutionary models used in our maximum likelihood analysis in RAxML ( Stamatakis , 2014 ) . To test the temporal structure of our ML trees , a root-tip-dated analysis was performed on Tempest 1 . 5 . 3 ( Rambaut et al . , 2016 ) for both DS2 ( B19V , HBV ) in the presence or absence of ancient sequences and without the sequences presented in this study ( Figure 3—figure supplements 5–6 ) . In the case of HBV , an additional analysis was performed only on the genotype A to find a higher temporal structure in the presence or absence of ancient sequences and without the HSJN194 HBV genome presented in this study ( Figure 3—figure supplement 5 ) . For the B19V_DS2 , the temporal structure suggested by root-tip distance analysis was corroborated using a date randomization test ( DRT ) with TipDatingBeast 1 . 0 . 5 ( Rieux and Khatchikian , 2017 ) and BEAST 2 . 5 . 1 ( Drummond et al . , 2012; Figure 3—figure supplement 6 ) . Since the DRT and the root-tip-dated analysis suggested a temporal structure for the B19V_DS2 , a coalescent dated tree was generated in BEAST 2 . 5 . 1 ( Drummond et al . , 2012 ) for B19V using a relaxed and strict clock; both with different priors ( coalescent constant , exponential , and Bayesian skyline population priors ) , with an a priori substitution rate interval of 1 × 10–3–1 × 10–7 s/s/y ( Mühlemann et al . , 2018b ) . For the Colonial genomes used in this study , a uniform sampling was indicated using the radiocarbon dates for HSJN240 ( 495 ± 166 ybp ) . When radiocarbon dating was not possible , an archeological date interval was set for HSJNC81 ( 332 . 5 ± 269 ybp ) and COYC4 ( 320 ± 400 ybp ) , based on the archeological estimates of both sites . The strict molecular clock analyses were performed with a 50 million MCMC sampled each 5000 generations , while the relaxed molecular clock with exponential population was run with a 250 million MCMC sampled each 5000 generations , and the relaxed molecular clock with coalescent constant and Bayesian Skyline population priors were run with 250 million MCMC and with 350 million MCMC sampled each 5000 generations . Both files were mixed with a 25% burn-in LogCombiner ( Drummond et al . , 2012 ) . All the Bayesian analyses were mixed and reached convergence ( >200 ESS ) as estimated in Tracer 1 . 7 ( Rambaut et al . , 2018; Supplementary file 1D ) . The first 25% of the generated trees were discarded ( burn in ) and a Maximum Clade Credibility Tree with median ages was created with TreeAnnotator ( Drummond et al . , 2012; Figure 3—figure supplements 3–4 ) . Radiocarbon analysis was conducted at the Physics Institute of the National Autonomous University of Mexico ( UNAM ) for the individuals in this study with complete skeletons ( HSJN194 and HSJN240 ) . From these individuals , phalanx bones ( left hand ) were cleaned , dried , and powdered to be digested in a HCl 0 . 5 M solution followed by a NaOH 0 . 01 M and HCl 0 . 2 M treatment . Collagen was then filtered ( >30 kDa ) and graphitization was performed on an AGEIII ( Ion Plus ) . 14C , 13C , and 12C isotopes were analyzed from graphite in a Tandetron ( High Voltage Engineering Europa B . V . ) mass spectrometer with a 1 V energy accelerator . Radiocarbon dates were estimated based on InCal13 ( Reimer et al . , 2013 ) calibration curve and corrected with OxCal v4 . 2 . 4 . ( Bronk Ramsey , 2013 ) . Tooth enamel was carefully extracted with the aid of dental tools . The material underwent several cleaning procedures before crushing to a 50 μm grain size with an agate mortar . Chemicals used for this purpose included 30% H2O2 and 1–1 . 5 N HNO3 . In between , rinses were performed with deionized water ( Milli-Q ) . Ultrasonic bath ( USB ) was used to accelerate these processes . After obtaining the desired grain size , samples were treated with 30% H2O2 , 1 N NH4Cl , and alternated with water washes . To get rid of any secondary contaminant or any postmortem external agent that could alter the Sr isotopic values , tooth samples were treated with a three-step leaching technique: the first leachate is obtained with 0 . 1 N acetic acid for 30 min ( USB ) . The solution is decanted and dried under infrared light ( Lix 1 ) . The residue was leached for 15 min in 1 N acetic acid ( USB ) and subsequently stored overnight for 12 hr in the same acid . The solution was decanted and dried to obtain the second leachate ( Lix 2 ) . The residue ( Res ) is dissolved in 8 N HNO3 as well as Enamel Lix 1 and Enamel Lix 2 in closed Teflon beakers on a hot plate at 90°C . A total of three aliquots from each molar were obtained from this leaching process . After sample digestion , Sr from teeth and bone samples was extracted with Sr-Spec ( EICHROM ) ion exchange column chemistry . Detailed analytical procedures are described in Solís-Pichardo et al . , 2017 . Sr isotope analysis was carried out with a Triton Plus ( Thermo Scientific ) thermal ionization mass spectrometer with 9 Faraday collectors at the ‘Laboratorio Universitario de Geoquímica Isotópica’ ( LUGIS , UNAM ) . Sr was measured as metallic ions with 60 isotopic ratios that were normalized for mass fractionation to 86Sr/88Sr = 0 . 1194 . The mean value for the NBS 987 Sr standard was 87Sr/86Sr = 0 . 710254 ± 0 . 000012 ( ±1 sdabs , n = 86 ) and the analytical blank yielded 0 . 23 ng Sr . 87Sr/86Sr ratios were performed on the tooth enamel ( crowns ) of individuals HSJNC81 and HSJN240 . Similar analyses were done for HSJNC81 and HSJN240 using the parietal and phalanx bone , respectively . In the case of the two individuals analyzed in this study , bone 87Sr/86Sr values 0 . 70672 ( HSJNC81 ) and 0 . 70755 ( HSJN240 ) ( Table S6 ) are comparable to those obtained from soil samples from the eastern TMVB rim in Veracruz with a mean 87Sr/86Sr of 0 . 70703 ( n = 6 ) ( Solís-Pichardo et al . , 2017 ) . For West African igneous and metamorphic rocks , a mean value 87Sr/86Sr of 0 . 71044 was obtained ( n = 20 , Figure 4—figure supplement 1 ) . Data are compiled in Supplementary file 2 with their corresponding references . Human-mapped reads ( BWA aln ) obtained from the pre-capture sequence data of viral-positive samples were used to infer the genetic ancestry of the hosts using PCA . The genomic alignments ( to hg19 ) of the four ancient individuals ( HSJNC81 , HSJN240 , HSJN194 , and COYC4 ) was intersected with the genotype data of 400 present-day individuals from eight populations ( 50 individuals per population ) in the 1000 Genomes Project ( 1000 Genomes Project Consortium , 2015; IBS: Iberian from Spain; CEU: Utah Residents with Northern and Western European Ancestry; CHB: Han Chinese in Beijing; CHS: Southern Han Chinese , YRI: Yoruba in Ibadan; MSL: Mende in Sierra Leone , MXL: Mexican Ancestry from Los Angeles; and PEL: Peruvians from Lima; Supplementary file 2A ) . Pseudo haploid genotypes were called by randomly selecting one allele at each intersected site , both in the reference panel and in the genomic alignments , and filtering by a base quality >30 in the latter . The merged dataset was processed using PLINK ( Purcell et al . , 2007 ) with the following parameters: a linkage disequilibrium filter ( --indep-pairwise 200 25 0 . 2 ) , genotype missingness filter of 5% ( --geno 0 . 05 ) , and minor allele frequency of 5% ( --maf 0 . 05 ) . This resulted in 904 , 258 SNVs passing the filters . PCA was then performed on with the program smartpca ( EIGENSOFT package ) ( Patterson et al . , 2006; Price et al . , 2006 ) using the option lsqproject to project the ancient individuals into the PC space defined by the modern individuals . A total of 58 , 670 SNPs intersected between the 1000 Genomes Project reference panel and the COYC4 ancient genome ( see previous section for details ) . The program ADMIXTURE ( Alexander and Lange , 2011 ) was run on these intersected data with K values between 2 and 5 , and 100 replicates for each K using a different random seed number . For each K , the ADMIXTURE run with the best likelihood was chosen to be plotted using AncestryPainter ( Feng et al . , 2018 ) . NGS reads were mapped to the human mitochondrial genome reference ( rCRS ) with BWA ( aln algorithm , -l default ) , the alignment file was then used to generate a consensus mitochondrial genome with the program Schmutzi ( Renaud et al . , 2015 ) The assignment of the mitochondrial haplogroup was carried out with Haplogrep ( Kloss-Brandstätter et al . , 2011; Weissensteiner et al . , 2016 ) using the consensus sequence as the input . Assignment of biological sex was inferred based on the number of reads mapped to the Y-chromosome ( Ry ) relative to those mapping to the Y and X-chromosome ( Skoglund et al . , 2013 ) . Ry <0 . 016 and Ry >0 . 075 were considered XX or XY genotype , respectively . The resulting XY sex was coherent with the one inferred morphologically ( Supplementary file 2A ) .	The arrival of European colonists to the Americas , beginning in the 15th century , contributed to the spread of new viruses amongst Indigenous people . This led to massive outbreaks of disease , and millions of deaths that caused an important Native population to collapse . The exact viruses that caused these outbreaks are unknown , but smallpox , measles , and mumps are all suspected . During these times , traders and colonists forcibly enslaved and displaced millions of people mainly from the West Coast of Africa to the Americas . The cruel , unsanitary , and overcrowded conditions on ships transporting these people across the Atlantic contributed to the spread of infectious diseases onboard . Once on land , infectious diseases spread quickly , partly due to the poor conditions that enslaved and ndigenous people were made to endure . Native people were also immunologically naïve to the newly introduced pathogens , making them susceptible to severe or fatal outcomes . The new field of paleovirology may help scientists identify the viruses that were circulating in the first years of colonization and trace how viruses arrived in the Americas . Using next-generation DNA sequencing and other cutting-edge techniques , Guzmán-Solís et al . extracted and enriched viral DNA from skeletal remains dating back to the 16th century . These remains were found in mass graves that were used to bury epidemic victims at a colonial hospital and chapel in what is now Mexico City . The experiments identified two viruses , human parvovirus B19 and a human hepatitis B virus . These viral genomes were recovered from human remains of first-generation African people in Mexico , as well as an individual who was an Indigenous person . Although the genetic material of these ancient viruses resembled pathogens that originated in Africa , the study did not determine if the victims died from these viruses or another cause . On the other hand , the results indicate that viruses frequently found in modern Africa were circulating in the Americas during the slave trade period of Mexico . Finally , the results provide evidence that colonists who forcibly brought African people to the Americas participated in the introduction of viruses to Mexico . This constant influx of viruses from the old world , led to dramatic declines in the populations of Indigenous people in the Americas .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "genetics", "and", "genomics" ]	2021	Ancient viral genomes reveal introduction of human pathogenic viruses into Mexico during the transatlantic slave trade
The process by which visual information is incorporated into the brain’s spatial framework to represent landmarks is poorly understood . Studies in humans and rodents suggest that retrosplenial cortex ( RSC ) plays a key role in these computations . We developed an RSC-dependent behavioral task in which head-fixed mice learned the spatial relationship between visual landmark cues and hidden reward locations . Two-photon imaging revealed that these cues served as dominant reference points for most task-active neurons and anchored the spatial code in RSC . This encoding was more robust after task acquisition . Decoupling the virtual environment from mouse behavior degraded spatial representations and provided evidence that supralinear integration of visual and motor inputs contributes to landmark encoding . V1 axons recorded in RSC were less modulated by task engagement but showed surprisingly similar spatial tuning . Our data indicate that landmark representations in RSC are the result of local integration of visual , motor , and spatial information . Spatial navigation requires the constant integration of sensory information , motor feedback , and prior knowledge of the environment ( Hardcastle et al . , 2015; McNaughton et al . , 2006; Taube , 2007; Valerio and Taube , 2012 ) . Visual landmarks are particularly advantageous for efficient navigation , representing information-rich reference points for self-location and route planning ( Etienne et al . , 1996; Gothard et al . , 1996b; Jeffery , 1998; Knierim et al . , 1995; McNaughton et al . , 1991 ) . Even in situations where the immediate surroundings may not be informative , distal landmarks can provide critical orientation cues to find goal locations ( Morris , 1981; Tolman , 1948 ) . Their importance is further underlined by the fact that salient visuo-spatial cues anchor almost every type of spatially-tuned cells observed in the mammalian brain to date , including head-direction cells ( Jacob et al . , 2017; Taube et al . , 1990; Yoder et al . , 2011 ) , hippocampal place cells ( Buzsáki , 2005; Jeffery , 1998 ) , and grid cells in the medial entorhinal cortex ( Hafting et al . , 2005; Pérez-Escobar et al . , 2016 ) . Even in scenarios where self-motion feedback is in conflict with external visual cues , landmarks exert powerful influence on the head-direction system ( Etienne and Jeffery , 2004; Valerio and Taube , 2012 ) . A number of theoretical studies have shown the importance of landmarks for error correction during spatial computations ( Burgess et al . , 2007; Fuhs and Touretzky , 2006; Monaco et al . , 2011; Sreenivasan and Fiete , 2011 ) . However , it remains poorly understood how visual information is integrated into spatial code for goal-directed behavior . Converging evidence points to the retrosplenial cortex ( RSC ) as an important locus for landmark computations . Studies in humans with damage to RSC , as well as functional imaging studies , suggest a key role for RSC in utilizing familiar visual cues for navigation ( Auger et al . , 2012; Epstein , 2008; Epstein and Vass , 2014; Maguire , 2001; Vann et al . , 2009 ) . Additionally , RSC exhibits some of the earliest measurable metabolic dysfunction in Alzheimer’s disease ( AD ) ( Minoshima et al . , 1997; Pengas et al . , 2010; Villain et al . , 2008 ) . This is consistent with the putative roles of RSC in general mnemonic processing ( Cooper et al . , 2001; Spiers and Maguire , 2006; Svoboda et al . , 2006; Valenstein et al . , 1987 ) and route-planning ( Spiers and Maguire , 2006 ) , both of which are hallmarks of cognitive decline in AD patients ( Vann et al . , 2009 ) . Lesion studies in rodents indicate that RSC is also important for navigating based on self-motion cues alone ( Elduayen and Save , 2014 ) . These findings are congruent with known RSC anatomy: situated at the intersection of areas that encode visual information , motor feedback , higher-order decision making , and the hippocampal formation ( van Groen and Wyss , 1992; Kononenko and Witter , 2012; Miyashita and Rockland , 2007; Sugar et al . , 2011 ) , RSC is ideally positioned to integrate these inputs to guide ongoing behavior . Electrophysiological recordings in freely moving rats have shown that individual RSC neurons conjunctively encode space in egocentric and allocentric spatial reference frames ( Alexander and Nitz , 2015 ) . When placed in a one-dimensional environment , RSC neurons exhibit single , spatially tuned receptive fields ( Mao et al . , 2017 ) , while in two-dimensional environments ( Alexander and Nitz , 2017 ) they were found to express multiple receptive fields . RSC neurons have further been shown to encode context as well as task-related cues such as goal location ( Smith et al . , 2012; Vedder et al . , 2017 ) . Recent studies have focused on understanding multimodal integration ( Minderer et al . , 2019 ) , accumulation of evidence ( Koay et al . , 2019 ) , and how locomotion is differentially represented in RSC and visual cortex ( Clancy et al . , 2019 ) . Finally , a subset of cells in RSC also encode head-direction in a way that is particularly sensitive to local environmental cues ( Jacob et al . , 2017 ) . A common theme across these studies is the importance of visual inputs for RSC function . While the role of proximal , non-visual cues , such as whisker stimulation , has not been thoroughly evaluated , it is clear that visual cues alone are sufficient to guide behavior ( Etienne et al . , 1996 ) . Together , these converging results strongly implicate RSC as an important neural substrate for landmark encoding . We set out to identify how visual cues that inform an animal about a goal location are represented in RSC . We focus in particular on the dysgranular part of RSC ( Brodmann Area 30 ) which is less well characterized compared to granular RSC ( A29 ) , but has been shown to express spatial receptive fields ( Mao et al . , 2017 ) . We developed a task in which animals learn the spatial relationship between a salient visual cue and a hidden rewarded zone on a virtual linear track . The visual cue serves as a landmark indicative of the animal’s distance to a reward . Studies investigating how spatial tuning is influenced by the environment generally use a single orienting cue ( Hafting et al . , 2005; Muller and Kubie , 1987; Taube et al . , 1990 ) , visual cues directly indicating the presence or absence of a reward ( Pakan et al . , 2018; Poort et al . , 2015 ) , or cue-rich environments where understanding the visual surrounding was not required to locate rewards ( Campbell et al . , 2018; Fiser et al . , 2016; Gauthier and Tank , 2018; Harvey et al . , 2009; Saleem et al . , 2018 ) . In contrast , our task requires mice to use allocentric inputs as reference points , and combine them with self-motion feedback to successfully execute trials . This task utilizes landmarks as an indicator of distance to a reward on a linear track , as opposed to as an orientation cue . While A30 does contain head-direction tuned neurons , they are unlikely to contribute to computations in this task . We found the majority of task-active neurons , as well as the population response , to be anchored by landmarks . Showing the same visual stimuli at a static flow speed while animals were not engaged in the task resulted in significantly degraded responses , suggesting that active navigation plays a crucial role in RSC function . Landmark responses were the result of supralinear integration of visual and motor components . To understand how visual information is translated into behaviorally-relevant representations in RSC , we recorded the activity of axons from the primary visual cortex ( V1 ) during task execution . V1 sends strong projections to RSC ( Oh et al . , 2014 ) which , in turn , sends a top-down projection back to V1 ( Makino and Komiyama , 2015 ) , creating a poorly understood cortico-cortical feedback loop between a primary sensory and associative cortex . Understanding this circuit could provide key insights into how sensory and contextual information combine to guide behavior . We found strikingly similar receptive fields as those expressed by RSC neurons , suggesting that V1 inputs may be key in shaping their receptive fields . Importantly , their activity was less modulated by active navigation , illuminating a key difference between primary sensory and associative cortex . We developed a behavioral task that required mice to learn the spatial relationships between visual cues and hidden rewards along a virtual linear corridor ( Figure 1A and B ) . Each trial began at a randomized distance ( 50–150 cm ) from one of two salient visual cues with a vertical or diagonal stripe pattern respectively . Along the rest of the corridor , which appeared infinitely long , a gray-and-black dot uniform pattern provided optic flow feedback but no spatial information . An unmarked 20 cm wide reward zone was located at fixed distances from the visual landmarks ( starting at 80 or 140 cm from the end of the landmark , respectively ) . At the end of each trial , the animal was ‘teleported’ into a ‘black box’ ( black screens ) for at least 3 s . A trial ended when an animal either triggered a reward by licking within the reward zone or received a ‘default reward’ when it passed the reward zone without licking . Default rewards were provided throughout the experiment but constituted only a small fraction of trials in trained animals ( mean ± SEM: 14 . 24 ± 2 . 65% of trials , n = 12 sessions , 11 mice ) . The 3 s black box timeout was included in the task to give the animals a salient signal for the end and start of trials . It further ensured that GCaMP6f signals underlying different aspects of behavior ( reward delivery and consumption versus the initiation of a new trial and changes in locomotion ) could decay before the start of the next trial . Mice learned to use the visual cues to locate the reward zones along the corridor ( Figure 1C , mean 32 . 3 ± 3 . 6 training sessions for n = 10 mice ) . In theory , animals could achieve a high fraction of successful trials by licking frequently but randomly , or in a uniform pattern . We tested if we could use licking as a behavioral assay for an animal’s understanding of the spatial relationship between visual cue and reward location by calculating a spatial modulation index using a bootstrap shuffle test ( Figure 1—figure supplement 1E , F and Materials and methods ) . This test randomly shifted licking locations relative to the track location for each trial and evaluated whether the animal would have still triggered a reward based on randomized licking ( at least one lick inside the reward zone ) . The overall fraction of successful trials is then calculated for the entire session using randomized licking locations . This process is repeated 1000 times . Finally , the z-score of the actual success rate relative to the shuffled distribution is calculated and checked whether it was significantly higher than the shuffled distribution . Licking behavior of expert animals was significantly spatially modulated ( mean ± SEM: 13 . 48 ± 1 . 23 , n = 12 session , 11 mice ) , making it an accurate behavioral assay . We quantified an animal’s ability to use landmarks for navigation by calculating the difference between the median location of first licks on short and long trials , expressed as a ‘task score’ ( mean ± SEM of n = 12 recording sessions: 34 . 2 ± 5 . 47 cm ) . Location of first licks as opposed to mean lick location or lick frequency was used as it provided the most conservative measure of where animals anticipated rewards . This task structure inherently minimizes the ability of mice using alternative strategies such as time or an internal odometer to locate rewards . We tested whether animals used the start of the trial and a fixed distance , time , or number of steps before they started probing for rewards ( Figure 1E ) . For each trial in one session , we plotted the location of the start vs . the location of the first lick and evaluated the linear regression coefficient showing no dependence of first lick on trial start location ( mean ± SEM of slope: 0 . 13 ± 0 . 043W , n = 12 sessions ) . To test if RSC was involved in task performance , we suppressed RSC activity by selectively activating inhibitory interneurons in expert animals during the task . We bilaterally injected a viral vector containing Cre-dependent channelrhodopsin-2 ( ChR2 ) in multiple locations along the anterior-posterior axis of RSC ( 2–3 injections per hemisphere , Figure 1—figure supplement 1H ) in VGAT-Cre mice . This restricted ChR2 expression to GABAergic neurons in RSC and allowed us to rapidly and reversibly inhibit local neural activity ( Lewis et al . , 2015; Liu et al . , 2014 ) . Ferrules were implanted on the surface of the skull over RSC to deliver light during behavior . Stimulation was delivered by a 470 nm fiber-coupled LED on a randomized 50% subset of trials . The stimulation light was turned on at the beginning of a given trial and lasted until the end of the trial or the maximum pulse duration of 10 s was reached . A masking light was shown throughout the session . Task scores on trials with stimulation was significantly lower compared to trials where only the masking light was shown within the same session ( 43 . 7 ± 7 . 4 cm vs . 24 . 3 ± 6 . 2 cm , n = 5 mice , paired t-test: p = 0 . 003 , Figure 1F ) , indicating that RSC activity contributes to successful execution of this behavior . The fraction of successful trials , in contrast , was not significantly different in the mask only and stim condition ( mask: 7 . 5 ± 2 . 0% , stim: 6 . 7 ± 1 . 9% , paired t-test: p=0 . 61 ) , showing that the decrease in task score was not a result of a diminished ability to trigger rewards . We sought to understand which task features were represented by neurons in dysgranular RSC ( A30 ) . Mice injected with AAV expressing the genetically encoded calcium indicator GCaMP6f were trained until they reliably used landmarks to locate rewards . On average , we recorded 120 . 0 ± 17 . 56 RSC neurons per mouse ( n = 7 mice ) . GCaMP signals of all active neurons ( >0 . 5 transients/min , n = 966 ) were tested for significant peaks of their mean response above a shuffled distribution ( z-score of mean trace >3 , see Materials and methods ) and for transients on at least 25% of trials . The calcium traces of individual neurons that met these criteria ( n = 491 ) were grouped by trial type and aligned to each of three points: trial onset , landmark , and reward ( Figure 2A , right; Figure 2B ) . The peak activity of the mean GCaMP trace of each neuron was then compared across alignment points and classified based on which task feature resulted in the largest mean response ( Figure 2B ) . This analysis was carried out for short trials and long trials independently . The majority of RSC neurons found to be task engaged were aligned to the visual landmark ( Figure 2C , n = 55 short trial onset and 62 long trial onset , 206 and 235 landmark , 101 and 118 reward neurons , respectively; seven mice; mean ± SEM fractions of aligned neurons: trial onset: 5 . 6 ± 0 . 3% , landmark: 22 . 3 ± 1 . 6% , reward: 11 . 6 ± 0 . 7%; one way ANOVAshort: p = 0 . 0023; ANOVAlong: p<0 . 001 , Tukey’s HSD post-hoc pairwise comparison with Bonferroni correction ) . A smaller but sizeable fraction of RSC neurons were aligned to the reward point , suggesting that RSC also encodes behavioral goals as well , and the onset of a trial , regardless of where an animal was placed on the track relative to the visual landmark . Consistent with previous findings ( Alexander and Nitz , 2015 ) , these data indicate that egocentric ( trial onset ) as well as allocentric ( landmark and reward ) variables are encoded in RSC during landmark-based navigation . The vast majority of neurons showed a single peak of activity in our task . Previous studies have found neurons with multiple peaks or sustained firing in RSC neurons ( Alexander and Nitz , 2015; Alexander and Nitz , 2017 ) . Our task , however , did not contain repeating sections found in a W-shaped or plus-shaped maze which may explain the discrepancy in our findings . The peak response of a given neuron did not necessarily need to happen directly at a given alignment point , but could also occur at some distance from it . A landmark-aligned neuron , for example , did not have to exhibit its peak response at the landmark . Instead , it could be active close to the reward zone but its calcium trace was still best aligned to the landmark ( Figure 2F ) . Surprisingly , we found no significant difference in transient amplitude or robustness ( probability of a transient ) between trials where mice operantly triggered rewards themselves compared to ‘unsuccessful’ trials , in which they received the default reward across all neuron types ( mean amplitude on successful trials: 1 . 79 ± 0 . 09 ∆F/F , unsuccessful trials: 1 . 69 ± 0 . 09 ∆F/F , n = 241 neurons , paired t-test: p=0 . 9 . Mean robustness on successful trials: 0 . 61 ± 0 . 02 , unsuccessful trials: 0 . 61 ± 0 . 09 , n = 210 neurons , paired t-test: p=0 . 07 . This analysis only included neurons in sessions with successful and unsuccessful trials on short or long track ) . However , we found neurons that were differentially active within a session when we split the responses into the most and least accurate 25% of trials . A subset ( 21 . 6% ) changed their activity by >0 . 5 ∆F/F . These changes were bidirectional: some neurons increased their activity while others decreased their activity based on how well an animal predicted the location of a reward , measured as the distance of the first lick in a given trial to the start of the reward zone ( Figure 4—figure supplement 1C , D ) . We employed a template matching decoder ( Montijn et al . , 2014 ) to analyze how well trial type could be decoded from neural activity alone ( Figure 2E ) . While trial onset or reward-aligned neurons provided only chance level decoding or slightly better ( trial onset: median 53 . 44% correct; reward: 62 . 1% ) , trial type decoding by landmark neurons was significantly higher ( 82 . 56% , Kruskal-Wallis test p=0 . 015; post-hoc Mann-Whitney U pairwise testing with Bonferroni correction for multiple comparisons ) . We examined how individual task features ( landmark , trial onset , and rewards ) are differentially represented in RSC layers 2/3 ( L2/3 ) and layer 5 ( L5 ) . Cortical layers were identified by their depth under the dura ( L2/3: 130 . 0 ± 4 . 0 µm , L5: 327 . 0 ± 15 . 2 ) , and confirmed post-hoc with histological sections in a subset of animals ( Figure 2—figure supplement 1A , Figure 2—figure supplement 2A ) . We found that superficial as well as deep layers contained trial onset , landmark , and reward neurons . However , L5 contained substantially fewer landmark neurons ( Figure 2—figure supplement 1B , C; One-way ANOVA , p<10–8 , Tukey HSD post-hoc test with Bonferroni correction ) . Finally , we asked whether the subpopulation of landmark encoding neurons showed a preference for visual cue identity . We calculated a landmark modulation index as the difference between peak activity divided by the sum of their activity [LMI = ( LMshort – LMlong ) / ( LMshort + LMlong ) ] . Peak activity for each trial type was calculated separately . Only a small number of neurons were found to be tuned to landmark identity ( Figure 2H ) , with most neurons showing no specific preference . Similarly , trial onset neurons and reward neurons did not show trial type selectivity ( Figure 3—figure supplement 1E ) . These results indicate that neurons in RSC encode a mix of task variables with a strong preference for visual cues informing the animal about goal locations . In this task , trial onset and visual landmarks provide egocentric and allocentric context , respectively . We tested which reference point anchored the neural representation of the animal’s location using population vector cross-correlation analyses for all task active neurons ( n = 491 , Gothard et al . , 1996a; Alexander and Nitz , 2015; Alexander and Nitz , 2017; Mao et al . , 2017 ) . Two activity vectors were constructed for each neuron by randomly taking data from one half of the trials for the first vector and the other half for the second vector . Data from the first vector was used to determine the location of a neuron’s maximum response while the second vector was used for correlation analysis . This process prevents introduction of artifactual spatial structure into the population code . Activity was binned into 5 cm wide bins and the mean across all included trials was calculated and normalized to 1 . We found largely even tiling of space from the trial onset until reward ( Figure 3A and D ) . To test the spatial specificity of the population code , we calculated a population vector cross-correlation matrix using the Pearson cross-correlation coefficient ( Figure 3B and E ) for each spatial bin . To ensure that randomly splitting data into halves didn’t lead to spurious results , we calculated the mean of 10 cross-correlation maps , each randomly drawing a different subset of trials . Slices of the cross-correlation matrix ( Figure 3C and F , taken at the dashed lines indicated in 3B and 3E ) , reveal that the spatial code is sharpest at the landmark . The cross-correlation at the animal’s true location ( i . e . along the diagonal from top left to bottom right ) significantly increases as the animal approaches the landmark and remains elevated until it reaches the reward ( Figure 3G ) . We tested how well we could reconstruct the animal’s location from the neural code by calculating how far the pixel with the highest cross-correlation was from the actual location for each row in the cross-correlation matrix . We observed a significantly lower location reconstruction error when neural activity was aligned to landmarks , rather than trial onset ( Figure 3H , mean ± SEM: 3 . 7 ± 0 . 61 vs . 5 . 27 ± 0 . 56 short trials; 3 . 23 ± 0 . 48 vs . 5 . 26 ± 0 . 42 , unpaired , 2-tailed t-test: p<0 . 016 ( short ) , p<0 . 001 ( long ) ) . Finally , we found no significant difference in reconstruction error between short track and long track trials , as most neurons are active on both trials ( Figure 3—figure supplement 1 ) . These results provide evidence for a spatial code in RSC that is strongly modulated by environmental cues to inform the animal about the location of its goal . To determine if goal-directed navigation , as opposed to visual input alone , was required to explain RSC activity , we recorded neurons while animals were shown the same virtual corridor without actively executing the task . During this decoupled stimulus presentation paradigm ( DC ) , the virtual corridor moved past the animals at two speeds: 10 cm/sec and 30 cm/sec ( Figure 4A , see Figure 1—figure supplement 1B , D for speed profiles during virtual navigation ) . Both trial types were interleaved in the pattern as during virtual navigation , but no rewards were dispensed when animals reached reward locations . We found a significant decrease in neuronal responses in this condition ( Figure 4B and C , mean ± SEM VR: 0 . 17 ± 0 . 005 , mean decoupled: 0 . 059 ± 0 . 004 , paired , two-tailed t-test: p<0 . 001 ) . This was true for neurons of all three categories: trial onset , landmark , and reward ( Figure 4D , median values VR: trial onset = 0 . 1 , landmark = 0 . 19 , reward: 0 . 13; decoupled: trial onset = 0 . 03 , landmark = 0 . 04 , reward = 0 . 02 ) . This result suggests that activity in RSC is strongly dependent on active task engagement . Congruent with this , population activity showed significantly less spatial specificity during decoupled stimulus presentation ( Figure 4E–H , mean reconstruction error ± SEM: 3 . 7 ± 0 . 61 vs . 7 . 88 ± 0 . 78 short trials; 3 . 23 ± 0 . 48 vs . 11 . 16 ± 1 . 18 , unpaired , Mann-Whitney U: p<0 . 001 ( short ) , p<0 . 001 ( long ) ) . This indicates that encoding of behaviorally-relevant variables in RSC is modulated by ongoing behavior , rather than being driven solely by sensory inputs . We note that not providing a reward and decoupling the stimulus presentation from animal locomotion constitute simultaneous changes that may both influence neural activity . However , if reward anticipation was a key driver in the change in neural activity we would expect neurons anchored by trial onset or landmark to be less affected than reward driven neurons . We find that all neuron types are similarly affected ( Figure 4D ) , suggesting that reward anticipation is not the major cause for the change in activity we observed . A second potential factor modulating neuronal responses is whether the animal is attending to the cue or not . We have addressed this issue by analyzing responses during quiet wakefulness and locomotion in the next section ( Figure 5 ) . Finally , attending to the stimuli and/or task may significantly impact the respective neural representation . We therefore recorded pupil dilation in three well trained animals . However , we found no significant difference in pupil dilation during active navigation and decoupled stimulus presentation ( Figure 4—figure supplement 1F–H ) , suggesting that attention is an unlikely explanation for our results . We sought to gain insight into potential mechanisms underlying the changes in neuronal activity during decoupled stimulus presentation by comparing GCaMP6f signals observed in both conditions . Individual events were detected when ∆F/F exceeded six standard deviations of a neuron’s baseline activity for at least two spatial bins ( bin size: 2 cm ) and lay within ±60 cm of the peak mean response ( Figure 4I ) . We found that the standard error of the distance of individual transients to the peak of the mean trace ( Figure 4J , median jitter ( cm ) : shortVR = 4 . 67 , shortDC = 8 . 66; longVR = 5 . 75 , longDC = 9 . 85 ) was lower during virtual navigation compared to decoupled stimulus presentation . In other words , transients were more tightly clustered around that neuron’s peak response when the animal was actively engaged in the task . Furthermore , we saw a significant reduction in the number of transients per trial during decoupled stimulus presentation ( Figure 4K , median values ( transients/trial ) : shortVR = 0 . 45 , shortDC = 0 . 14; longVR = 0 . 45 , longDC = 0 . 14 ) , but only very little change in the amplitude of individual transients ( Figure 4L , median values ( ∆F/F ) : shortVR = 1 . 48 , shortDC = 1 . 36; longVR = 1 . 4 , longDC = 1 . 33 , Kruskal-Wallis test p<0 . 001; Mann-Whitney-U pairwise comparisons with Bonferroni correction results indicated , ***=p < 0 . 001 ) . These results show that the changes during decoupled stimulus presentation is due to poorer spatial anchoring of activity and fewer instances of a given neuron to exhibit a transient . Nonlinear integration of synaptic inputs dramatically enhances the computational power of individual neurons and neural networks ( Miller and Cohen , 2001; London and Häusser , 2005; Mante et al . , 2013; Rigotti et al . , 2013; Jadi et al . , 2014; Ranganathan et al . , 2018 ) . For a single neuron , the integration of multiple input streams may engage mechanisms of supralinear integration to produce complex , conjunctive responses ( Bittner et al . , 2015; Larkum et al . , 1999; Takahashi et al . , 2016; Xu et al . , 2012 ) . In contrast , neural networks may express conjunctive representations through high dimensional codes ( Murray et al . , 2017; Rigotti et al . , 2013; Stringer et al . , 2019 ) . We therefore evaluated the evidence for nonlinear integration in landmark-anchored RSC neurons . During decoupled stimulus presentation , mice were free to spontaneously locomote on the treadmill or watch passively . Trials within a session were separated based on whether the animal locomoted as the virtual environment passed a neuron’s receptive field ( Figure 5A and B , ‘+ motor’ trials , running speed >3 cm/sec in a ± 50 cm window ) . ‘No input’ and ‘motor only’ conditions were measured while the animal was in the black box in between trials using similar criteria as before but with a 1 . 5 s time window ( instead of a spatial window ) for locomotion . Thus , any brief spontaneous movements that may have occurred within that window , but were below threshold , were labeled passive viewing . Despite the possibility that such small movements impact neural responses , we find a striking contrast between locomoting and passively viewing animals . When landmark presentation occurred during locomotion , activity was significantly increased ( Figure 5C , n = 127 neurons , five mice , mean ± SEM number of trials/neuron: 13 . 0 ± 0 . 41 resting , 9 . 3 ± 0 . 3 running; peak mean ∆F/F ± SEM: VR: 0 . 93 ± 0 . 06 , landmark + motor: 0 . 57 ± 0 . 06 , landmark , no motor: 0 . 22 ± 0 . 04 , Kruskal-Wallis test: p<0 . 001 , Mann-Whitney U post-hoc comparison with Bonferroni correction: p<0 . 001 for all shown comparisons ) . However , we found that visual inputs alone or visual inputs plus locomotion did not elicit the same response as virtual navigation ( Figure 5C , D , Kruskal-Wallis test: p<0 . 001; post-hoc Mann-Whitney-U and Bonferroni correction: p<0 . 001 for all shown comparisons ) . We then evaluated the responses while animals were locomoting or stationary while in the ‘black box’ between trials to obtain estimates of population activity in ‘no input’ ( neither visual nor motor inputs ) and ‘motor only’ conditions ( Figure 5D , ∆F/F ± SEM: black box + motor = 0 . 16 ± 0 . 03 , black box , no motor = −0 . 04 ± 0 . 03 ) . Finally we asked whether the linear sum of ‘motor only’ and ‘landmark , no motor’ added up to ‘landmark + motor’ ( Figure 5D and E ) . We found that the linear sum approached , but remained lower than the mean amplitude recorded during ‘landmark + motor’ . Analysis of GCaMP transient patterns during ‘no motor’ and ‘+ motor’ conditions revealed that neurons show significantly more transients while locomoting ( Figure 5H , median values ( transients/trial ) : no motor = 0 . 06 , + motor = 0 . 33; VR = 0 . 42 ) , however , transient amplitude and jitter were broadly similar ( Figure 5G and I , median jitter ( cm ) : no motor = 13 . 82 , + motor = 10 . 7; VR = 5 . 69; median amplitude ( ∆F/F ) : no motor = 1 . 59 , + motor = 1 . 89; VR = 2 . 28 ) . Our results suggest that motor input drives RSC neurons , however it does not aid in anchoring their activity or modulating the number of spikes produced once it has reached firing threshold . We note that GCaMP6f may not provide linear translation from underlying spikes to fluorescence signal . However , our analysis focuses on relative differences within the same neuron under different conditions and thus nonlinearities of the calcium indicator are unlikely explain these results . Together these results provide evidence for substantial nonlinear integration of visual and motor inputs in RSC neurons during goal-directed virtual navigation as well as decreased , but still significant , nonlinear processing during decoupled stimulus presentation . A possible explanation for this result is a correlation of between running speed and transient amplitude . To test this , we analyzed the modulation of transient amplitude by running speed ( Figure 4—figure supplement 1A , B ) . We found only a small number of neurons showing modulation of transient amplitude by running speed across the population of task-active neurons ( Figure 4—figure supplements 1B , 0 . 5% on short trials n = 29 neurons and 15 . 6% on long trials n = 51 neurons ) . Of these neurons , 22 were positively and 58 negatively correlated . As we find changes in the vast majority of landmark-anchored neurons ( Figure 5C ) , it is unlikely that running speed modulation explains these results . During decoupled stimulus presentation , locomotion no longer influences the virtual environment or reward availability , which may change animals’ internal state . We therefore performed experiments imaging the same neurons before and after learning in a separate group of 3 mice ( mean ± SEM task score on expert session: 47 . 21 ± 2 . 6 ) ; animals were required to perform the same behavior , with the same stimulus and reward contingencies , in both conditions . We found some neurons that previously had not expressed discernible spatially tuned activity , establishing spatial receptive fields , while showed amplified responses ( Figure 6A and B ) . When we identified task active neurons in expert animals and calculated the population vector cross correlation during naïve and expert sessions , we found that the encoding of space in the naïve animal was much degraded ( Figure 6C–E ) . However , this result could be explained by different subsets of neurons being active in naïve and expert animals , as opposed to a robust spatial code emerging during learning . To test this , we identified task active neurons in naïve and expert sessions independently and calculated the population vector cross correlation . Neurons in the naïve animal showed significantly worse representation of the animal’s location ( Figure 6E ) . The observed neural activity in expert animals is thus the result of a code that develops in RSC as animals learn to associate the visual cue with a reward location , turning it into a spatial landmark . Finally , we sought to dissect how dysgranular RSC produces landmark representations by identifying what information it receives from primary visual cortex ( V1 ) , a major input source to RSC ( Vogt and Miller , 1983 ) . To this end , we injected GCaMP6f into V1 in a separate group of trained animals and recorded the responses of axonal boutons in RSC ( Figure 7A and B ) . Use of a passive pulse splitter ( Ji et al . , 2008 ) in the laser path allowed us to image axons continuously during self-paced behavioral sessions with no photobleaching or toxicity . To prevent overrepresentation of axons with multiple boutons in a given FOV , highly cross-correlated boutons were collapsed and represented as a single data point ( see Materials and methods and Figure 7—figure supplement 1 ) . In a separately injected and trained group of 4 animals we found a total of 77 unique , task-active axons . Unexpectedly , we found receptive fields that were strikingly similar to those we observed in RSC neurons ( Figure 7C and D ) . Boutons also tiled space along the virtual linear track in a parallel manner to RSC neurons ( Figure 7E and F ) . Furthermore , we found a similar preference of V1 boutons to be anchored by landmarks ( Figure 7G ) . However , when we quantified how active task engagement modulates activity in RSC neurons versus V1 boutons , we found that the former were significantly more modulated compared to the latter ( Figure 7H and I , Mann-Whitney U: p<0 . 001 ) . These results are consistent with a number of recent studies that describe the encoding of non-visual stimuli in V1 ( Ji and Wilson , 2007; Niell and Stryker , 2010; Pakan et al . , 2018; Poort et al . , 2015; Saleem et al . , 2018 ) . The specificity of these responses suggests that at least a subpopulation of RSC-projecting neurons in V1 is tuned to behaviorally-relevant visual cues to a previously unknown extent . Despite their specificity , however , they represent visual inputs more faithfully and are less modulated by context than RSC neurons themselves , pointing to the local computations performed in RSC . In this study , we introduce a novel behavioral paradigm in which mice learned the spatial relationships between salient environmental cues and goal locations ( Figure 1 ) . The task required animals to discriminate visual cues , use them to localize themselves in space , and navigate to a rewarded zone based on self-motion feedback . Using this paradigm , we found that landmarks anchored the majority of task-active neurons in dysgranular RSC ( Figure 2C and D ) and significantly sharpened the representation of the animal’s current location in the population code ( Figure 3 ) . This spatial representation largely changed 2during learning ( Figure 6 ) . Landmark responses were not the result of simple visual and/or motor drive: showing the same visual stimuli while the animals were not engaged in the task elicited significantly attenuated responses ( Figure 4 ) . Further dissection of neuronal activity provided evidence for supralinear integration of visual and motor information in RSC . Coinciding visual input and motor feedback during decoupled stimulus presentation did not elicit the same response amplitudes as observed during active navigation ( Figure 5 ) . Interestingly , receptive fields expressed by V1 axonal boutons in animals executing the same behavior were strikingly similar to those recorded from RSC neurons ( Figure 7 ) . However , they were less modulated by active task engagement ( Figure 7H and I ) , indicating a hierarchy of sequential processing . A major challenge in understanding how computations in RSC contribute to behavior is the diversity and complexity of functions attributed to this area ( Maguire , 2001; Vann et al . , 2009 ) . Studies in humans suggest that RSC is key for utilizing environmental cues during navigation ( Epstein , 2008; Ino et al . , 2007; Maguire , 2001 ) , while experiments in rodents found deficits in path integration ( navigation based on self-motion cues ) when RSC was lesioned or inactivated ( Cooper et al . , 2001; Cooper and Mizumori , 1999; Elduayen and Save , 2014 ) . We describe a task that combines both of these navigational strategies: animals are required to use visual landmarks for self-localization , followed by path integration to successfully find rewards . Using optogenetic inactivation on a randomized subset of trials we found a deficit in the animal’s ability to use landmarks to guide localizing rewards ( Figure 1F ) . This could potentially be explained by a pure path integration deficit . However , if this is the case , RSC should exhibit a purely ego-centric representation of space , that is aligned to the start of a given trial . In contrast , we find that spatial representations in RSC are anchored by allocentric ( landmark ) cues and maintained by self-motion feedback after the animal has passed the visual cue ( Figure 3 ) . We note that in our task , trial onset is not an optimal reference point to test whether RSC acts as an internal odometer , represents sensory inputs , or , as we argue , integrates both . We found a degradation as a function of distance from the onset of the task . However , the onset of a trial is a poor reference point due to the randomized location at which the animal is placed on the track ( 50–150 cm before the landmark ) . Significantly more neurons are anchored by the landmark , compared to the trial onset . Therefore , what may look like a neural code that gets noisier as a function of distance from the trial start , may indeed only reflect the fact that landmarks are at varying distances from the trial start in a given session . Therefore , the landmark is a better reference point to test the accumulation of noise is therefore the landmark . In Figure 3G , we show that population vector cross correlation decreases slowly after the landmark with an uptick shortly before the animal gets to the reward zone . While this is not an optimal measure , and indeed this task was not designed to test error accumulation , we believe that this is evidence for error accumulation in VR . Numerous experimental and theoretical studies have emphasized the importance of landmarks for anchoring spatially tuned cells during navigation ( Burak and Fiete , 2009; Campbell et al . , 2018; Funamizu et al . , 2016; Gothard et al . , 1996a; Harvey et al . , 2012; Jeffery , 1998; Knierim et al . , 1998; Muller and Kubie , 1987 ) , yet the mechanisms that combine inputs from different modalities to represent landmarks remain poorly understood . We found that simple linear summation of visual and motor inputs was insufficient to explain landmark encoding in RSC . Instead , a nonlinear mechanism ( or multiple mechanisms ) underlie the integration of these variables to produce robust visuo-spatial responses during navigation . Active navigation ( in virtual reality ) sharpened spatial tuning and increased robustness of encoding in RSC compared to viewing the movie without being engaged in the task . Interestingly , the amplitude of recorded transients was unchanged , suggesting the presence of a thresholding process in the circuit ( Figure 4I–L ) . Locomotion during decoupled stimulus presentation significantly increased the robustness of encoding while having very little effect on the fidelity of spatial tuning or transient amplitude . This indicates that motor input broadly pushes neurons towards spiking but does not contribute to its spatial tuning ( Figure 5F–I ) . Furthermore , our data show that visual input alone is insufficient to explain the fidelity of spatial tuning we observed during virtual navigation ( Figure 5D ) . Our results indicate that the most likely mechanism underlying supralinear integration in RSC is the multiplicative effects of significantly improved fidelity of spatial tuning and increased likelihood of emitting transients within a neurons receptive field . While the latter seems to be mediated by motor inputs , the nature and source of an anchoring signal is unclear but may originate in the hippocampal formation where landmarks have been found to sharpen spatial tuning of neurons ( Campbell et al . , 2018; Gothard et al . , 1996a; Knierim et al . , 1995 ) . Animals developed a more robust code for space during task acquisition ( Figure 6E and F ) , suggesting that this supralinear mechanism may be the result of learning the spatial significance of a visual landmark cue . Understanding how learning shapes the integrative properties of individual neurons is an exciting avenue for future studies . The spatial tuning we observed in task active neurons in dysgranular RSC appears similar to those of place cells ( see Figure 2B and Figure 3A and D ) . This is consistent with findings in Mao et al . ( 2017 ) , who report that spatial tuning in RSC is somewhat degraded when tactile or visual cues are removed from a belted treadmill . However , RSC may not exhibit spatial representations that differ from CA1 when sensory information regarding goals is absent . Consistent with this , we find that RSC is strongly biased to encode behaviorally relevant visual cues that inform the animal about the location of a reward . This robust code only emerges after learning the spatial significance of the visual cues ( Figure 6 ) . These findings are complementary to previous studies showing that RSC conjunctively encodes information in egocentric and allocentric reference frames ( Alexander and Nitz , 2015; Alexander and Nitz , 2017 ) as well as other variables ( Smith et al . , 2012; Vedder et al . , 2017 ) . RSC’s bias to encode behaviorally relevant stimuli is particularly interesting in light of its relationship with axonal inputs from V1 ( Figure 7 ) . Using the same landmark-dependent navigation task , we found that V1 axons exhibited comparable receptive field tunings as RSC neurons . However , these responses were substantially less modulated by task engagement ( Figure 7I ) , suggesting that V1 axons encode visual features more faithfully . Previous studies have shown that neurons in V1 are themselves locomotion modulated ( Niell and Stryker , 2010; Saleem et al . , 2013 ) . We report less locomotion modulation in V1 compared to RSC . The exact nature of this difference may be either of a qualitative nature , in which individual neurons are impacted differently by locomotion , or quantitative , in which across the population fewer neurons are locomotion modulated in V1 compared to RSC . We cannot disambiguate between these possibilities , as the overall fraction of locomotion-modulated neurons in V1 has not yet been established . Furthermore , it is not clear if only a specific functional subset projects to RSC , which may be more or less locomotion-modulated relative to the rest of V1 . The modulation we did observe may be the result of strong top-down inputs from RSC itself ( Makino and Komiyama , 2015 ) or from other regions ( Zhang et al . , 2014 ) . This is congruent with a recent study showing that activity in RSC is more correlated with V1 during locomotion compared to quiescent periods ( Clancy et al . , 2019 ) . Indeed , before learning the behavioral significance of visual features in a novel environment , RSC may initially receive purely visual inputs from V1 . As the animal learns to navigate in the new environment , feedback from RSC to V1 ( as well as other areas such as ACC ) may lead to modulated responses based on behavioral relevance in primary visual cortex , as reported by an increasing number of studies ( Attinger et al . , 2017; Pakan et al . , 2018; Poort et al . , 2015; Saleem et al . , 2018 ) . Our data provide evidence that RSC may act as a critical processing node that gates behaviorally relevant visual inputs and relays them to the entorhinal cortex and other areas involved in spatial navigation , where its readout may be used to anchor spatially tuned neurons such as grid cells ( Burak and Fiete , 2009; Campbell et al . , 2018 ) or head-direction cells ( Jacob et al . , 2017 ) . While RSC receives inputs from V2 , M2 , and other cortical areas ( Oh et al . , 2014; Sugar et al . , 2011 ) , functional imaging studies show RSC to be uniquely engaged during landmark-based navigation ( Auger et al . , 2012; Epstein , 2008; Epstein and Vass , 2014; Maguire , 2001; Vann et al . , 2009 ) , suggesting RSC is indeed a key locus for integrating visual and spatial information compared to other association areas . Finally , this work provides novel insights into the neural mechanisms underlying cognitive computations . Our results are consistent with data from humans with RSC lesions who show an impaired ability to use environmental cues for navigation , as well as neuroimaging studies that show increased activity in RSC during spatial behaviors ( Cho and Sharp , 2001; Ino et al . , 2007; Julian et al . , 2018; Maguire , 2001; Robertson et al . , 2016; Vann et al . , 2009 ) . Leveraging RSC to unravel how multiple input streams are integrated during higher level associative processes like navigation may in the future provide novel insights into the mechanisms of cognition and its dysfunction in Alzheimer’s disease and other currently intractable brain disorders . All animal procedures were carried out in accordance with NIH and Massachusetts Institute of Technology Committee on Animal care guidelines . Male and female mice were singly housed on a 12/12 hr ( lights on at 7 am ) cycle . C57BL/6 mice ( RRID: IMSR_JAX:000664 ) were implanted with a cranial window and headpost at 7–10 weeks of age . First , the dorsal surface of the skull was exposed and cleaned of residual connective tissue . This was followed by a 3 mm wide round craniotomy centered approximately 2 . 5 mm caudal of the bregma . To minimize bleeding , particularly from the central sinus , the skull was thinned along the midline until it could be removed in two pieces . AAV1 . Syn . GCaMP6f . WPRE . SV40 was injected at 2–6 injection sites , 350–600 µm lateral of the midline in boluses of 50–100 nl at a slow injection rate ( max . 50 nl/min ) to prevent tissue damage . Following injections , a cranial window was placed over the craniotomy and fixed with cyanoacrylate glue ( Krazy Glue , High Point , NC , USA ) . The windows consisted of two 3 mm diameter windows and one 5 mm diameter window stacked on top of each other ( Warner instruments CS-3R and CS-5R , Hamden , CT , USA ) . The windows were glued together with optical glue ( Norland Optical Adhesive #71 , Edmund Optics , Barrington , NJ , USA ) . Cranial windows consisted of 3 ( instead of 2 ) stacked windows to account for increased bone thickness around the midline and minimize brain motion during behavior . Subsequently , the headplate was attached using cyanoacrylate glue and Meatbond ( Parkell Inc NY , USA ) mixed with black ink to avoid light leaking into the objective during recordings . Mice prepared for imaging of V1 boutons in RSC had GCaMP6f injected into V1 ( ~2 . 49 mm lateral , 3 . 57 caudal ) through small burr holes at a depth of 600 µm to target primarily layer 5 neurons . For the rest of the procedure , the same steps as for imaging of RSC neurons were followed . For the optogenetic inactivation during behavior experiment , VGAT-Ires-Cre knock-in mice ( VGAT is encoded by Slc32a1 , RRID: IMSR_JAX:028862 ) on a C57BL/6 background ( The Jackson Laboratory ) were injected with flexed channelrhodopsin-2 ( ChR2 , AAV5 . ef1a . DIO . ChR2 . eYFP , University of Pennsylvania Vector Core ) in 2–3 locations along the AP axis of RSC ( 50–100 nl per injection ) . Prior to injection the location of the central sinus was identified by placing saline on the skull and waiting until it was translucent . This was done because the overlying sagittal suture can be inaccurate in identifying the midline of the brain . One ferrule was placed centrally on each hemisphere over RSC . Each ferrule was calibrated prior to implantation to ensure the same light intensity was provided into each hemisphere . Head-fixed mice were trained to run down a virtual linear corridor by locomoting on a polystyrene cylinder measuring 8 cm in width and 20 cm in diameter ( Graham Sweet Studios , Cardiff , UK ) . The cylinder was attached to a stainless-steel axle mounted on a low-friction ball bearing ( McMaster-Carr #8828T112 , Princeton , NJ , USA ) . Angular displacement of the treadmill was recorded with an optical encoder ( US Digital E6-2500 , Vancouver , WA , USA ) . A custom designed head-restraint system was placed such that animals were comfortably located on the apex of the treadmill . Rewards were provided through a lick spout ( Harvard Apparatus #598636 ) placed within reaching distance of the mice’s mouth . Timing and amount were controlled using a pinch valve ( NResearch 225PNC1-21 , West Caldwell , NJ , USA ) . Licking behavior was recorded using a capacitive touch sensor ( SparkFun #AT42QT1010 , CO , USA ) connected to the lick spout . The virtual environment was created and rendered in real time in Matlab using the software package ViRMeN ( Aronov and Tank , 2014 ) as well as custom written code . Two 23 . 8’ computer screens ( U2414H , Dell , TX , USA ) were placed in a wedge configuration to cover the majority of the mice’s field of view . After mice had undergone preparatory surgery , they were given at least one week to recover before water scheduling began . Initially , mice received 3 ml of water per day in the form of 3 g of HydroGel ( ClearH2O , Watertown , MA , USA ) , which was gradually reduced to 1 . 0–1 . 5 g per day . During this period , mice were handled by experimenters and habituated to being head restrained as well as running on a cue-less version of the virtual corridor . During habituation , mice were given small water rewards to allow them to acclimate to receiving 10% sugar-water rewards through a spout during head-restraint . Behavioral training began once mice were locomoting comfortably , as assessed by posture and gait . Initially , mice were trained on one trial type alone ( short track ) . Each trial started at a randomized distance from the landmark ( 150–50 cm , drawn from a uniform distribution ) . The wall pattern consisted of a uniform pattern of black dots against a dark gray background to provide generic optic flow information . The view-distance down the corridor was not limited . The landmark cues were 40 cm wide and extended above the walls of the corridor ( see Figure 1B ) . After passing the landmark , mice were able to trigger rewards by licking a fixed distance from the landmark . The reward zone was 20 cm long but not indicated in any way so that the animals had to use self-motion cues and the location of the landmark to locate it . If an animal passed through the reward zone without licking , an automatic ‘reminder’ reward was dispensed . Each reward bolus consisted of 4–6 µl of 10% sucrose water . Sucrose was added to maximize training success ( Guo et al . , 2014 ) . Reward delivery marked the end of a trial and animals were ‘teleported’ into a ‘black box’ for at least 3 s . In some training and recording sessions , animals were required to not lick or run for 3 s , however that requirement was later removed . Training using only one trial type was carried out daily in 30 to 60 min sessions until licking behavior was reliably constrained to after the landmark . At that point , the second trial type ( long track ) was introduced . Training using two tracks was carried out until the licking behavior of mice indicated that they used landmark information to locate the reward ( ‘experts’ , typically 2–4 weeks ) . An empirical bootstrap shuffle test was used to calculate confidence intervals and evaluate whether or not mean first lick locations where significantly different . At that point , mice were transferred to the 2-photon imaging rig . In some instances , a small number of training sessions with the recording hardware running were carried out on the imaging setup to acclimatize animals . The spatial modulation z-score ( SMZ ) was calculated by randomly rotating the location of licks within each trial by a random amount . The fraction of correctly triggered trials within this new , shuffled session was calculated by evaluating whether at least one lick was within the rewarded zone . This process was repeated 1000 times and a null distribution of fraction successful from random licking was calculated . Optogenetic inactivation was carried out on mice that had been trained to expert level . Once they reached proficiency at using landmarks to locate rewards , the masking light was introduced . Animals were allowed a small number of sessions to habituate to the masking light before inactivation trials were introduced . The masking stimulus was provided by two 465 nm wavelength LEDs mounted on top of the computer screens facing the animal ( Thorlabs LED465E , Thorlabs , NJ , USA ) . Optogenetic stimulation light was provided by a 470 nm fiber coupled LED ( Thorlabs M470F3 ) powered by a Cyclops LED driver ( Newman et al . , 2015 ) . Stimulation consisted of a solid light pulse with a maximum duration of 10 s ( Lewis et al . , 2015 ) . Stimulation was provided on half of the trials in a random order , with the only exception that no two consecutive trials could be stimulation trials . Light intensity ranged from 1 to 10 mW and was calibrated individually for each animal . Each animal was observed during stimulation trials and checked for no visible effects on behavior such as change in posture or gait . No difference was found in mean running speed or licks per trial when the stimulation light on compared to when only the masking stimulus was shown ( Figure 1—figure supplement 1 ) . The task scores on mask only trials were compared to the task scores on mask + stimulation trials to assess deficits in the mice’s ability to use the landmark as a cue to locate rewards . A Neurolabware 2-photon microscope coupled to a SpectraPhysics Insight DeepSee II were used for GCaMP6f imaging . To prevent photodamage or bleaching during extended recording periods , a 4x pulse splitter was placed in the light path ( Ji et al . , 2008 ) . The virtual reality software ran on a separate computer that was connected to the image acquisition system . Start and end of recording sessions were controlled by the virtual reality software to ensure synchrony of behavior and imaging data . Animals were placed in the head restraint and had a custom-designed 3D printed opaque sleeve placed over their cranial window to block light from the VR screens from leaking into the objective . The scope was lowered and suitable FOV identified before recordings began . Neurons in RSC were recorded at a wavelength of 980 nm . During V1 bouton recordings , the wavelength was switched to 920 nm . This was done to minimize autofluorescence from the dura mater , which is more pronounced at 980 nm excitation , especially during superficial recordings . Images were acquired at a rate of either 15 . 5 Hz or 31 Hz . In a subset of recordings , an electronically tunable lens was used to record from multiple FOVs in the same animal and session . In all but one cases , dual-plane imaging was used at a rate of 31 Hz , resulting in 15 . 5 Hz per plane acquisition . In a single recording session , six planes were acquired at 5 . 1 Hz . The two planes with most somas where included in this study . Recordings were acquired continuously throughout each session as opposed to epoch-based on trials . Custom written Mathworks Matlab code was used for image registration , segmentation and signal extraction . Each recording session was stabilized using an FFT based rigid algorithm to register each frame to a template created from a subset of frames drawn at random from the whole session . This was followed by creating a pixel-by-pixel local cross-correlation and global cross-correlation maps . Regions of interest were drawn semi-automatically based on local cross-correlation from an experimenter defined seed-point . In addition to cross-correlation , global PCA , mean intensity , and other maps were created to aid identification of neurons and axonal boutons . Since the FOV was the same for virtual navigation and decoupled stimulus presentation , the same ROI maps created during virtual navigation could be used for decoupled stimulus presentation . During signal extraction , the mean brightness value of all pixels within a single ROI was calculated . A neuropil ‘donut’ was automatically generated around each ROI to allow for subtraction of local brightness from ROI signal . ∆F/F was calculated using a 60 s sliding time window . For neurons , F0 was calculated from the bottom 5th percentile of data points within the sliding window . For boutons , the bottom 50th percentile was used to calculate F0 . Neuropil signal was subtracted from ROI signal prior to calculating ∆F/F . Each ROI time course was manually inspected prior to inclusion into analysis . ROIs were excluded if they had few transients ( <0 . 5/min ) . Transients were identified as detected whenever the ∆F/F signal was above six standard deviations for at least 500 ms . The ROI time course was then aligned and re-sampled to match behavioral data frame-by-frame using custom code and the Scipy signal processing toolbox ( Jones et al . , 2001 ) . To test for long term imaging side effects despite using a pulse splitter , we tested for baseline drift of mean frame brightness for each included recording session ( Figure 4—figure supplement 2 ) . Experiments in which the same neurons were recorded in naïve and expert animals , field of views ( FOV ) were matched manually at recording time . For signal extraction , ROIs drawn on the naïve FOV were transferred to the expert FOV and , where necessary , manually adjusted overlay on the same neuron . The time course of each neuron was split into individual trials and aligned to one of three anchor points: trial onset , landmark , and reward . For the neuron to be considered task engaged it had to fulfill the following criteria: 1 ) ∆F/F had to exceed three standard deviations of the ROI’s overall activity on at least 25% of all trials; 2 ) the mean ∆F/F across trials had to exceed a peak z-score of 3 at its peak . The z-score for each ROI was determined by randomly rotating its ∆F/F time course with respect to its behavior 500 times and the peak value of the mean trace was then used to calculate the peak z-score; 3 ) the minimum of the mean trace amplitude ( i . e . highest – lowest value ) had to exceed 0 . 2 ∆F/F . Criteria for axonal boutons were the same with the exception that the minimum mean trace amplitude was 0 . 1 ∆F/F . For the neurons that passed these criteria , the alignment point that resulted in the largest mean response was determined . To avoid edge-cases at the beginning and end of the track , the mean trace was only calculated for bins where at least 50% of trials were present . For the comparison of peak amplitude in Figure 4C and D , the peak amplitude as a function of space , rather than time , was used . The landmark selectivity index was calculated for all neurons that were classified as landmark aligned on at least one trial type as LMI = ( LMshort – LMlong ) / ( LMshort + LMlong ) , where LMx refers to the peak response to the respective landmark . Only neurons that were classified as landmark-aligned neurons were included in that analysis . The fraction of neurons classified as trial onset , landmark , or reward were calculated from the total number of neurons with a baseline activity of at least 0 . 5 transients/min . A template matching decoder was used to assess the accuracy by which the trial type could be identified based on the activity of the different categories of neurons ( trial onset , landmark , reward ) . First , template vectors were constructed for each trial type by calculating the mean response across trials within a session . The responses of the same neurons in individual trials were then compared to the template vectors , resulting in a similarity index for each trial type:Θθ= Σi=1N Rit RiθRt⋅Rθ Here , Θ is the similarity index for trial type θ ( short or long ) . Rθ is the template vector for trial type θ , Rt is a vector of the responses of all neurons in a given trial , and i…N are the indices of all neurons of a given category . Whichever similarity index was higher for a given trial was considered the decoded trial type and compared to the trial type the animal was actually on . Population plots were created by binning the activity of each neuron as a function of space . Each bin was 5 cm wide and all data points falling within a bin were averaged to calculate the mean activity at that location . The first bin started at 100 cm distance from the landmark such that it contained data from at least 50% of trials on average . The activity in each bin was normalized to the bin with peak activity of the same neuron such that all data ranged from 0 to 1 . To plot the mean activity of all neurons in this study , the data was split . Half the trials were randomly drawn to calculate the bin with peak activity . The other half of the trials was used to calculate activity to be plotted . The population vector cross-correlation was calculated similarly by randomly drawing half of the trials to construct one vector and using the other 50% trials to construct the other vector . For each spatial bin , the Pearson correlation coefficient was calculated . The location reconstruction error was calculated as the distance between the spatial bin with the highest cross-correlation value to the animals’ actual location . As randomly splitting trials into halves can lead to spurious cross-correlation maps , this process was repeated 10 times and the mean cross-correlation coefficient and position reconstruction error for each spatial bin was calculated . The peak response during decoupled stimulus presentation was evaluated by aligning each neuron or bouton to its preferred alignment point during virtual navigation . The response was then measured at the same point relative to that alignment point ( in space ) where it showed its peak response during virtual navigation . To allow for small shifts in peak activity during decoupled stimulus presentation , a window of ± 20 cm was introduced and the peak value within that window was used for analysis . Transients were identified whenever ∆F/F in a given trial ( binned into 2 cm spatial bins ) rose above six standard deviations of that neurons baseline activity ( 70th percentile of data points ) for at least two consecutive bins . Transients located outside ± 60 cm of the mean peak during virtual navigation were excluded . Jitter was calculated as the standard error of the difference between mean peak and transient peak locations . The effect of concurrent motor and visual inputs during decoupled stimulus presentation was assessed by grouping trials based on whether the animal was running or stationary . A trial was considered a ‘running’ trial if its average speed exceeded 3 cm/sec in a ± 50 cm time window around its peak response relative to the landmark during virtual navigation . Only neurons with at least three running and non-running trials were included . To allow for slight mismatches between a neuron’s peak response during virtual navigation and decoupled stimulus presentation , the peak within 20 cm of the VR peak was used . Activity in the black box was calculated as a neuron’s response 1 . 5 s after onset of showing black screens with a movement time window of ±1 s . The relative response amplitudes were calculated by normalizing the activity of each neuron to its activity during virtual navigation . Only sessions in which a given animal ran and was stationary were included in this analysis . Running speed modulation analysis was carried out by calculating the average running speed during the transient ( e . g . if the transient was 500 ms long , the average running speed during those 500 ms was calculated ) and correlating it to the peak amplitude of the transient . Only neurons where at least 10 transients where present on the respective trial type were included in this analysis . To assess differences in neuronal responses of superficial vs deep neurons in RSC , the depth of the recordings was used to determine which layer neurons belonged to . RSC does not possess a layer 4 , and layers 2/3 and 5 are separated by a section of relatively few cell bodies . In addition , layer five is comparatively superficial , starting at only 300 µm below the pia ( Lein et al . , 2007 ) . This made identification of cortical layers during in vivo 2-photon imaging possible . We split recordings into layer 2/3 and layer five recordings based on depth below the pia . In one recording , a Rbp4-Cre positive animal , which expresses Cre in many layer 5 cells , was used in conjunction with a flexed GCaMP6f construct . The recording depth for this animal was congruent with other recordings in which we located layer five based on recording depth alone .	When moving through a city , people often use notable or familiar landmarks to help them navigate . Landmarks provide us with information about where we are and where we need to go next . But despite the ease with which we – and most other animals – use landmarks to find our way around , it remains unclear exactly how the brain makes this possible . One area that seems to have a key role is the retrosplenial cortex , which is located deep within the back of the brain in humans . This area becomes more active when animals use visual landmarks to navigate . It is also one of the first brain regions to be affected in Alzheimer's disease , which may help to explain why patients with this condition can become lost and disoriented , even in places they have been many times before . To find out how the retrosplenial cortex supports navigation , Fischer et al . measured its activity in mice exploring a virtual reality world . The mice ran through simulated corridors in which visual landmarks indicated where hidden rewards could be found . The activity of most neurons in the retrosplenial cortex was most strongly influenced by the mouse’s position relative to the landmark; for example , some neurons were always active 10 centimeters after the landmark . In other experiments , when the landmarks were present but no longer indicated the location of a reward , the same neurons were much less active . Fischer et al . also measured the activity of the neurons when the mice were running with nothing shown on the virtual reality , and when they saw a landmark but did not run . Notably , the activity seen when the mice were using the landmarks to find rewards was greater than the sum of that recorded when the mice were just running or just seeing the landmark without a reward , making the “landmark response” an example of so-called supralinear processing . Fischer et al . showed that visual centers of the brain send information about landmarks to retrosplenial cortex . But only the latter adjusts its activity depending on whether the mouse is using that landmark to navigate . These findings provide the first evidence for a “landmark code” at the level of neurons and lay the foundations for studying impaired navigation in patients with Alzheimer's disease . By showing that retrosplenial cortex neurons combine different types of input in a supralinear fashion , the results also point to general principles for how neurons in the brain perform complex calculations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Representation of visual landmarks in retrosplenial cortex
Polyunsaturated fatty acids ( PUFAs ) in phospholipids affect the physical properties of membranes , but it is unclear which biological processes are influenced by their regulation . For example , the functions of membrane arachidonate that are independent of a precursor role for eicosanoid synthesis remain largely unknown . Here , we show that the lack of lysophosphatidylcholine acyltransferase 3 ( LPCAT3 ) leads to drastic reductions in membrane arachidonate levels , and that LPCAT3-deficient mice are neonatally lethal due to an extensive triacylglycerol ( TG ) accumulation and dysfunction in enterocytes . We found that high levels of PUFAs in membranes enable TGs to locally cluster in high density , and that this clustering promotes efficient TG transfer . We propose a model of local arachidonate enrichment by LPCAT3 to generate a distinct pool of TG in membranes , which is required for normal directionality of TG transfer and lipoprotein assembly in the liver and enterocytes . Polyunsaturated fatty acids ( PUFAs ) have important biological functions in health and disease . They are related to cardiovascular diseases , neuronal functions , metabolic syndromes , and many other pathophysiological events ( Poudyal et al . , 2011; Baum et al . , 2012; Jiao et al . , 2014 ) . Phospholipids containing PUFAs render the membrane more flexible , which reduces the energy required for deformation ( Rawicz et al . , 2000; Pinot et al . , 2014 ) . Although PUFAs in phospholipids affect membrane physical properties , the biological significance related to these different properties remain unclear . A recent elegant study showed that PUFAs in phospholipids facilitate endocytosis by reducing the energy required for membrane deformation and fission ( Pinot et al . , 2014 ) . However , the requirement for PUFAs in other biological processes accompanying membrane deformation has to be demonstrated . Arachidonate ( C20:4; indicative of a fatty acid of 20 carbons with four double bonds ) is one of the most studied PUFAs , since it serves as a precursor for biologically active eicosanoids ( prostaglandins , leukotrienes , etc ) ( Shimizu , 2009; Narumiya and Furuyashiki , 2011 ) . The functions of membrane arachidonate that are not related to eicosanoid synthesis remain largely unknown . We reported previously that arachidonate in membrane phosphatidylcholine ( PC ) suppresses Akt signaling in cultured cells ( Koeberle et al . , 2013 ) , but it is unknown whether this occurs in vivo . Studies of mice lacking lysophosphatidylinositol acyltransferase 1 revealed that a decrease in phosphatidylinositol ( PI ) arachidonate has deleterious effects in neuronal development , probably related to PI poly-phosphate dysfunction ( Lee et al . , 2012; Anderson et al . , 2013 ) . However , the in vivo functions of arachidonate in the most abundant membrane phospholipids such as PC and phosphatidylethanolamine ( PE ) , as well as its effect on membrane physical properties , remained to be solved . We recently reported that the levels of several PUFAs ( such as linoleate [C18:2] and docosahexaenoate [C22:6] ) in PC are affected by the substrate selectivity of lysophosphatidic acid acyltransferases ( LPAATs ) during de novo biosynthesis ( Harayama et al . , 2014 ) , but arachidonate levels seem to be regulated in a different manner . Pioneering studies suggested that arachidonate is incorporated into phospholipids after de novo synthesis during the remodeling pathway ( also termed Lands' cycle ) , which consists of a phospholipid deacylation–reacylation cycle ( Hill and Lands , 1968 ) . The reacylation is catalyzed by lysophospholipid acyltransferases ( LPLATs ) ( Hishikawa et al . , 2014 ) , among which lysophosphatidylcholine acyltransferase 3 ( LPCAT3 ) is the strongest candidate to incorporate arachidonate into membranes , due to its substrate selectivity in vitro ( Gijón et al . , 2008; Hishikawa et al . , 2008; Matsuda et al . , 2008; Zhao et al . , 2008 ) . Indeed , knockdown of LPCAT3 by RNA interference reduced the levels of several arachidonate-containing PC species , but additional PUFA-containing PC species were differently affected between studies ( Hishikawa et al . , 2008; Li et al . , 2012; Rong et al . , 2013 ) . Therefore , it remains unclear whether LPCAT3 is specific for the regulation of arachidonate , or it has a broader impact on membrane PUFA levels . In addition , the biological functions of LPCAT3 and membrane arachidonate remain to be fully established , since residual activity of LPCAT3 might affect the overall results in RNA interference experiments . Since PUFAs affect the bending rigidity of the membrane , it is possible that the regulation of phospholipid arachidonate has effects on membrane shape change . When triglycerides ( TGs ) are synthesized by diacylglycerol acyltransferases , they accumulate between the two leaflets of the lipid bilayer and deform the membrane ( Yen et al . , 2008; Thiam et al . , 2013 ) . Molecular dynamics simulation showed that TGs between leaflets generate a ‘blister-like’ shape , where the membrane on its surface is highly curved ( Khandelia et al . , 2010 ) . Therefore , the ability of the membrane to form curved structures might affect the properties of the TG pool between the leaflets . This pool serves as a precursor for cytosolic lipid droplets in most cells ( Thiam et al . , 2013 ) . In hepatocytes and enterocytes , this pool is also used for transport by microsomal triglyceride transfer protein ( MTP ) into the endoplasmic reticulum ( ER ) lumen to generate lipoproteins and luminal lipid droplets ( Sturley and Hussain , 2012 ) . MTP transfers TGs to nascent apolipoprotein B ( apoB ) to form primordial lipoproteins ( Abumrad and Davidson , 2012 ) . MTP also transfers TGs to generate apoB-free lipid droplets in the ER lumen ( Kulinski et al . , 2002 ) , which are required for TG enrichment in nascent lipoproteins . MTP deficiency in humans leads to an impaired absorption of dietary lipids , which is called abetalipoproteinemia ( Wetterau et al . , 1992 ) . Therefore , the normal luminal transfer of TG is critical in vivo , but factor ( s ) and environment enabling an efficient transport by MTP remain poorly understood ( Yao et al . , 2013 ) . Here , using LPCAT3-deficient cells and mice , we report that LPCAT3 is critical and relatively specific for the regulation of arachidonate levels in membrane phospholipids in vivo . In addition , we show that LPCAT3 regulates the directionality of TG transfer into lipoproteins , preventing the over-accumulation of cytosolic lipid droplets in hepatocytes and enterocytes . Additional analyses suggest that membrane PUFAs facilitate TG local clustering , which enables efficient transport by MTP . This study identifies LPCAT3 as a major regulator of membrane phospholipid arachidonate , which is a critical factor affecting luminal TG transport . To investigate whether the remodeling pathway contributes to arachidonate incorporation into phospholipids , we analyzed the enzymatic properties of LPCAT3 . We first examined whether LPCAT3 regulates the acyl-chain composition of the major phospholipid PC . We established rat hepatoma RH 7777 cells that stably overexpress FLAG-tagged murine LPCAT3 ( Figure 1—figure supplement 1A ) . Control cells had an LPCAT activity selective for linoleoyl- and arachidonoyl-CoA ( Figure 1A , B ) . The overexpression of LPCAT3 doubled LPCAT activity without affecting acyl-CoA selectivity ( Figure 1A , B ) . The acyl-chain composition of PC in LPCAT3-overexpressing cells was analyzed using liquid chromatography-tandem mass spectrometry ( LC-MS ) . Chromatograms of 38:4 PC ( PC species containing 38 carbons and four double bonds as a sum of the two acyl-chains ) revealed two isomers with different retention times ( Figure 1—figure supplement 2 ) . The levels of 36:4 PC and the later-eluting isomer of 38:4 PC , but not 34:2 PC , were increased in LPCAT3-overexpressing cells ( Figure 1C ) . Additional selected reaction monitoring ( SRM , see ‘Materials and methods’ for detail ) analyses were performed to resolve the two acyl-chains , and showed that 34:2 PC , 36:4 PC , and the later-eluting isomer ( peak 2 ) of 38:4 PC correspond to 16:0–18:2 PC , 16:0–20:4 PC , and 18:0–20:4 PC , respectively ( Figure 1—figure supplement 2 and data not shown ) . Therefore , LPCAT3 overexpression increases arachidonate , but not linoleate levels in PC . 10 . 7554/eLife . 06328 . 003Figure 1 . LPCAT3 regulates arachidonate levels in PC . ( A and B ) LPCAT activity and selectivity in mock- or LPCAT3-overexpressing RH 7777 cells . LPC and five acyl-CoA species ( 16:0- , 18:1- , 18:2- , 20:4- , and 22:6-CoA ) were used as substrates of LPCAT assay to generate DPPC ( 16:0–16:0 PC ) , POPC ( 16:0–18:1 PC ) , PLPC ( 16:0–18:2 PC ) , PAPC ( 16:0–20:4 PC ) , and PDPC ( 16:0–22:6 PC ) . The total of five products ( A ) and the relative amount of each PC species , equivalent to acyl-CoA selectivity ( B ) are shown . The experiment was performed in technical triplicate . ( C ) Levels of selected PC species in mock- or LPCAT3-stable cells ( n = 8 ) . Data are % of the total signals from all PC species detected . ( D–G ) Total LPCAT activity products ( D and F , in technical triplicate ) and PC levels of the selected species ( E ( n = 3 ) and G ( n = 5 ) ) were measured in mock- or LPCAT3-null RH 7777 cells ( D and E ) or the null cells rescued with LPCAT3 or a mutant lacking activity ( F and G ) . ( D and E ) The results of two different clones are separately indicated . Error bars are SD ( A , B , D , and F ) or SEM ( C , E , and G ) . sgLPCAT3: single guide RNA targeting Lpcat3 . p < 0 . 01 , *p < 0 . 0001 . See also Figure 1—figure supplements 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 00310 . 7554/eLife . 06328 . 004Figure 1—figure supplement 1 . Confirmation of stable expression . ( A and B ) Stable overexpression of transfected FLAG-LPCAT3 was detected by western blot analysis using an anti-FLAG antibody . Ponceau S staining was used as a loading control . ( A ) Detection of FLAG-LPCAT3 in RH 7777 cells . ( B ) Detection of wild type FLAG-LPCAT3 or a H374A mutant lacking activity . The parental cells are LPCAT3-null RH 7777 cells . Control cells that were stably transfected with an empty vector were also established . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 00410 . 7554/eLife . 06328 . 005Figure 1—figure supplement 2 . Annotation of LC-MS signals . Using methods that detect head group information at Q3 ( top , in this case , 38:4 PC is detected using a precursor scan of m/z 184 . 1 , a characteristic fragment of PC ) multiple peaks that might have different combinations of sn-1 and sn-2 fatty acids ( e . g . , 16:0–22:4 PC , 18:0–20:4 PC , or 18:1–20:3 PC ) can be detected with different retention times . When fatty acid fragments are selected at Q3 ( bottom , in this case , a stearate fragment of 38:4 PC is selected at Q3 using SRM ) , the peaks can be attributed to a single acyl-chain combination ( since stearate is detected , only 18:0–20:4 PC is detected ) . By comparing the retention times of the two chromatograms , we can estimate the acyl-chain composition of phospholipids eluted differently ( in this case , peak 2 of 38:4 PC is annotated as 18:0–20:4 PC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 00510 . 7554/eLife . 06328 . 006Figure 1—figure supplement 3 . Generation of LPCAT3-null cells . ( A ) Strategy for genome editing of rat Lpcat3 locus using the CRISPR/Cas9 system . The WHG sequence required for LPCAT3 activity was removed by cleavage of two adjacent regions , leading to a small genomic deletion . Arrowheads: PCR primers for analysis of deletion; sgRNA: single guide RNA; NHEJ: non-homologous end joining . ( B ) Genome editing of Lpcat3 in RH 7777 cells was confirmed by PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 006 Next , LPCAT3-null RH 7777 cells were generated using the CRISPR/Cas9 ( Clustered Regularly Interspaced Short Palindromic Repeat/Cas9 [Ran et al . , 2013] ) system for loss of function studies ( Figure 1—figure supplement 3A , B ) , and two clones per group were analyzed . LPCAT3-null cells had almost no lysophosphatidylcholine acyltransferase ( LPCAT ) activity ( Figure 1D ) . LPCAT3 deficiency largely decreased arachidonate-containing 36:4 PC and 38:4 PC levels , but not linoleate-containing 34:2 PC ( Figure 1E ) . The changes in LPCAT activity and levels of arachidonate-containing PC were restored when deficient cells were stably transfected with wild type murine LPCAT3 , but not with a mutant H374A ( Shindou et al . , 2009 ) lacking enzymatic activity ( Figure 1F , G , and Figure 1—figure supplement 1B ) . Taken together with overexpression studies , despite its enzymatic preference for both linoleate and arachidonate in vitro , LPCAT3 has a high selectivity on the accumulation of arachidonate-containing PC at the cellular level . Since the cell culture experiments established LPCAT3 as a target molecule to investigate the functions of arachidonate in membrane phospholipids , we next analyzed the functions of this enzyme in vivo . We first examined the expression pattern of LPCAT3 at the late embryonic stages ( between embryonic days 18 . 5 ( E18 . 5 ) and E19 . 5 ) . LPCAT3 mRNA was ubiquitously detected , with high expression in intestine , followed by liver ( Figure 2A ) , consistent with the protein level ( Figure 2B ) . Intestinal LPCAT3 mRNA expression was strongly induced during the late developmental stages , while the changes in the liver were relatively small ( Figure 2C ) . To examine intestinal LPCAT3 expression more in detail , we divided the whole intestine into three parts: the proximal small intestine , where lipid absorption is high ( Abumrad and Davidson , 2012 ) , distal small intestine , and colon ( Figure 2D ) . LPCAT3 expression was highest in the proximal small intestine ( Figure 2E , F ) . Enzymatic assays revealed that LPCAT activity is highly correlated with LPCAT3 expression ( Figure 2B , F , G ) . We generated LPCAT3-deficient mice by gene targeting and confirmed the loss of protein expression ( Figure 2—figure supplement 1A–C ) . LPCAT activity was almost completely absent in tissues of LPCAT3-deficient mice ( Figure 2G ) . Therefore , LPCAT3 is the major LPCAT enzyme in vivo . 10 . 7554/eLife . 06328 . 007Figure 2 . Tissue distribution of LPCAT3 . ( A–F ) The distribution of LPCAT3 mRNA ( A , C , and E , by quantitative PCR ) or protein ( B and F , by western blot analysis ) was analyzed in different tissues at E18 . 5 ( A , B , E , and F ) or at various perinatal stages ( C ) . Ponceau S staining was performed as a loading control for western blot analysis . ( D ) Illustration of the intestine separated into the proximal- , distal small intestine , and colon . ( G ) LPCAT activity was determined in tissues obtained from mice of the indicated genotype at E18 . 5 . See legend of Figure 1A for description of the method . Error bars are SEM ( n = 3 ) . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 00710 . 7554/eLife . 06328 . 008Figure 2—figure supplement 1 . Generation of LPCAT3-deficient mice . ( A ) Strategy for gene targeting to generate LPCAT3-deficient mice and for the identification of targeted clones . Neo: Neomycin-resistance cassette; DT-A: diphtheria toxin subunit A cassette . ( B and C ) LPCAT3 deficiency was confirmed by PCR using tail tip genomic DNA ( B ) and by western blot analysis of proteins from small intestine ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 008 We next analyzed how LPCAT3 deficiency affects the lipidome in mice at E18 . 5–E19 . 5 . We measured tissue PC levels and found a slight reduction in the proximal small intestine of LPCAT3-deficient mice , but not in other tissues ( Figure 3A ) . Next , we analyzed the changes in acyl-chain composition of phospholipids from tissues of LPCAT3-deficient mice . LC-SRM-MS lipidomics analysis of four phospholipid classes ( PC , PE , phosphatidylserine ( PS ) , and PI ) was carried out . The total of signals from all the detected molecules were comparable between genotypes , and were used for normalization ( Figure 3—figure supplement 1 ) . The slight reduction in the proximal small intestine of LPCAT3-deficient mice was consistent with the reduced PC amount ( Figure 3A and Figure 3—figure supplement 1 ) . To detect the global differences in PC , we analyzed the ratio ( as % of wild type ) for each PC species in individual tissues , and then averaged the ratio values from all tissues ( Figure 3B ) . This analysis enabled us to detect the most common changes that occur under LPCAT3 deficiency . The most decreased species were 34:4 PC , 36:4 PC , 36:5 PC , and 38:4 PC ( shown in green in Figure 3B ) , suggesting that the reduction in arachidonate levels is most prominent . These decreases occurred in almost all tissues analyzed ( Figure 3C–E ) . The changes in species that might contain linoleate ( 34:2 PC and 36:2 PC ) were small or absent ( Figure 3B ) . On the other hand , the most increased PC species were 40:4 PC , 40:5 PC , 42:4 PC , and 42:5 PC ( shown in magenta in Figure 3B ) , which might contain fatty acids that arise from arachidonate metabolism ( carbon chain elongation , desaturation , and partial β-oxidation [Schmitz and Ecker , 2008] [Figure 3B , F , G] ) . To investigate the acyl-chain composition of the changed PC species ( 34:4 PC , 36:4 PC , 36:5 PC , 38:4 PC , 40:4 PC , 40:5 PC , 42:4 PC , and 42:5 PC ) , we performed SRM analyses of PC from the proximal small intestine , using transitions that discriminate fatty acids ( Table 1 ) . Each PC subspecies did not have a single acyl-chain composition , but was a mixture of various combinations . Interestingly , most of arachidonate-containing species were decreased in LPCAT3-deficient mice , as well as some species containing C18 PUFAs ( C18:2 and C18:3 ) and other C20 PUFAs ( C20:3 and C20:5 ) . On the other hand , PC species containing C22 PUFAs ( C22:4 and C22:5 ) and C24 PUFAs ( C24:4 and C24:5 ) were increased , suggesting that the excess of arachidonoyl-CoA caused by LPCAT3 deficiency increases its metabolism ( such as elongation ) into C22 and C24 PUFAs , which were then incorporated into PC ( Figure 3—figure supplement 2 ) ( Schmitz and Ecker , 2008 ) . In addition to the changes in PC , we found that lysophosphatidylcholine ( LPC ) levels are increased in the proximal small intestine , distal small intestine , and liver of LPCAT3-deficient mice ( Figure 3H ) . 10 . 7554/eLife . 06328 . 009Figure 3 . LPCAT3 regulates PC arachidonate levels in vivo . Lipidomics analyses were performed in various tissues of mice at E18 . 5 . ( A ) Total PC amounts in tissues from wild type and LPCAT3-deficient mice . ( B ) Heat map showing the ratio of each PC species in LPCAT3-deficient mice . Each value is the average of ratio values ( % of wild type ) for tissues analyzed in ( C–H ) . ( C–G ) The levels of the indicated PC species in wild type and LPCAT3-deficient mice were measured by LC-MS . ( H ) The levels of 16:0 LPC were measured in wild type and LPCAT3-deficient mice . Data were normalized to total MS signals shown in Figure 3—figure supplement 1 . Prox . : Proximal; Dist . : Distal . ( A , and C–H ) Error bars are SEM ( n = 5 ) . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , **p < 0 . 0001 . See also Figure 3—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 00910 . 7554/eLife . 06328 . 010Figure 3—figure supplement 1 . Total signals used for normalization of lipidomic profiling analysis . Total MS signals from all the SRM channels of the lipidomic profiling analysis , which were used for normalization ( n = 5 ) . Note that the amount of spleen lipids analyzed was not normalized to tissue weight , making if difficult to compare the total values ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 01010 . 7554/eLife . 06328 . 011Figure 3—figure supplement 2 . Metabolism of arachidonate . Metabolism of arachidonate ( elongation , desaturation , and partial β-oxidation ) that might have been enhanced due to the excess of arachidonoyl-CoA in LPCAT3-deficient mice . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 01110 . 7554/eLife . 06328 . 012Table 1 . Acyl-chain composition of PC species with changed levels in LPCAT3-deficient miceDOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 012PC subclassAcyl-chain compositionLPCAT3+/+ peak areaLPCAT3−/− peak areaKO/WT ( % ) p valueMeanSEMMeanSEM34:414:0–20:432 , 607125311 , 60983536<0 . 000116:1–18:342 , 145257441 , 3404555980 . 881636:416:0–20:41 , 037 , 88721 , 426917 , 6428844880 . 000818:1–18:3122 , 575842161 , 4101726500 . 000118:2–18:2366 , 67517 , 961250 , 35026 , 185680 . 006436:516:0–20:5147 , 0548925387210953<0 . 000116:1–20:4104 , 768777379 , 8514883760 . 026518:2–18:336 , 381366916 , 7621790460 . 001338:416:0–22:481 , 0936150169 , 8499408209<0 . 000118:0–20:4670 , 48823 , 121624 , 85318 , 289930 . 160218:1–20:3230 , 51710 , 302108 , 788800947<0 . 000118:2–20:229 , 789158714 , 56770149<0 . 000140:416:0–24:410 , 392161720 , 86530722010 . 016718:0–22:4176 , 3254788293 , 8326785167<0 . 000118:1–22:322 , 450270462 , 1453805277<0 . 000120:0–20:412 , 13112656590899540 . 007320:1–20:332 , 19125914991174016<0 . 000140:516:0–24:515351258538111623500 . 054918:0–22:5110 , 7527361163 , 72314 , 1861480 . 010618:1–22:439 , 6032862117 , 3625425296<0 . 000120:1–20:458 , 465376732 , 5952075560 . 000342:418:0–24:4190740312 , 41217626510 . 000418:1–24:316637779584639576<0 . 000118:2–24:244979501587605350 . 073622:0–20:411 , 18093050391154450 . 003242:518:0–24:5309262914 , 907882482<0 . 000118:1–24:4125840511 , 476858912<0 . 000122:1–20:425 , 450279011 , 7281853460 . 0035SRM analyses for discrimination of fatty acids were performed for the indicated PC species ( n = 5 ) . Anions of the fatty acids indicated at the second position ( e . g . , 20:4 for 14:0–20:4 PC ) were used for selection at Q3 . Sample concentration was adjusted based on tissue weight . Analysis of other phospholipid classes showed that 38:4 PE , but not 38:4 PI levels are decreased , and that 40:4 PE and 40:5 PE are increased in LPCAT3-deficient mice ( Figure 4A–D ) . Similarly with PC , SRM analyses for fatty acid discrimination revealed that PE species containing C20 PUFAs are decreased , while those containing C22 PUFAs are increased in LPCAT3-deficient mice ( Table 2 ) . This is consistent with the LPEAT activity of LPCAT3 with arachidonate preference ( Figure 4E , F ) . The number of detected molecules for PS was low , thus we did not analyze these species in detail . However , preliminary data suggested that arachidonate-containing PS species are decreased in LPCAT3-deficient mice ( data not shown ) . 10 . 7554/eLife . 06328 . 013Figure 4 . LPCAT3 regulates PE arachidonate levels in vivo . ( A–D ) The levels of the indicated PE species in wild type and LPCAT3-deficient mice were measured by LC-MS and normalized using the values in Figure 3—figure supplement 1 . Error bars are SEM ( n = 5 ) . ( E and F ) LPEAT activity was examined in mock- or LPCAT3-transfected RH 7777 cells ( error bars are SD , technical triplicate ) ( E ) and proximal small intestine from wild type or LPCAT3-deficient mice ( error bars are SEM , n = 3 ) ( F ) . 17:1 LPE was used instead of LPC in a similar experiment to Figure 1A , and peak areas of each LPEAT product are illustrated . Prox . : Proximal; Dist . : Distal . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , **p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 01310 . 7554/eLife . 06328 . 014Table 2 . Acyl-chain composition of PE species with changed levels in LPCAT3-deficient miceDOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 014PE subclassAcyl-chain compositionLPCAT3+/+ peak areaLPCAT3−/− peak areaKO/WT ( % ) p valueMeanSEMMeanSEM38:416:0–22:4373 , 45820 , 2391 , 073 , 90250 , 565288<0 . 000118:0–20:48 , 394 , 810381 , 4994 , 349 , 617102 , 27152<0 . 000118:1–20:31 , 503 , 04138 , 884689 , 13114 , 846460 . 000240:418:0–22:41 , 831 , 03280 , 3193 , 803 , 42198 , 424208<0 . 000118:1–22:3124 , 6384846236 , 22423 , 3571900 . 001620:0–20:4187 , 358868745 , 797428624<0 . 000120:1–20:335 , 837301911 , 7212388330 . 000240:518:0–22:5523 , 41646572 , 056 , 845111 , 780393<0 . 000118:1–22:4731 , 94743 , 0881 , 934 , 98435 , 746264<0 . 000120:1–20:4224 , 759357264 , 708306629<0 . 0001SRM analyses for discrimination of fatty acids were performed for the indicated PE species ( n = 5 ) . Anions of the fatty acids indicated at the second position ( e . g . , 22:4 for 16:0–22:4 PE ) were used for selection at Q3 . Sample concentration was adjusted based on tissue weight . To gain further insights in the fatty acid compositions , we performed gas chromatography with flame ionization detection ( GC-FID ) . Fatty acids were detected as methyl esters derivatized from total lipids ( including free fatty acids , neutral lipids , and phospholipids ) . The total amounts of fatty acids were slightly decreased in proximal small intestine of LPCAT3-deficient mice , as expected from the reduced PC amount ( Figure 3A and Figure 5—figure supplement 1 ) . Unexpectedly , they increased in liver of LPCAT3-deficient mice ( Figure 5—figure supplement 1 ) . GC-FID confirmed that arachidonate levels are reduced in all tissues of LPCAT3-deficient mice ( Figure 5A ) . Linoleate levels were decreased only in some tissues ( Figure 5B ) , showing that the changes seen in proximal small intestine PC ( Table 1 ) are not universal , consistently with Figure 3B . We observed increased docosahexaenoate levels , but only in some tissues ( Figure 5C ) . Instead , adrenate ( C22:4 n-6 ) and docosapentaenoate ( C22:5 n-6 ) levels were increased in almost all tissues , confirming that the excess of arachidonoyl-CoA caused by LPCAT3 deficiency increases its use for further metabolism , as has been suggested by the results of LC-MS ( Figure 5D , E , and Figure 3—figure supplement 2 ) . C24 PUFAs could not be analyzed under the GC-FID condition used . Due to these increases , the total amounts of PUFAs that have four or more double bonds were only slightly affected in LPCAT3-deficient mice ( Figure 5F ) . 10 . 7554/eLife . 06328 . 015Figure 5 . Arachidonate levels and metabolism are changed in LPCAT3-deficient mice GC-FID was performed using various tissues to analyze fatty acid amounts and compositions in total lipids , as well as those of lipid subclasses . ( A–F ) The levels of the indicated PUFAs ( A–E ) , and the total of PUFAs ( F , sum of PUFAs with four or more double bonds ) in wild type and LPCAT3-deficient mice were measured by GC-FID . Data are % of the total levels shown in Figure 5—figure supplement 1 . ( G ) The total amounts of fatty acids in neutral lipids and phospholipids that were fractionated from the indicated tissues in wild type and LPCAT3-deficient mice were analyzed by GC-FID . Prox . : Proximal; Dist . : Distal . Error bars are SEM ( n = 5 ) . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , **p < 0 . 0001 . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 01510 . 7554/eLife . 06328 . 016Figure 5—figure supplement 1 . Total fatty acids in tissues . Total signals from GC-FID analyses that were used for normalization in Figure 5A–F . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 016 To investigate the mechanism of the unexpected increased fatty acids in liver from LPCAT3-deficient mice , total liver lipids were divided into neutral lipid and phospholipid fractions by solid phase extraction . We found that fatty acid amounts of neutral lipids are increased in liver of LPCAT3-deficient mice ( Figure 5G ) . A similar experiment was performed using proximal small intestine , and fatty acid amounts of phospholipids are decreased in this tissue of LPCAT3-deficient mice ( Figure 5G ) , consistent with the decreased PC levels ( Figure 3A ) . Since it was proposed that LPLATs promote the reacylation of free arachidonate and reduce eicosanoid production ( Zarini et al . , 2006; Gijón et al . , 2008 ) , we investigated eicosanoid levels , but found no major change in LPCAT3-deficient mice , at least under normal conditions ( Figure 6A–C and data not shown ) . In conclusion , LPCAT3 regulates membrane phospholipid arachidonate levels without largely affecting the total degree of membrane unsaturation and eicosanoid levels . Therefore , the phenotypes of LPCAT3-deficient mice under normal conditions would reveal the functions of membrane arachidonate that are independent of eicosanoids . 10 . 7554/eLife . 06328 . 017Figure 6 . Eicosanoid levels are not affected in tissues of LPCAT3-deficient mice . ( A–C ) The levels of 6-keto PGF1α ( A ) , PGE2 ( B ) and 15-HETE ( C ) in wild type and LPCAT3-deficient mice at E18 . 5 . Error bars are SEM ( n = 5 ) . PG: prostaglandin; HETE: hydroxyeicosatetraenoic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 017 Next , we analyzed liver of LPCAT3-deficient mice at E18 . 5–E19 . 5 , because fatty acid levels were increased in neutral lipids , suggesting an accumulation of TGs ( Figure 5G ) . GC-FID analysis of the phospholipid fraction confirmed that arachidonate levels were decreased in liver of LPCAT3-deficient mice ( Figure 7A ) . On the other hand , C22 n-6 PUFAs and docosahexaenoate were increased . Although they were minor components , GC-FID analysis of the neutral lipid fraction showed that arachidonate and C22 n-6 PUFAs ( C22:4 and C22:5 ) are increased in neutral lipids of LPCAT3-deficient mice , while docosahexaenoate was unchanged ( Figure 7B ) . 10 . 7554/eLife . 06328 . 018Figure 7 . TGs accumulate in the embryonic liver of LPCAT3-deficient mice . ( A and B ) Liver phospholipids ( A ) and neutral lipids ( B , only selected PUFAs are shown ) were obtained by solid-phase extraction and used for the analysis of acyl-chain composition using wild type and LPCAT3-deficient mice . The percentage of fatty acids detected by GC-FID is illustrated . Error bars are SEM ( n = 5 ) . ( C and D ) Levels of TG ( C ) and cholesterol ( D ) in liver were measured using wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 . ( E–H ) The amounts of lipid droplets in liver of wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 were detected by oil red O and hematoxylin staining ( E and F ) , or by electron microscopy ( G and H ) . Arrowheads: lipid droplets . p < 0 . 05 , p < 0 . 01 , *p < 0 . 0001 . See also Figure 7—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 01810 . 7554/eLife . 06328 . 019Figure 7—figure supplement 1 . Histological analysis of embryonic liver . ( A–D ) Histological images of liver sections from wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 stained by hematoxylin/eosin ( A and B ) or PAS/hematoxylin ( C and D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 01910 . 7554/eLife . 06328 . 020Figure 7—figure supplement 2 . The machinery for VLDL assembly is present in LPCAT3-deficient liver . ( A and B ) The protein expression levels of MTP ( A ) and PDI ( B ) in liver of wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 were examined by western blot . Ponceau S staining was used as a loading control . ( C ) VLDL-like particles ( arrowhead ) were detected in the Golgi apparatus of hepatocytes in LPCAT3-deficient mice at E18 . 5–E19 . 5 , showing that VLDL assembly is not absent . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 020 We found that liver TG levels are increased in LPCAT3-deficient mice , while liver cholesterol levels were only slightly changed ( Figure 7C , D ) . Histological analysis revealed that a massive amount of glycogen is present at this period ( irrespectively of the genotype ) , and we could not distinguish wild type and LPCAT3-deficient mice by hematoxylin/eosin , or periodic acid-Schiff ( PAS ) /hematoxylin staining ( Figure 7—figure supplement 1A–D ) . On the other hand , oil red O staining revealed an accumulation of lipid droplets in embryonic liver of LPCAT3-deficient mice ( Figure 7E , F ) , which was confirmed by electron microscopy ( Figure 7G , H ) . The loss of very low-density lipoprotein ( VLDL ) assembly might explain such a phenotype , but the protein levels of MTP and protein disulfide isomerase ( PDI , an interacting partner of MTP required for TG transfer [Kulinski et al . , 2002] ) appeared normal in LPCAT3-deficient mice ( Figure 7—figure supplement 2A , B ) . In addition , VLDL-like particles were detected in the Golgi apparatus of LPCAT3-deficient mice ( Figure 7—figure supplement 2C ) . Therefore , LPCAT3-deficient mice accumulate TGs in the liver at embryonic stages , while the components of VLDL assembly are present . LPCAT3-deficient mice were born at a Mendelian distribution ( n = 24 [LPCAT3+/+] , 45 [LPCAT3+/−] , 21 [LPCAT3−/−] ) , but died within a few days after birth ( Figure 8A ) . Although they looked indistinguishable from control mice at birth and had normal suckling ( Figure 8B ) and breathing , they did not gain weight and became smaller than wild type mice during the first days ( Figure 8C , D ) . LPCAT3-deficient mice had normal blood glucose levels at birth ( at postnatal day ( P ) 0 . 5 ) , but became severely hypoglycemic at P1 . 5 ( Figure 8E , F ) . Hypoglycemia was not due to hyperinsulinemia or an insufficient hepatic glycogenolysis ( Figure 8—figure supplement 1A–D ) . Histological analysis of multiple tissues from mice at P1 . 5 was performed , and the most significant difference between wild type and LPCAT3-deficient mice was seen in the small intestine . The proximal small intestine of LPCAT3-deficient mice had a ‘vacuolated’ appearance ( Figure 9A , B ) . Oil red O staining revealed that neutral lipids are normally transported in wild type mice , while they drastically accumulate in intestinal epithelial cells of LPCAT3-deficient mice ( Figure 9C , D ) . Massive lipid droplet accumulation was confirmed by electron microscopy ( Figure 9E–H ) . In addition to lipid droplet accumulation , enterocytes of LPCAT3-deficient mice were severely damaged , as judged from the loss of microvilli and deformed mitochondria ( Figure 9G , H ) . Therefore , this severe damage probably caused a malabsorption of nutrients leading to hypoglycemia . On the other hand , although the arachidonate content was decreased in phospholipids of the small intestine from LPCAT3-deficient embryos at E18 . 5–E19 . 5 ( Figure 9I ) , their histological appearance was normal ( data not shown ) . Therefore , the intestinal damages in LPCAT3-deficient mice develop postnatally , most likely due to the epithelial accumulation of lipids from breast milk intake soon after birth . 10 . 7554/eLife . 06328 . 021Figure 8 . LPCAT3-deficient mice are neonatally lethal . ( A–D ) The survival rate ( A ) , gross appearance ( B and C ) , and body weight change ( n = 9–62 ) ( D ) were evaluated in neonatal mice of the indicated genotypes . Arrowhead: normal milk intake in both genotypes . ( E and F ) Blood glucose was measured in wild type and LPCAT3-deficient mice at P0 . 5 ( E ) and P1 . 5 ( F ) . Error bars are SEM . p < 0 . 05 , *p < 0 . 01 . See also Figure 8—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 02110 . 7554/eLife . 06328 . 022Figure 8—figure supplement 1 . Plasma insulin and liver glycogen in LPCAT3-deficient mice . ( A–D ) Plasma insulin ( A and B ) and liver glycogen ( C and D ) were measured in wild type and LPCAT3-deficient mice at P0 . 5 and P1 . 5 . Notice that insulin is not higher in LPCAT3-deficient mice than in wild type mice , and that liver glycogen levels are reduced at P1 . 5 in both genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 02210 . 7554/eLife . 06328 . 023Figure 9 . TG accumulation in enterocytes of LPCAT3-deficient mice . ( A–H ) Histology of the proximal small intestine of mice at P1 . 5 was analyzed by light microscopy after staining with hematoxylin and eosin ( A and B ) , oil red O and hematoxylin ( C and D ) , or by electron microscopy ( E–H ) . ( H ) Shortened microvilli ( arrow ) and a mitochondrion with disrupted outer membrane ( arrowhead ) are seen in LPCAT3-deficient mice . ( I ) Phospholipids from proximal small intestine were obtained by solid-phase extraction and used for the analysis of acyl-chain composition using wild type and LPCAT3-deficient mice . The percentage of fatty acids detected by GC-FID is illustrated . ( J and K ) Levels of TG ( J ) and cholesterol ( K ) in small intestine samples from wild type and LPCAT3-deficient mice at P1 . 5 . Error bars are SEM ( n = 5 ) . p < 0 . 05 , *p < 0 . 01 . See also Figure 9—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 02310 . 7554/eLife . 06328 . 024Figure 9—figure supplement 1 . Distal small intestine of LPCAT3-deficient mice absorbs lipids . ( A and B ) Oil red O/hematoxylin staining of distal small intestine from wild type ( A ) and LPCAT3-deficient mice ( B ) at P1 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 02410 . 7554/eLife . 06328 . 025Figure 9—figure supplement 2 . The machinery for chylomicron assembly is present in LPCAT3-deficient mice . ( A and B ) The protein levels of MTP ( A ) and PDI ( B ) were examined in proximal small intestine of wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 by western blot analysis . Ponceau S staining was used as a loading control . ( C ) Chylomicron-like particles ( arrowheads ) were detected in the Golgi apparatus of enterocytes from the proximal small intestine of LPCAT3-deficient mice at P0 . 5 , showing that chylomicron assembly is not absent . ( D and E ) MTP ( D ) and villin ( E ) mRNA levels were quantified in small intestine of wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 or P1 . 5 by quantitative PCR ( n = 6 ) . Error bars are SEM . p < 0 . 05 , p < 0 . 01 , *p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 025 Next , we analyzed tissue neutral lipid levels in small intestine of LPCAT3-deficient mice at P1 . 5 . In the proximal small intestine , TG levels were high due to the massive absorption of lipids at this region , but tended to be increased in LPCAT3-deficient mice ( Figure 9J ) . On the other hand , cholesterol did not accumulate in LPCAT3-deficient mice at this region ( Figure 9K ) . In addition , we found an increase in TG and cholesterol levels in distal small intestine of LPCAT3-deficient mice at P1 . 5 ( Figure 9J , K ) . Oil red O staining revealed that this region compensatorily absorbs lipids in LPCAT3-deficient mice ( Figure 9—figure supplement 1A , B ) , probably due to the inability to absorb them at the proximal small intestine . Similarly to the case of the embryonic liver , the accumulation of lipid droplets was not due to MTP deficiency , since the protein levels of MTP and PDI in proximal small intestine were normal in LPCAT3-deficient mice at E18 . 5–E19 . 5 ( Figure 9—figure supplement 2A , B ) . Also , we detected chylomicron-like particles in the Golgi apparatus of proximal small intestine enterocytes at P0 . 5 ( Figure 9—figure supplement 2C ) . Although the mRNA level of MTP was decreased in proximal intestine of LPCAT3-deficient mice at P1 . 5 , this was explained as a secondary event by the damage of enterocytes , since villin levels also decreased ( Figure 9—figure supplement 2D , E ) . On the other hand , the mRNA level of MTP was increased in distal small intestines of LPCAT3-deficient mice at P1 . 5 , being consistent with the acquisition of an absorptive phenotype as stated above ( Figure 9—figure supplements 1A , B , 2D , E ) . Enterocytes and hepatocytes with lipid droplet accumulation secrete neutral lipids into plasma as apoB-containing lipoproteins ( Abumrad and Davidson , 2012; Sturley and Hussain , 2012 ) . Although lipoproteins were detected by electron microscopy ( Figure 7—figure supplement 2C and Figure 9—figure supplement 2C ) , their composition might be altered . Therefore , we analyzed plasma lipids at various stages . We found that TG levels , but not apoB , cholesterol , or PC levels , are significantly decreased in plasma of LPCAT3-deficient mice at E18 . 5–E19 . 5 ( Figure 10A–D ) . Plasma lipids were further analyzed by gel filtration chromatography . TGs are usually detected in chylomicron or VLDL fractions in adult plasma . However , in our samples , the elution profiles of both TGs and cholesterol suggested an association with particles of an intermediate size between VLDL , intermediate-density lipoprotein ( IDL ) , and low-density lipoprotein ( LDL ) ( Figure 10—figure supplement 1A–H ) . This is consistent with the elution profiles in previous studies ( van Straten et al . , 2009 ) , and thus is a characteristic of this neonatal period . Therefore , we will regard this ‘VLDL/IDL/LDL’ fraction as the lipoproteins secreted from the liver . TGs in the VLDL/IDL/LDL fraction decreased in LPCAT3-deficient mice at all periods analyzed ( Figure 10E and Figure 10—figure supplement 1A–D ) . In contrast , cholesterol levels were not decreased ( Figure 10F and Figure 10—figure supplement 1E–H ) . After P0 . 4 , TGs and cholesterol were detected in the chylomicron fractions ( Figure 10G , H , and Figure 10—figure supplement 1A–H ) . Chylomicron TG levels were normal in LPCAT3-deficient mice at P0 . 4 , but tended to be decreased at later periods ( Figure 10G and Figure 10—figure supplement 1A–D ) . Since the chylomicron lipid levels vary largely due to the difficulty to synchronize suckling behaviors , the differences did not reach statistical significance . Chylomicron cholesterol levels were not largely affected in LPCAT3-deficient mice at all periods examined ( Figure 10H and Figure 10—figure supplement 1E–H ) . Therefore , a reduction in TGs , but not in cholesterol , was observed commonly in lipoproteins secreted from both liver and small intestine . Based on this observation , we calculated the ratio between TGs and cholesterol in chylomicron and VLDL/IDL/LDL fractions at different time points ( Figure 10I , J ) . In the chylomicron fraction , the TG/cholesterol ratio increased gradually after birth in wild type , but remained constant after P0 . 4 in LPCAT3-deficient mice ( Figure 10I ) . A similar trend was observed in the VLDL/IDL/LDL fraction ( Figure 10J ) . These results suggest that in LPCAT3-deficient mice , the assembly of TG into lipoproteins is normal until some degree , but becomes severely inhibited when high amounts of TGs have to be transported . 10 . 7554/eLife . 06328 . 026Figure 10 . The levels of TGs are decreased in plasma lipoproteins of LPCAT3-deficient mice . ( A–D ) The levels of apoB ( A ) , TG ( B ) , cholesterol ( C ) , and PC ( D ) were measured in plasma of wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 ( n = 10 ) . ( E–J ) Plasma lipoproteins from wild type and LPCAT3-deficient mice were fractionated by gel-filtration chromatography ( n = 5 ) . TG ( E and G ) and cholesterol ( F and H ) levels in VLDL/IDL/LDL fractions ( E and F ) or chylomicron fractions ( G and H ) were calculated from the raw data shown in Figure 10—figure supplements 1–H . The differences in ( E ) did not reach statistical significance , but the changes in some subfractions in the raw data were significant . ( I and J ) The ratio of TG to cholesterol was calculated in chylomicron ( I ) and VLDL/IDL/LDL ( J ) fractions . ( K–M ) TG ( K ) , cholesterol ( L ) , and PC ( M ) concentration in apoB-containing lipoproteins precipitated from plasma of wild type and LPCAT3-deficient mice ( n = 3 ) . Values were normalized to apoB to estimate the amount of each component per particle . See also Figure 10—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 02610 . 7554/eLife . 06328 . 027Figure 10—figure supplement 1 . Fractionation of plasma lipoproteins by gel-filtration chromatography . Plasma lipoproteins were fractionated by gel-filtration chromatography ( n = 5 ) . ( A–H ) TG and cholesterol concentrations were analyzed in plasma lipoprotein subfractions from wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 ( A and E ) , P0 . 4 ( B and F ) , P0 . 8 ( C and G ) , P1 . 5 ( D and H ) . The data are raw data used for calculations in Figure 9E–J . ( I ) PC concentration in plasma lipoprotein subfractions of wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 . The annotation of fractions is based on the elution profile of a control plasma . CM: chylomicron; HDL: high-density lipoproteins . Error bars are SEM . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , **p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 027 To analyze lipoprotein composition more in detail , we used dextran sulfate to precipitate apoB-containing lipoproteins based on their surface charge ( Warnick et al . , 1982 ) . We quantified TG , cholesterol , and PC levels relative to apoB in the precipitates . Consistent with the results from gel filtration chromatography , apoB-associated TG , but not cholesterol or PC , was decreased in LPCAT3-deficient mice ( Figure 10K–M ) . The normal PC level was confirmed by gel filtration chromatography ( Figure 10—figure supplement 1I ) . Taken together , we conclude that in LPCAT3-deficient mice , TGs are not efficiently used for lipoprotein assembly when high TG amounts have to be transported , leading to cytosolic lipid droplet accumulation and cellular damages . Finally , we investigated how LPCAT3 might affect TG transport . It was unlikely that PC synthesized by LPCAT3 is a major source of lipoprotein PC , since their levels were unchanged in LPCAT3-deficient mice ( Figure 10D , M , and Figure 10—figure supplement 1I ) . In addition , although the acyl-chain composition of PC in apoB-containing lipoproteins was slightly different in LPCAT3-deficient mice , PC species with arachidonate were relatively minor components of lipoproteins , irrespectively of the genotype ( Figure 11—figure supplement 1 ) . In addition to these observations , the normal levels of proteins involved in lipoprotein assembly ( Figure 7—figure supplement 2 and Figure 9—figure supplement 2 ) suggested that a previously unknown mechanism should explain the impaired TG transport . Since the membrane environment around LPCAT3 should be enriched in PUFAs ( see ‘Discussion’ ) , we analyzed whether a PUFA-rich membrane affects the clustering and transport of TG . When present at a high concentration in a limited space , the fluorescence of the fluorophore 7-nitro-2 , 1 , 3-benzoxadiazole ( NBD ) decreases due to its self-quenching property ( Brown et al . , 1994 ) . Therefore , the quenching of NBD-labeled TG ( NBD-TG ) in PC liposomes was measured to analyze the local clustering between leaflets ( Figure 11A ) . The total fluorescence ( without quenching ) was measured by disassembling liposomes in isopropanol ( Figure 11A , B ) . In the absence of PC , we did not observe NBD-TG fluorescence in the samples , showing that free NBD-TG is lost during liposome preparation ( Figure 11B ) . To prepare liposomes , we used egg PC , which contains only a slight amount of PUFAs ( Figure 11—figure supplement 2 ) , or egg PC supplemented with 30% of synthetic PC containing palmitate at the sn-1 position , and oleate , linoleate , arachidonate , or docosahexaenoate at the sn-2 position . We calculated the self-quenching of NBD-TG from the fluorescence of liposomes in buffer or isopropanol ( Figure 11B ) , and found that a membrane rich in PUFAs such as arachidonate or docosahexaenoate increases quenching ( Figure 11C ) , equivalent to local TG clustering . 10 . 7554/eLife . 06328 . 028Figure 11 . A PUFA-rich membrane promotes TG clustering and efficient transfer . ( A ) Outline of TG clustering analysis based on fluorescence quenching . When TG clustering is higher , quenching increases and fluorescence decreases . Isopropanol disrupts liposomes and TG clustering , leading to fluorescence without quenching . ( B and C ) Fluorescence ( B ) and the quenching rate ( C ) of NBD-TG were analyzed in PC liposomes of different compositions . ( D ) Outline of the MTP assay , measuring the transfer of NBD-TG from ‘light’ donor liposomes to ‘heavy’ acceptor liposomes . ‘Heavy’ acceptor liposomes are prepared in the presence of sucrose , and can be pelleted by centrifugation . ( E ) Fluorescence of NBD-TG in acceptor liposomes , after incubation for the indicated time with donor liposomes and liver extracts ( or boiled extracts ) . NBD-TG is transferred to acceptor liposomes after incubation with liver extracts for 1 hr . The difference is used for calculation of TG transfer efficiency . ( F ) TG transfer efficiency was measured using donor PC liposomes of different compositions . Error bars are SEM ( n = 3 ) . p < 0 . 05 , p < 0 . 01 , *p < 0 . 001 . See also Figure 11—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 02810 . 7554/eLife . 06328 . 029Figure 11—figure supplement 1 . PC acyl-chain composition of lipoproteins . PC acyl-chain composition of apoB-containing lipoproteins precipitated from plasma of wild type and LPCAT3-deficient mice at E18 . 5–E19 . 5 ( n = 3 ) . Note that the major species can be estimated to contain at least one monounsaturated fatty acid ( e . g . , 16:0–18:1 for 34:1 PC , 18:0–18:1 for 36:1 PC , and 18:1–18:1 for 36:2 PC ) , and that species that might contain arachidonate ( 36:4 PC and 38:4 PC ) are minor . Error bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 02910 . 7554/eLife . 06328 . 030Figure 11—figure supplement 2 . Acyl-chain composition of egg PC . Acyl-chain composition of egg PC used in the TG clustering and TG transfer assays ( single experiment ) . Species containing PUFAs are minor . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 030 Next , we analyzed whether this local pool of clustered TGs facilitates their transfer by MTP from donor liposomes to acceptor liposomes ( Figure 11D ) . The acceptor liposomes , which contained sucrose and were made heavier , were separated from donors by centrifugation , and the fluorescence in the pellet was measured to detect the transfer ( Figure 11D , E ) . Incubation of liver extracts with donor and acceptor liposomes for 1 hr increased the fluorescence of NBD-TG in acceptor liposomes , reflecting TG transfer by MTP activity ( Figure 11E ) . Using this assay , we found that high levels of PUFAs in donor liposomes facilitate TG transfer by MTP ( Figure 11F ) . Therefore , the local enrichment of membrane PUFAs by LPCAT3 might facilitate the clustering of TGs and their transport by MTP . Although LPC levels were increased in LPCAT3-deficient mice , LPC ( 1–5% of PC in liposome ) did not affect TG clustering and transfer , at least under the present assay conditions ( data not shown ) . In this study , we first analyzed in detail how LPCAT3 affects LPLAT activity and lipid composition . Although multiple LPLATs have been identified ( Hishikawa et al . , 2014 ) , we found that LPCAT3-deficient mice lack most of the activity required for the remodeling of PC , PE , and possibly PS ( Figures 2G , 4F , and data not shown ) . The remodeling pathway has been thought to modify the acyl-chain composition of phospholipids synthesized de novo and to be especially important for PUFA accumulation ( Hill and Lands , 1968; MacDonald and Sprecher , 1991; Lands , 2000 ) . Analyses of LPCAT3-deficient mice revealed complex changes in the lipidome , but among the major fatty acids , the most commonly decreased one was arachidonate ( Figures 3–5 ) . Our results show that the remodeling by LPCAT3 is highly selective for arachidonate accumulation and that many PUFA species persist in membranes of LPCAT3-deficient mice ( Figures 3–5 ) . Therefore , the remodeling by LPCAT3 does indeed modulate the acyl-chain composition of de novo synthesized phospholipids , but is not completely requisite for PUFA accumulation . The observation that C22 and C24 n-6 PUFAs that arose after arachidonate elongation accumulate in LPCAT3-deficient mice ( Figure 3—figure supplement 2 ) shows that this enzyme utilizes arachidonoyl-CoA competitively with other enzymes . C22 n-6 PUFAs increased not only in phospholipids but also in neutral lipids ( Figure 7B ) . Since TG and phospholipids share phosphatidic acid as the same precursor ( Coleman and Lee , 2004 ) , this suggests that in LPCAT3-deficient mice , the C22 n-6 PUFA-CoAs are utilized by LPAATs during de novo synthesis , together with one part of the excess of arachidonoyl-CoA ( Figure 12 ) . The unchanged docosahexaenoate in neutral lipid fractions ( Figure 7B ) suggests that its increase in phospholipids ( Figure 7A ) is not due to an accumulation by LPAATs , but rather suggests that docosahexaenoate-containing phospholipids were not utilized for acyl-chain remodeling in LPCAT3-deficient mice . Therefore , we propose that the increased n-6 C22 PUFAs and C24 PUFAs in phospholipids of LPCAT3-deficient mice are due to increased incorporation during de novo synthesis ( Figure 12 ) , although we cannot exclude the possibility that unknown LPCAT enzymes selective for these PUFAs exist . In addition to the changes in phospholipid acyl-chain composition , LPCAT3 deficiency induced changes in LPC ( in liver and small intestine ) or total PC ( in proximal small intestine ) levels , but only in tissues where LPCAT3 expression is high ( Figures 2B , 3A , H ) . LPCAT3 deficiency might have caused a pronounced imbalance between LPC and PC in these tissues . 10 . 7554/eLife . 06328 . 031Figure 12 . Proposed model of arachidonoyl-CoA shunting in LPCAT3-deficient mice . During de novo synthesis , TG and PC share the same precursor , diacylglycerol ( DG ) . Therefore , the fatty acid profile of neutral lipids is suggestive of incorporation during de novo synthesis of both TG and phospholipids , assuming that TG has a negligible degree of acyl-chain remodeling . Based on the enzymatic assays and the fatty acid profiles of phospholipids and neutral lipids , we propose the following explanation for the results . ( A ) In wild type mice , arachidonoyl-CoA is largely utilized by LPCAT3 to be incorporated into phospholipids . ( B ) In LPCAT3-deficient mice , an excess of arachidonoyl-CoA occurs , which is utilized by LPAATs for de novo synthesis of both phospholipids and triglycerides , either directly or after being metabolized into C22 and C24 n-6 PUFAs ( Figure 3—figure supplement 2 ) . This pathway is less utilized in wild type mice . The monoacylglycerol acyltransferase pathway of TG synthesis is ignored in this figure , since differences in TG profiles should originate from those synthesized by the embryo and not those provided by the mother bloodstream . G3P: glycerol 3-phosphate; LPA: lysophosphatidic acid; PA: phosphatidic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 031 It is of note that when comparing different tissues , the extent of arachidonate decrease in LPCAT3-deficient mice does not correlate with the expression levels of this enzyme ( Figures 2A , 5A ) . Even in tissues such as the brain and heart , where LPCAT3 expression is relatively small , arachidonate levels were prominently reduced in LPCAT3-deficient mice . Therefore , LPCAT3 is required for accumulating arachidonate , but the expression level of this enzyme is not the single factor to explain different arachidonate levels in tissues . Rather , the different levels should be explained by both LPCAT3 activity and supply of arachidonoyl-CoA , suggesting that tissue arachidonate levels are regulated in the remodeling pathway with dependence on substrate supply , as we recently proposed ( Harayama et al . , 2014 ) . In addition , the observation that eicosanoids are unchanged in LPCAT3-deficient mice ( Figure 6 ) suggests that the phenotypes of LPCAT3-deficient mice are unrelated to these lipid mediators . Therefore , analyses of LPCAT3-deficient mice revealed novel functions of membrane arachidonate that are not attributed to eicosanoids . In the future , it is of interest to determine eicosanoid levels in stimulated cells , where phospholipase A2 is activated ( Shimizu , 2009; Dennis et al . , 2011 ) . In addition , since changes in linoleate and docosahexaenoate levels were also observed in some tissues ( Figure 5B , C ) , it will be interesting to investigate other lipid mediators , such as hydroxyoctadecadienoic acids , resolvins , and protectins ( Obinata et al . , 2005; Buckley et al . , 2014 ) . We next showed that LPCAT3-deficient mice accumulate cytosolic lipid droplets ( Figures 7 , 9 ) in hepatocytes and enterocytes due to an inefficiency of luminal TG transport by MTP . The quenching assay ( Figure 11C ) suggests that this inefficiency is caused by the differences in TG clustering , which is facilitated by the enrichment of PUFAs in a local membrane by LPCAT3 ( Figure 13A ) . In this section , we will describe how we reached this conclusion from the literature and our results . 10 . 7554/eLife . 06328 . 032Figure 13 . A proposed mechanism of TG transport facilitated by LPCAT3 . ( A ) Differences in PUFA-containing phospholipids distribution between wild type and LPCAT3-deficient mice . Since PUFAs are shunted into de novo synthesis ( Figure 12 ) , the total levels of PUFAs are similar in both genotypes . However , since the remodeling by LPCAT3 is highly selective for arachidonate , arachidonate-containing phospholipids are concentrated around LPCAT3 , generating a local membrane area rich in PUFAs . ( B and C ) Proposed mechanism of TG transport facilitated by LPCAT3 . ( B ) LPCAT3 locally enriches PUFA-containing phospholipids . When TG synthesis occurs in the proximity , this PUFA-enriched domain favors the formation of a blister-like structure with a high capacity and high surface curvature . TG in the blister-like structure has a high flip-flop rate , enabling efficient supply to MTP at the luminal side . ( C ) The substrate selectivity of the final step of PC synthesis does not favor the generation of a PUFA-enriched domain . LPL: lysophospholipid; DG: diacylglycerol; DGAT: diacylglycerol acyltransferase; CPT: diacylglycerol cholinephosphotransferase . DOI: http://dx . doi . org/10 . 7554/eLife . 06328 . 032 Intestine-specific MTP deficiency leads to ( at least partial ) neonatal lethality , which was explained by lipotoxicity ( see the tamoxifen-treated apoB48 Mttp-IKO mice in [Xie et al . , 2007] ) . LPCAT3-deficient mice display similar lethality and enterocyte lipid accumulation , and have less TGs in plasma lipoproteins ( Figures 8–10 ) . Therefore , we attribute the lethality of LPCAT3-deficient mice to the intestinal lipotoxicity caused by the intake of large amounts of milk TG under MTP inefficiency . Indeed , enterocyte damage was obvious in LPCAT3-deficient mice , and probably resulted in an inability to absorb nutrients , not limited to sugars and lipids . One difference between LPCAT3-deficient mice and MTP-deficient mice is that the former assemble cholesterol normally in lipoproteins ( Figure 10F , H ) . This suggests that LPCAT3 deficiency does not directly inhibit MTP , but may affect the properties of the MTP substrate TG . The quenching assay supports this , and one of these properties is the clustering of TG between leaflets ( Figure 11C ) . It is of note that TG absorption is normal in LPCAT3-deficient mice at P0 . 4 ( Figure 10G ) . This suggests that the difference in transport efficiency is apparent only when TGs are in high amounts and require to be densely clustered in the membrane for efficient transport . Therefore , it is possible that cholesterol absorption is also affected when feeding a high cholesterol diet . Comparison with reports of LPCAT3 knockdown in liver led us to focus on the synthesis of PUFA-containing phospholipids in a ‘local membrane microenvironment’ . In a previous report , LPCAT3 knockdown did not impair TG transport , but rather increased VLDL secretion by the liver , due to the induction of MTP caused by increased LPC ( Li et al . , 2012 ) . Despite this opposite result , the changes in the lipidome had similarities; LPC increased and arachidonate-containing PC decreased ( Figure 3 ) ( Li et al . , 2012; Rong et al . , 2013 ) . Therefore , these similar changes cannot explain the impaired TG transport in LPCAT3-deficient mice . LPCAT3 is ideal to generate a local high PUFA concentration , since the enzyme is very selective for linoleoyl- and arachidonoyl-CoA ( Figures 1B , 13A , B ) ( Hishikawa et al . , 2008; Zhao et al . , 2008 ) . On the other hand , the enzyme catalyzing the last step of PC de novo synthesis , diacylglycerol cholinephosphotransferase ( CPT ) is not selective for diacylglycerol species containing PUFA ( Mantel et al . , 1993 ) , and thus is unable to do the same ( Figure 13C ) . Therefore , local PUFA-rich domains arise continuously around LPCAT3 in wild type mice , but not in LPCAT3-deficient mice ( Figure 13A ) . Residual LPCAT3 in knockdown experiments might be sufficient for this local PUFA enrichment . The results of Figure 11 suggest that if TG synthesis by diacylglycerol acyltransferases occurs in proximity of a local PUFA-rich domain , TG clustering will occur , providing a pool for efficient transport by MTP ( Figure 13B ) . Although further studies are required for a complete understanding , the different results between the LPCAT3 knockdown study ( Li et al . , 2012; Rong et al . , 2013 ) and our present study might be due to the ability to accumulate PUFAs locally . The observation that not only arachidonate but also docosahexaenoate promote TG clustering and transfer , as well as the fact that total PUFA levels are not largely changed in LPCAT3-deficient mice , again suggest the necessity of considering a local high concentration . In the presence of LPCAT3 , PUFAs might be unevenly distributed in the membrane , and this should be required to reach a sufficient high PUFA concentration for efficient TG clustering and transfer . Therefore , although docosahexaenoate can substitute arachidonate in in vitro TG transfer assays , its inability to be locally concentrated by LPCAT3 make this fatty acid incompetent for lipoprotein assembly . Other mouse models that have abnormal PUFA metabolism are mice that are deficient in fatty acid desaturase 1 ( FADS1 ) or FADS2 . These enzymes induce double bonds at different steps of PUFA generation , and mice lacking either gene encoding them could not synthesize arachidonate from linoleate . Knockout mice of both FADS1 and FADS2 have been generated , and displayed multiple abnormalities such as lethality , decreased intestinal crypt proliferation , and sterility ( Stoffel et al . , 2008; Stroud et al . , 2009; Fan et al . , 2012 ) . However , the accumulation of intestinal lipid droplets was not reported . It is possible that these mice might receive enough arachidonate from breast milk during the lactating period , making it difficult to analyze intestinal arachidonate functions at the perinatal period . Indeed , most studies report abnormalities of FADS1/2-deficient mice after weaning . We speculate that LPCAT3 is required for intestinal TG absorption especially when TG levels in food are high , as is the case in milk . Therefore , it is intriguing to observe phenotypes by feeding these mice with high-fat diet . It is of note that lipid droplet accumulation in the liver is seen in FADS2-deficient mice , and that arachidonate feeding reverses this phenotype ( Stoffel et al . , 2014 ) , which may fit with our proposed model . It will be important to compare these different models in detail to unveil more functions of membrane PUFAs in future studies . The quenching assay shows that a PUFA-rich membrane enables a dense clustering of TGs . This composition may favor the deformation of the membrane surrounding TGs in a ‘blister-like’ morphology , with high local curvature ( Khandelia et al . , 2010 ) . When compared with a membrane rich in monounsaturated fatty acids , a PUFA-rich membrane is more easily deformed when it is bent ( Vanni et al . , 2013; Pinot et al . , 2014 ) . Based on this theory , it can be imagined that when TG synthesis occurs intensively , there is a demand to generate a highly curved blister-like structure to store more TGs in a limited place for efficient transport by MTP , and that a PUFA-rich environment generated by LPCAT3 helps this process by making the membrane more flexible and mobile ( Figure 11A ) . Our model suggests that one benefit of regulating acyl-chain composition is to adapt membrane flexibility to a required curvature . When we analyzed plasma lipoprotein PC , we detected many monounsaturated species ( Figure 11—figure supplement 1 ) . This composition is similar to that reported for PC surrounding cytosolic lipid droplets ( Tauchi-Sato et al . , 2002 ) . Therefore , in the case of a membrane surrounding TGs with a large diameter , the composition is shifted to one with more monounsaturated species , which makes the membrane less flexible . This suggests that membranes surrounding TGs acquire acyl-chains that fit well with the curvature required . It is unknown how the local clustering affects TG transfer . Molecular dynamics simulations suggested that when TG is clustered , the efficiency of its flip-flop between leaflets is accelerated ( Khandelia et al . , 2010 ) . Since MTP is present only in the luminal side of the bilayer , it is likely that the accelerated flip-flop enables TG to be present in the leaflet that is accessible by MTP ( Figure 13B ) . Future studies would clarify how MTP is acting at the membrane interface to answer how membrane PUFAs enable efficient TG transfer . In cases where many lipids are ingested ( as in the neonatal period ) , the efficient export of TG is important not only for nutrition , but also to prevent lipotoxicity on enterocytes ( Xie et al . , 2007 ) . The generation of an arachidonate-enriched membrane microenvironment by LPCAT3 might be required for efficient clustering of TGs and their correct transport ( Figure 13B ) . Previously , the functions of membrane arachidonate that are not related to eicosanoids were largely unknown . In addition , the mechanism and factors that enable MTP to efficiently transfer TGs unidirectionally into the ER lumen were poorly understood . Our study reveals for the first time that membrane arachidonate and TG clustering are critical factors to regulate the correct luminal directionality of TG transport , which has lethal consequences when improperly regulated . Therefore , the present study identifies LPCAT3 as a novel key molecule involved in normal lipoprotein assembly . During the final editing of the revised manuscript , Rong et al . reported LPCAT3-deficient mice with similar conclusions . In the future , conditional ablation of LPCAT3 in different contexts will reveal other functions of arachidonate in phospholipids and will provide further critical insights into membrane biology . Expression plasmids for FLAG-LPCAT3 and FLAG-LPCAT3 H374A were generated previously in a pCXN2 expression vector , which expresses the inserted construct under the control of a CAG promoter and contains a neomycin-resistance gene ( Niwa et al . , 1991; Hishikawa et al . , 2008; Shindou et al . , 2009 ) . The empty vector for the expression of single guide RNAs , pX459 was obtained from Addgene . Oligos that are listed in a supplementary table were annealed and inserted into the Bbs I sites of pX459 ( Addgene , Cambridge , MA , plasmid 48139 ) by Golden Gate assembly ( Engler et al . , 2008 ) as previously described ( Ran et al . , 2013 ) , using Bpi I ( Thermo Fisher Scientific K . K . , Kanagawa , Japan ) and T4 DNA ligase ( New England Biolabs , Ipswich , MA ) . Plasmids were transformed in ECOS competent Escherichia coli JM109 ( Nippon Gene Co . , Ltd . , Toyama , Japan ) . Sequences were verified by a sequencing service ( Eurofins Genomics K . K . , Tokyo , Japan ) , and plasmids were purified using a Plasmid Midi Kit ( Qiagen K . K . , Tokyo , Japan ) . RH 7777 cells were cultured on IWAKI collagen-coated dishes ( Asahi Glass Co . , Ltd . , Tokyo , Japan ) in Dulbecco's Modified Eagle Medium ( Nacalai Tesque , Inc . , Kyoto , Japan ) supplemented with 10% GIBCO fetal bovine serum ( Life Technologies Japan Ltd . , Tokyo , Japan ) , in a humidified incubator at 37°C with 5% CO2 . For lipid analysis , cells were cultured for 24 hr in the presence of 10 μM each of linoleate , arachidonate , and docosahexaenoate ( Cayman Chemical Company , Ann Arbor , MI ) to make the changes in PUFA-containing PC more easily distinguished ( although most of the changes reported in the manuscript can be seen without this supplementation ) . Lipofectamine 3000 ( Life Technologies ) was used for transfection . For the establishment of stable transfectants , cells were selected in 2 mg/ml G418 ( Life Technologies ) for 1 week , starting at 24 hr post transfection . Selected cells were maintained in medium containing 0 . 3 mg/ml G418 . LPCAT3-null cells were established by transfecting a pair of single guide RNAs , inserted in the expression vector pX459 ( Addgene , plasmid 48 , 139 ) . Transfection with this vector leads to the simultaneous expression of single guide RNAs , Cas9 , and a puromycin resistance gene . Single guide RNAs were designed to flank the region coding the WHG sequence of rat LPCAT3 and cause a ∼100 bp deletion in Lpcat3 gene ( Figure 1—figure supplement 3A ) . The WHG sequence is conserved in all members of the membrane-bound O-acyltransferase family ( Hofmann , 2000 ) , and is required for LPCAT3 activity ( Shindou et al . , 2009 ) . Cells that were transfected transiently were selected by puromycin ( 10 μg/ml , InvivoGen , San Diego , CA ) for 24 hr . Clones were obtained by limiting dilution on 96 well plates . Clones that contained deletion in the Lpcat3 locus were screened by PCR using genomic DNA of each clone as a template , and ExTaq HS DNA polymerase ( Takara Bio Inc . , Shiga , Japan ) ( Figure 1—figure supplement 3A ) . Cultured cells were scraped in ice-cold T20 buffer ( 20 mM Tris-HCl [pH7 . 4 , Wako Pure Chemical Industries , Ltd . , Osaka , Japan] , 300 mM sucrose [Wako] , and a proteinase inhibitor mixture , Complete [Roche Diagnostics K . K . , Tokyo , Japan] ) and sonicated using a probe sonicator ( Ohtake Works , Tokyo , Japan ) . Frozen tissues ( 1–100 mg ) were homogenized in ice-cold T100 buffer ( 100 mM Tris-HCl ( pH7 . 4 ) , 300 mM sucrose , and Complete ) using a Physcotron homogenizer ( Microtec Co . Ltd . , Chiba , Japan ) . The homogenate was centrifuged at 800×g for 10 min . The supernatant was used for western blot analysis of MTP and PDI . The same supernatant was centrifuged at 100 , 000×g for 1 hr to obtain membrane fractions . The pellet was resuspended in TSE buffer ( 20 mM Tris-HCl [pH7 . 4] , 300 mM sucrose , and 1 mM EDTA [Dojindo Laboratories , Kumamoto , Japan] ) . This membrane fraction was used for western blot analysis of LPCAT3 and the enzymatic assays . Protein concentration was measured using the Bio-Rad Protein Assay ( Bio-Rad Laboratories , Inc . , Hercules , CA ) . Protein samples were snap frozen in liquid nitrogen and stored at −80°C until use . Protein samples were resolved on 10% SDS-polyacrylamide gels and electrophoretically transferred to nitrocellulose membranes ( GE Healthcare UK Ltd . , Buckinghamshire , England ) using a Trans-Blot SD semi-dry transfer cell ( Bio-Rad ) . Membranes were stained with Ponceau S ( Sigma–Aldrich Co . LLC . , St . Louis , MO ) , and then blocked overnight with 5% skim milk ( BD Biosciences , Franklin Lakes , NJ ) in Tris-buffered saline with 0 . 1% Tween 20 ( Wako ) ( TBST ) . Primary antibodies were diluted in 5% skim milk/TBST as following: anti-LPCAT3 ( 40 ng/ml ) , anti-FLAG M2 antibody ( 5 μg/ml , Sigma–Aldrich ) , anti-MTP antibody ( 50 ng/ml , BD ) , or anti-PDI antibody ( 1:1000 dilution , Cell Signaling Technology , Inc . , Danvers , MA ) . Horseradish peroxidase-conjugated secondary antibodies ( GE Healthcare ) were used at a 1:2000 dilution in 5% skim milk/TBST . TBST was used for washing steps and changed at least three times between incubation steps . ECL select western blot detection system ( GE Healthcare ) was used for chemiluminescence , and detected using ImageQuant LAS500 ( GE Healthcare ) . LPCAT assays were performed as previously described , in a condition that provides linearity ( Harayama et al . , 2014; Martin et al . , 2014 ) . Briefly , membrane proteins ( 0 . 01 μg/tube ) were mixed with 25 μM deuterium-labeled 16:0 LPC or non-labeled 17:1 LPE and 1 μM each of 16:0- , 18:1- , 18:2- , 20:4- , and 22:6-CoA at 37°C for 10 min ( all from Avanti Polar Lipids , Inc . , Alabaster , AL ) . Reaction mixtures contained 110 mM Tris-HCl ( pH 7 . 4 ) , 1 . 5 mM EDTA , 2 mM CaCl2 ( Wako ) , 0 . 015% Tween-20 , and 150 mM sucrose in a total volume of 100 μl . Reactions were stopped by the addition of 300 μl chloroform/methanol ( 1/2 , Wako ) . Internal standards ( dilauryl-PC for LPCAT assay and dimyristoyl-PE for LPEAT assay , Avanti ) were added , lipids were extracted by the method of Bligh and Dyer ( Bligh and Dyer , 1959 ) , dried using a centrifugal evaporator ( Sakuma Seisakusho Ltd . , Tokyo , Japan ) , and reconstituted in methanol . Products were measured by LC-MS . Products were separated on an ACQUITY UPLC BEH C8 column ( 1 . 7 μm , 2 . 1 × 30 mm , Waters Corporation , Milford , MA ) using a linear gradient of solvent B ( acetonitrile , Wako ) over solvent A ( 20 mM NH4HCO3/water , Wako ) , using an ACQUITY ultra performance liquid chromatography ( UPLC ) system ( Waters ) . The flow rate was 800 μl/min . The gradient started at 55% solvent B , was linearly increased to 95% solvent B in 4 . 5 min , and maintained for 1 . 5 min . Detection was done on a TSQ Vantage triple stage quadrupole mass spectrometer ( Thermo Fisher Scientific ) by selected reaction monitoring ( SRM ) . Transitions were [M + H]+ → 184 . 1 for PC and [M + H]+ → [M + H-141]+ for PE ( both in the positive ion mode electrospray ionization ) . Signals of LPCAT products were compared to calibration curves of nonlabeled standards for quantification . When total LPCAT activity is shown , it illustrates the sum of the five products generated during the assay . Standards for LPEAT products were not available , and peak areas normalized to the internal standard were used as a measure of enzymatic activity . Lipids from cultured cells were obtained using methanol . Lipids from precipitated lipoproteins were extracted by the method of Bligh and Dyer . Egg PC ( Sigma–Aldrich ) in ethanol ( Wako ) was used after dilution in methanol . Lipids were separated on an ACQUITY UPLC BEH C8 column ( 1 . 7 μm , 1 . 0 × 100 mm ) using a linear gradient of solvent B ( acetonitrile ) over solvent A ( 20 mM ammonium bicarbonate ) , using an ACQUITY UPLC system . The flow rate was 100 μl/min . The gradient started at 20% solvent B , was linearly increased to 95% solvent B in 20 min , and maintained for 15 min . PC was detected using a precursor ion scanning for m/z 184 . 1 in the positive ion mode electrospray ionization using a TSQ Vantage triple stage quadrupole mass spectrometer . This method does not resolve the acyl chains at the sn-1 and sn-2 positions . To characterize the acyl-chain composition of selected peaks , SRM was additionally performed in the negative ion mode . The transitions were [M + HCO3]− → 255 . 2 for species containing palmitate , and [M + HCO3]− → 283 . 2 for those containing stearate . Although this method is more specific in the discrimination of acyl-chains , we obtained a better linear dynamic range in the positive ion mode , and thus used the precursor ion scanning for the figure in this manuscript . Peak areas of all detected diacyl-PC species were summed , and the ratio of individual species ( as % of total ) was calculated . See Figure 1—figure supplement 2 for an example of peak annotation using different detection methods . When using LC-MS methods that detect only the sum of both sn-1 and sn-2 fatty acids in phospholipids , the molecule is illustrated as XX:Y , where XX is carbon number and Y is double bond number , as a sum of both acyl-chains . When additional detection of fatty acid fragments is performed to confirm the acyl-chain composition , the molecule is illustrated as AA:B-CC:D , where AA and CC are carbon numbers , and B and D are double bond numbers . Note that the methods do not discriminate which sn- position each fatty acid resides ( meaning that the above phospholipid might also have been CC:D-AA:B ) . All animal experiments were approved by and performed in accordance with the guidelines of the Animal Research Committee of National Center for Global Health and Medicine ( 12 , 053 , 13 , 009 , 14 , 045 ) , and the animal experimentation committee of the University of Tokyo ( H09-144 , P08-042 ) . Newborn mice and embryos were euthanized by rapid decapitation and had laparotomy . Tissues were harvested and were snap frozen in liquid nitrogen and stored at −80°C until RNA and protein extraction . For histological analyses , tissues were fixed as described in each section . Frozen tissues ( 0 . 5–100 mg ) were homogenized using a Physcotron homogenizer in QIAzol Lysis Reagent ( Qiagen ) and total RNA was extracted using an RNeasy Mini Kit ( Qiagen ) . Complementary DNA synthesis was carried out using SuperScript III reverse transcriptase ( Life Technologies ) using 1 μg total RNA as a template . Real-time quantitative PCR was performed using Fast SYBR Green Master Mix and the Step One Plus real-time PCR system ( Life Technologies ) , using the primers listed in a supplementary table . Anti-LPCAT3 antibody was produced by Sigma–Aldrich . Rabbits were immunized with a C-terminal LPCAT3 peptide ( CHKAMVPRKEKLKKRE ) . Blood was collected and antiserum was obtained . The same C-terminal peptide was immobilized on activated thiol-Sepharose 4B ( GE Healthcare ) , and anti-LPCAT3 antibody was affinity purified using this column . LPCAT3-floxed mice ( Accession No . CDB0653K: http://www . cdb . riken . jp/arg/mutant%20mice%20list . html ) were generated as described ( http://www . cdb . riken . jp/arg/Methods . html ) using the HK3i embryonic stem cell line ( Kiyonari et al . , 2010 ) . To generate a targeting vector , genomic fragments of the Lpcat3 locus were obtained from the RP23-388D4 BAC clone ( BACPAC Resources Children's Hospital Oakland , St . Oakland , CA ) . A 903 bp region containing exons 10 , 11 , and 12 of the Lpcat3 gene was flanked by loxP sites ( Figure 2—figure supplement 1A ) . Targeted ES clones were microinjected into ICR 8-cell stage embryos , and injected embryos were transferred into pseudopregnant ICR females . The resulting chimeras were bred with C57BL/6 mice , and heterozygous offsprings were identified by PCR . Exons 10 to 12 of Lpcat3 were removed by mating these mice with C57BL/6 telencephalin-Cre mice ( Nakamura et al . , 2001; Fuse et al . , 2004 ) . For genotyping , DNA was extracted from tail tips and subjected to PCR using Ex Taq HS DNA polymerase ( Takara ) . Primer sequences are described in a supplementary table . Frozen tissues ( 0 . 5–100 mg ) were pulverized and 0 . 8 ml methanol was added . The tissue suspensions were centrifuged at 15 , 000×g , and the collected supernatant was used for the measurement of phospholipids . PC concentration was measured by an enzyme-based fluorescent assay with minor modifications from a previous report ( Morita et al . , 2010 ) . Samples were dried in a centrifugal evaporator and dissolved in 0 . 16 ml 1% Triton X-100 ( Wako ) per mg wet tissue weight . 10 μl of sample was reacted with 40 μl mixture C1 ( PC-Specific Phospholipase D [1:200 dilution , Cayman] , 1 . 5 mM CaCl2 , 50 mM NaCl [Wako] , and 50 mM Tris-HCl [pH 7 . 4] ) at 37°C for 30 min . After generation of free choline from PC , 50 μl mixture C2 ( 4 U/ml choline oxidase from Alcaligenes sp . [Wako] , 5 U/ml horseradish peroxidase [Oriental Yeast Co . , ltd . , Tokyo , Japan] , 0 . 3 mM Amplex Red ( Life Technologies ) , 0 . 2% Triton X-100 , 50 mM NaCl and 50 mM Tris-HCl [pH7 . 4] ) were added and incubated at room temperature for 30 min . The hydrogen peroxide generated by choline oxidase was fluorescently detected using Amplex Red reagent and measured with ARVO X3 ( PerkinElmer Inc . , Waltham , MA ) . For phospholipid analysis , the methanol extracts were further diluted with methanol to adjust the concentration to 10 mg tissue/ml ( liver , brain , and lung ) , 7 mg tissue/ml ( kidney ) , 5 mg tissue/ml ( proximal small intestine ) , 3 . 5 mg tissue/ml ( heart ) , 3 mg tissue/ml ( distal small intestine and stomach ) , 1 . 5 mg tissue/ml ( thymus ) , 1 mg tissue/ml ( colon ) , and 0 . 5–2 . 7 mg tissue/ml ( spleen ) . Embryonic spleens were very small with high variability , and were used without adjustment of concentration . This might have led to the difference in total signal in Figure 3—figure supplement 1 . LC-SRM-MS analysis was performed using a Nexera UHPLC system and triple quadrupole mass spectrometers LCMS-8040 or LCMS-8050 ( Shimadzu Corporation , Kyoto , Japan ) . For phospholipid analysis , an Acquity UPLC BEH C8 column ( 1 . 7 μm , 2 . 1 mm × 100 mm , Waters ) was used with the following ternary mobile phase compositions: 5 mM NH4HCO3/water ( mobile phase A ) , acetonitrile ( mobile phase B ) , and isopropanol ( mobile phase C ) . Pump gradient [time ( %A/%B/%C ) ] was programmed as follows: 0 min ( 75/20/5 ) –20 min ( 20/75/5 ) –40 min ( 20/5/75 ) –45 min ( 5/5/90 ) –50 min ( 5/5/90 ) –55 min ( 75/20/5 ) . Flow rate was 0 . 35 ml/min and column temperature was 47°C . Injection volume was 5 μl . SRM analysis with phospholipid class discrimination was performed with the following transitions: [M + H]+ → 184 for PC , [M + H]+ → [M + H-141]+ for PE , [M-H]− → [M-H−87]− for PS , [M-H]− → 241 for PI . Peak areas of all the detected species were summed to obtain the total signal . Peak areas of individual species were normalized with this sum , and are illustrated as % of total . For the comparison of Figure 3B , PC species that were detected in all samples were selected . The ratio between LPCAT3-deficient mice and wild type mice ( as % of wild type ) was calculated for each species in every tissues . These values were further averaged for all tissues to detect the changes that occurred globally . As in the case of lipid analysis for cultured cells , this method does not resolve the acyl chains at the sn-1 and sn-2 positions . For relevant phospholipid species , additional SRM analyses with selection of fatty acid fragments at Q3 were performed to confirm that they contain the fatty acid of interest . Detection was performed at the negative ion mode using the following transitions: [M + HCO3]− → [FA-H]− for PC , [M-H]− → [FA-H]− for PE , where [FA] is the monoisotopic mass of the fatty acid of interest . We set up SRM channels for all possible acyl-chain composition , based on the assumption that the following fatty acids are present in vivo: C12:0 , C12:1 , C14:0 , C14:1 , C16:0 , C16:1 , C17:0 , C17:1 , C18:0 to C18:3 , C20:0 to C20:5 , C22:0 to C22:6 , C24:0 to C24:6 . The data presented in the manuscript are derived from SRM channels detecting the fatty acid fragment with more unsaturation ( or the longer one when both have the same unsaturation ) , but only molecules that were detected in SRM channels for both fatty acid fragment with consistent retention times were analyzed . For fatty acid analysis by GC-FID , lipid samples were extracted from tissues by the method of Bligh and Dyer . C23:0 ( Supelco n-Tricosanoic acid , Sigma–Aldrich ) was added to the extracted samples as an internal standard . Then , the fatty acids were methylated with the Fatty Acid Methylation Kit ( Nacalai Tesque ) , and purified using the Fatty Acid Methyl Ester Purification Kit ( Nacalai Tesque ) following the manufacturer's instructions . Fatty acid methyl ester samples were characterized with GC-2010 Plus system ( Shimadzu ) equipped with an FID . The flow rate of carrier gas ( He ) was set at 45 cm/s linear velocity . The temperature of the injection unit and the detector were 240°C and 250°C , respectively . The oven temperature was initiated at 140°C , then raised to 200°C at a rate of 11°C/min , then increased to 225°C at a rate of 3°C/min , and finally elevated to 240°C at a rate of 20°C/min and held at this temperature for 5 min . The injection volume was 2 μl in the split injection mode . For separation , a capillary column ( FAMEWAX , 30 m , 0 . 25 mm ID , 0 . 25 μm; Restek Corporation , Bellefonte , PA ) was used . Fatty acid methyl esters were identified and quantified using a mixture of fatty acid methyl ester standards ( Supelco 37 Component FAME Mix and DPA ( n-3 ) from Sigma–Aldrich; DPA ( n-6 ) from Nu-Chek Prep , Inc . , Elysian , MN; DTA ( n-6 ) from Cayman ) for calibration . Neutral lipids and phospholipids were separated by solid phase extraction using InertSep NH2 aminopropyl columns ( GL Sciences Inc . , Tokyo , Japan ) . Lipids were extracted using the method of Bligh and Dyer , dried in a centrifugal evaporator , dissolved in chloroform , and applied to the columns . Flow-through was collected and combined with the fraction eluted by chloroform:isopropanol ( 2:1 by volume ) as the neutral lipid fraction . After the removal of free fatty acids by 2% acetic acid in diethyl ether , phospholipids were eluted using 2 . 8% ammonia in methanol . For quantification of eicosanoids , internal standards were spiked in methanol extracts , and samples were purified with solid phase extraction with an Oasis HLB column ( Waters ) as previously described ( Kita et al . , 2005 ) . Eicosanoids were quantified by a triple quadrupole mass spectrometer LCMS-8040 ( Shimadzu ) . Separation was performed on a Kinetex C8 column ( 2 . 6 μm , 2 . 1 × 150 mm , Phenomenex ) with a binary mobile phase of the following compositions: 0 . 1% formic acid/water ( mobile phase A ) and acetonitrile ( mobile phase B ) . Pump gradient [time ( %A/%B ) ] was programmed as follows: 0 min ( 90/10 ) –5 min ( 75/25 ) –10 min ( 65/35 ) –20 min ( 25/75 ) –20 . 1 min ( 5/95 ) –28 min ( 5/95 ) –28 . 1 min ( 90/10 ) –30 min ( 90/10 ) . Flow rate was 0 . 4 ml/min and column temperature was 40°C . SRM transitions were: 369 . 3 → 245 . 2 for 6-keto-PGF1α , 373 . 3 → 249 . 2 for 6-keto-PGF1α-d4 , 351 . 2 → 271 . 2 for PGE2 , 355 . 2 → 275 . 2 for PGE2-d4 , 319 . 2 → 219 . 2 for 15 ( S ) -HETE , and 327 . 2 → 226 . 2 for 15 ( S ) -HETE-d8 . Signals were compared to those of standard curves for quantification as previously described ( Kita et al . , 2005 ) . Tissue lipids were obtained as described above in the lipid fractionation ( before column separation ) . The dried lipid samples were dissolved in 5% Nonidet P-40 ( Nacalai tesque ) , and heated at 95°C for 5 min followed by vortexing . This process was repeated to increase the solubility of neutral lipids . TG and cholesterol levels were measured using LabAssay Triglyceride Kit and LabAssay Cholesterol Kit ( Wako ) , respectively . Collected tissues were fixed using 10% neutral buffered formalin ( Wako ) or 4% paraformaldehyde ( PFA , Wako ) for 24 to 48 hr . After fixation , tissues were embedded in Tissue-Tek paraffin wax II 60 ( Sakura Finetek Japan , Tokyo , Japan ) and sectioned by a sliding microtome REM-700 ( Yamato Kohki Industrial , Saitama Japan ) . Otherwise , tissues were embedded into NEG50 frozen section medium ( Thermo Scientific ) , frozen , and sectioned by Cryostat CM3050 S ( Leica Biosystems GmbH , Nussloch , Germany ) . The following dyes were used for staining the sections: hematoxylin ( Wako ) /eosin ( Muto Pure Chemicals Co . , LTD . , Tokyo , Japan ) , Periodic acid-Schiff ( Muto Pure Chemicals ) /hematoxylin ( Wako ) , or oil red O ( Sigma–Aldrich ) /hematoxylin ( Vector ) . Some of the microscopic examinations were performed at BoZo Research Center Inc . Tokyo , Japan . Samples were pre-fixed with 2% PFA and 2% glutaraldehyde ( GA , in 30 mM HEPES buffer ( pH7 . 4 ) ) for overnight at 4°C , followed by post-fixation with aldehyde–OsO4 mixture ( 1 . 25% GA , 1% PFA , 0 . 32% K3[Fe ( CN ) 6] , and 1% OsO4 in 30 mM HEPES buffer ( pH 7 . 4 ) ) for 2 hr at room temperature as previously described ( Shirato et al . , 2006 ) . Fixed samples were washed with Milli Q water ( Merck Millipore Corporation , Tokyo , Japan ) three times , dehydrated in ethanol series , infiltrated with propylene oxide , and embedded in Quetol 812 ( Nisshin EM Corporation , Tokyo , Japan ) . Resin blocks were sectioned at 80 nm thickness with an ultramicrotome ( Leica EM UC7 , Leica ) , contrasted with the EM stainer ( Nisshin EM ) and lead citrate , and observed with a transmission electron microscope ( JEM-1400 , JEOL Ltd . , Tokyo , Japan ) . The number of newborns was counted every day for 1 week after birth . The genotypes were analyzed at P0 and P7 for living pups , or at the day of death . Survival was analyzed only for mice that received normal parenting ( lactation and heating ) . Blood was collected using Heparinized Microhematocrit Capillary Tubes ( Thermo Fisher Scientific ) and centrifuged at 15 , 000×g to obtain plasma . Blood glucose was measured using the blood glucose meter Glutest Neo Super ( Sanwa Kagaku Kenkyusho Co . , Ltd . , Aichi , Japan ) . Plasma insulin levels were measured using the ultra sensitive mouse insulin ELISA Kit ( Morinaga Institute of Biological Science , Inc . , Kanagawa , Japan ) . Plasma lipoproteins were analyzed using a lipoprotein profiling service , LipoSEARCH ( Skylight Biotech Inc , Akita , Japan ) . This analysis consists of gel filtration chromatography followed by an on-line enzymatic method for simultaneous quantification of cholesterol , TG , and PC , according to the procedure described by Usui et al . ( 2002 ) with slight modifications . The nomenclature of the fractions was based on the elution pattern of control plasma that was used for quality check . Glycogen content in liver was analyzed using Glycogen Colorimetric/Fluorometric Assay Kit ( BioVision , Inc . , Milpitas , CA ) . ApoB-containing lipoproteins were precipitated from pooled plasma ( approximately 40 μl ) using the LDL/VLDL and HDL Purification Kit ( Cell Biolabs , Inc . , San Diego , CA ) according to the manufacturer's protocol . Only the LDL/VLDL purification method was used . Precipitated samples were used for apoB , TG , cholesterol , and PC measurements . ApoB levels in plasma were measured by ELISA . Chicken monoclonal anti-apoB antibody ( HUC20 , Hiroshima Bio-Medical Co . , Ltd . , Hiroshima , Japan ) was used for capture and rabbit anti-apoB antiserum ( Abcam plc . , Cambridge , UK ) was used for detection . Capture antibody was coated on a Costar 96 well EIA/RIA Easy Wash Clear Flat Bottom High Binding Plate ( Corning Incorporated , Corning , NY ) overnight at 4°C , and a blocking buffer ( 1× Reagent Diluent Concentrate 2 , R&D Systems , Inc . , Minneapolis , MN ) was plated to inhibit nonspecific binding for 2 hr at room temperature . All the following steps were performed at room temperature . Diluted samples were then added to the plate and incubated for 2 hr . Next , the detection antibody was incubated for 2 hr , followed by an incubation of horseradish peroxidase-conjugated anti-rabbit IgG antibody ( GE Healthcare ) for 2 hr . Wells were washed with PBS plus 0 . 05% Tween 20 for three times after all the incubation steps . TMB ( 3 , 3′ , 5 , 5′-tetramethylbenzidine , Kirkegaard & Perry Laboratories , Inc . , Gaithersburg , MD ) substrate was added to the plate , and the reaction was stopped with 2 N sulfuric acid ( Nacalai tesque ) . Absorbance at 450 nm was measured on a plate reader ARVO X3 , and absorbance at 544 nm was subtracted to normalize background . A standard curve was prepared with plasma from mice at E18 . 5–E19 . 5 for the relative quantification . The quenching assay based on a previous report ( Athar et al . , 2004 ) . 180 nmol of PC ( eggPC from Sigma–Aldrich , DPPC , POPC , PLPC , PAPC , and PDPC from Avanti ) and 5 . 6 nmol of NBD-TG ( Setareh Biotech , LLC , Eugene , OR ) were mixed and dried in glass vials under a stream of nitrogen gas . Then , drying was completed in a centrifugal evaporator for 15 min . The dried lipid film was resuspended in 400 μl assay buffer ( 10 mM Tris-HCl ( pH7 . 4 ) , 150 mM NaCl , and 2 mM EDTA ) , freeze-thawed five times , and then filtered 21 times through a 50 nm pore size polycarbonate membrane ( Nucleopore Track-Etch Membrane , GE Healthcare ) using a mini-extruder ( Avanti ) . The solution was centrifuged at 16 , 100×g ( 18°C , 10 min ) , and the supernatant was used as PC/NBD-TG vesicles . Fluorescence of vesicles was measured with a plate reader ( ARVO X3 ) using 485 nm excitation and 535 nm emission wavelengths . Total fluorescence was determined by adding 97 μl of isopropanol to 3 μl of vesicles . Quenching of NBD-TG was calculated as follows: % quenching = ( total fluorescence—fluorescence of vesicles ) /total fluorescence × 100 . We noted that the TG to PC ratio largely affects the quenching efficiency , thus the optimal concentration ( which is around 3% TG in our experience ) might vary between vials . Approximately , 1 g of mouse liver was homogenized in 5 ml of ice-cold hypotonic buffer ( 1 mM Tris-HCl [pH7 . 4] , 1 mM MgCl2 [Wako] , and 1 mM EGTA [Sigma–Aldrich] ) using a Potter-Elvehjem ( Teflon/Glass ) tissue grinder . After ultracentrifugation at 218 , 800×g ( 10°C , 1 hr ) , the supernatant was collected and used as an MTP source . Protein concentration was measured using the Bio-Rad Protein Assay . The supernatant of boiled liver homogenate ( 95°C , 5 min ) centrifuged at 16 , 100×g ( 18°C , 10 min ) was used as a negative control . MTP assay was performed by a modification of previous methods ( Atzel and Wetterau , 1993 , 1994; Athar et al . , 2004 ) . MTP activity was measured as the transfer of NBD-TG from donor to acceptor vesicles . PC/NBD-TG vesicles prepared for the quenching assay were used as donor vesicles . Acceptor vesicles were prepared with egg yolk PC ( 4 . 385 μmol ) dried in glass vials under nitrogen gas stream and then under vacuum for 15 min . The dried PC was rehydrated in 500 μl of sucrose buffer ( 180 mM sucrose , 10 mM Tris-HCl ( pH7 . 4 ) , 150 mM NaCl , and 2 mM EDTA ) , freeze-thawed five times , and then filtered 21 times through a 400 nm pore size polycarbonate membrane ( Nucleopore Track-Etch Membrane , GE Healthcare ) using a mini-extruder ( Avanti Polar Lipids ) . Vesicles were diluted 10-fold in assay buffer ( 10 mM Tris-HCl ( pH7 . 4 ) , 150 mM NaCl , and 2 mM EDTA ) , pelleted at 16 , 100×g ( 18°C , 10 min ) , and washed with assay buffer . For each reaction , donor vesicles ( containing 2 . 7 pmol of PC and 0 . 084 pmol of NBD-TG ) , acceptor vesicles ( containing 43 . 5 pmol of PC ) , liver homogenate ( 100 μg ) , and 0 . 05 mg BSA ( Sigma–Aldrich ) were combined and adjusted to a total volume of 50 μl with assay buffer . Donor vesicle-free reactions were also prepared and used for blanks . The reactions were incubated in a Deep Well Maximizer incubator/mixer ( Taitec Co . , Ltd . , Saitama , Japan ) for 0 or 1 hr ( 60 rpm , 37°C ) and then centrifuged at 16 , 100×g ( 18°C , 10 min ) . The supernatants ( which contained the donor vesicles ) were collected , and the acceptor vesicle-containing pellets were washed with 100 μl of assay buffer and then resuspended and collected in 10 μl of assay buffer . 100 μl of 10% assay buffer/90% isopropanol was applied to wash the tube walls and also collected . The collected supernatant and pellet fractions were diluted 10-fold with isopropanol and total fluorescence of supernatant ( donor ) , pellet ( acceptor ) , and tube wall washes were measured with a plate reader ( ARVO X3 ) using 485 nm excitation and 535 nm emission wavelengths . The fluorescence value of each fraction was corrected for background by subtracting blank well values . NBD-TG transfer was calculated by the differences in fluorescence between 0 and 1 hr incubation . Transfer was calculated as follows: % transfer = fluorescence of pellet fraction/ ( fluorescence of pellet + fluorescence of supernatant + fluorescence of tube wall wash ) × 100 . Unpaired t-tests were used to compare two groups . Multiple comparisons were performed with Dunnett's multiple comparison tests , Tukey's multiple comparison tests , or Bonferroni's multiple comparison tests , depending on the combinations of comparisons , after one-way or two-way ANOVA . All analyses were done with GraphPad Prism 5 or 6 for Mac OS X software ( GraphPad Software , Inc . , La Jolla , CA ) .	Membranes made of molecules called lipids surround every living cell and also form compartments inside the cell . There are hundreds of different lipid molecules that can be found in membranes . The amount of each type within the membrane can vary , which affects the flexibility and other physical properties of the membrane . One type of lipid found in membranes is called arachidonic acid . It is involved in cell communication and other processes , and is required for young animals to grow and develop properly . An enzyme called LPCAT3 is thought to incorporate arachidonic acid into membranes , but this has not yet been proven to occur in living animals . Here , Hashidate-Yoshida , Harayama et al . studied the role of LPCAT3 in newborn mice . The experiments show that this enzyme is found at high levels in the intestine and liver . Mice that lacked LPCAT3 had much lower levels of arachidonic acid compared with normal mice . These mice also showed signs of severe intestinal damage due to the build up of lipids from their mother's milk , and died within a few days of being born . The mice that lacked LPCAT3 had different amounts of another type of lipid—called triacylglycerols—in their intestine and liver . Normally , these lipids would be assembled into larger molecules called lipoproteins that are released into the blood stream and used in the muscles and other parts of the body . However , Hashidate-Yoshida , Harayama et al . found that in the mice missing LPCAT3 , the triacylglycerols did not get assembled into lipoproteins and so they accumulated inside the intestine and liver cells . The experiments also show that high levels of arachidonic acid and other similar lipids in the membrane enable triacylglycerol molecules to cluster together , which increases the production of lipoproteins . Hashidate-Yoshida , Harayama et al . 's findings suggest that LPCAT3 incorporates arachidonic acid into the membrane of intestine and liver cells , which enables triacylglycerols to be assembled into lipoproteins . The next challenge will be to find out if LPCAT3 is also important for the production of lipoproteins in humans . If it is , then developing new therapies that alter the activity of this enzyme might be beneficial for patients with abnormal levels of lipids in the blood ( known as dyslipidemia ) .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	Fatty acid remodeling by LPCAT3 enriches arachidonate in phospholipid membranes and regulates triglyceride transport
Intermediate filament ( IF ) proteins , including nuclear lamins and cytoplasmic IF proteins , are essential cytoskeletal components of bilaterian cells . Despite their important role in protecting tissues against mechanical force , no cytoplasmic IF proteins have been convincingly identified in arthropods . Here we show that the ancestral cytoplasmic IF protein gene was lost in the entire panarthropod ( onychophoran + tardigrade + arthropod ) rather than arthropod lineage and that nuclear , lamin-derived proteins instead acquired new cytoplasmic roles at least three times independently in collembolans , copepods , and tardigrades . Transcriptomic and genomic data revealed three IF protein genes in the tardigrade Hypsibius dujardini , one of which ( cytotardin ) occurs exclusively in the cytoplasm of epidermal and foregut epithelia , where it forms belt-like filaments around each epithelial cell . These results suggest that a lamin derivative has been co-opted to enhance tissue stability in tardigrades , a function otherwise served by cytoplasmic IF proteins in all other bilaterians . Tardigrades , also known as water bears , are microscopic invertebrates that live in marine , freshwater and semi-aquatic/limno-terrestrial environments ( Kinchin , 1994; Nelson , 2002 ) ( Figure 1 ) . Although tardigrades have become renowned for their ability to survive extreme conditions ( Møbjerg et al . , 2011; Ramløv and Westh , 2001 ) , including exposure to space ( Persson et al . , 2011; Rebecchi et al . , 2011 ) , only little is known about the actual mechanisms that allow their cells to resist severe mechanical stress caused by desiccation and freezing ( Møbjerg et al . , 2011; Ramløv and Westh , 2001; Wright , 2001; Tanaka et al . , 2015; Yamaguchi et al . , 2012 ) . The integrity and plasticity of tardigrade tissues might be achieved by specialised cytoskeletal components , such as IF proteins , which are known to be essential for stress resilience of cells ( Herrmann et al . , 2009; Coulombe and Wong , 2004; Kim and Coulombe , 2007 ) . While lamins , a group of IF proteins found in the nucleus , occur in most eukaryotes , including social amoebae ( Krüger et al . , 2012 ) and all metazoans ( Dittmer and Misteli , 2011 ) , cytoplasmic IF proteins are thought to have evolved from an ancestral lamin gene by duplication in the bilaterian lineage ( Erber et al . , 1999; Herrmann and Strelkov , 2011 ) . Genomic and biochemical studies have revealed that the cytoplasmic IF proteins are present in all bilaterian taxa excluding arthropods ( Bartnik and Weber , 1989; Goldstein and Gunawardena , 2000; Erber et al . , 1998 ) [but see a contradictory report ( Mencarelli et al . , 2011 ) of a putative cytoplasmic IF protein in a collembolan] . The apparent loss of cytoplasmic IF proteins in the arthropod lineage might correlate with the acquisition of an exoskeleton ( Herrmann and Strelkov , 2011; Goldstein and Gunawardena , 2000; Erber et al . , 1998 ) , which provides mechanical support to the arthropod skin . However , this hypothesis has never been tested , as it is unknown whether or not onychophorans and tardigrades , the soft-bodied relatives of arthropods , possess cytoplasmic IF proteins . 10 . 7554/eLife . 11117 . 003Figure 1 . Light micrograph of a specimen of the tardigrade Hypsibius dujardini in dorsal view . Anterior is left . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 003 To clarify whether or not onychophorans and tardigrades possess cytoplasmic IF proteins , we analysed Illumina-sequenced transcriptomes of five distantly related onychophoran species and the freshwater tardigrade H . dujardini . Although this model tardigrade ( Gabriel et al . , 2007; Gross et al . , 2015 ) shows only a limited ability to tolerate desiccation ( anhydrobiosis ) , it clearly survives immediate freezing ( cryobiosis; see Video 1 ) . Our analyses revealed only one putative IF protein transcript in each of the five onychophoran species but three in H . dujardini . Additional screening of the recently sequenced genome ( Boothby et al . , 2015; Koutsovoulos et al . , 2015 ) of H . dujardini confirms that the identified transcripts correspond to the three potential IF protein-coding genes of this species . According to sequence comparisons with well-characterized IF proteins from humans and the nematode Caenorhabditis elegans , all corresponding proteins of the three identified genes share a similar α-helical rod domain organization with three coiled coil-forming segments ( coil 1A , coil 1B , coil 2; Figure 2A , B and Figure 2—figure supplements 1 and 2 ) ( review Chernyatina et al . , 2015 ) . This , in conjunction with the highly conserved intermediate filament consensus motifs ( review Herrmann and Aebi , 2004 ) at the beginning and end of the rod domain of all three tardigrade proteins , classify them as intermediate filament proteins . The three tardigrade IF proteins further possess 42 residues in the coil 1B ( Figure 2B and Figure 2—figure supplement 1 ) — a feature that is shared between all eukaryote lamins and protostome cytoplasmic IF proteins but that must have been deleted from the ancestral cytoplasmic IF protein gene in chordates ( Herrmann et al . , 2009; Peter and Stick , 2015 ) . In contrast to the similar organization of the rod domain , the flanking sequences vary between all three tardigrade IF proteins . One of them , which we named lamin-2 , possesses all major regions known from other eukaryote lamins ( Herrmann et al . , 2009; Dittmer and Misteli , 2011; Burke and Stewart , 2013 ) , including the nuclear localization signal ( NLS ) — which generally mediates the import of proteins into the nucleus ( Mical et al . , 2004 ) — the lamin-tail domain ( LTD ) , and the carboxy-terminal CaaX motif ( Figure 2A , B ) . In contrast , the second tardigrade IF protein ( named lamin-1 ) lacks the CaaX box , whereas the third IF protein from H . dujardini is missing all three motifs , including the NLS , indicating this protein may instead localize in the cytoplasm rather than the nuclear lamina of H . dujardini cells . Since the structure of the latter resembles the domain composition of known bilaterian cytoplasmic IF proteins ( Figure 2—figure supplements 1 and 2 ) we consequently named it cytotardin . 10 . 7554/eLife . 11117 . 004Video 1 . The tardigrade Hypsibius dujardini survives freezing . This time-lapse video shows thawing specimens of the tardigrade H . dujardini after 4 days frozen in ice . One specimen starts with minuscule movements of one leg after 20 min of thawing and fully recovers locomotion within 120 min . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 00410 . 7554/eLife . 11117 . 005Figure 2 . Structure and organization of the IF proteins of the tardigrade Hypsibius dujardini . ( A ) Protein sequence alignment of H . dujardini IF proteins with human ( Homo sapiens ) lamins A and B1 . The names of H . dujardini IF proteins were chosen according their structural similarities to known IF proteins ( lamin-1 and lamin-2: lamin-like; cytotardin: cytoplasmic IF-like; see text for details ) . Note the sequence similarities of the rod domains ( coil 1A , L1 , coil 1B , linker L12 and coil 2 ) and the intermediate filament consensus motifs ( highlighted in red with highly conserved parts in a box ) among all three proteins . The positions of rod sub-domains are placed as described for human IF proteins ( review Chernyatina et al . , 2015 ) . The nuclear localization signal in lamin-1 , lamin-2 , lamin A and lamin B1 is highlighted in green and the immunoglobulin fold ( Ig fold ) is marked in light orange . Note the absence of an Ig fold in cytotardin . Predictions ( Jpred3 , JNetPRED ) of α-helices and β-sheets are indicated by red waved underlines and solid underlines , respectively . The C-terminal prenylation motif of lamins ( CaaX ) is marked in purple . The alignment ( Clustal Omega ) has been performed using Analysis Tool Web Services from the EMBL-EBI ( McWilliam et al . , 2013 ) . ( * ) indicates positions which have a single , fully conserved residue . ( : ) Indicates conservation between groups of strongly similar properties — scoring > 0 . 5 in the Gonnet PAM 250 matrix . ( . ) Indicates conservation between groups of weakly similar properties — scoring ≤ 0 . 5 in the Gonnet PAM 250 matrix . ( B ) Organization of the three IF proteins of H . dujardini . Dark blue colour in the 1B coil of the rod domain indicates six heptads that have been lost in the chordate lineage of cytoplasmic intermediate filament proteins . The numbers denote the amino acid positions of the beginning and end of each rod sub-domain . 1A , 1B and 2 , coiled-coil segments of the rod domain; CaaX , isoprenylation motif at the carboxyl terminus; LTD , lamin tail domain; NLS , nuclear localization signal . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 00510 . 7554/eLife . 11117 . 006Figure 2—figure supplement 1 . Protein sequence alignment of Hypsibius dujardini cytotardin with selected human ( Homo sapiens ) cytoplasmic IF proteins . Human type I keratins K14 ( epidermis ) and K18 ( simple epithelia ) are forming IFs from obligatory heterodimers with type II keratins K5 ( epidermis ) and K8 , respectively . The type III IF protein vimentin is able to form homopolymeric IFs . The position of the cytotardin rod domain , containing the coiled-coil and linkers ( coil 1A , L1 , coil 1B , L12 , coil 2; predicted from the protein sequence alignment in Figure 2 ) , is indicated by blue lines . Note the sequence similarities of the rod domain , especially at the rod domain-flanking intermediate filament consensus motifs ( highlighted in red ) . Note also 42 amino acids in the coil 1B of cytotardin , which have been deleted from the ancestral cytoplasmic IF protein gene in chordates . The alignment ( Clustal Omega ) has been performed using Analysis Tool Web Services from the EMBL-EBI ( McWilliam et al . , 2013 ) . ( * ) indicates positions which have a single , fully conserved residue . ( : ) Indicates conservation between groups of strongly similar properties — scoring > 0 . 5 in the Gonnet PAM 250 matrix . ( . ) Indicates conservation between groups of weakly similar properties — scoring ≤ 0 . 5 in the Gonnet PAM 250 matrix . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 00610 . 7554/eLife . 11117 . 007Figure 2—figure supplement 2 . Protein sequence alignment of Hypsibius dujardini IF proteins with selected IF proteins from Caenorhabditis elegans . C . elegans ifa-1 and ifb-1 are epithelial cytoplasmic intermediate filament proteins , whereas lmn-1 represents the single lamin of C . elegans . The position of the cytotardin rod domain , containing the coiled-coil and linkers ( coil 1A , L1 , coil 1B , L12 , coil 2; predicted from the protein sequence alignment in Figure 2 ) , is indicated by blue lines . The intermediate filament consensus motifs are highlighted in red , the immunoglobulin fold ( Ig fold ) is marked in light orange , and the C-terminal prenylation motif of lamins ( CaaX ) is marked in purple . Note the presence of a prenylation motif in H . dujardini lamin-2 and C . elegans lmn-1 and its absence in the other proteins . The alignment ( Clustal Omega ) has been performed using Analysis Tool Web Services from the EMBL-EBI ( McWilliam et al . , 2013 ) . ( * ) indicates positions which have a single , fully conserved residue . ( : ) Indicates conservation between groups of strongly similar properties — scoring > 0 . 5 in the Gonnet PAM 250 matrix . ( . ) Indicates conservation between groups of weakly similar properties — scoring ≤ 0 . 5 in the Gonnet PAM 250 matrix . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 007 To firmly place the onychophoran and tardigrade IF homologs on the evolutionary tree , we reconstructed the phylogeny of broadly sampled metazoan lamin and cytoplasmic IF protein genes . In our phylogenetic analyses , all identified tardigrade and onychophoran sequences cluster within the bilaterian lamin clade , whereas none of them groups with cytoplasmic IF protein-coding genes ( Figure 3 and Figure 3—figure supplements 1 and 2 ) . Surprisingly , the three identified IF sequences of H . dujardini , together with two sequences from Milnesium tardigradum , form a strongly supported monophyletic clade of tardigrade lamins ( GTR+G: Bootstrap support BS=77 , LG+G: BS=80 ) in our analyses ( Figure 3 and Figure 3—figure supplements 1 and 2 ) . This implies that there were at least two duplication events in the tardigrade lineage that gave rise to lamin-1 , lamin-2 and cytotardin genes — consequently characterizing the tardigrade IFs ( including cytotardin ) , for example , as co-orthologous to nematode lamins rather than orthologous to nematode cytoplasmic IFs ( Figure 3 and Figure 3—figure supplements 1 and 2 ) . Our results further show that the isomin sequence of the collembolan Isotomurus maculatus does in fact cluster with other identified collembolan transcripts ( GTR+G: BS=74 , LG+G: BS=70 ) within the clade of arthropod lamins ( Figure 3 and Figure 3—figure supplements 1 and 2 ) ; it had previously been interpreted ( Mencarelli et al . , 2011 ) as closely related to cytoplasmic IF proteins of nematodes and therefore as an ortholog of the bilaterian cytoplasmic IF proteins , likely due to the narrower dataset used for their phylogenetic analysis . These results clearly challenge the identity of isomin as a member of the bilaterian cytoplasmic IF protein clade ( Mencarelli et al . , 2011 ) and suggest that orthologs of genes encoding these proteins are entirely missing in arthropods , at least in those with known genomic sequences . In fact , besides the putative IF proteins from chelicerates , crustaceans and hexapods obtained from publicly available databases ( e . g . GenBank ) , our transcriptomic and genomic analyses , which included screening of the genome of the centipede Strigamia maritima ( see Chipman et al . , 2014 ) , the water flea Daphnia pulex ( see Colbourne et al . , 2011 ) , and more than 70 transcriptomes from hexapod species sequenced as part of the 1KITE project ( Misof et al . , 2014 ) , strongly suggest that these genes were already lost in the panarthropod lineage , since all of these panarthropod IF proteins cluster within a well-supported monophyletic clade of bilaterian lamins ( GTR+G: BS=84 , LG+G: BS=76; Figure 3—figure supplements 1 and 2 ) . In this respect , even if the metazoan lamins are polyphyletic , as recently proposed by Kollmar , 2015 based on the finding of putative nematocilin homologs in Bilateria , our results favour the tardigrade , copepod , and collembolan IF proteins as members of the bilaterian lamins rather than the bilaterian cytoplasmic IFs or nematocilins ( Figure 3—figure supplements 1 and 2 ) . 10 . 7554/eLife . 11117 . 008Figure 3 . Phylogeny of the metazoan intermediate filament proteins illustrating the position of the three tardigrade IF proteins ( highlighted in red ) . The tree was obtained from a Maximum likelihood analysis under a dataset-specific GTR+G substitution model of 447 eukaryotic intermediate filament proteins ( see Figure 3—figure supplement 1 for the full tree ) . Note that all tardigrade as well as collembolan ( green ) and copepod IF proteins ( purple ) belong to the bilaterian lamin clade ( light blue ) . Hence , cytotardin and isomin are closer related to , for example , nematode lamins ( light brown ) than to nematode cytoplasmic IF proteins ( dark brown ) , which are orthologs of the bilaterian cytoplasmic IF proteins ( yellow ) . Selected bootstrap support values are given at particular nodes . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 00810 . 7554/eLife . 11117 . 009Figure 3—figure supplement 1 . Maximum likelihood tree under a dataset-specific GTR+G substitution model and accession numbers of 447 eukaryotic intermediate filament proteins and the placement of the IF protein genes of the tardigrade Hypsibius dujardini ( highlighted in red ) . Lamins of onychophorans are highlighted in orange , copepods in purple , collembolans in green , nematodes in light brown and cytoplasmic IF proteins of nematodes in dark brown . Note the position of isomin of the collembolan Isotomurus maculatus , which has been interpreted as a putative cytoplasmic IF protein ( Mencarelli et al . , 2011 ) , within a group of collembolan lamins ( asterisk ) . The domain structure of each protein is depicted on the right . Note the absence of a nuclear localization signal ( NLS ) in cytotardin of H . dujardini as well as in the coloured sequences of copepods ( purple ) and collembolans ( green ) . Bootstrap values from 1 , 000 pseudoreplicates ≥50% are given at the nodes . Scale bar indicates the number of substitutions per site . Abbreviations: CaaX , isoprenylation motif at the carboxyl terminus; LTD , lamin tail domain; NLS , nuclear localization signal . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 00910 . 7554/eLife . 11117 . 010Figure 3—figure supplement 2 . Maximum likelihood tree under the empirical LG+G substitution model and accession numbers of 447 eukaryotic intermediate filament proteins and the placement of the IF protein genes of the tardigrade Hypsibius dujardini ( highlighted in red ) . Lamins of onychophorans are highlighted in orange , copepods in purple , collembolans in green , nematodes in light brown and cytoplasmic IF proteins of nematodes in dark brown . Note the position of isomin of the collembolan Isotomurus maculatus , which has been interpreted as a putative cytoplasmic IF protein ( Mencarelli et al . , 2011 ) , within a group of collembolan lamins ( asterisk ) . The domain structure of each protein is depicted on the right . Note the absence of a nuclear localization signal ( NLS ) in cytotardin of H . dujardini as well as in the coloured sequences of copepods ( purple ) and collembolans ( green ) . Bootstrap values from 1000 pseudoreplicates ≥50% are given at the nodes . Scale bar indicates the number of substitutions per site . Abbreviations: CaaX , isoprenylation motif at the carboxyl terminus; LTD , lamin tail domain; NLS , nuclear localization signal . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 010 To determine their subcellular localization and organization , we generated antisera against the three IF proteins of H . dujardini and confirmed their specificity ( see Figure 4—figure supplement 1 and Figure 5—figure supplement 1 ) . Immunolocalization of lamin-1 , lamin-2 and cytotardin proteins in whole-mount preparations and cryosectioned specimens , in conjunction with confocal laser-scanning microscopy , revealed a highly specific subcellular distribution and tissue-restricted expression of these proteins in H . dujardini . As predicted , lamin-1 and lamin-2 proteins both display a typical , lamin-like distribution within all nuclei ( Figure 4A–D and Figure 4—figure supplement 2 ) . While lamin-1 is localised throughout the nucleoplasm ( Figure 4A ) , lamin-2 is restricted to the nuclear periphery ( Figure 4B ) , although there is a small overlap region between these two proteins ( Figure 4C ) . The intranuclear distribution of lamin-1 corresponds to its lack of the CaaX motif , which is responsible for the association of lamins with the nuclear envelope ( Kitten and Nigg , 1991 ) . In contrast to the two lamins , cytotardin of H . dujardini is not localised within the nucleus , but in the peripheral cytoplasm of all epidermal and foregut cells , where it appears to be closely associated or aligned with the plasma membrane ( Figure 4E–G and Figure 4—figure supplement 3 and Figure 6—figure supplement 1 ) . In this setting , it encircles the apical regions of the cells in a belt-like , filamentous array ( Figure 4E–G and Figure 4—figure supplement 3 ) . 10 . 7554/eLife . 11117 . 011Figure 4 . Immunofluorescence labelling of IF proteins in the tardigrade Hypsibius dujardini . Confocal laser-scanning micrographs . ( A–D ) Triple labelling of lamin-1 ( green ) , lamin-2 ( red ) and DNA ( cyan ) . Note the localization of lamin-1 within the nucleoplasm and that of lamin-2 at the nuclear periphery . ( E , F ) Double labelling of cytotardin ( glow-mode ) and DNA ( cyan ) on cryosections . ( E ) Tangential section of dorsolateral body wall . Arrow points to the dorsal midline . ( F ) Cross-section of a specimen . Dorsal is up . ( G ) Whole-mount preparation of a contracted specimen in ventral view . Anterior is up . Inset shows detail of the tip of a leg . cl , claw; ep , epidermis; hd , head; lg , leg; lg1–lg4 , legs 1 to 4; nu , nucleus; vs , ventral body surface . Scale bars: ( A–D ) 1 µm , ( E ) 5 µm , ( F , G ) 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 01110 . 7554/eLife . 11117 . 012Figure 4—figure supplement 1 . Western blots of lamin-1 , lamin-2 and cytotardin antisera . ( A ) Western blot analysis of lamin-1 protein expression in Hypsibius dujardini . Anti-lamin-1 antibody stains a band of ≈70 kDa , as expected . ( B ) Western blot analysis of lamin-2 protein expression in H . dujardini . Anti-lamin-2 antibody stains a band of ≈72 kDa , as expected . ( C ) Western blot analyses of cytotardin protein expression in H . dujardini , transformed Escherichia coli ( E . coli ) and transfected MCF-7 cells . Anti-cytotardin antibody stains a band of ≈57 kDa , as expected . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 01210 . 7554/eLife . 11117 . 013Figure 4—figure supplement 2 . Immunofluorescence labelling of lamin-1 and lamin-2 in the tardigrade Hypsibius dujardini . Confocal laser-scanning micrographs . Sagittal sections of specimens . Anterior is left , dorsal is up . Note the presence of lamin-1 and lamin-2 inside all nuclei . Cuticular structures of the foregut are autofluorescent . ( A , B ) Double labelling of lamin-1 ( glow-mode ) and DNA ( cyan ) on a cryosection . ( C , D ) Double labelling of lamin-2 ( glow-mode ) and DNA ( cyan ) on a cryosection . br , position of the brain; bt , buccal tube; cg , claw gland; lg1–lg4 , legs 1 to 4; mi , position of the midgut; nu , nuclei; ov , ovary; ph , pharynx; sg , salivary gland . Scale bars: ( A–D ) 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 01310 . 7554/eLife . 11117 . 014Figure 4—figure supplement 3 . Immunofluorescence labelling of cytotardin in the tardigrade Hypsibius dujardini with focus on the epidermis . Confocal laser-scanning micrographs . ( A ) Double labelling of cytotardin ( glow-mode ) and DNA ( cyan ) on a cryosection . Sagittal section of a specimen . Anterior is left , dorsal is up . Note the exclusive presence of cytotardin filaments in the epidermis and tissues of the foregut . ( B ) Anti-cytotardin immunolabelling on a cryosection . Tangential section of dorsolateral body wall . Arrows point to the dorsal midline . ( C , D ) Double labelling of cytotardin ( glow-mode ) and DNA ( cyan ) on a whole-mount preparation of a contracted specimen . Detail of the third pair of legs in ventral view . Anterior is up . Note the apical position of cytotardin-filament belts in the epidermal cells ( arrowheads ) and their dense arrangement in the tips of each leg . br , brain; bt , buccal tube; cg , claw gland; cl , claw; ep , epidermis; lg , leg; lg1–lg4 , legs 1 to 4; mc , mouth cone; mi , position of the midgut; nu , nuclei; oe , oesophagus; ov , ovary; ph , pharynx . Scale bars: ( A–D ) 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 014 In order to investigate the filament-forming capacity of cytotardin , we transiently transfected human mammary epithelial MCF-7 cells with a corresponding cDNA ( Figure 5A–F and Figure 5—figure supplement 1 ) . Immunofluorescence analyses using cytotardin antibody revealed that this protein is located exclusively in the cytoplasm of transfected MCF-7 cells , where it forms both short filaments and extensive cytoskeletal networks which most likely are homopolymeric ( Figure 5A–F ) . Notably , some of the cytotardin arrays display cage-like perinuclear structures , while others are located in the periphery close to the cell membrane ( Figure 5E , F ) . Double labelling for cytotardin and desmoplakin , a desmosomal protein mediating membrane attachment of mammalian IF proteins ( Simpson et al . , 2011 ) , shows that cytotardin occurs close to desmosomes but is not co-localised with desmoplakin ( Figure 5C , D ) . To examine whether this arrangement was mediated by interactions between cytotardin and keratins endogenously expressed in MCF-7 cells , we double-labelled the transfectants for cytotardin and keratin-8 . Our data show that these two proteins are not co-localised and that the endogenous keratin networks are displaced in cells with dense cytotardin arrays ( Figure 5E , F ) . These findings strongly support an intrinsic ability of the cytotardin protein of H . dujardini to both form homopolymeric filaments and cytoplasmic networks — both properties that are functionally analogous to mammalian cytoplasmic IFs ( Bohnekamp et al . , 2015 ) . 10 . 7554/eLife . 11117 . 015Figure 5 . Immunolocalization of exogenous cytotardin in human MCF-7 epithelial cells . ( A , B ) Double labelling of cytotardin ( glow-mode ) and DNA ( cyan ) . Note the cytoplasmic cytotardin filamentous network surrounding the nucleus and extensions close to cell borders . Short cytotardin filaments are aligned along the plasma membrane , different from the arrangement seen in tardigrade epithelial cells . ( C , D ) Triple labelling of cytotardin ( glow-mode ) , desmoplakin ( green ) and DNA ( cyan ) . Note that cytotardin forms a cytoplasmic filamentous network extending from the perinuclear area to the cell membrane . Note also that it is not co-localised with desmoplakin . ( E , F ) Triple labelling of cytotardin ( glow-mode ) , keratin-8 ( green ) and DNA ( cyan ) . Endogenous keratin networks have been displaced by cytotardin filaments from the perinuclear region without being disrupted ( asterisk ) . dp , desmosomal plaque; nu , nucleus . Scale bars: ( A–F ) 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 01510 . 7554/eLife . 11117 . 016Figure 5—figure supplement 1 . Characterization of cytotardin antiserum . ( A–C ) Triple labelling of cytotardin ( glow-mode ) , HA-tag ( green ) and DNA ( cyan ) in MCF-7 cells transfected with pcDNA3-HA-cytotardin plasmids . Anti-cytotardin and anti-HA-tag staining show similar signals confirming the specificity of anti-cytotardin antibodies . HA-cytotardin is cytoplasmic and forms short filaments able to localize close to cell borders . ( D ) Western blot analysis of cytotardin in MCF-7 cells transfected with pcDNA3-cytotardin , pcDNA3-HA-cytotardin , pEGFP-cytotardin and the corresponding empty vectors . Left panel shows a Ponceau red-staining of the blotted cell extract as loading control . Anti-tag ( right panel ) compared to anti-cytotardin ( middle panel ) staining shows the expression of the exogenous protein and specificity of the antibodies for immunoblotting . nu , nucleus . Scale bars: ( A–C ) 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 01610 . 7554/eLife . 11117 . 017Figure 5—figure supplement 2 . Plasmids , primers , and restriction enzymes used for the cloning of tardigrade lamin and cytotardin genes . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 017 Our data on tissue-specific distribution of cytotardin in H . dujardini further show that its belt-like arrays are confined to the ectodermal epithelia , including the epidermis , buccal tube , pharynx , and oesophagus ( Figures 4E–G and 6A , B and Figure 4—figure supplement 3 and Figure 6—figure supplement 1 ) . Thus , the entire tardigrade body is ensheathed by a grid of belt-like filaments formed by the cytotardin protein , which retain their integrity even in contracted specimens ( Figure 4E–G and Figure 4—figure supplement 3 and Figure 6—figure supplement 1 ) . The most prominent anti-cytotardin immunoreactivity is found in areas exposed to considerable physical stress in living specimens , including the bases of claws and the stylet apparatus ( Figures 4F , G and 6B and Figure 4—figure supplement 3 and Figure 6—figure supplement 1 ) . We therefore anticipate that much of the resistance of the tardigrade body to extreme conditions , such as cryobiosis ( Wright , 2001; Hengherr et al . , 2009 ) , might be attributable to the dense , fibre-like cytotardin meshwork . The belt-like structures encircling each epidermal cell might help to resist the shearing forces that arise during freezing and thawing cycles , whereas the dense meshwork at the basis of each claw and around the stylets might provide the tissue stability necessary for locomotion and feeding . 10 . 7554/eLife . 11117 . 018Figure 6 . Distribution of cytotardin within the cell and across the tissues in Hypsibius dujardini and the evolutionary history of cytoplasmic intermediate filament proteins . ( A ) Diagram of an epidermal cell of H . dujardini with a belt-like arrangement of cytotardin . ( B ) Diagram of H . dujardini showing the distribution of cytotardin ( red ) , which is confined to the ectodermal tissues . ( C ) Scenario of the independent origin of lamin-derived cytoplasmic intermediate filaments in tardigrades , collembolans , and copepods . Note that the cytoplasmic IFs in these three lineages ( indicated in red , purple , and green , respectively ) evolved independently from the cytoplasmic IFs of other bilaterians ( highlighted in orange ) . ba , buccal apparatus; br , brain; cb , cytotardin filament belt; cg , claw gland; cm , cell membrane; ep , epidermis; mi , midgut; nu , nucleus; oe , oesophagus; ov , ovary; ph , pharynx . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 01810 . 7554/eLife . 11117 . 019Figure 6—figure supplement 1 . Immunofluorescence labelling of cytotardin in the tardigrade Hypsibius dujardini with focus on the foregut . Confocal laser-scanning micrographs . Double labelling of cytotardin ( glow-mode ) and DNA ( cyan ) on sagittal ( A–C , E ) and cross sections ( D ) of cryosectioned specimens . Cuticular structures of the foregut are autofluorescent . Anterior is left ( A–C , E ) ; dorsal is up ( in A–E ) . ( A ) Overview of the foregut . Note the dense arrangement of cytotardin arrays in the epithelia of the foregut , especially those surrounding the stylets . ( B ) Detail of the buccal tube . ( C ) Detail of the pharynx . ( D ) Cross-section of the pharynx . ( E ) Detail of the oesophagus . bt , buccal tube; mc , mouth cone; nu , nucleus; oe , oesophagus; pc , placoid; ph , pharynx; pm , pharyngeal myoepithelium; st , epithelium surrounding the stylets . Scale bars: ( A–E ) 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11117 . 019 Together , our data demonstrate that cytotardin of H . dujardini is a cytoplasmic IF protein , predominantly expressed in ectodermal epithelia , that has evolved independently from the cytoplasmic IF proteins of other bilaterians by gene duplication and subsequent neofunctionalization ( Figure 6C ) . This process was most likely triggered by an initial loss of the CaaX motif and nuclear localization signal from the ancestral lamin gene in the tardigrade lineage . Expression of the cytotardin protein in tissues that are subject to mechanical stress might have served as a pre-adaptation for the ability of tardigrades to survive extreme environmental conditions . Similar duplication and neofunctionalization events might have occurred in the collembolan and copepod lineages , as our findings demonstrate that these two taxa also show duplicated lamins that have lost their nuclear localization signals ( Figures 3 and 6C and Figure 3—figure supplements 1 and 2 ) . While isomin might stabilize the intestinal epithelial cells in collembolans ( Mencarelli et al . , 2011 ) , the localization and function of the putative lamin-derived cytoplasmic IF protein in copepods has yet to be clarified . Specimens of Hypsibius dujardini ( Doyère , 1840 ) were purchased from Sciento ( Manchester , UK ) and cultured as described by Mayer et al . , 2015 . The Illumina-sequenced transcriptomes of H . dujardini and five onychophoran species from Hering et al . , 2012 and Hering and Mayer , 2014 were screened for expressed intermediate filament genes , including lamins , by BLAST searches ( Altschul et al . , 1997 ) with known metazoan lamins and cytoplasmic IF genes as bait sequences . Three putative IF protein genes from H . dujardini and one from each of the onychophoran species studied were verified afterwards by reciprocal BLAST searches against the nr database of GenBank ( http://blast . ncbi . nlm . nih . gov/Blast . cgi ) . In addition , BLAST searches of the putative tardigrade intermediate filament genes in the genome of H . dujardini ( http://badger . bio . ed . ac . uk/H_dujardini/; Koutsovoulos et al . , 2015 ) yielded nearly identical sequences as obtained in our transcriptomic data , although the automatically predicted transcripts from the genome seem to be erroneous as previously reported by Hering and Mayer , 2014 and had to be corrected manually . Furthermore , publicly available resources and databases [nr , TSA and EST databases of GenBank , Compagen ( Hemmrich and Bosch , 2008 ) ] as well as the genome of the centipede Strigamia maritima ( see Chipman et al . , 2014 ) , the water flea Daphnia pulex ( see Colbourne et al . , 2011 ) , and more than 70 transcriptomes from hexapod species sequenced as part of the 1KITE project ( Misof et al . , 2014 ) were comprehensively screened for putative eukaryotic intermediate filament genes . In total , 447 eukaryotic lamin and cytoplasmic IF genes were manually curated and selected for further analyses . Each of the sequences was verified by screening for the presence of a coiled-coil-forming domain ( Pfam PF00038 ) , a typical feature of intermediate filament proteins , and the presence or absence of a lamin tail domain ( LTD; Pfam PF00932 , SCOP d1ifra_ , PDB 2lll/2kpw/1ufg/1ivt ) using the webserver of Pfam 27 . 0 ( Finn et al . , 2014 ) and SMART ( Schultz et al . , 1998; Letunic et al . , 2015 ) , respectively . In addition , the occurrence of a nuclear localization signal ( NLS motif ) was predicted for all sequences by using the webserver of NucPred ( Brameier et al . , 2007 ) and cNLS Mapper ( Kosugi et al . , 2009 ) ( cut-off score = 2 . 0 ) . The protein structures ( α-helices and β-sheets ) were predicted for the tardigrade lamin-1 , lamin-2 and cytotardin using Jpred3 and JNetPRED ( Cole et al . , 2008 ) . For phylogenetic analyses , the rod domains of all sequences were aligned using the Mafft online version v7 . 245 ( Katoh and Standley , 2013 ) with the most accurate option L-INS-i and default parameters . To remove homoplastic and random-like positions , the alignment was afterwards masked with the software Noisy rel . 1 . 15 . 12 ( Dress et al . , 2008 ) ( -cutoff=0 . 8 , -seqtype=P , -shuffles=20 , 000 ) . Two Maximum likelihood analyses were conducted with the Pthreads version of RAxML v8 . 1 . 15 ( Stamatakis , 2014 ) . For each run , the best tree was obtained from 10 independent inferences and GAMMA correction of the final tree under either the empirical LG substitution model or a dataset-specific GTR substitution matrix . The LG model was automatically selected by RAxML ( PROTGAMMAAUTO option ) as best-fitting substitution model , which is in line with the best-fitting model obtained with ProtTest v3 . 4 . 1 ( Darriba et al . , 2011 ) ( LG+G ) according to the Akaike information criterion ( Akaike , 1974 ) ( AIC ) , Bayesian Information Criterion ( Schwarz , 1978 ) ( BIC ) , corrected AIC ( Sugiura , 1978; Hurvich and Tsai , 1989 ) , and Decision Theory Criterion ( Minin et al . , 2003 ) ( DT ) . Notably , the analysis using the dataset-specific GTR+G model yielded a better log likelihood score ( −174 , 216 . 74 ) for the best tree than using the best obtained empirical model LG+G ( −178 , 092 . 26 ) . Bootstrap support values for both trees were calculated using the rapid bootstrapping algorithm implemented in RAxML from 1 , 000 pseudoreplicates . The protein domain structures of the sequences analysed were mapped on the trees using iTol v2 ( Letunic and Bork , 2011 ) . Total RNA from several hundred specimens of H . dujardini was extracted and purified using TRIzol Reagent ( Life Technologies , Carlsbad , CA ) and RNeasy MinElute Cleanup Kit ( Qiagen , Hilden , Germany ) according to the manufacturers’ protocols . First strand cDNA synthesis was performed using random hexamer primer and SuperScript II Reverse Transcriptase ( Life Technologies ) and afterwards used as template for amplification of the whole coding sequence ( CDS ) of lamin-1 , lamin-2 and cytotardin using gene specific primers . The cytotardin specific primers contained restriction sites that were required for subsequent cloning into bacterial or mammalian expression vectors ( see Figure 5—figure supplement 2 ) . The amplicons of lamin-1 and lamin-2 were cloned into the pGEM-T Vector System ( Promega , Madison , WI ) to generate the plasmids pGEM-T-lamin-1 and pGEM-T-lamin-2 . Cytotardin CDS was cloned into expression vectors pET15b ( Novagen Merck Millipore , Darmstadt , Germany ) , pcDNA3 . 1/Zeo ( + ) ( Life Technologies ) and pEGFP-C3 ( Clontech Laboratories , Mountain View , CA ) to generate the following plasmids: pET15b-cytotardin , pcDNA3-cytotardin , pcDNA3-HA-cytotardin and pEGFP-cytotardin . All PCR-amplified constructs were verified by Sanger sequencing and have been deposited in GenBank ( http://www . ncbi . nlm . nih . gov/genbank ) under accession numbers KU295460–KU295467 . Polyclonal antibodies against HPLC-purified synthetic peptides ( C-terminal ) of lamin-1 , lamin-2 and cytotardin of H . dujardini were newly generated , following coupling to KLH ( key limpet hemocyanin ) ( Peptide Specialty Laboratories GmbH , Heidelberg , Germany ) . Anti-lamin-1 ( antigen: SNLDIHNDSVRDSPRSAG-C ) and anti-cytotardin ( antigen: EQKITETFKASGRVGPRTDW-C ) were purified from sera of immunised guinea pigs and anti-lamin-2 ( antigen: REMTQSSTRDDSYLGPSGLPKR-C ) from the serum of immunised rabbits ( Peptide Specialty Laboratories GmbH ) . For cryosectioning , specimens of H . dujardini were concentrated by filtering the culture medium through a polyamide mesh ( pore size: 30 µm ) . The concentrated specimens were transferred into an embedding medium for cryosectioning ( Tissue-Tek O . C . T Compound; Sakura Finetek , Staufen , Germany ) . Single drops of the medium containing the tardigrades were immediately frozen in dry ice-cooled 2-methylbutane . The frozen drops were cryosectioned into 5 µm thick sections . The sections were attached to SuperFrost Plus slides ( Menzel , Braunschweig , Germany ) , then dried at room temperature and stored at −80°C . The ice-cooled cryosections were fixed on slides with a 4% solution of formalin ( FA ) freshly prepared from paraformaldehyde in phosphate-buffered saline ( PBS; 0 . 1 M , pH 7 . 4 ) for 15 min . The sections were then shortly washed with an ammonium chloride solution ( 50 mM in PBS ) and rinsed in Tris-buffered saline ( TBS; 0 . 01 M , pH 7 . 6 ) two times for 3 min each . The anti-lamin-1 serum was applied to cryosections at a concentration of 1 . 9 µg/mL in TBS containing 1% bovine serum albumin ( BSA ) . Equally applied were the anti-lamin-2 serum with a concentration of 1 . 8 µg/mL and the anti-cytotardin serum with a concentration of 2 . 1 µg/mL . The incubation with the primary antibody solution was performed at room temperature for one hour . After two 3-min washing steps in TBS , a secondary antibody , either donkey anti-guinea pig Alexa Fluor 488 ( 3 µg/mL in TBS with 1% BSA; Jackson ImmunoResearch Laboratories , Hamburg , Germany ) for anti-lamin-1 and anti-cytotardin or goat anti-rabbit Alexa Fluor 568 ( 4 µg/mL in TBS with 1% BSA; Molecular Probes , Darmstadt , Germany ) for anti-lamin-2 , was applied at room temperature for one hour . The sections were then washed in TBS two times for 3 min and shortly rinsed with distilled water and afterwards with ethanol ( 100% ) . For whole-mount immunohistochemistry , specimens were first concentrated as described above and then pipetted into a 1 mL Eppendorf tube . After carefully removing excess water , the specimens were flash-frozen by placing each tube on dry ice . Frozen specimens were fixed immediately by applying a 4% solution of formaldehyde ( FA ) in PBS with 1% dimethyl sulfoxide ( DMSO ) . The fixative was applied at room temperature overnight . The whole-mounts were then washed in PBS two times for 10 min , two times for 30 min and two times for an hour . The specimens were cleared by dehydration in an ascending ethanol series ( 70% , 90% , 95% , 100% , 100% ) , applying xylene as a clearing agent two times for 3 min and rehydration in a descending ethanol series ( 100% , 100% , 90% , 70% , 50% ) . The whole-mount preparations were then washed at 37°C in PBS two times for 10 min . A mixture of collagenase/dispase ( each 1 mg/mL; Roche Diagnostics , Mannheim , Germany ) and hyaluronidase ( 1 mg/mL; Sigma-Aldrich , Munich , Germany ) diluted in PBS was applied at 37°C for 10 min . After the incubation with enzymes , a 15-min post-fixation with 4% FA in PBS followed at room temperature . The specimens were washed in an ammonium chloride solution ( 50 mM ) two times for 15 min , afterwards in PBS with 1% Triton X-100 ( Sigma-Aldrich ) two times for 10 min , two times for 30 min , once for an hour , overnight and then two times for 1 min . The whole-mount preparations were incubated in a blocking solution containing 10% normal goat serum ( NGS; Sigma-Aldrich ) , 1% Triton X-100 and 1% DMSO in PBS at room temperature for one hour . The anti-cytotardin serum with a concentration of 4 . 2 µg/mL in PBS with 1% NGS , 1% DMSO and 0 . 02% sodium azide was applied at room temperature for three days . After washing in PBS with 1% Triton X-100 and 1% DMSO three times for 5 min , two times for 15 min , four times for two hours , overnight and two times for 15 min a secondary antibody ( donkey anti-guinea pig Alexa Fluor 488; Jackson ImmunoResearch Laboratories ) with a concentration of 3 µg/mL in PBS with 1% NGS , 1% DMSO and 0 . 02% sodium azide was applied at room temperature for three days . Finally the specimens were washed in PBS with 1% Triton X-100 and 1% DMSO three times for 5 min , three times for 15 min and two times for one hour and afterwards in PBS ( without Triton X-100 and DMSO ) four times for 15 min . Either DAPI ( 1 µg/mL; Carl Roth , Karlsruhe , Germany ) , SYBRGreen I ( diluted 1:10 , 000; Molecular Probes ) , propidium iodide ( 0 . 5 µg/mL; Carl Roth ) or TO-PRO-3 iodide ( diluted 1:1 , 000; Molecular Probes ) were used as nuclear counterstain markers . Each of these fluorescent dyes was simply added to the solution containing the secondary antibody . Cryosections and whole-mount preparations were mounted in Prolong Gold antifade reagent ( Molecular Probes ) . The whole-mount preparations were mounted between two cover slips using petrolatum at all corners as spacer . All slides and cover slips were sealed at the edges with transparent nail polish for long time storage . Specimens were analysed with the confocal laser-scanning microscopes Zeiss LSM 780 ( Carl Zeiss Microscopy , Göttingen , Germany ) and Leica TCS STED ( Leica Microsystems CMS , Wetzlar , Germany ) . Escherichia coli BL21 ( DE3 ) were transformed with pET15b or pET15b-cytotardin vectors . Expression of recombinant protein was induced with 0 . 1 mM IPTG for three hours in Luria Bertani ( LB ) medium . Bacterial suspensions were centrifuged at 3 , 000 g for 10 min at 4°C . Bacterial pellets were then washed twice in PBS and lysed in 5x Laemmli buffer ( Volume in µL = OD600/10xVolume of culture in mL ) containing protease inhibitor ( #78439; Thermo Scientific , Waltham , MA ) , heated at 95°C for 10 min , briefly sonicated and heated again . Lysate of cells transformed with pET15b-cytotardin was diluted at 1:50 . Recombinant cytotardin protein expression was first assessed by Coomassie blue stained SDS-PAGE and further by Western blotting . Cells from the human cell line MCF-7 ( ATCC HTB-22 ) were cultured in Dulbecco’s modified Eagle’s medium ( DMEM , GE Healthcare , Little Chalfont , UK ) supplemented with 10% foetal bovine serum ( FBS , GE Healthcare ) , 100 U/mL penicillin , and 100 μg/mL streptomycin . Cells were transfected with pcDNA-cytotardin , pcDNA-HA-cytotardin , pEGFP-cytotardin and the corresponding empty vectors using Xfect Transfection reagent ( Clontech Laboratories ) according to the manufacturer’s protocol . For immunofluorescence microscopy , cells were grown on glass slides and fixed 24 hr after transfection . MCF-7 cells were washed twice with PBS and fixed either in methanol or in 4% FA freshly prepared from paraformaldehyde in PBS . For methanol fixation , cells grown on a coverslip were put in ice-cold methanol for 10 min , acetone for 1 min and dried at room temperature for 30 min . For FA fixation , cells were put in a 4% FA solution in PBS for 15 min at room temperature . After fixation , cells were washed in PBS and used for immunolabelling . FA-fixed cells were first permeabilised in PBS 0 . 1% Triton X-100 for 5 min and blocked in PBS 5% bovine serum albumin for 30 min . Methanol- and FA-fixed cells were then incubated overnight at 4°C with primary antibody solutions , washed three times , incubated in secondary antibodies + DAPI ( Sigma-Aldrich ) solution for one hour , washed two times in PBS , once in water , once in 100% ethanol , dried for 30 min and mounted in mounting medium ( Dianova , Hamburg , Germany ) . Primary antibodies were used at the following dilutions: guinea pig anti-cytotardin , 1:1 , 000; mouse anti-desmoplakin ( 11-5F ) , 1:150; mouse anti-keratin-8 ( Ks 8 . 7 ) , 1:100; mouse anti-HA-tag ( MMS-101P; Covance , Princeton , NJ ) , 1:100 . Images were taken using a confocal laser-scanning microscope ( LSM 780; Carl Zeiss Microscopy ) with 63×/1 . 46 NA oil immersion objective and Z-stack images were assembled by 'Maximum Intensity Projection' of the ZEN 2012 software ( Carl Zeiss Microscopy ) . Western blots were performed either on lysates of transfected MCF-7 cells or multiple specimens of H . dujardini . Transfected MCF-7 cells grown in 10 cm2 dishes were lysed in 200 µL 5x Laemmli sample buffer , boiled for 10 min at 98°C and briefly centrifuged prior to use . Concentrated specimens of H . dujardini were lysed in 800 µL 5x Laemmli buffer containing protease inhibitor , boiled for 10 min , sonicated and boiled again for 10 min . SDS-PAGE and Western blotting were performed as described previously ( Löffek et al . , 2010 ) . Primary antibodies , diluted in Tris-buffered saline containing 0 . 05% Tween 20 ( A4974; AppliChem , Darmstadt , Germany ) , were used at the following concentrations: anti-lamin-1 , 1:10 , 000; anti-lamin-2 , 1:5 , 000; anti-cytotardin , 1:10 , 000 , anti-HA tag ( MMS-101P; Covance ) , 1:1 , 000 and anti-GFP ( sc-5385; Santa Cruz Biotechnology , Dallas , TX ) , 1:1 , 000 . Anti-lamin-1 and anti-lamin-2 antibodies were only tested in Western blots based on lysates of tardigrade specimens .	Different proteins exist to support the stability of animal cells . The intermediate filament proteins are an important example . One type – called lamins – stabilizes the nucleus ( the structure within an animal cell that stores most of its DNA ) , while another forms scaffold-like structures in the rest of cell . The second type , referred to as “cytoplasmic” intermediate filaments , are not found in many hard-bodied creatures including insects and their closest relatives . This is probably because these animals , which are collectively known as arthropods , are instead supported by their tough external skeleton . The soft-bodied animals called tardigrades ( also known as water bears or moss piglets ) are closely related to the arthropods . These microscopic animals can endure extreme environmental conditions such as freezing . The tardigrade’s endurance is likely to require some way to stabilize the animal’s cells . This might involve cytoplasmic intermediate filaments , but nothing was known about these proteins in tardigrades . Now , Hering , Bouameur , Reichelt et al . have investigated if , and where , intermediate filaments are found in the cells of tardigrades . First , the complete set of active genes was analyzed for a species of tardigrade called Hypsibius dujardini; this revealed that three genes for intermediate filament proteins were active . Staining tissue slices or whole tardigrades with a marker that binds to intermediate filament proteins revealed that two of the three proteins were lamins and located within the nucleus . The third protein , which has been named "cytotardin" , was found outside of the nucleus . However , unlike well-known cytoplasmic intermediate filaments , this protein did not form scaffold-like structures throughout the cell . Instead , cytotardin formed belt-like filaments that encircled each cell in the skin of the tardigrades . Hering , Bouameur , Reichelt et al . then discovered that cytotardin seems to be more closely related to lamins than it is to cytoplasmic intermediate filaments . This suggests that cytotardin actually evolved from a tardigrade lamin and then acquired a new role in building filaments outside of the nucleus . The fact that cytotardin is only found in the skin of the tardigrade and in those tissues that experience mechanical stress ( for example , the mouth and legs ) hints that it might help stabilize these cells . This could mean that the protein also helps these animals to resist extreme conditions . Further studies should focus on clarifying cytotardin’s role in stabilizing cells , in particular if it is required for the tardigrades' tolerance to environmental stress .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology" ]	2016	Novel origin of lamin-derived cytoplasmic intermediate filaments in tardigrades
SUMO-modification of nuclear proteins has profound effects on gene expression . However , non-toxic chemical tools that modulate sumoylation in cells are lacking . Here , to identify small molecule sumoylation inhibitors we developed a cell-based screen that focused on the well-sumoylated substrate , human Liver Receptor Homolog-1 ( hLRH-1 , NR5A2 ) . Our primary gene-expression screen assayed two SUMO-sensitive transcripts , APOC3 and MUC1 , that are upregulated by SUMO-less hLRH-1 or by siUBC9 knockdown , respectively . A polyphenol , tannic acid ( TA ) emerged as a potent sumoylation inhibitor in vitro ( IC50 = 12 . 8 µM ) and in cells . TA also increased hLRH-1 occupancy on SUMO-sensitive transcripts . Most significantly , when tested in humanized mouse primary hepatocytes , TA inhibits hLRH-1 sumoylation and induces SUMO-sensitive genes , thereby recapitulating the effects of expressing SUMO-less hLRH-1 in mouse liver . Our findings underscore the benefits of phenotypic screening for targeting post-translational modifications , and illustrate the potential utility of TA for probing the cellular consequences of sumoylation . SUMO-modification or sumoylation with the Small Ubiquitin-like Modifier ( SUMO ) is a prevalent post-translational modification of many transcription factors and is generally associated with transcriptional repression ( Gill , 2005 ) . Similar to other ubiquitin-like modifications , the sumoylation cycle is multi-stepped , as reviewed in Gareau and Lima ( 2010 ) , and is initiated by E1 ( SAE1 ) , which forms a thioester bond with either SUMO-1 , 2 , or 3 . The single E2 ( UBC9 ) facilitates the hand-off and covalent conjugation of SUMO on a given protein substrate . Although E3s are believed to guide substrate selectivity in cells , only E1 and E2 are required for in vitro sumoylation ( IVS ) . Sumoylation is then reversed by sentrin-specific proteases ( SENPs ) . Genetic manipulations that either eliminate the sumoylation machinery or permanently disrupt the normal sumoylation cycle of a substrate can result in embryonic lethality or impaired organogenesis ( Flotho and Melchior , 2013; Kang et al . , 2010; Lee et al . , 2011a; Nacerddine et al . , 2005; Wang et al . , 2014 ) . Our lab and others find that sumoylation represents an important , ligand-independent mode to regulate the NR5A nuclear receptor subfamily that includes Steroidogenic Factor 1 ( NR5A1/SF-1 ) ( Campbell et al . , 2008; Lee et al . , 2011a; Lee et al . , 2005 ) and Liver Receptor Homolog 1 ( NR5A2/LRH-1 ) ( Chalkiadaki and Talianidis , 2005; Stein et al . , 2014; Venteclef et al . , 2010; Ward et al . , 2013; Yang et al . , 2009 ) . For instance , knocking-in an unsumoylatable or SUMO-less mutant SF-1 allele leads to profound changes in endocrine physiology and tissue development ( Lee et al . , 2011a ) . Importantly , even in the presence of one wild-type ( WT ) allele , the phenotypic effects of SUMO-less SF-1 dominate . Mechanistically , we find that SUMO-less SF-1 can regulate select downstream targets , which we refer to as SUMO-sensitive ( Campbell et al . , 2008 ) . Indeed , sonic hedgehog ( SHH ) signaling is ectopically activated after expressing SUMO-less SF-1 in both cells and tissues . This gain-of-function or dominance of the SUMO-less SF-1 mutant leads to expansion of select cell types and hormone imbalance , illustrating how disrupting the normal cycle of substrate sumoylation results in disease states . Consistent with our findings , gain-of-function heterozygous missense mutations in the sumoylation site of the transcription factor microphthalmia-associated transcription factor are tightly linked with some forms of human melanoma ( Bertolotto et al . , 2011 ) . SUMO-less variants of the androgen ( AR ) and glucocorticoid hormone ( GR ) receptors are also reported to activate new transcriptional programs linked to cellular proliferation ( Paakinaho et al . , 2014; Sutinen et al . , 2014 ) . These studies suggest that successful efforts to chemically target substrate sumoylation could be used to alter transcription factor activity , to either promote or attenuate SUMO-sensitive genetic programs . Thus far , efforts to drug sumoylation using in vitro target-based assays and in silico screens have identified different classes of SUMO inhibitors with IC50s in the micromolar range ( Table 1 ) . In vitro target-based screens rely exclusively on defined components ( E1 , UBC9 and a test substrate ) , and as such , exclude the contributions by E3s and other unidentified obligate cofactors on substrate sumoylation . An in situ cell-based screen in permeabilized , fixed cells partially overcomes this limitation but still requires the addition of exogenous SUMO components for the assay ( Hirohama et al . , 2013 ) . Currently , inhibitors that target E1 include ginkgolic acid ( GA ) , davidiin , and kerriamycin B . Inhibitors of UBC9 include 2-D08 , GSK145A , and spectomycin B1 . While GA remains the most widely used and commercially available chemical probe targeting general sumoylation , its efficacy as an inhibitor can vary greatly depending on the assay and substrate being tested ( Bogachek et al . , 2014; Kim et al . , 2013; Tossidou et al . , 2014 ) . 10 . 7554/eLife . 09003 . 003Table 1 . List of sumoylation inhibitors identified by screen type and reported IC50 values . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 003CompoundClassScreenLibraryAssaySubstrateTargetIC50 ( µM ) Reference2-D08FlavonoidTarget500 FlavonesIVSAR PeptideUBC96 . 0 ( Kim et al . , 2013 ) DavidiinEllagitanninTarget750 ExtractsIn SituRanGap1E10 . 15 ( Takemoto et al . , 2014 ) Ginkgolic acidAlkylphenolTarget500 ExtractsIn SituRanGap1E13 . 0 ( Fukuda et al . , 2009a ) GSK145ADiamino-pyrimidineTargetGSK LibraryIVSTRPS1 PeptideUBC912 . 5 ( Brandt et al . , 2013 ) Kerriamycin BAntibioticTarget1800 BrothsIn SituRanGap1E111 . 7 ( Fukuda et al . , 2009b ) Spectomycin B1AntibioticTargetChemical LibraryIn SituRanGap1UBC94 . 4 ( Hirohama et al . , 2013 ) C#21Phenyl UreaVirtualMaybridgeDockingRanGap1E114 . 4 ( Kumar et al . , 2013 ) Tannic acidGallotanninPhenotypicPharmakonqPCRhLRH-1E112 . 8 ( This Study ) Phenotypic cell-based screens offer an alternative approach to in vitro target-based screens for finding new molecular entities ( Swinney and Anthony , 2011 ) . Here , we set out to identify nontoxic chemical probes that would modulate substrate sumoylation using a cell-based gene-expression screen , which assayed two human LRH-1 ( hLRH-1 ) SUMO-sensitive transcripts as the primary readout . hLRH-1 is an ideal test substrate for evaluating any potential hits because one has the ability to test how candidate hits affect hLRH-1 activity in both immortalized hepatocellular carcinoma cells and in primary mouse hepatocytes ( Lee et al . , 2011b; Mataki et al . , 2007; Oosterveer et al . , 2012 ) . In addition , as shown in this study , hLRH-1 is well-sumoylated in vitro , in cells , and in vivo . From the initial phenotypic screen of the FDA- and European-approved Pharmakon 1600 drug library , the commercial plant extract , tannic acid ( TA ) was identified as a nontoxic general sumoylation inhibitor , which was effective in multiple platforms , including primary mouse hepatocytes . Sumoylation of hLRH-1 occurs primarily in the flexible hinge domain on two major conserved acceptor lysines K192 and K270 , with a minor site located in the N-terminal region at K44 ( Figure 1A ) . Similar to our prior results obtained with SF-1 , hLRH-1 is efficiently sumoylated ( ~30% ) in human placental choriocarcinoma JEG3 and hepatocellular carcinoma HepG2 cells expressing Flag-hLRH-1 ( Figure 1A ) . Importantly , multiple sumoylated hLRH-1 species are readily detected with only the endogenous SUMOylation machinery and without the need to add exogenous SUMO or UBC9 . Substituting lysines K192 and K270 with arginines ( K192R/K270R or 2KR ) eliminates nearly all sumoylated LRH-1 species , as previously noted ( Chalkiadaki and Talianidis , 2005; Lee et al . , 2005; Yang et al . , 2009 ) and Figure 1—figure supplement 1 . In vivo sumoylation of hLRH-1 is also equally robust , as observed in mouse liver humanized to express wild-type hLRH-1 ( WT ) ( Figure 1B ) , following viral-mediated infection with recombinant adeno-associated virus serotype 8 ( AAV8 ) ( Cotugno et al . , 2012; Ill et al . , 1997 ) and Figure 1C , D—figure supplement 2 . Impressively , the extent and pattern of hLRH-1 sumoylation in mouse liver are identical to those found in cultured cell lines , demonstrating for the first time that hLRH-1 is efficiently sumoylated at multiple lysines in vivo . By contrast , expressing SUMO-less hLRH-1 ( 2KR ) eliminates nearly all hLRH-1 sumoylation ( Figure 1B ) . These collective data establish that hLRH-1 is robustly sumoylated in several platforms , making it an excellent test substrate for assessing both the biochemical and functional effects of small molecule inhibitors of sumoylation . 10 . 7554/eLife . 09003 . 004Figure 1 . Human LRH-1 is efficiently sumoylated in cells and in vivo . ( A ) Schematic of hLRH-1 protein ( NR5A2 isoform 1 ) showing the location of major sumoylation sites at K192 and K270 , and the minor K44 site ( top panel ) . WT and SUMO-less forms of hLRH-1 ( 3KR and 2KR ) expressed in JEG3 and HepG2 cells are indicated as detected with anti-Flag antibody . Unsumoylated ( hLRH-1 ) as well as sumoylated hLRH-1 species ( 1x , 2x , and 3x ) are indicated in bottom panel by arrows . Additional bands observed in HepG2 cells that persist after mutating both K192 and K270 are indicated with asterisk ( * ) . Strategy used to humanize mouse liver for expression of wild type or SUMO-less ( 2KR ) hLRH-1 . ( B ) Sumoylated hLRH-1 species detected by anti-Flag in harvested livers after first infecting with AAV8-vectors expressing eGFP , WT or 2KR . ( C ) Relative transcripts levels of hLRH-1 transcripts in mouse livers infected with recombinant AAV8-vectors expressing eGFP , wild-type hLRH-1 ( WT ) or SUMO-less hLRH-1 ( 2KR ) . ( D ) Staining for tagged-hLRH-1 as detected by immunofluorescence using anti-Flag ( white arrows ) . Hepatocytes are prepared as described in ‘Materials and methods’ from harvested , perfused livers 2 weeks post retro-orbital viral-mediated infection . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 00410 . 7554/eLife . 09003 . 005Figure 1—figure supplement 1 . Mutating individual acceptor lysines in hLRH-1 establishes the importance of K192 and K270 in SUMO modification of hLRH-1 . Schematic of hLRH-1 with position of 3X-Flag epitope tag and three sumoylation sites in hLRH-1 ( upper panel ) . Sumoylation pattern in HepG2 cells expressing WT or mutant forms ( K192R , K270R , or 2KR ) of Flag-Tagged hLRH-1 . * = non-specific bands observed in all mutants forms of hLRH-1 in HepG2 cells ( lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 00510 . 7554/eLife . 09003 . 006Figure 1—figure supplement 2 . Human LRH-1 transcripts and protein are expressed in liver after AAV8-TBG viral infection . ( A ) Relative expression of hLRH-1 or endogenous mLrh-1 transcripts in mouse liver 14 days post-infection with either AAV8-eGFP ( GFP ) or AAV8 hLRH-1 ( hLRH-1 ) at a vector genome titer of 1 x 1011 ( genome copies/ml or GC/ml ) . ( B ) Human LRH-1 detected by anti-Flag antibody in heart and liver tissue collected from mice expressing Flag-hLRH-1 ( hLRH-1 ) or eGFP as described in ‘Materials and methods’ . Human Flag-tagged LRH-1 protein expressed in HepG2 cells is indicated with arrow ( far left lane ) and loading controls for each sample ( Gapdh ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 006 Rather than assessing substrate sumoylation directly , we used a cell-based screen that monitored two SUMO-sensitive transcripts as the endpoint ( diagrammed in Figure 2A ) . The JEG3 cell line was initially chosen because it performed well in all steps of the primary screen and has been used previously to study NR5A activity ( Campbell et al . , 2008 ) . Profiling was carried out on JEG3 cells stably expressing hLRH-1 or the SUMO-less hLRH-1 mutant , or after siRNA knock down of UBC9 ( siUBC9 ) to identify the most robust SUMO-sensitive genes . To ensure that minor sumoylation on hLRH-1 was eliminated we used the 3KR mutant that disrupts K44 , as well as the two major acceptor lysines in the hinge region . Two genes , APOC3 and MUC1 , were identified by microarray as the readout transcripts for the screen . These genes are highly induced by either SUMO-less LRH-1 or siUBC9 ( Figure 2B ) and were chosen as two SUMO-sensitive genes in the primary screen assay . Interestingly , APOC3 can be directly regulated by LRH-1 ( Hwang-Verslues and Sladek , 2008 ) , whereas MUC1 is regulated by the androgen receptor ( Rajabi et al . , 2011 ) . In contrast , expression of a well-known NR5A downstream target gene , CYP11A1 is unaffected by both SUMO-less LRH-1 and siUBC9 knockdown and is thus designated as a SUMO-insensitive LRH-1 target ( Figure 2B ) . 10 . 7554/eLife . 09003 . 007Figure 2 . A Phenotypic screen identifies TA as a small-molecule sumoylation modulator . ( A ) Schematic outline of the primary screen using JEG3 cells expressing wild-type hLRH-1 and downstream filtering steps to identify small molecules that modulate sumoylation . Individual amplification profiles for MUC1 transcripts are shown for each drug treatment using the Pharmakon 1600 library ( upper right panel ) . Highlighted in red is the amplification curve of MUC1 obtained with TA , the top hit from the primary screen . ( B ) Relative levels of SUMO-sensitive transcripts APOC3 , MUC1 and the SUMO-insensitive transcript CYP11A1 in JEG3 cells expressing wild-type hLRH-1 or SUMO-less hLRH-1 ( 3KR ) ( top panel ) . Relative levels of transcripts as above are shown after 72 hr siControl ( Con ) or siUBC9 ( UBC9 ) treatment in JEG3 cells expressing hLRH-1 . Results represent values obtained for triplicate samples . Statistical significance: **p<0 . 0001 , p<0 . 01 , p<0 . 05 . ( C ) Scatter plot from the primary screen showing normalized Z-scores for APOC3 and MUC1 , calculated as described in ‘Materials and methods’ . All compounds yielding Z-scores greater than +2 or less than -2 are shown within red dashed boxes . Positive Z-scores correspond to increased expression of transcripts relative to the control housekeeping gene , TBP . The Z-score obtained for TA is indicated as red dot . TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 007 A gene-expression-based screen adapted from ( Arany et al . , 2008 ) assayed APOC3 and MUC1 with a 1600-compound drug library ( Pharmakon 1600 ) . JEG3 hLRH-1 cells were cultured in a 384-well format , treated with 10 µM of each drug , and measured for APOC3 and MUC1 transcripts . Robust Z-scores for each drug treatment were obtained by normalizing the amplification cycle number ( CT ) of APOC3 or MUC1 to TBP as an internal control ( △CT ) , and then to the DMSO external control ( △△CT ) . A scatter plot of Z-scores ± 2SD from the primary screen shows 13 drugs producing significant changes in MUC1 and APOC3 expression ( up and down ) , representing a 0 . 8% hit rate ( Figure 2C and Source data 1 ) . Drugs resulting in cytotoxicity or changing TBP expression were then filtered out leaving six potential hits listed in Table 2 . 10 . 7554/eLife . 09003 . 008Table 2 . List of top six hits from primary screen ( Z-scores ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 008DrugAPOC3 MUC1 Tannic acid4 . 874 . 31Trifluridine4 . 012 . 71Taxol2 . 93-0 . 26Vincristine-3 . 61-1 . 01Colforsin1 . 24-5 . 70Ouabain-5 . 14-0 . 38 All potential nontoxic candidates were repurchased and tested in a dose response for induction of the two SUMO-sensitive genes used in the primary assay . Only TA emerged as a valid hit from the primary screen , showing significant induction of APOC3 and MUC1 , but not CYP11A1 ( Figure 3A ) . We then asked if hLRH-1 sumoylation is required to observe the stimulatory effects of TA on APOC3 and MUC1 . As expected , TA had no effect on CYP11A1 regardless of the hLRH-1 variant tested ( Figure 3B ) . On the other hand , the dose-dependent effects of TA on APOC3 are lost when tested with the SUMO-less hLRH-1 ( 3KR ) , demonstrating that activation of APOC3 by TA depends on the ability of hLRH-1 to be sumoylatable . Further , APOC3 upregulation by TA depends on hLRH-1; TA failed to change APOC3 alone ( Figure 3—figure supplement 1 ) . On the other hand , the effects of TA on MUC1 levels are largely independent of hLRH-1 , implying that TA can act more broadly on non-hLRH-1 SUMO-sensitive targets . 10 . 7554/eLife . 09003 . 009Figure 3 . TA enhances SUMO-sensitive gene expression in cells . ( A ) Relative expression of CYP11A1 , APOC3 , and MUC1 in JEG3 cells expressing wild-type ( WT ) hLRH-1 following 24 hr treatment with increasing TA . ( B ) Relative expression of CYP11A1 , APOC3 , and MUC1 in JEG3 cells transiently expressing WT or SUMO-less hLRH-1 ( 3KR ) following 6 hr treatment with increasing concentrations of TA as indicated . Vehicle DMSO control is shown ( - ) . ( C ) Levels of unsumoylated ( hLRH-1 ) and sumoylated hLRH-1 species in JEG3 cells detected with anti-Flag following 5 hr treatment with increasing TA concentrations ( 1–50 µM ) . Shorter exposure of 1x-sumoylated hLRH-1 is shown in bottom panel ( 1× ) . Vehicle control ( DMSO ) is shown ( - ) . ( D ) Levels of unsumoylated ( hLRH-1 ) and sumoylated hLRH-1 species in JEG3 cells detected with anti-Flag after TA , trifluidine ( Tri . ) , 2-D08 , and ginkgolic acid ( GA ) treatment ( 30 µM each ) following 5 hr treatment . Shorter exposure of unsumoylated hLRH-1 ( hLRH-1 ) is shown in bottom panel . ( E ) Cell viability is shown for JEG3 cells expressing WT hLRH-1 following 5 hr treatment with increasing GA or TA . Statistical significance: **p<0 . 0001 , p<0 . 05 . TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 00910 . 7554/eLife . 09003 . 010Figure 3—figure supplement 1 . Upregulation of APOC3 by TA in JEG3 cells depends on presence of hLRH-1 . Relative levels of endogenous APOC3 in JEG3 cell lines expressing either a control vector ( iEV ) or wild-type hLRH-1 ( WT ) after treatment with no ( - ) or increasing concentrations of TA for 6 hr . TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 01010 . 7554/eLife . 09003 . 011Figure 3—figure supplement 2 . Significant cell toxicity in JEG3 wtLRH-1 cells after 24 hr treatment with GA but not TA . Percent cell viability in JEG3 cells expressing wild-type hLRH-1 ( JEG3 + hLRH-1 ) following 24 hr treatment with increasing concentrations of either TA ( red ) or GA ( blue ) . GA: Ginkgolic acid; TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 011 The transcriptional effects of TA correlated well with the dose-dependent inhibition of hLRH-1 sumoylation in cells , as evidenced by a ~40–50% decrease in all hLRH-1 sumoylated species at 10 µM TA ( Figure 3C ) . Diminished hLRH-1 sumoylation is observed only after TA and GA treatment; the other published sumoylation inhibitor , 2’ , 3’ , 4’-trihydroxyflavone ( 2-D08 ) ( Kim et al . , 2013 ) , and another top screening hit , trifluridine ( Tri ) , failed to show similar results ( Figure 3D ) . However , while GA appears more effective than TA at reducing hLRH-1 sumoylation in JEG3 cells ( Figure 3D ) , this compound leads to significant cytotoxicity beginning at 10 µM after 5 hr ( Figure 3E ) and 24 hr exposure ( Figure 3—figure supplement 2 ) . By contrast , TA is nontoxic even at higher concentrations . Next , we asked whether TA would also inhibit IVS of recombinant full-length human LRH-1 ( FL-hLRH-1 ) protein . In our assay conditions , the pattern of sumoylated FL-hLRH-1 in vitro is identical as that found in cells and in vivo , and collapses down to a single unmodified band after addition of SENP1 ( Figure 4A and refer back to Figure 1 ) . Using IVS conditions that achieve ~50% sumoylation of FL-hLRH-1 , TA is the most effective sumoylation inhibitor when compared to 2-D08 and GA , as well as other candidate hits , ( Figure 4B ) , with an apparent IC50 of 12 . 8 µM ( Figure 4C and Figure 4—figure supplement 1 ) . As predicted from our initial cellular data , TA inhibits and impairs the rate of sumoylation of other substrates , including recombinant hinge-LBD SF-1 , full-length IκBα , and an AR peptide ( Figure 4—figure supplement 2 ) . As found for davidiin , another tannin sumoylation inhibitor ( Takemoto et al . , 2014 ) , TA impairs E1 thioester formation in non-reducing conditions ( Figure 4D ) . 10 . 7554/eLife . 09003 . 012Figure 4 . TA is a detergent-resistant inhibitor of substrate sumoylation in vitro . ( A ) In vitro sumoylation ( IVS ) of recombinant full length ( FL ) -hLRH-1 , without ATP , with ATP or with ATP and recombinant SENP1 added to IVS reactions as described in ‘Materials and methods’ . ( B ) IVS assays with increasing tannic acid ( TA ) , trifluidine ( Tri . ) , 2-D08 , and ginkgolic acid ( GA ) . Sumoylated and unsumoylated FL-hLRH-1 are indicated with arrows as detected with anti-LRH-1 antibody . ( C ) IC50 of TA in FL-hLRH-1 IVS assay . Data are represented as mean ± SEM from at least three independent replicates . ( D ) Formation of E1 thioester with or without TA , in non-reducing conditions ( -DTT ) . Effects of TA ( 10 µM ) on formation of SUMO-E1 complex ( SUMO-SAE1 , top band ) compared to reducing conditions without TA ( -DTT , last lane ) . SUMO1 dimers are formed in non-reducing conditions ( SUMO1 Dimer ) . E1 thioester formation assays are initiated by addition of freshly prepared ATP ( 10 mM ) and described in ‘Materials and methods’ . Anti-SUMO1 antibody was used to detect SUMO1 species . ( E ) Levels of sumoylated and unsumoylated FL-LRH-1 in IVS assay with TA ( 15 and 30 μM ) and in the presence or absence of Triton X-100 in vitro ( left panel ) . Bar graph of quantified data showing percent inhibition of hLRH-1 sumoylation by TA and with or without Triton X-100 ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 01210 . 7554/eLife . 09003 . 013Figure 4—figure supplement 1 . Effects of TA , other candidate hits , and published sumoylation inhibitors in an IVS assay of full-length hLRH1 . ( A ) Dose-dependent inhibition of full-length ( FL ) -hLRH-1 by TA compared to other top candidate hits from primary screen as assayed by IVS . IVS assay and immunoblotting conditions used to detect hLRH-1 species are described in Materials and Methods . Sumoylated hLRH-1 ( 1x , 2x , 3x ) and unmodified LRH-1 ( LRH-1 ) species are indicated by arrows . ( B ) IVS assays of FL-hLRH-1 were performed with increasing concentrations of TA are shown ( left panel ) and plotted as normalized values in graph ( right panel ) . Effects of two other published sumoylation inhibitors , 2-D08 and GA are also shown in graph . IVS data was normalized to DMSO control for each compound , and then plotted as percent conversion per log10 [µM] concentration . Curve fitting of data is described in ‘Materials and methods’ . GA: Ginkgolic acid; IVS: In vitro sumoylation; TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 01310 . 7554/eLife . 09003 . 014Figure 4—figure supplement 2 . IVS of multiple substrates inhibited by TA . ( A ) Coomassie staining of recombinant hinge-LBD mSF-1 protein ( aa178-462 ) with DMSO ( 0 ) or with 10 µM TA . This protein fragment of mSF-1 contains only one of the two conserved sumoylation consensus sites at K194 . Migration of SUMO1-SF-1 and unmodified SF-1 hinge-LBD ( SF-1 ) are indicated by arrows . ( B ) IVS of full length IκBα without ( 0 ) or with increasing concentrations of TA , as indicated , with SUMO-IκBα and unmodified IκBα indicated by arrows . ( C ) Dose-dependent inhibition of IVS of fluorescently labeled AR peptide by TA with the IC50 provided . Data are plotted as percent conversion versus TA concentration ( log10 [µM] , left panel ) . Real-time sumoylation of AR peptide ( % Conversion ) are plotted at different concentrations of TA ( right panel ) . IVS conditions and detection of AR peptide sumoylation by electrophoretic mobility shift assay are previously described in Kim et al . , ( 2013 ) . IVS: In vitro sumoylation; TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 014 Given that TA and other polyphenols are prone to aggregate formation which can lead to non-specific or promiscuous inhibition ( Feng and Shoichet , 2006 ) , we assessed the performance of TA in the presence of a non-ionic detergent ( Triton X-100 ) , which limits aggregate formation ( Pohjala and Tammela , 2012 ) . Even in the presence of 0 . 01% Triton X-100 , FL-LRH-1 sumoylation was inhibited by TA treatment , as shown in Figure 4E . That TA maintains its inhibitory activity with this detergent suggests that in these assay conditions , the inhibitory effects of TA are not due to non-specific aggregate formation . Next , we tested whether TA is also effective in hepatocellular carcinoma HepG2 cells , a relevant cell line for studying hLRH-1 ( Figure 5A ) . When compared to a nontoxic dose of GA ( 10 µM ) or 2-D08 , TA is much more efficient at decreasing levels of sumoylated hLRH-1 and shows little cytotoxicity in HepG2 cells after longer exposure times ( Figure 5B ) . We also directly compared the effects of TA versus knockdown of UBC9 ( siUBC9 ) on hLRH-1 and other sumoylated proteins in HepG2 cells . Surprisingly , despite a substantial loss of UBC9 transcripts ( 95% ) and protein ( 60% ) following siRNA-mediated knockdown for 72 hr ( Figure 5—figure supplement 1 ) , hLRH-1 remained fully sumoylated ( Figure 5C ) . On the other hand , levels of hLRH-1 sumoylation decreased in a dose-dependent manner with TA ( Figure 5C ) . The higher migrating sumoylated hLRH-1 species in HepG2 cells that diminish with TA were authenticated as SUMO1 or SUMO2 species ( Figure 5D ) . Interestingly , 1x-Su-hLRH-1 appears to be exclusively modified by SUMO1 , whereas higher Su-hLRH-1 species are modified by SUMO2 . Both siUBC9 and TA inhibit global sumoylation , but in this setting , siUBC9 is slightly more effective ( Figure 5E ) . As expected , neither TA nor siUBC9 changes the pool of ubiquitinated proteins . Lastly , we wanted to know if TA similarly modulates sumoylation of endogenously expressed LRH-1 . Unfortunately , our ability to cleanly detect or efficiently pulldown endogenous sumoylated LRH-1 species in cells/tissues is difficult with available anti-LRH-1 antibodies , including both commercial and non-commercial sources . Instead , we used human adrenal carcinoma H295R cells that express high levels of endogenous hSF-1 and found that similar to exogenously expressed hLRH-1 , TA decreases levels of sumoylated hSF-1 in H295S cells ( Figure 5F ) . Similar results were also obtained for endogenous RanGap ( Figure 5—figure supplement 2 ) . Taken together , these data show that TA is a nontoxic potent global sumoylation inhibitor . 10 . 7554/eLife . 09003 . 015Figure 5 . TA inhibits exogenous and endogenous NR5A sumoylation , as well as general sumoylation in cells . ( A ) Levels of unsumoylated ( LRH-1 ) and sumoylated hLRH-1 species in HepG2 cells detected with anti-Flag following TA , 2-D08 , or GA at specified concentrations after 24 hr treatment . Vehicle control ( DMSO ) is shown ( - ) . ( B ) Cell viability for HepG2 cells expressing wild-type hLRH-1 following 24 hr treatment with increasing concentrations of GA or TA . ( C ) Sumoylation of wild-type hLRH-1 expressed in HepG2 cells and detected with anti-Flag after siCont or siUBC9 knockdown for 72 hr ( + ) , and with increasing TA ( 6 h ) . Shorter exposure of 1x-sumoylated hLRH-1 is shown in panel below ( 1x ) , as well as loading control ( GAPDH ) . Vehicle control ( DMSO ) is shown ( - ) . ( D ) Flag-tagged hLRH-1 protein immunoprecipitated by anti-Flag in HepG2 cells treated with vehicle or TA ( 30 or 50 μM ) followed by immunoblot with either anti-SUMO1 ( top panel ) or anti-SUMO2 ( bottom panel ) . Arrows indicated migration of 1x , 2x , and 3x sumoylated hLRH-1 species . ( E ) Levels of total sumoylated or ubiquitinated proteins in HepG2 cells following siControl and siUBC9 ( 72 hr ) or TA treatment ( 6 hr ) , as detected by anti-SUMO1 , -SUMO2 , or -ubiquitin . ( F ) Effects of TA ( 6 hr ) on endogenous SF-1 sumoylation in H295R cells and detected anti-SF-1 antibody . GA: Ginkgolic acid; TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 01510 . 7554/eLife . 09003 . 016Figure 5—figure supplement 1 . UBC9 transcripts and protein levels following siUBC9 knockdown in HepG2 hLRH-1 cells . ( A ) Relative expression of UBC9 ( left panel ) and hLRH-1 ( right panel ) in HepG2 cells following 72 hr exposure to siControl ( siCont ) or siUBC9 as described in ‘Materials and methods’ . ( B ) Immunoblot of UBC9 protein following 72 hr exposure to siCont or siUBC9 with percentage decrease of UBC9 protein shown in bar graph ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 01610 . 7554/eLife . 09003 . 017Figure 5—figure supplement 2 . TA attenuates endogenous RanGap sumoylation . Endogenous sumoylated ( Su-RanGap 1x ) and unmodified RanGap ( RanGap ) in HepG2 hLRH-1 cells after siCont , siUBC9 , or after TA treatment ( 6 hr ) with concentrations indicated . Note that the anti-RanGap rabbit monoclonal antibody from GeneTex recognizes both 1x Sumoylated RanGap and unmodified RanGap . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 017 To ask if TA alters SUMO-sensitive targets in HepG2 cells , we identified a small subset of target genes that is ( 1 ) upregulated by hLRH-1 , ( 2 ) bound by LRH-1 as detected by chromatin-immunoprecipitation high-throughput sequencing ( ChIP-Seq ) and ( 3 ) altered by TA in the presence of hLRH-1 after profiling HepG2 cells ( Source data 2 ) . Of the 42 genes in this small subset , we identified three genes that were considered SUMO-sensitive and also found to be putative LRH-1 target genes by ChIP-Seq analyses ( Source data 2 ) , CYP24A1 , PFKFB3 and SERPINE1 ( Figure 6A , B ) . TA regulates all three genes in a dose-dependent manner ( Figure 6C ) . Moreover , using ChIP-qPCR we find a significant increase in hLRH-1 occupancy on these LRH-1 binding sites , as shown for a site in the SERPINE1 proximal promoter region and also for an intronic site in CYP24A1 ( Figure 6D ) . These cellular data suggest that TA modulates and increases recruitment to SUMO-sensitive hLRH-1 target genes in a relevant HepG2 cellular model system . 10 . 7554/eLife . 09003 . 018Figure 6 . TA increases expression and promotes hLRH-1 occupancy on target genes . ( A ) Venn diagram representing the overlap between transcripts changed by induction of hLRH-1 ( +Dox ) ( Blue ) and hLRH-1+TA ( +Dox , +TA 30 µM ) ( Yellow ) , as well as hLRH-1 binding sites identified by ChIP-Seq in HepG2 cells ( +Dox , Orange ) . Heat map of top three genes from overlapping set of 42 genes: PFKFB3 , SERPINE1 ( PAI1 ) , and CYP24A1 showing changes after induction of hLRH-1 ( Dox/EtOH ) and then after TA treatment ( TA/DMSO ) . ( B ) ChIP-Seq binding profiles of the three hLRH-1 targets from panel A . Representative views for ChIP-Seq peaks called by MACS are shown along with genomic location and consensus sequence of putative hLRH-1 binding sites ( red text ) . ( C ) Relative expression of three hLRH-1 targets in HepG2 cells from ( A and B ) before ( -Dox ) and after induction of hLRH-1 ( +Dox ) and following treatment with TA for 6 hr . ( D ) ChIP-qPCR results in HepG2 cells expressing hLRH-1 for regions identified in panel B , with vehicle control ( - ) and TA ( 6 hr ) . Statistical significance for panels C and D: *p<0 . 001 , p<0 . 01 , p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 01810 . 7554/eLife . 09003 . 019Figure 6—figure supplement 1 . Quantification of transcriptional changes in HepG2 cells after TA or siUBC9 treatment and in presence of hLRH-1 . ( A ) Venn diagram representing overlap between transcriptionally responsive genes by siUBC9 ( Blue ) or TA ( Red ) ; data analyses are described in ‘Materials and methods’ . ( B ) Motif search of HepG2 WT hLRH-1 binding sites as described in ‘Materials and methods’ yields a consensus sequence with the frequency distribution quantified as bits and the E-value ( probability ) . The 8 bp consensus site obtained after Motif search is shown after induction of WT hLRH-1 ( +Dox ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 019 We then determined if TA functions as a nontoxic sumoylation inhibitor in primary mouse hepatocytes and if TA is able to recapitulate gene expression changes observed with SUMO-less hLRH-1 . Limited profiling was carried out to identify SUMO-sensitive genes in infected mouse liver overexpressing WT or SUMO-less hLRH-1 ( 2KR ) . Both WT and SUMO-less hLRH-1 were expressed equivalently with levels ~10-fold higher than that of endogenous mLRH-1 , as judged by immunoblots using an anti-LRH-1 antibody that detects both mouse and hLRH-1 ( Figure 7A ) . Expression of endogenous mLrh-1 is unaffected after infecting mice with AAV8-hLRH-1 ( refer back to Figure 1—figure supplement 2 ) . Although WT and SUMO-less hLRH-1 upregulate the classic LRH-1 target Cyp8b1 ( Lee et al . , 2008 ) , expression of SUMO-less hLRH-1 leads to robust activation of adiponectin ( Adipoq ) and sonic hedgehog ( Shh ) , in mouse liver ( Figure 7B ) . The pattern of hLRH-1 sumoylation is preserved in cultured hepatocytes , but lost in primary cultures expressing the SUMO-less hLRH-1 mutant ( Figure 7C , left panel ) . Importantly , when tested in primary mouse hepatocytes TA ( 10 µM , 5 hr ) diminishes levels of sumoylated WT hLRH-1 , with nearly all sumoylated species absent including 1x-SUMO-hLRH-1 ( Figure 7C , right panel ) ; thus closely resembling the SUMO-less 2KR mutant . 10 . 7554/eLife . 09003 . 020Figure 7 . TA mimics SUMO-less hLRH-1 when expressed in humanized mouse primary hepatocytes . ( A ) Relative expression of hLRH-1 in mouse livers 2 weeks post-infection with amount of vector indicated ( GC/ml ) . Hepatic expression of endogenous mouse mLRH-1 ( red arrow ) in control C57BL/6 mice , in LKO mice ( mLRH-1f/f;Alb-Cre ) , or in C57BL/6 mice 2 weeks post infection with recombinant AAV8-eGFP ( eGFP ) , AAV8-WT-hLRH-1 ( WT ) , or AAV8-2KR-hLRH-1 ( 2KR ) ( black arrow ) . Endogenous mLRH-1 and exogenous hLRH-1 are detected with an anti-LRH-1 antibody . Note that non-specific bands detected with the anti-LRH-1 antibody obscure sumoylated LRH-1 species . ( B ) Relative expression in mouse liver of the classic LRH-1 target , Cyp8b1 or SUMO-sensitive LRH-1 targets Adipoq and Shh ( and its downstream target Gli2 ) following infection with either AAV8-eGFP , AAV8 hLRH-1 or a SUMO-less AAV8-2KR-hLRH-1 . Each bar represents values obtained from three livers . Values below the threshold of detection ( qPCR >40 cycles ) are indicated as ND . ( C ) Sumoylation pattern of AAV8 hLRH-1 ( WT ) and AAV8-2KR hLRH-1 ( 2KR ) in cultured mouse primary hepatocytes ( left panel ) . Sumoylation of hLRH-1 in infected primary mouse hepatocytes treated with increasing TA for 5 hr ( right panel ) . Sumoylated species and unsumoylated Flag-hLRH-1 are indicated by arrows and detected by anti-Flag antibody . ( D ) Relative expression of genes shown in Panel B measured in uninfected primary mouse hepatocytes treated for 5 hr with increasing concentrations of TA . Statistical significance for panels B and D: p<0 . 01 , p<0 . 01 . TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 02010 . 7554/eLife . 09003 . 021Figure 7—figure supplement 1 . No cellular toxicity in primary hepatocytes by TA . Cell viability ( % ) is shown for primary hepatocytes treated with increasing TA concentrations as indicated for 24 h . TA: Tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 09003 . 021 SUMO-sensitive genes identified in whole liver were then tested with TA in mouse primary hepatocytes that express only endogenous mLRH-1 . As found in the liver , Shh , its downstream effector Gli2 , and Adipoq are essentially switched on by TA ( Figure 7D ) . This trigger-like effect of TA and SUMO-less hLRH-1 on Shh and Gli2 is consistent with our previous findings that hedgehog signaling can be ectopically activated in mouse organs after knocking-in SUMO-less SF-1 ( Lee et al . , 2011a ) . Importantly , cell viability was unchanged by TA after 24 hr exposure ( Figure 7—figure supplement 1 ) . These data establish that in primary cells , TA abrogates hLRH-1 sumoylation and induces hLRH-1 SUMO-sensitive downstream target genes . Here , using a phenotypic , gene-expression-based screen we identify the polyphenol , TA as a potent inhibitor of hLRH-1 sumoylation in multiple platforms . In vitro assays confirm that TA impairs substrate sumoylation . When tested n several cellular model systems , an acute , nontoxic dose of TA treatment markedly reduces levels of sumoylated hLRH-1 , as well as hSF-1 , and modulates expression of hLRH-1 target genes . Impressively , in primary cultures of mouse hepatocytes , TA inhibits hLRH-1 sumoylation and is able to mimic the transcriptional output of the SUMO-less LRH-1 mutant by switching on genes that have been associated with hepatic injury . Our gene-expression-based screen was designed to sample the transcriptional output of substrate sumoylation , which has not been possible with in vitro target-based screens . We reasoned that assessing transcriptional endpoints makes it possible to integrate other mechanisms that promote substrate-specific sumoylation or desumoylation , including possible synergistic effects of multiple transcription factor sumoylation ( Holmstrom et al . , 2008; Komatsu et al . , 2004 ) . Several factors make hLRH-1 an excellent test substrate in initial and follow-up assays . First , our earlier work established that blocking the NR5A sumoylation cycle results in robust transcriptional changes that could be leveraged in a phenotypic screen . Second , hLRH-1 is well-sumoylated with a stereotypic pattern of multiple sumoylated species requiring only the endogenous sumoylation machinery when it is expressed in HEK293S , JEG3 , and HepG2 cells , in primary hepatocytes and in liver tissue . These findings differ from the need for exogenous SUMO1/2 and E3s required to detect 1x-sumoylated V5-tagged mLRH-1 in HEK293 cells or primary hepatocytes ( Stein et al . , 2014 ) . In addition , when compared to hFXR , another sumoylated hepatic NR , the number and intensity of hLRH-1 sumoylated species far exceed the faint , single FXR sumoylated band detected when hFXR is expressed in mouse liver ( Kim et al . , 2015 ) . Finally , the pattern of hLRH-1 sumoylation is readily duplicated in vitro , allowing one to also test candidate sumoylation modulators in a defined cocktail . We posit that NR5As might be particularly good substrates for sumoylation because their major acceptor lysines reside in the flexible hinge domain , possibly promoting strong protein–protein interactions previously observed between Ubc9 and both SF-1 and LRH-1 ( Lee et al . , 2005 ) . Indeed , the residual UBC9 protein following siUBC9 knockdown likely accounts for the persistent levels of 1x SU-LRH-1 and suggests that once UBC9 is charged with SUMO1/2 , sumoylation of hLRH-1 proceeds efficiently . Interestingly , as with siUBC9 , inhibiting E2 by 2-D08 is also ineffective at blocking hLRH-1 sumoylation . Regardless of the underlying mechanisms that confer relatively high basal levels of NR5A sumoylation in cells and in tissues , the consistent and robust sumoylation observed for hLRH-1 were instrumental in facilitating follow-up studies on candidate small molecule hits from our primary screen . While polyphenols such as TA are easily dismissed as promiscuous inhibitors and false positives in high-throughput screens ( Feng and Shoichet , 2006; Pohjala and Tammela , 2012 ) , the effects of TA in our sumoylation assays are extremely reproducible in multiple cell lines and in primary hepatocytes . Interestingly , although other colloid-forming pehnolic compounds , such as bergapten and coumarin 153 ( Pohjala and Tammela , 2012 ) , are present in the Pharmakon library , both failed to emerge as hits in the primary screen . When compared to other sumoylation inhibitors TA performs well . Indeed , in our IVS assay conditions , TA is more effective than either 2-D08 or GA and can inhibit sumoylation of multiple substrates in vitro , including the hinge-LBD of SF-1 , an androgen receptor peptide , and full-length IκBα . The fact that 2-D08 fails to inhibit hLRH-1 sumoylation but is effective on other substrates ( AR peptide and FL-IκBα ) might reflect the fact that 2-D08 fails to block the strong interactions between Ubc9 and NR5As , as mentioned above . These data imply that mechanistically distinct sumoylation inhibitors act on different classes of substrates . We also find that unlike GA , which decreases cell viability as shown here and reported by ( Liu and Zeng , 2009 ) , TA appears to be well-tolerated in both immortalized and primary cell cultures . Hence , while GA might decrease sumoylation as an adaptive response to cell death , the utility of GA in assessing the transcriptional responses of substrate sumoylation is potentially quite limiting . Our data suggest strongly that TA blocks substrate sumoylation by inhibiting E1 thioesterization , as found for the ellagitannin , Davidiin ( Takemoto et al . , 2014 ) . The known aggregate formation and antioxidant properties of TA appear to be less important in inhibiting substrate sumoylation . Indeed , TA inhibits FL-hLRH-1 sumoylation even in the presence of detergent . Polyphenols , including TA , are also antioxidants and can scavenge reactive oxygen species ( ROS ) during oxidative stress ( Chen et al . , 2007; Yazawa et al . , 2006 ) , which might also directly affect the equilibrium between sumoylation-desumoylation ( Bossis and Melchior , 2006 ) . In this regard , we find that two other antioxidants , ellagic acid and EGCG , are ineffective at inhibiting hLRH-1 sumoylation ( data not shown ) . Furthermore , conditions in our IVS assays are highly reducing making it unlikely that TA inhibits LRH-1 sumoylation via its antioxidant properties in this setting . That TA is effective at blocking hLRH-1 sumoylation in humanized primary hepatocytes greatly strengthens the validity of TA as a useful chemical tool to assess the cellular effects of sumoylation . Interestingly , TA is more effective at blocking hLRH-1 sumoylation in primary hepatocytes as compared to HepG2 cells where 1x SUMO-hLRH-1 persists even at the highest dose of TA; a similar trend was noted for endogenous hSF-1 in H295R cells . The lower efficacy of TA in immortalized cell lines may reflect an increase in the general sumoylation machinery in immortalized versus primary cells , as noted by ( Bellail et al . , 2014 ) . The use of humanized mouse hepatocytes and the dramatic changes we observed in adiponectin and sonic hedgehog transcripts may begin to provide new insights into the in vivo function of LRH-1 sumoylation . The ectopic activation of SHH signaling observed here in primary hepatocytes after overexpressing SUMO-less hLRH-1 and after TA treatment confirms our earlier work showing that elimination of SF-1 sumoylation activates hedgehog signaling in endocrine tissues ( Lee et al . , 2011a ) . Others have noted that hyperactivation of hedgehog signaling in liver is associated with non-alcoholic steatohepatitis ( NASH ) progression and responses to liver injury ( Grzelak et al . , 2014; Guy et al . , 2012; Hirsova and Gores , 2015 ) . Adiponectin is an adipocyte-derived protein that reduces fatty liver ( Xu et al . , 2003 ) and appears protective against NASH ( Asano et al . , 2009 ) . Indeed , while adiponectin is normally never expressed in liver , hepatic adiponectin transcripts are observed in rats after chemically induced hepatotoxicity ( Yoda-Murakami et al . , 2001 ) and in patients with fatty liver or fully progressed NASH ( Uribe et al . , 2008 ) . The finding that SUMO-less hLRH-1 and TA switch on hepatic adiponectin and hedgehog signaling leads us to speculate that tipping the balance of the hLRH-1 sumoylation cycle toward desumoylation might initiate adaptive responses to liver injury , and eventually a pro-inflammatory response , as suggested by others ( Venteclef et al . , 2010 ) . Interestingly , a global knock-in of a single SUMO mutation ( K289R ) in mouse LRH-1 , which is equivalent to K270R in hLRH-1 , has no strong phenotype on its own , but mitigates aortic plaque formation in Ldlr-/- arteriosclerosis-prone mice ( Stein et al . , 2014 ) . Hence , revealing the full physiological consequences of LRH-1 sumoylation could require the elimination of both major sumoylation sites in the flexible hinge domain and the use of conditional knock-in strategies that are specific for the adult liver . In summary , using a novel cell-based assay , we report that the commercially derived , plant extract TA is a useful , nontoxic chemical tool for assessing the transcriptional and cellular effects of sumoylation in both immortalized and primary cell cultures . Based on our collective studies that have focused on the sumoylation of NR5As , we propose that the ratio of sumoylated to desumoylated substrate can be chemically manipulated to switch on and off sumo-sensitive transcriptional programs . Clearly , continued efforts are needed to determine whether more selective chemical tools can be found that promote or block sumoylation of a given substrate . To generate tetracycline ( TET ) -inducible Flp-In T-REx stable JEG3 cells , 3x Flag-tagged WT and 3KR ( K44R/K192R/K270R ) hLRH-1 were cloned into pcDNA5/FRT/TO expression vectors ( Life technologies , South San Francisco , CA ) , followed by selection with 100 or 125 μg/ml Hygromycin B ( Gemini Bio-Products , Sacramento , CA ) . JEG3 hLRH-1 cells were treated with tetracycline ( 100 ng/ml , Teknova Laboratory , Hollister , CA ) for 6 hr to induce WT or SUMO-less LRH-1 proteins . Doxycycline ( Dox ) -inducible HepG2 3G stable cells were made by cloning 3x Flag-tagged WT and 2KR ( K192/270R ) hLRH-1 into pTRE 3G vectors ( Clontech , Mountain View , CA ) , followed by selection with 250 µg/ml Hygromycin B ( Gemini Bio Products , Sacramento , CA ) . The TET-On 3G HepG2 parental cell line was a generous gift from Dr . Stephen Hand ( Li et al . , 2012 ) . For detecting WT or SUMO-less LRH-1 expression , HepG2 3G cells were treated with 200 ng/ml Dox ( Sigma-Aldrich , St . Louis , MO ) for 6 hr . For siUBC9 knockdowns , Ubc9 ( SI04185937 , SI04368420 ) and non-silencing control ( SI03650318 ) siRNA were purchased from Qiagen , Hilden , Germany . SiRNA at 5 nM final concentration was reverse-transfected into JEG3 or HepG2 WT hLRH-1 stable cells by RNAiMax ( Life Technologies ) for 72 hr followed by induction of hLRH-1 expression by addition of 100 ng/ml TET for 24 hr to JEG3 cells or by addition of 250 ng/ml Dox for 6 hr to HepG2 cells . For cell viability assays , JEG3 hLRH-1 or HEPG2 hLRH-1 cells were plated on 24-well plates in 0 . 5 ml of media . Primary hepatocytes were seeded on 96-well plates in 0 . 1 ml of media . The following day , fresh media was applied with compounds or DMSO control . After 5 or 24 hr treatment , cell viability was assayed using CelltiterGlo ( Promega , Madison , WI ) according to manufacturer’s instructions . Relative luminescence was measured with Veritas Microplate Luminometer ( Turner BioSystems , Sunnyvale , CA ) with an integration time of 1 . 0 s and normalized to the DMSO control . Cells were lysed in RIPA buffer ( 6 mM Na2HPO4 , 4 mM NaH2PO4 , 2 mM EDTA pH 8 . 0 , 150 mM NaCl , 1% NaDOC , 0 . 1% SDS and 1% NP-40 ) and tissues were lysed in Tris–SDS buffer ( 2% SDS , 0 . 6 M Tris-Cl pH 8 . 0 and 0 . 1 M DTT ) supplemented with protease inhibitors ( Sigma-Aldrich ) and 20 mM N-Ethylmaleimide ( NEM; Sigma-Aldrich ) and sonicated using the Diagenode Bioruptor . Lysates were clarified by centrifugation and protein concentration was measured using Protein Assay Dye reagent concentrate ( Bio-Rad , Hercules , CA ) according to the manufacturer’s protocol . The following antibodies and concentrations were used: anti-Flag M2; 1:7500 ( Sigma-Aldrich ) , anti-LRH-1; 1:3000 for mouse liver and 1:10 , 000 for in vitro assay ( R&D , Minneapolis , MN ) , anti-SF-1; 1:1000 ( Upstate , EMD Millipore , Billerica , MA ) , anti-UBC9; 1:1000 ( Cell Signaling , Danvers , MA ) , anti-SUMO1; 1:1000 ( Developmental Studies Hybridoma Bank , Iowa City , IA ) , anti-SUMO2; 1:1000 ( Life technologies ) and Ubiquitin monoclonal P4D1 antibody; 1:1000 ( Cell Signaling ) , HRP-conjugated anti-βactin 1:2500 ( Cell Signaling ) , and anti-GAPDH 1:5000 ( Santa Cruz Biotechnology , Santa Cruz , CA ) . Total RNA from cells and tissues were isolated using Trizol Reagent ( Life Technologies ) and PureLink RNA mini kit ( Life Technologies ) , respectively . DNase-treated 1 µg total RNA was used to generate cDNA using High-Capacity cDNA Reverse Transcription kits ( Life Technologies ) . RT-qPCR was performed using SYBR Green , High ROX ( Biotool , Houston , TX , or Quanta , Gaithersburg , MD ) and data analyzed essentially as described ( Kurrasch et al . , 2007 ) . Sequences for all primer pairs used for qPCR reactions are listed in Supplementary file 1 . JEG3 WT hLRH-1 cells ( 5000 cells per well ) were plated into 384-well by cell dispenser Wellmate ( Thermo Scientific , Waltham , MA ) for 24 hr . Using an FDA- and European-approved Pharmakon library of 1600 compounds ( MicroSource Discovery Systems , Gaylordsville , CT ) , drugs were pinned at a concentration of 10 µM in 0 . 1% DMSO using a Biomek FXP ( Beckman , Pasadena , CA ) . At the same time , cells were treated with Tet ( 100 ng/ml ) for inducing WT hLRH-1 . Twenty-four hours later , cells were washed once in PBS and then lysed in 25 µl of lysis buffer provided in the TurboCapture 384 mRNA Kit ( Qiagen ) using EL406 microplate washer ( BioTek , Costa Mesa , CA ) . After a 10 min incubation at 37°C , 20 µl of cell lysate was transferred to 384-well oligo ( dT ) -coated plate ( Qiagen ) using Biomek FXP and incubated at room temperature for 90 min with shaking . Plates were washed three times with washing buffer and reverse transcription was performed in the same well using High-Capacity cDNA Reverse Transfection kits ( Life Technologies ) , according to the manufacturer’s instruction , with a total volume of 20 µl . Aliquots of cDNA was delivered to 384-well qPCR plates using a Biomek FXP Liquid handler and stored at -20°C for subsequent qPCR assays . See Supplementary file 2 for further details . Retesting of top candidates was carried out with repurchased drugs from sources listed in Supplementary file 3 . For qPCR assays , the master-mix buffer ( 8 µl ) containing PCR oligos and Quanta qScript cDNA SYBR Mix ( Quanta Biosciences ) was added to cDNA ( 2 µl ) . All qPCR assays were performed using an ABI 7900HT instrument . Data were analyzed using the ΔΔCT method . Average of △CT from DMSO-treated samples ( N = 58 ) was used as external controls . MUC1 and APOC3 genes were used as the endpoint read-outs for the SUMO-sensitive genes . Calculations: △CT = Gene of interest ( APOC3 or MUC1 ) - CT House Keeping gene ( TBP ) CT △△CT = △CT ( Drug ) - △CT ( DMSO ) For Reference: 1CT change = 2-fold change Z-score: - ( Drug △△CT – Average △△CT ( for all 1600 Drugs ) /SD △△CT ( for all 1600 Drugs ) For Reference: Positive Z-score = Upregulation of MUC1/APOC3 , Negative Z-score = Downregulation of MUC1/APOC3 Full-length LRH-1 ( aa1-541; UniprotKB entry: O00482 ) was subcloned into pRSF-2 vector ( Novagen , Madison , WI ) and grown in Escherichia coli BL21Star ( DE3 ) cells ( Invitrogen ) at 16°C for 16–18 hr to an OD 0 . 6–0 . 8 and induced with IPTG ( 0 . 2 mM ) . Cells were resuspended in lysis buffer A ( 20 mM Tris–HCl pH 8 , 300 mM NaCl , 10% glycerol , 40 mM Imidazole , 5 mM β-mercaptoethanol [BME] , 1 mM CHAPS ) supplemented with protease inhibitors ( Roche , Indianapolis , IN ) . hLRH-1 protein was purified using Ni-nitrilotriacetate beads ( Qiagen ) and eluted with Buffer A and 300 mM imidazole . Eluted hLRH-1 was bound to 24 bp duplex region of the Inhibin-A promoter ( 5’-GGAGATAAGGCTCATGGCCACAGA-3’ ) to stabilize protein and was further purified by size exclusion chromatography in Superdex 200 ( 16/60 ) . Native gel electrophoresis confirmed that the hLRH-1/DNA complex eluted as a monomer . His-tagged components of sumoylation reactions including hE1 , hUBC9 , and hSUMO1 were grown to an OD 0 . 3–0 . 7 and induced with IPTG ( 0 . 35 mM ) for 5 hr at 22°C . Cells were lysed in 50 mM Tris–HCl , pH 8 . 0 , 150 mM NaCl , and 5% glycerol supplemented with protease inhibitors and proteins purified as described ( Reverter and Lima , 2009; Yunus and Lima , 2009 ) . FL-hLRH-1 was prepared and lysed in 20 mM Tris–HCl pH 8 . 0 , 1 mM CHAPS , 10% glycerol , 5 mM BME , 20 mM imidazole , and 300 mM NaCl supplemented with protease inhibitors and then eluted with lysis buffer with 300 mM Imidazole . IVS reactions were performed at 37°C for 1 hr using 0 . 1 µM E1 , 10 µM UBC9 , 30 µM SUMO1 , and 1 µM FL-hLRH-1 substrate ( Ward et al . , 2013 ) in 50 mM Tris–HCl , 100 mM NaCl , 10 mM MgCl2 , 2 mM DTT and initiated by addition of freshly made 10 mM ATP . Aggregation assays used 0 . 01% Triton X-100 ( Sigma ) . IVS reactions were quenched with 4x Laemmli Buffer with BME , boiled for 5 min and loaded onto a Novex Nupage 4–12% Bis-Tris gel and transferred to nitrocellulose membranes followed by incubation with mouse anti-LRH-1 ( 1:7500 , R&D ) or mouse anti-SUMO1 ( 1:325 , DSHB ) . Proteins were visualized using LiCor Odyssey system and goat anti-mouse 800 ( 1:20 , 000 LiCor , Pierce , Lincoln , NE ) and quantitated by Image Studio Lite . Percent conversion was calculated by the ratio of sumoylated protein over total signal per reaction normalized to DMSO control . Concentration curves were derived from at least three independent reactions and fit with nonlinear fitting of log10 [µM TA] versus variable slope using Prism graphing software ( GraphPad , La Jolla , CA ) . IVS of full-length IκBα and fluorescent AR peptide was performed as previously described ( Kim et al . , 2013 ) . Conditions for the thioester assay were as described above , but with only E1 and SUMO1 proteins added to IVS reactions . Human Exonic Evidence Based Open-source ( HEEBO ) arrays were printed at the UCSF Center for Advanced Technology ( CAT ) . Hybridization conditions were carried out in Flp-In T-REx JEG3 cells as previously described ( Kurrasch et al . , 2007 ) to identify top genes upregulated by expression of SUMOless hLRH-1 versus WT hLRH-1 , or after Ubc9 knockdown as described above . For siRNA experiments , HepG2 hLRH-1 cells were reverse-transfected with 5 nM of pooled siRNA directed against human siUBC9 or siRNA control from Qiagen , with RNAiMAX transfection reagent ( Life technologies ) according to the manufacturer’s protocol . Seventy-two hours after siRNA transfection , WT hLRH-1 was induced with Dox for 6 hr . Total RNA was purified using RNAeasy kit ( Life technologies ) according to the manufacturer’s protocol . Hybridizations were performed at 65°C for 16 hr using mixers compatible with the MAUI hybridization systems ( BioMicro Systems , Salt Lake City , UT ) . Arrays were scanned using an Axon Scanner 4000B , and data analyzed by GenePix 6 . 0 software ( Molecular Devices ) . Heat maps were generated using open-access TreeView software . HepG2 hLRH-1 cells were seeded ( 4 x 106 ) on 10 cm plates overnight , induced with 250 ng/ml Dox and treated with DMSO or 50 µM TA for 6 hr . Cells were cross-linked with 1% formaldehyde for 3 min at room temperature and quenched by addition of 400 mM glycine . Cells were harvested in 50 mM HEPES-KOH pH 7 . 4 , 1 mM EDTA , 150 mM NaCl , 10% glycerol , and 0 . 5% Triton X-100 , swelled for 40 min at 4°C , then nuclei were pelleted at 600 x g for 5 min and resuspended in RIPA buffer ( 10 mM Tris–HCl pH 8 . 0 , 1 mM EDTA , 150 mM NaCl , 5% glycerol , 0 . 1% sodium deoxycholate , 0 . 1% SDS , 1% Triton X-100 ) ( Watson et al . , 2013 ) . Lysates were sonicated for a total of 30 min ( 30 s on , 30 s off , 5 min intervals ) with a Diagenode Biorupter UD-200 on High setting at 4°C . Sonicated chromatin was clarified by centrifugation then IP’d with 1 µg anti-Flag M2 antibody pre-conjugated to 10 µl Protein G Dynabeads ( Invitrogen ) for 2 hr at 4°C . Bound protein were washed with 500 mM NaCl and LiCl buffer before reverse cross-linking and proteinase K digested overnight . DNA was isolated using Zymogen ChIP DNA Clean and Concentrator columns and pooled for deep sequencing . ChIP DNA was sent to Hudson Alpha Genomic Services Laboratory for library preparation using Illumina TruSeq Kit . Triplicates of hLRH1 ChIP-Seq ( WT hLRH-1 ) and a control ( Input ) were sequenced on the Illumina HiSeq 2000 platform using 50 bp , single-end reads . Reads were mapped to the hg19 human reference using bowtie and de-duplicated using Samtools . Final data compilation includes a total of 3 . 36 , 10 . 77 , and 7 . 7 million aligned sequence reads for WT hLRH-1 and 2 . 67 million reads from Input . Quality control and ChIP-signal strength assessment was performed via CHANCE ( Diaz et al . , 2012 ) . CHANCE called both experiments as successful ( via a comparison with the distribution of ChIP-strengths observed in the ENCODE repository ) at a combined , positive false discovery rate ( FDR ) of FDR = 2 . 1X10-4 for WT hLRH1 . Reads from replicates were then pooled , and peaks were called via MACS ( Zhang et al . , 2008 ) , using the default parameter settings . This generated 18 , 884 peaks from the wthLRH-1 samples . Genes up-regulated and down-regulated in response to TA was called by setting a probe-intensity threshold at the 95th percentile or 5th percentile , respectively , of array-wide probe intensities . Motif searches were done by MEME-chip ( Machanick and Bailey , 2011 ) and NR5A binding sequences were discovered by PROMO ( Farre et al . , 2003; Messeguer et al . , 2002 ) . Expression of 3X Flag-tagged WT or 2KR hLRH-1 in mouse liver was achieved using adeno-associated virus serotype-8 ( AAV8 ) by cloning each respective cDNA into the viral gateway vector pAAV2 . 1-TBG ( Penn Vector Core , Philadelphia , PA ) downstream of the liver-specific promoter thyroxine-binding globulin ( TBG ) . AAV8 virus expressing WT hLRH-1 , 2KR or enhanced green fluorescent protein ( eGFP ) ( AAV8 hLRH1 , AAV8 hLRH1-2KR , or AAV8-eGFP ) was amplified at the University of Pennsylvania Gene Therapy Vector Core . Tissue specificity and efficiency of infection was assessed in 8-week-old C57BL/6 male mice ( The Jackson Laboratory , Bar Harbor , ME ) infected via retro-orbital injection with AAV8-eGFP at concentration of 1 x 1011 genome copies per ml ( GC/ml ) , this concentration was used for subsequent experiments using AAV8-hLRH-1 and AAV8-2KR . Tissues were collected 14 days post-infection , as previously described ( Lu et al . , 2012 ) , and analyzed for eGFP fluorescence by fluorescent microscopy or extracted for mRNA as described below . Mice were euthanized in accordance with the UCSF Institutional Animal Care and Use Committee under Ingraham lab protocol . Mice were perfused with PBS prior to collection of liver tissue for all subsequent biochemical and gene expression studies . Primary hepatocytes were isolated from mice as previously described ( Silver et al . , 2000 ) . Briefly , mice were anesthetized with Avertin ( 250 mg/kg ) and perfused with pre-warmed perfusion buffer ( HBSS supplemented with HEPES , pH 7 . 4 ) followed by perfusion with 75 ml digest buffer ( HBSS/HEPES , pH 7 . 4 , 0 . 3 mg/ml collagenase type 1 , EDTA-free protease inhibitors ) . Digested liver was then dispersed on a 100 mm cell culture dish containing 25 ml cold Dulbecco's modified Eagle's medium ( DMEM ) ( DMEM , 10% FBS , Penicillin/Streptomycin ) and filtered through a 200-μm nylon membrane into a 50 ml Falcon tube . Hepatocytes were isolated by mixing filtrate with 24 ml Percoll solution ( 1x HBSS , 4 mM NaHCO3 , pH 2 . 2 ) and centrifuged at 100 x g for 7 min at 4°C . Pellet containing hepatocytes was then washed with 40 ml DMEM and resuspended in desired media volume . Cells were plated on 6-well plates coated with collagen-1 and were allowed to attach overnight . The following morning DMEM was replaced with fresh DMEM supplemented with TA at the indicated concentration . Data are represented as mean +/- SEM : p<0 . 05; p<0 . 005; p<0 . 001; **p<0 . 0001 . Statistical analyses were performed using Prism 5 ( GraphPad ) software . Statistical significance was determined by unpaired Student’s T-test unless otherwise indicated .	Proteins in cells carry out diverse tasks . One way in which this diversity is achieved by proteins is through the attachment of molecular tags . SUMO is one such tag that can reversibly attach to proteins and alter their activity . The modification of proteins by SUMO is known as sumoylation , and it regulates many processes that are essential for living cells . In particular , transcription factors—the proteins that bind to DNA to switch genes on or off—are highly modified by SUMO . However , the consequences of sumoylation are not fully understood , and current research into this area has been hindered by a lack of effective and non-toxic chemicals that stop or slow down sumoylation . Suzawa , Miranda , Ramos et al . have now screened a large collection of compounds , which had already been approved for medical use , to find one that could inhibit sumoylation without toxic effects . The compounds were tested for their ability to alter the activity of a transcription factor called human Liver Receptor Homolog-1 . This protein , which is referred to as LRH-1 for short , is an ideal candidate to test SUMO inhibitors because it is highly modified by multiple SUMO tags . This screen identified a compound from plants called tannic acid as a non-toxic and potent inhibitor of sumoylation . Further experiments confirmed that tannic acid prevented the modification of LHR-1 as well a number of different proteins that also commonly modified by SUMO . Inhibiting the sumoylation of LRH-1 led to an increase in the expression of genes that are normally silenced by SUMO-modified LRH-1 . Similar results were obtained when tannic acid was tested using human cells and “humanized” liver cells from mice that had been engineered to express human LRH-1 . The next big challenge is to find new chemical probes that can be used to specifically promote or inhibit SUMO modification of just one particular protein .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "tools", "and", "resources" ]	2015	A gene-expression screen identifies a non-toxic sumoylation inhibitor that mimics SUMO-less human LRH-1 in liver
We describe the computational design of proteins that bind the potent analgesic fentanyl . Our approach employs a fast docking algorithm to find shape complementary ligand placement in protein scaffolds , followed by design of the surrounding residues to optimize binding affinity . Co-crystal structures of the highest affinity binder reveal a highly preorganized binding site , and an overall architecture and ligand placement in close agreement with the design model . We use the designs to generate plant sensors for fentanyl by coupling ligand binding to design stability . The method should be generally useful for detecting toxic hydrophobic compounds in the environment . Fentanyl is a potent agonist of the μ-opioid receptor ( MOR ) , with an affinity of approximately 1 nM and a potency 100-times that of morphine ( Volpe et al . , 2011 ) . It is used both pre- and post-operatively as a pain management agent . The fast acting nature and strength of fentanyl have been attributed to its high degree of lipophilicity ( Peckham and Traynor , 2006 ) . Fentanyl has become a widespread drug of abuse , and has played a central role in the growing opioid epidemic . Reports of illegal manufacturing and fentanyl-related deaths across the country and other parts of the world have increased significantly in recent years ( Drug Enforcement Administration , 2017 ) . Custom-designed ligand-binding proteins offer the possibility of both detecting and counteracting toxins such as fentanyl . Antibodies raised against small molecules generally require mammalian expression systems and conjugation of the compound ( hapten ) to an immunogenic carrier protein . In addition , elicitation of antibodies by immunization does not provide control over the interactions that the protein makes with the ligand . In contrast , computationally designed proteins can be readily expressed in bacterial and other low-cost expression systems , and specific interactions can be directly programmed . However , computational design of precise protein–ligand interactions for flexible , predominantly hydrophobic compounds is challenging . As these molecules are in some sense ‘featureless’ , due to their overall hydrophobic character , binding depends heavily on the protein-ligand shape complementarity . We previously reported a method for generating binders for relatively rigid molecules containing hydrogen bonding functional groups , where the focus was on solutions with optimal hydrogen bonding geometry ( Tinberg et al . , 2013 ) . However , this approach is not well suited for flexible , nonpolar compounds such as fentanyl . We pursued a two-step approach to designing fentanyl binders ( Figure 1—figure supplement 1 ) . Fentanyl contains 6 rotatable bonds , which increases the combinatorial complexity of possible protein–ligand interactions to be considered . Starting from the structure of a fentanyl-citrate toluene solvate ( Peeters et al . , 1979 ) , we generated 11 conformers plus an additional hydrated model of fentanyl , based on the small molecule structure , with non-covalently bound water atoms at both the tertiary amine ( 3 Å nitrogen to water distance , 109° carbon-nitrogen-water angle ) and the carbonyl oxygen ( 3 Å oxygen to water distance , 120° carbon-oxygen-water angle ) ( Figure 1a ) . For each fentanyl conformer , we identified a large number of shape complementary placements of fentanyl within protein scaffolds from the MOAD database ( Hu et al . , 2005 ) using the fast docking algorithm PatchDock , which identifies shape complementary interactions between binding partners ( Duhovny et al . , 2002 ) . Multipose binding has been observed in many naturally occurring protein-ligand complexes ( Kulp et al . , 2012; Blum et al . , 2011; Barelier et al . , 2015 ) , but for our approach we sought precise control of the fentanyl pose by considering only a single conformer per protein scaffold . In the second design step , we selected the top 20 scoring docks from PatchDock for each scaffold and optimized the identities and rotamer conformations of amino acids within 8 Å of fentanyl for shape complementarity and specific protein–ligand interactions . Similar to other MOR agonists , fentanyl possesses a charged tertiary amine , one of only two sites capable of making electrostatic interactions . We sought to exploit the tertiary amine to confer directionality and allow atomic level control over the placement of the otherwise hydrophobic molecule . Two design strategies were pursued: ( 1 ) The introduction of specific side chain–fentanyl interactions , either acidic ( Asp or Glu ) or cation-pi ( Phe , Tyr , Trp ) with the tertiary amine , and ( 2 ) the use of the hydrated fentanyl , as described above , for bridging indirect fentanyl-protein interactions . Designs were filtered based on shape complementarity , protein–fentanyl interface energy and the solvent-accessible-surface-area ( SASA ) , and 62 were selected for experimental characterization . The designs were expressed using yeast surface display and probed for binding with a bovine serum albumin-fentanyl ( Fen-BSA ) conjugate . Sixty-one of the 62 designs expressed well , and three bound fentanyl with low micromolar to high nanomolar affinities . Fen49 , the strongest binder ( 500 nM affinity for Fen-BSA ) on yeast ( Figure 2a ) , and Fen21 ( 10 μM ) were chosen for further experimental characterization , as they represent two different scaffold classes . Of these two designs , recombinantly expressed Fen49 proved to be more stable and amenable to crystallization ( see below ) . Following the placement of the hydrated fentanyl into the binding pocket via PatchDock , RosettaDesign introduced 9 mutations to the Fen49 scaffold to optimize the protein–ligand interactions ( Figure 1b ) . Yeast–binding experiments of individual Fen49 point mutants corresponding to the computationally substituted positions showed that most are crucial for recognizing fentanyl ( Figure 2b ) . Fentanyl does not bind the unmodified Fen49 scaffold ( Figure 2a ) , a glycoside hydrolase ( PDB 2QZ3 ) . Purified Fen49 displayed an affinity of 6 . 9 μM for a fentanyl-Alexa-488 conjugate by fluorescence polarization ( Figure 2c ) . We chose to conjugate the Alexa-488 fluorophore at the 4-phenyl position , as this site is compatible with the designed binding mode , and is also the site of fentanyl conjugation to the BSA probe used in our initial yeast display experiments ( see Materials and Methods ) . 2QZ3 was cocrystallized with xylotetraose ( only 3 of the 4 xylose molecules were placed in the final 2QZ3 model ) , a sugar molecule with a high degree of polarity compared with fentanyl ( Figure 1b ) ( Vandermarliere et al . , 2008 ) . Such a dramatic repurposing of a sugar–binding protein is possible because the initial low-resolution docking step is agnostic to the polar character of the scaffold–binding cavity , as shape complementarity is the primary focus . We solved an atomic resolution ( 1 . 00 Å ) X-ray crystal structure of Fen49 in the apo state , one of the first examples of an original ( non-optimized ) computational design that has been structurally characterized ( Figure 3a ) . The structure reveals a highly preorganized binding cavity ( 28 of 30 non-alanine/non-glycine side chains within ~8Å of fentanyl adopt the designed rotamer ) and an overall structure in very close agreement with the design model; the r . m . s . d . of the design model to the parent structure is 0 . 26 over 184 of 185 residues ( TM_align ( Zhang and Skolnick , 2005 ) score of 0 . 99 ) . The Fen49 apo crystals were obtained from a condition containing 25% polyethylene glycol ( PEG ) 3350 as the precipitant . During model building , a well-ordered portion of PEG was observed in the binding cavity ( Figure 3—figure supplement 1 ) . Soaking experiments with fentanyl tended to crack the crystals and destroy X-ray diffraction , likely as a result of PEG being displaced from the binding cavity . This , coupled with a lack of alternate crystal forms , prevented us from obtaining a structure of the parent Fen49-fentanyl complex . To obtain a detailed map of the sequence determinants of folding and binding , we carried out site-saturation mutagenesis ( SSM ) on 184 of the 185 Fen49 residues , with the exception of the start methionine . At each position , each of the 20 amino acids were allowed , resulting in 3680 unique , single-mutant sequences ( 184 × 20 = 3680 ) . Next-gen sequencing ( millions of sequence reads ) was carried out after each of 4 rounds of affinity enrichment ( Figure 2—figure supplement 1 ) . The majority of the binding site residues were preserved during selection , suggesting that Fen49 was designed with a near-optimal binding cavity ( Figure 2b ) . Exceptions to this were three alanine residues , A67 , A78 and A172 , at the base of the binding pocket that were frequently substituted with larger hydrophobic residues , which provide additional packing for fentanyl . Two positions above the binding cavity enriched to amino acids that could reduce steric hindrance ( Arg 112 to smaller aliphatic amino acids ) or function as a hydrophobic lid over the binding site ( Pro 116 to larger side chains ) . Charged amino acids , which might be expected to destabilize the hydrophobic cavity of Fen49 , were disfavored during selection . However , a modest enrichment for glutamate at position 37 was observed in the second round of selection , suggesting an E37-tertiary amine salt bridge and the possibility of alternative poses of fentanyl within the binding site . This substitution was depleted in later rounds as a hydrophobic pocket was ultimately selected . We identified a combination of 2 substitutions , A78V plus A172I , that produced a Fen49 variant with an approximate 100-fold affinity improvement for fentanyl , to 64 nM ( Figure 2c ) . These substitutions increase packing in the binding site , and likely require a modest positional adjustment of fentanyl to avoid a steric clash with I172 ( Figure 2—figure supplement 2 ) . From the SSM experiments , we identified a Fen49 Y88A point mutant , termed Fen49* , that proved to be more suitable for complex structure determination ( Figure 3 and Figure 3—figure supplement 2 ) . The 1 . 79 Å Fen49-apo structure again revealed a highly preorganized binding site , and an overall structure in close agreement with the Fen49 design ( 0 . 72 r . m . s . d . for Fen49 compared with the design model over 184 of 185 residues ( TM_align score of 0 . 98 ) ) . The majority of Fen49* side chains adopt the design conformations ( 25 of 30 non-alanine/non-glycine residues within ~ 8 Å of fentanyl are correct ) and the structure shows minimal backbone rearrangements . The only significant deviation from the parent Fen49 is in the loop region Thr87 – Thr93 , which contains the Y88A substitution ( Figure 3 and Figure 3—figure supplement 2 ) . In addition , a 3-residue polar network between Arg89 , Asp106 and Tyr108 on the backside of the binding cavity is disrupted in Fen49* ( Figure 3—figure supplement 3 ) . As a consequence of the altered loop , tryptophans 63 and 90 adopt non-designed rotamers and collapse inwards towards the center of the binding cavity , with the designed Trp90-fentanyl stacking interaction replaced by a Trp63-fentanyl dipole-quadrupole ( Figure 3b , Figure 3—figure supplement 2 and Figure 3—figure supplement 4 ) . Unlike the parent Fen49 , Fen49-apo produced crystals with an empty binding cavity that proved to be useful for soaking experiments . We solved a 1 . 67 Å Fen49-fentanyl complex structure , which exhibits a high degree of similarity both with the designed model ( r . m . s . d . of 0 . 64 over 184/185 residues , TM_align score of 0 . 99 ) , and the Fen49-apo structure ( r . m . s . d . of 0 . 420 over all 185 residues , TM_align score of 0 . 99 ) . The Thr87 – Thr93 loop adopts the same structure found in Fen49-apo . With the exception of Trp63 , which is flipped nearly 180° in the complex , fentanyl does not induce any significant changes to the active site upon binding ( Figure 3—figure supplement 5 ) . Fentanyl appears to stabilize the binding site; Fen49-apo Trp63 and the Thr87 – Thr93 loop exhibit higher than average B-factors when compared both with the Fen49-apo structure overall and with the corresponding residues in the Fen49* complex ( Figure 3—figure supplement 6 ) . Despite the divergent Thr87 – Thr93 loop , the parent Fen49 and Fen49* have virtually identical affinities for fentanyl , suggesting that this loop , and more specifically the differential Trp63-90 interaction with fentanyl , do not substantially lower the free energy of fentanyl binding . Instead , preorganization of the inner binding cavity residues appears to be the main determinant for binding . Fen49 was designed to bind a solvated fentanyl . The water modeled at the fentanyl tertiary amine was introduced to bridge an indirect protein–ligand interaction with Tyr80 . During structure refinement , a strong electron density peak was observed at this location ( 3 Å distance and 109 . 2° angle ) . Refinement with water at this position produced a strong positive signal in the Fo-Fc difference map , and it became clear that the density corresponded instead to a chloride ion ( Figure 3 ) . The chloride occupies the site of the designed water; it is coordinated by the tertiary amine , which is protonated at the crystallization pH ( 7 . 5 ) , Tyr80 and a nearby water , a trigonal planar arrangement for chloride typically found in the PDB ( Carugo , 2014 ) ( Figure 3—figure supplement 7 ) . To address the role of the chloride in binding , we carried out binding experiments using potassium phosphate ( pH 7 . 4 ) , free of chloride , as the assay buffer ( protein was prepared in KPi as well ) . Nearly identical affinities were observed for Fen49 . 1 , while Fen49* showed a modest 3-fold ( ~20 μM ) reduction in affinity ( data not shown ) , suggesting that a chloride may be preferred , but is not required for binding . We speculate that in the absence of chloride , a water molecule takes its place as in the design model . The Tyr80–chloride interaction observed in Fen49* is mimicked by a Tyr80-PEG hydrogen bond in the Fen49 parent structure ( Figure 3—figure supplement 1 ) . A second water molecule is observed bound to the fentanyl carbonyl oxygen at the designed position ( 2 . 7 Å distance , 135 . 2° angle ) . A fentanyl detector would have applications in both medicine and public health . To this end , we incorporated our fentanyl binders into a transcription factor ( TF ) -based biosensor system , which couples ligand binding to transcription activation by stabilizing the protein against degradation ( Banaszynski et al . , 2006; Feng et al . , 2015 ) . This sensor can be placed in any system; we chose plants as they offer inexpensive on-site sensors for public health workers . We first tested our sensors in isolated plant cells ( Arabidopsis protoplasts ) . The sensors were cloned into a plasmid containing a Gal4-responsive promoter-driving expression of a luciferase reporter , and expressed transiently in protoplasts . Fentanyl was added to the liquid media in which the protoplasts were incubated . In the presence of fentanyl , both the Fen21 and Fen49 TFs activated luciferase expression ( Figure 4a ) . The Fen21-based sensor , which proved to be the better of the two , produced an 8-fold increase in luciferase expression over background when treated with 250 μM fentanyl . Fen49-expressing protoplasts displayed a modest background signal in the absence of ligand , suggesting that Fen49 is too stable in its apo form to function effectively as a sensor without additional engineering . To demonstrate that our computationally designed fentanyl sensor could function in a multicellular organism , we stably transformed Arabidopsis plants with the Fen21 TF directing luciferase expression . Plants were incubated in liquid plant culture media supplemented with 100 μM fentanyl ( 250 μM fentanyl was toxic to plants; data not shown ) . As early as 24 hr , the Fen21 TF transgenic plants showed an approximately 5-fold increase in luciferase activity , which increased to 10-fold at 72 hr post incubation ( Figure 4b ) . Neutralization of toxic compounds , either through binding or enzymatic breakdown , is an area of great interest for medical and environmental purposes . Our computational approach to designing environmental detectors lays the foundation for engineering practical plant-based sensors that are , in theory , able to detect and respond to any given small molecule . Unlike previous computationally designed ligand binders ( Tinberg et al . , 2013 ) , the method that generated Fen49 did not involve any manual intervention , and hence could be rapidly applied to many ligands . Fen49 does not possess any of the conserved sequence elements of MOR , a membrane bound GPCR , that are required for binding of agonists and antagonists ( Surratt et al . , 1994 ) . Fentanyl is likely to make a direct salt bridge with MOR via its tertiary amine and a conserved aspartate in the third transmembrane helix of the receptor ( Manglik et al . , 2016; Manglik et al . , 2012 ) . In contrast , we have designed an entirely orthogonal , soluble protein that exploits indirect protein–ligand interactions and shape complementarity as the primary drivers of binding . With Fen49 , we have expanded the repertoire of small molecules that are amenable to computational design to include predominantly hydrophobic , flexible ligands . Binders targeting toxic small molecules such as fentanyl should find useful applications as environmental sensors and antidotes . Eleven conformers of fentanyl were generated based on an earlier investigation ( Subramanian et al . , 2000 ) . To model solvation of the positive charge of the fentanyl tertiary amine and the polar carbonyl , explicit oxygen atoms were added at idealized distances and angles ( 3 . 0 Å N-O distance , 109° C-N-O angle for the tertiary amine; 3 . 0 Å O-O distance , 120° C-O-O angle for the carbonyl ) . These values were chosen based on the small molecule crystal structure of a fentanyl-citrate toluene solvate ( Peeters et al . , 1979 ) . Scaffold proteins were comprised primarily of the 2010 MOAD set of high-resolution protein–ligand structures ( Hu et al . , 2005 ) ( 2454 PDBs ) , as well as a curated set of homologous proteins that had been shown to express well in the laboratory or that were suitable for computational design ( 399 PDBs ) . Also included was a set of 83 PDBs from the Pfam group of ketosteroid isomerases . The full list of PDB scaffolds used in this study is given in Supplementary Table 1 ( Supplementary file 1 ) . To generate the fentanyl parameters for use in Rosetta ( RRID:SCR_015701 ) , an auxiliary script , which is distributed with the Rosetta software package , was used to convert the mol2/mol format to Rosetta parameters , using the command shown below . Execution of the command generates a parameter file in full atom as well as in centroid mode with a formal charge of +1 for fentanyl . Python ~/Rosetta/main/source/src/python/apps/public/molfile_to_params . py Fentanyl_cambridge . mol -n CFN -c --recharge=1 Scaffold proteins were used for initial docking of the set of fentanyl conformers . Docking was restricted to binding pockets identified by preexisting ligands in the crystal structure , or by RosettaHoles ( Sheffler and Baker , 2009 ) , which was used to define a position file of residues in the pocket . The 2 . 0 drug module of PatchDock ( Schneidman-Duhovny et al . , 2005 ) was used with the default settings for docking . The top 20 scoring poses for each scaffold were selected for subsequent RosettaDesign . ~/Rosetta/main/source/bin/gen_apo_grids . linuxgccrelease -s PDBFILE -database ~/Rosetta/main/database @flags Where the flags file contained the following ( text following # are comments ) :-mute all -unmute apps . pilot . wendao . gen_apo_grids -chname off -constant_seed -ignore_unrecognized_res -packstat:surface_accessibility -packstat:cavity_burial_probe_radius 3 . 0 # if the cavity ball can be touched by probe r>3 , then it is not in a pocket -packstat:cluster_min_volume 90 # minimum size of a pocket , smaller voids will not be considered -packstat:min_cluster_overlap 1 . 0 # cavity balls must overlap by this much to be clustered -packstat:min_cav_ball_radius 1 . 0 # radius of the smallest void-ball to consider -packstat:min_surface_accessibility 1 . 4 # voids-balls must be at least this exposed These positions were set by using the receptorActiveSite in the PatchDock parameter file by initially converting the position file into PatchDock format:python splitfile . py PDBPOSITIONFILE . pos Next , the parameters were generated using the supplied scripts: buildParams . plperl buildParams . pl PDBFILE FENTANYL_CONFORMATION 2 . 0 drug The receptor active site was added to the parameter file:echo "receptorActiveSite POSITIONFILEPATCHDOCK" >> params . txt Lastly , fentanyl was docked into each scaffold protein:patch_dock . Linux params . txt patchdock . out For a small subset of the final 62 fentanyl binder designs , the RosettaMatch algorithm ( Zanghellini et al . , 2006 ) was used to introduce specific protein–fentanyl interactions to the subset of ketosteroid isomerase scaffold proteins ( 83 PDBs ) . Polar residues were used to introduce hydrogen bonds to the fentanyl carbonyl , while acidic ( Asp and Glu ) and aromatic residues ( Phe , Trp and Tyr ) were used to make charge–charge or dipole–quadrupole interactions , respectively , with the tertiary amine . A summary of the designs , their sequences and the method used to generate them is given in Supplementary Table 2 ( Supplementary file 2 ) . For each docked or matched pose , residues within 8 Å of fentanyl were designed using the RosettaDesign ( Leaver-Fay et al . , 2011 ) algorithm to optimize the sequence around fentanyl for shape complementarity ( SC ) and protein–ligand interface energy . Initially , the designs were filtered based on the orientation of the ligand to allow egress of the chemical linker from the binding cavity for yeast display . For poses where the ligand had been placed according to the matching procedure , restraints were added and minimized in the context of an alanine backbone . For the initial round of designs , the catalytic residues were fixed and not allowed to change rotamer or amino acid identity . This was repeated 10 times and the lowest restraint score was kept for further design . The native scaffold residues were given a bonus of 1 . 5 REU , and the sequence was optimized with the matched residues fixed for 10 iterations . Again , the designs with the lowest interface energy ( IFE ) were kept . In order to limit the number of designs , we applied filters to remove poses that scored that did not meet the following thresholds: SC > 0 . 5 , IFE < −10 REU , dsasa ≥ 0 . 8 of the ligand , a geometrical restraint score ≤ 5 , packing of the rotamers in the binding pocket with an RMSD < 1 Å , ddG between the protein and the ligand of < −10 REU , and finally that the electrostatic field of the charged hydrogen and the oxygen of the carbonyl were negative . We furthermore made a greedy optimization of the residues in the interface such that they should contribute with at least an average energy to the interface energy , full-atom Dunbrack score , total score of the residue as well as its solvation score which we had computed from the CSAR 2010 high quality docking set using the enzdes score function . During the design process , we alternated between a linear version and an r6-r12 version of the Lennard-Jones potential . ~/Rosetta/main/source/bin/rosetta_scripts . linuxgccrelease @flags -database ~/Rosetta/main/database -extra_res_fa FEN . FA . PARAMETERS @flags -parser:protocol match_enzdes_then_greedy . xml/refinement_empirical_w_cst . xml -in:file:s PDBFILE For designs generated from Patchdock , no geometrical restraints were employed and the following flags were used for both runs:-run::preserve_header -enzdes::minimize_ligand_torsions 5 . 0 -packing::use_input_sc -packing::extrachi_cutoff 1 -packing::ex1 -packing::ex2 -linmem_ig 10 -parser_read_cloud_pdb 1 -ignore_unrecognized_res -enzdes::lig_packer_weight 1 . 5 -hackelec_min_dis 1 -hackelec_max_dis 2 . 5 -no_optH false -enzdes -detect_design_interface -cstfile match1 . cst -cst_opt -cst_min -flip_HNQ -no_his_his_pairE -hbond_params sp2_params -lj_hbond_hdis 1 . 75 -lj_hbond_OH_donor_dis 2 . 6 -correct -chemical:exclude_patches LowerDNA UpperDNA Cterm_amidation VirtualBB ShoveBB VirtualDNAPhosphate VirtualNTerm CTermConnect sc_orbitals pro_hydroxylated_case1 pro_hydroxylated_case2 ser_phosphorylated thr_phosphorylated tyr_phosphorylated tyr_sulfated lys_dimethylated lys_monomethylated lys_trimethylated lys_acetylated glu_carboxylated cys_acetylated tyr_diiodinated N_acetylated C_methylamidated MethylatedProteinCterm -nblist_autoupdate -mute all In order to confirm that the designs possessed the minimum energy fentanyl poses , we performed RosettaDock and examined the energy and RMSD from the design model . ~/Rosetta/main/source/bin/ligand_dock . linuxgccrelease @flags -database ~/Rosetta/main/database -extra_res_fa CFN . FA . PARAMS The flags file contained the following information:-run -constant_seed -rng mt19937 -in -file -s PDBFILE -native PDBFILE -extra_res_fa CFN . fa . params -out -nstruct 10000 -file -packing -ex1 -ex2 -ex1aro -flip_HNQ -docking -randomize2 -ligand -improve_orientation 10 # Evaluate coulomb between protein and ligand -old_estat -ligand -improve_orientation 1000 -minimize_ligand -harmonic_torsions 10 -minimize_backbone -harmonic_Calphas 0 . 3 -soft_rep Fen-BSA was purchased from CalBioreagents ( 10 mg/mLin dH2O , ~20 fentanyl molecules per BSA , catalog # C149 ) . Fentanyl monoclonal antibody was purchased from CalBioreagents ( catalog # M571 ) . Chicken anti-C-Myc-FITC conjugated polyclonal antibody was purchased from ICL Lab ( catalog # CMYC-45F ) . Streptavidin-R-phycoerythrin conjugate was purchased from Invitrogen ( catalog # S866 ) . Fentanyl citrate was purchased from Sigma ( catalog # F3886 ) . Arabidopsis thaliana , ecotype Columbia Col-0 seeds were purchased from Lehle seeds ( http://www . arabidopsis . com ) and used to establish the protoplast and plant lines . Fen21 , Fen49 and 2qz3 were purchased from Genscript with the coding sequence cloned into the NdeI and XhoI sites of vector pETCON , a modified version of pCTCON2 ( Fleishman et al . , 2011 ) that contains a c-terminal fusion to the c-myc epitope . Fen49 variants were generated by PCR using the megaprimer method ( Ke and Madison , 1997 ) using the primers listed in Supplementary Table 3 ( Supplementary file 3 ) . Oligos were ordered from Integrated DNA Technologies , Inc . Fen-BSA ( 10 mg/mL in dH2O ) was first diluted to 2 mg/mL in PBS pH 7 . 4 . Biotinylated Fen-BSA was prepared by reacting 14 . 3 μl of a 10 mM EZ-link-Sulfo-NHS-LC-Biotin solution ( prepared in PBS pH 7 . 4 , 10 eq ) with 500 μl of Fen-BSA in an eppendorf tube shielded from light , on ice . After 4 hr the solution was dialyzed at 4°C against 500 mL of PBS in order to remove unreacted biotin reagent . The dialysis buffer was exchanged for an additional round . The extent of biotinylation was determined using a Pierce biotinylation quantitation kit . Reactions resulted in 1–3 molecules of biotin per BSA . All designs , derivatives and controls were transformed into S . cerevisiae EBY100 cells following the protocol outlined by Gietz and Schiestl , but without the single-stranded carrier DNA ( Gietz and Schiestl , 2007 ) . Transformants were plated on selective media ( C –ura –trp ) and incubated at 30°C for 48 hr . Colonies were picked and grown overnight in 1 mL of SDCAA ( Chao et al . , 2006 ) at 30°C , 225 RPM . The following day , 1e7 cells were harvested by centrifugation at 1000 x g for 2 min at RT in an Eppendorf microcentrifuge . The supernatant was removed , and the cells were resuspended in 1 mL of SGCAA induction media supplemented with 0 . 2% glucose . Protein expression was carried out at 18–22°C . Following 36–48 hr of protein expression , 2e6 cells were collected into 96-well plates ( Corning #3363 ) . Cells were pelleted at 1000 x g for 2 min at 4°C and washed twice with 200 μl of ice-cold PBSF pH 7 . 4 ( PBS supplemented with 1 g/L of BSA ) . Cells were resuspended in a 20 μl PBSF solution containing Fen-BSA at various concentrations . 1 μl of anti-fentanyl antibody ( CalBioreagents ) was added to positive control cells expressing two tandem Z domains of protein A ( ZZ domain ) ( Mazor et al . , 2007; Nilsson et al . , 1987 ) . Plates were incubated at 4°C for 4 hr on a Heidolph Tetramax 1000 plate shaker at 1350 RPM . Unbound Fen-BSA was removed by centrifugation and washing the cells once with ice-cold PBSF . Cells were labeled with 0 . 5 μl of anti-C-myc-FITC conjugated antibody ( 1 mg/ml ) and 0 . 5 μl streptavidin-phycoerythrin ( SAPE , 3 . 3 μM ) in a 20 μl volume of PBSF for 10 min at 4°C with shaking . Cells were washed once with 200 μl of ice-cold PBSF to remove unbound C-myc and SAPE . Cell pellets were resuspended in 100 μl of ice-cold PBSF immediately prior to use . Protein expression and binding were measured on an Accuri C6 flow cytometer ( 488 nm excitation , 575 nm emission ) by monitoring the FITC and PE fluorescence contributions , respectively . A single site-saturation mutagenesis ( SSM ) library was generated for the entire Fen49 coding sequence , with the exception of the start methionine , using a 2-step overlapping PCR method and pETCON-Fen49 as the template . The first step involved 2 separate PCR reactions to generate the 5’ and 3’ fragments flanking the site of interest . The reaction conditions were as follows: 16 . 125 μl ddH2O , 5 μl of HF Buffer ( 5X ) , 0 . 125 of Phusion High-Fidelity DNA Polymerase ( NEB ) , 1 . 25 μl of 10 mM dNTPs , 0 . 5 μl of template DNA at 10 ng/μl , 1 μl each of either the 3’-MCS primer ( 5’-GTACGAGCTAAAAGTACAGTGGGAAC-3’ ) plus the forward NNK-containing primer , or the 5’-MCS primer ( 5’-TGACAACTATATGCGAGCAAATCCCCTCAC-3’ ) plus a reverse primer designed to have a partial overlap with the NNK primer . The second PCR step reconstituted the full Fen49 gene containing a single SSM site , plus 5’- and 3’-flanking sequences derived from the pETCON vector . The reactions conditions were as follows: 33 . 25 μl ddH2O , 10 μl HF Buffer ( 5X ) , 0 . 25 μl DNA Polymerase , 2 . 5 μl dNTPs , 1 μl each of the 5’- and 3’-MCS primers , 1 μl each of both PCR products described in step one . All primers were at 20 μM in dH2O . All amplifications were carried out by 30 cycles of PCR ( 98°C 15 s , 54°C 30 s , 72°C 60 s ) , with an initial 30 s melting step at 98°C and a final 5 min extension step at 72°C . The NNK degenerate primers used to generate the SSM library are listed in Supplementary Table 4 ( Supplementary file 4 ) . All full-length Fen49 PCR products were pooled and purified by gel extraction ( Qiagen ) using ddH2O . Library DNA was used to electroporate S . cerevisiae EBY100 cells in triplicate , following a slightly modified version of the protocol detailed by Benatuil et al . ( 2010 ) . , using 2 μg of NheI/XhoI/BamHI linearized pETCON and 6 μg of Fen49-SSM library DNA . Cells were electroporated using a Gene Pulser Xcell ( Bio-Rad ) at 2 . 5 Kv , 25 μF and 200 Ω . The library complexity was determined to be 5e8 . Following the recovery step in YPD-sorbitol , cells were grown in 100 mL C -Trp -Ura media plus 100 μg/mL carbenicillin at 30°C for 24 hr , 225 RPM . Cells were pelleted at 1000 x g for 3 min , resuspended in 100 mL of fresh C -Trp -Ura and incubated for another 24 hr . The 3 100 mL library cultures were pooled , and 4e8 cells were pelleted and resuspended in 25 mL of SGCAA media supplemented with 0 . 2% glucose , 100 μg/mL carbenicillin and 50 μg/mL kanamycin . The library was expressed overnight at 22°C , 225 RPM . In order to identify Fen49 variants with a greater affinity for fentanyl than the parent design , a fentanyl-SAPE tetramer label was used in place of Fen-BSA ( 20 molecules of fentanyl per BSA ) to reduce the dependency of the system on avidity . One-hundred-million cells from the naïve library were pelleted at 1000 x g , washed twice with 1 mL of ice-cold PBSF and labeled for 3 hr at 4°C , shielded from light , with 4 μM Fen-SAPE in a 1 mL volume of PBSF plus 20 μg/mL FITC conjugated anti-CMYC . Labeled cells were pelleted , washed once with 1 mL of ice-cold PBSF and resuspended in 4 mL PBSF . Cells with strongest signal in the PE channel ( 488 nm excitation , 585/30 nm optical filter ) were collected with a SONY SH800 series cell sorter . Collected cells were grown in C –trp –ura . Fen49 SMM library DNA was analyzed on an Illumina MiSeq , using the v3 kit ( 600 cycles ) , which produces 300-base reads . For full sequence coverage , the FEN49 library was split into 305 and 295 paired-end reads , respectively , where the overlapping sequence of the two portions was 42 base pairs . The next-generation sequencing of SSM libraries produced 5 , 017 , 520 forward and 5 , 093 , 932 reverse reads . The paired-end reads were assembled and filtered for quality ( average Phred score ≥ 18 and a minimum position Phred score of ≥ 12 ) . The paired-ends were further filtered based on a minimum of 11 overlapping reads . This resulted in 3 , 992 , 873 full length DNA reads ( Supplementary file 5 ) that were translated to their corresponding protein sequence and mutation frequencies were determined using the Enrich software package ( Fowler et al . , 2011 ) . The relative enrichment values , calculated as described below , were determined for mutations with >15 counts . ( 1 ) pijselect=fijselect∑i=1N∑j=121fijselect ( 2 ) pijnaive=fijnaive∑i=1N∑j=121fijnaive ( 3 ) Eij=log2PijselectPijnaive The frequency of an observed mutantion at position i for mutation j in the selected ( fijselect ) and naive ( fijnaive ) libraries were determined from the full length reads . Mutation frequencies over all positions and mutation types were summed to determine the total frequency of mutants in a selected and naive library for the full protein sequence length ( N ) . The possible mutations include the 19 standard amino acids ( wild-type amino acid identities were not considered ) and a sequence termination encoded by stop codon . The probability of a mutation in a sequence for the selected ( pijselect ) and naive library ( pijnaive ) were calculated as shown in Equation 1 and Equation 2 , respectively . The enrichment ( Eij ) value was determined as shown in Equation 3 . SSM heat maps were generated using the MatrixPlot function in Mathematica . All chemical reagents and anhydrous solvents for synthesis were purchased from commercial suppliers ( Sigma-Aldrich , Fluka , Acros ) and were used without further purification or distillation . The composition of mixed solvents is given by the volume ratio ( v/v ) . 1H and 13C nuclear magnetic resonance ( NMR ) spectra were recorded on a Bruker DPX 400 ( 400 MHz for 1H , 100 MHz for 13C , respectively ) , Bruker AVANCE III 400 Nanobay ( 400 MHz for 1H , 100 MHz for 13C , respectively ) , with chemical shifts ( δ ) reported in ppm relative to the solvent residual signals of CDCl3 ( 7 . 26 ppm for 1H , 77 . 16 ppm for 13C ) , CD3OD ( 3 . 31 ppm for 1H , 49 . 00 ppm for 13C ) , DMSO-d6 ( 2 . 50 ppm for 1H , 39 . 52 ppm for 13C ) . Coupling constants are reported in Hz . High-resolution mass spectra ( HRMS ) were measured on a Micromass Q-TOF Ultima spectrometer with electrospray ionization ( ESI ) or Bruker MicroTOF with ESI-TOF ( time-of-flight ) . LC-MS was performed on a Shimadzu MS2020 connected to a Nexerra UHPLC system equipped with a Waters ACQUITY UPLC BEH C18 1 . 7 µm 2 . 1 × 50 mm column . Buffer A: 0 . 05% HCOOH in H2O Buffer B: 0 . 05% HCOOH in acetonitrile . Analytical gradient was from 5% to 95% B within 5 . 5 min with 0 . 5 ml/min flow . Preparative RP-HPLC was performed on a Dionex system equipped with an UVD 170U UV-Vis detector for product visualization on a Waters SunFire Prep C18 OBD 5 µm 10 × 150 mm Column ( Buffer A: 0 . 1% TFA in H2O Buffer B: acetonitrile . Typical gradient was from 0% to 90% B within 30 min with 4 ml/min flow . ) . After lyophilization of HPLC purified compounds , the solid residue was generally dissolved in dry DMSO . Alexa 488 carboxylic acid ( Thermofisher scientific , 5 mg , 6 . 0 μmol ) was dissolved in dry DMSO ( 250 μl ) , treated with diisopropylethylamine ( 5 μl , 23 . 1 μmol ) and TSTU ( 2 . 8 mg , 9 . 3 μmol ) . The mixture was incubated for 5 min at r . t . and 1 , 8-diamino-3 , 6-dioxaoctane ( 5 . 7 mg , 38 . 5 μmol ) was added . After 20 min , the reaction was purified by RP-HPLC and lyophilized , yielding an orange-red solid ( 2 . 3 mg , 58% yield ) . HRMS ( ESI ) calcd for C27H29N4O12S2 [M+] 665 . 1218; found 665 . 1239 . A solution of compound 1 in DMSO ( 7 . 0 mM , 250 μl , 1 . 7 μmol ) was mixed with fentanyl isothiocyanate ( Tocris Bioscience ) in solution in DMSO ( 40 mM , 50 μl , 2 . 0 μmol ) . Diisopropylethylamine ( 13 μl , 75 μmol ) was added and the solution was incubated for 2 hr at r . t . The product was purified by RP-HPLC and lyophilized , yielding an orange-red solid ( 1 . 0 mg , 59% yield ) . HRMS ( ESI ) calcd for C50H56N7O13S3 [M+] 1058 . 3098; found 1058 . 3093 . Biotin- ( PEO ) 4-amine ( SCBT ) ( 17 . 0 mg , 40 . 6 μmol ) and fentanyl isothiocyanate ( Tocris Bioscience ) ( 5 mg , 12 . 7 μmol ) were dissolved in DMSO ( 200 μl ) and incubated for 1 hr at r . t . The product was purified by RP-HPLC and lyophilized , yielding a white solid ( 2 . 6 mg , 25% yield ) . HRMS ( ESI ) calcd for C41H62N7O6S2+ [M + H]+ 812 . 4197; found 812 . 4191 . Expression and purification methods refer to Fen49 and all derivatives . Coding sequences were subcloned from their pETCON constructs into the NdeI and BamHI sites of a modified version of pET28a ( Novagen ) , which replaces the N-terminal thrombin cleavage site with a PreScission Protease ( GE Healthcare Life Sciences ) site . Expression clones were transformed into BL21 ( DE3 ) cells and grown overnight in 2 mL of Terrific Broth II ( TB-II , MP Biomedicals ) supplemented with 150 μg/mL carbenicillin without first plating for colony selection . Overnight cultures were used to inoculate 1 L of TB-II and subsequently grown at 37°C until an OD₆₀₀ of 0 . 8–1 . 0 , at which point the shaker temperature was dropped to 18°C and protein expression carried out for 16–20 hr by the addition of IPTG to a final concentration of 0 . 1 mM . Cells were harvested by centrifugation at 4°C , 7500 x g for 20 min and the pellets from 2 L of culture were resuspended in ~30 mL of Nickel Buffer A ( 500 mM NaCl , 20 mM Tris pH 8 . 0 , 30 mM imidazole and 5% glycerol ) and stored at −80°C . Purifications were performed from 6 L of cells . Cells were lysed , while in an ice-water bath , by sonication using a Sonic Dismembrator Model 505 ( Fisher Scientific ) at 70% amplitude ( 4 × 1 min cycles of 5 s pulses followed by 10 s rest , with 1 min in an ice-water bath in between cycles ) . Lysates were clarified by centrifugation at 4°C , 43 , 000 x g for 30 min . The supernatant was loaded onto a 5 mL HisTrap FF column , charged with NiSO4 , at 2 . 5 mL/min using an ÄKTA Pure fast-protein-liquid-chromatography ( FPLC ) system ( GE Healthcare Life Sciences ) . The column was washed with Nickel Buffer A until a baseline absorbance was achieved . Fen49 was eluted from the column by performing a linear gradient to 100% Nickel Buffer B ( 500 mM NaCl , 20 mM Tris pH 8 . 0 , 200 mM imidazole and 5% glycerol ) over 25 mL , and 2 mL fractions were collected . Fractions containing Fen49 were pooled , PreScission Protease was added at a ratio of 1:20 with Fen49 , and the protein was dialyzed at 4°C against 2 × 1L of 50 mM NaCl , 20 mM Tris pH 8 . 0 , 5% glycerol . Cleavage with PreScission Protease leaves a vector derived gly-pro-his sequence on the N-terminus of the Fen49 sequence . Cleaved Fen49 was passed over a 5 mL GST HiTrap column and Q HiTrap column in series at 1 ml/min . The flow through containing Fen49 , but free from contaminating proteins , was collected . Fen49 purity was estimated to be > 95% . All FPLC steps were carried out at RT . Fen49 was concentrated at 4000 x g using an Amicon Ultra-15 10K Centrifugal Filter ( EMD Millipore ) to ~200 μl . Protein buffer was exchanged to 10 mM NaCl 5 times by dilution to 15 mL and re-concentration to 200 μl . Fluorescence polarization experiments were performed as previously described ( Rossi and Taylor , 2011 ) . All experiments were conducted at 25°C in a SpectraMax M5e microplate reader ( Molecular Devices ) at excitation and emission wavelengths of 485 nm and 538 nm , respectively , using a 515 nm emission cutoff filter . Experiments were carried out in 40 μl reaction volumes using High efficiency 96-well black opaque microplates ( Molecular Devices ) . Fentanyl-PEG-Alexa488 ( Fen-A488 ) was used as the fluorescent ligand in all experiments . All protein and ligand dilutions were made in PBS , pH 7 . 4 . For all experiments , the concentration of Fen-A488 was held at 500 nM , while the protein concentration was varied from 60 μM to 20 nM . Anisotropy values were collected over a period of 15 min , and the equilibrium dissociation constants ( Kd ) were determined as previously described ( Tinberg et al . , 2013 ) . Protein was spun at 4°C , 20 , 817 x g for 20 min to remove insoluble material prior to crystallization . Typically none was observed . Protein concentration was determined using the Bradford Protein Assay ( Bio-rad ) and BSA to generate the standard curve . Crystallization trials were conducted using a variety of 96-condition spare matrix suites from Qiagen and Hampton Research . All crystallization trials were conducted with the sitting drop vapor diffusion method at 20°C , using 3-well MRC crystallization plates ( Swissci ) . Fen49 crystallization trials were conducted using a Mosquito Crystal nanoliter robot ( TTP Labtech ) . Fen49 at 30 mg/mL was mixed in 1:2 , 1:1 and 2:1 protein to crystallization solution ratios in 400 nl drops . Crystals displaying a shard-like cluster morphology were obtained after ~1 week from a solution containing 0 . 1M citric acid pH 3 . 5% and 25% ( w/v ) PEG-3350 . Microseeding was employed in order to obtain crystals suitable for diffraction experiments . A drop containing Fen49 crystals was added to 50 μL of crystallization solution in a microfuge tube containing a Seed Bead ( Hampton ) . This solution was vortexed for 30 s . Fresh drops were set up using 1 μL of Fen49 at 20 mg/mL plus 1 μL crystallization solution , to which 0 . 2 μL of a 1:100 dilution of the seed stock was added . Large , single crystals were observed overnight , which grew to a maximum size of over 300 μm in length within ~3 days . Crystals were briefly dipped in a solution of 0 . 085M citric acid pH 3 . 5 , 21 . 25% ( w/v ) PEG-3350% and 15% glycerol and flash frozen in liquid nitrogen . Crystals of Fen49-apo were obtained manually by mixing 0 . 5 μL of protein at 10 mg/mL with 0 . 5 μL and 0 . 8 M sodium phosphate , 0 . 8 M potassium phosphate and 0 . 1 M HEPES pH 7 . 5 . Rod-like crystals appeared after ~3 days and grew to a maximum of 200 μm in length . Crystals were briefly soaked in a solution of 0 . 6M sodium phosphate , 0 . 6 M potassium phosphate , 0 . 075 M HEPES pH 7 . 5% and 25% glycerol , then flash frozen in liquid nitrogen . The Fen49-fentanyl complex was obtained by soaking Fen49-apo crystals overnight in mother liquor plus 20 mM fentanyl citrate ( solution made in dH2O to ~50 mM ) . The soaked crystals were sealed in a well containing mother liquor to allow excess water from the added fen-citrate to diffuse . Fen49-complex crystals were cryo-protected the same as Fen49-apo . Crystals were flash frozen the same as for Fen49-apo . All datasets were collected at the Advanced Light Source ( Berkeley , CA ) beam line 8 . 2 . 2 . using an ADSC Q315R CCD area detector . The Fen49 and Fen49-Apo datasets were processed in HKL2000 ( Otwinowski and Minor , 1997 ) . The Fen49-Complex dataset was processed in XDS ( Kabsch , 2010 ) . Fen49 – Diffraction data were collected over 220° with 1° oscillations , 1 s exposures , at 100K and at a wavelength of 0 . 75141 Å and a crystal-to-detector distance of 156 mm . Images were processed to 1 . 00 Å in space group P21 . Fen49-apo – Diffraction data were collected over 220° with 1° oscillations , 1 s exposures , at 100K , a wavelength of 0 . 976246 Å and a crystal-to-detector distance of 210 mm . Images were processed to 1 . 79 Å in space group P212121 . Fen49-complex – Diffraction data were collected over 135° with 0 . 5° oscillations , 1 s exposures , at 100K , a wavelength of 0 . 999878 Å and a crystal-to-detector distance of 190 mm . Images were processed to 1 . 67 Å in space group P212121 . All structures were solved by molecular replacement ( MR ) using PHASER ( McCoy et al . , 2007 ) in the PHENIX software suite ( Adams et al . , 2010 ) . Iterative rounds of manual building and refinement were conducted in Coot ( Emsley et al . , 2010 ) and Phenix . refine ( Afonine et al . , 2012 ) , respectively . Hydrogens were added for all refinement jobs . The geometric quality of the final models was verified using the MolProbity server ( Chen et al . , 20092010 ) . Resolution cutoffs were determined by monitoring the refinement statistics in the context of the reflection data completeness and the CC ½ and I/σI values ( Karplus and Diederichs , 2012 ) . Fen49 – The Fen49 design model , with residues 63 , 85–95 and 116–122 omitted , was used as search model for MR . Two copies of Fen49 were placed in the asymmetric unit ( AU ) . An initial model was generated using the PHENIX Autobuild module . All defaults were used , with the following exceptions: ‘Build-in-place’ was set to ‘False’ , simulated annealing was used for refinement , and prime-and-switch maps were used during model building to remove search model bias . All atoms except hydrogen were refined with anisotropic atomic displacement parameters . Fen49-apo – PDB 2QZ3 , the parent scaffold from which Fen49 was designed , was used as the MR search model . Three copies were placed in the AU . Manual rebuilding was conducted directly from the MR solution . Poor electron density was observed for the third Fen49 copy in the AU , suggesting a large degree of disorder for this domain . Fen49-complex – Fen49-apo was used as the MR search model . Manual rebuilding was conducted directly from the MR solution . Restraints for fentanyl were generated in Phenix . elbow ( Moriarty et al . , 2009 ) from the SMILES string using the eLBOW AM1 geometry optimization option . Data collection and refinement statistics are given in Supplementary Table 6 ( Supplementary file 6 ) . The Fen49 and Fen21 transcription factors were engineered by N-terminal fusion of the yeast MATα gene degron and the Gal4 DNA binding domain and C-terminal fusion of the VP16 transcriptional activator to either Fen49 or Fen21 . The resulting gene sequence was codon-optimized for optimal expression in Arabidopsis thaliana plants and cloned downstream of the CaMV35S promoter to drive constitutive expression in plants , and upstream of the octopine synthase ( ocs ) transcriptional terminator sequence . To quantify the transcriptional activation function of the Fen49 and Fen21 transcription factors , the luciferase gene from Photinus pyralis ( firefly ) was placed downstream of a synthetic plant promoter consisting of five tandem copies of a Gal4 upstream activating sequence ( UAS ) fused to the minimal ( −46 ) CaMV35S promoter sequence . Transcription of luciferase is terminated by the E9 terminator sequence . These sequences were cloned into a pSEVA 141 plasmid and used for transient expression assays in Arabidopsis protoplasts . We next inserted the genetic circuit for Fen21 transcription and luciferase reporting into the pCAMBIA 2300 plant transformation vector and stably transformed them into Arabidopsis thaliana ecotype Columbia plants using a standard Agrobacterium tumefaciens floral dip protocol . Primary transgenic plants were screened in vivo for fentanyl-dependent luciferase production using a Stanford Photonics XR/MEGA-10Z ICCD Camera and Piper Control Software System , and responsive plants were allowed to set seed for further testing . Second generation transgenic plants ( T1 , heterozygous ) were tested for fentanyl-dependent induction of luciferase expression using the same system described above .	Many small molecules , including toxins and some medicines , have flexible structures , which makes it difficult to detect and/or neutralize them . The pain medication fentanyl , for example , can rotate to adopt many shapes . In recent years , fentanyl drug abuse has become increasingly common , and the drug is often illegally produced . The number of deaths caused by fentanyl has risen greatly , which provides a strong reason to find new ways to detect this and other drugs . Now , Baker et al . have created new sensors that are able to detect fentanyl . First , the 11 most likely shapes that fentanyl could adopt were identified based on known information about the structure of the molecule . Then , a computer program was used to design proteins that were predicted to strongly bind to these most common shapes . Next , genes that coded for these proteins were synthesized in the laboratory and introduced into bacteria , which read the genes to build the proteins . Similar to a well-fitted lock and key , the shape of the newly designed protein had to complement a likely shape of the fentanyl molecule . Baker et al . used a technique called X-ray crystallography to visualize the proteins in atomic detail and confirm that these fentanyl-binders matched their corresponding computational models . Those proteins that bound fentanyl best were then engineered into plant cells , and later into whole plants , together with reporter systems that gave signals when the sensors detected fentanyl . In future , these specifically synthesized proteins could be integrated into entire panels of plants or other systems to detect toxins and other harmful chemicals . Such systems would be of interest in a medical setting and for detecting environmental contamination .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "computational", "and", "systems", "biology" ]	2017	Computational design of environmental sensors for the potent opioid fentanyl
Existing genetic methods of neuronal targeting do not routinely achieve the resolution required for mapping brain circuits . New approaches are thus necessary . Here , we introduce a method for refined neuronal targeting that can be applied iteratively . Restriction achieved at the first step can be further refined in a second step , if necessary . The method relies on first isolating neurons within a targeted group ( i . e . Gal4 pattern ) according to their developmental lineages , and then intersectionally limiting the number of lineages by selecting only those in which two distinct neuroblast enhancers are active . The neuroblast enhancers drive expression of split Cre recombinase fragments . These are fused to non-interacting pairs of split inteins , which ensure reconstitution of active Cre when all fragments are expressed in the same neuroblast . Active Cre renders all neuroblast-derived cells in a lineage permissive for Gal4 activity . We demonstrate how this system can facilitate neural circuit-mapping in Drosophila . An essential step in mapping brain circuits is identifying the function of the individual neurons that comprise them . This is commonly achieved by manipulating neuronal function using effectors encoded by transgenes whose expression is targeted to small subsets of cells using the regulatory elements of neurally-expressed genes ( Gohl et al . , 2017; Luo et al . , 2018 ) . While it has proved relatively easy to target large groups of neurons for cellular manipulation by this means in genetic model organisms using binary expression systems , such as the Cre-lox system of mice or the Gal4-UAS system of fruit flies , highly-specific targeting of neurons requires combinatorial methods . Genetic combinatorial methods typically use either the regulatory elements of two neurally-expressed genes or exploit stochastic events to limit transgene targeting to a subpopulation of a larger group of neurons . In fruit flies , both types of method have been used to target single cells under optimal conditions ( Aso et al . , 2014; Gao et al . , 2008; Gordon and Scott , 2009; Kohatsu et al . , 2011; Lee and Luo , 1999; Luan et al . , 2012; Pool et al . , 2014; Sen et al . , 2019; Shang et al . , 2008; Yu et al . , 2010 ) , and combinatorial approaches using three regulatory elements have achieved some success ( Dolan et al . , 2017; Pankova and Borst , 2017; Shirangi et al . , 2016; Sullivan et al . , 2019 ) . However , general limitations apply to all such approaches: stochastic methods are , by nature , poorly reproducible , while combinatorial methods are labor-intensive , often requiring the characterization of many neurally active enhancer elements ( Dionne et al . , 2018; Tirian et al . , 2017 ) . Simpler methods of targeting small populations of brain cells are therefore desirable in the effort to comprehensively map neural function . An attractive approach to increase the specificity of neuronal targeting is to identify neurons based not only on the genes they express in the terminally differentiated state ( i . e . terminal effector genes , TEG ) , but also on their developmental history ( Awasaki et al . , 2014; Dymecki et al . , 2010; Huang , 2014 ) . Most neuronal lineages produce diverse neuron types , and while some striking correspondences have been found ( Lacin et al . , 2019 ) , lineage identity , in general , correlates poorly with neuronal identity as defined by gene expression ( Hobert et al . , 2016; Zeng and Sanes , 2017 ) . Conversely , gene expression is often correlated across neurons that differ in identity as defined by their function , morphology , and neuroanatomical location ( Hobert , 2016; Hobert and Kratsios , 2019 ) . This is because neuronal identities are defined not by single genes , but by the expression of often overlapping batteries of TEGs . An intersection of lineage with the expression of a specific TEG may thus , in general , include fewer neurons than an intersection of the expression patterns of two TEGs . In addition , because neurons from a given lineage typically remain regionally localized , intersections made using lineage information will tend to restrict neuronal targeting anatomically . Recombinase-based intersectional methods that combine information about lineage and cell type have been developed in both mice and fruit flies and have been shown to substantially restrict targeting to cell groups of interest ( Brust et al . , 2014; Ren et al . , 2016 ) . However , the use of such methods has remained largely limited to specific cases—in mice , sublineages of brainstem serotonergic neurons ( Okaty et al . , 2015 ) , and in flies , subtypes of Type II transit-amplifying neural stem cells ( i . e . neuroblasts , NBs ) of the central brain ( Ren et al . , 2018; Ren et al . , 2017 ) . This is because of the paucity of lineage-restricted enhancers . Just as there are few TEG enhancers that are active in small numbers of mature neurons , there are also few identified enhancers that exhibit lineage-specific activity . In the fly , a systematic analysis of some 5000 neural enhancer domains identified 761 with activity in embryonic NBs , but 99 of these expressed in most or all lineages ( Manning et al . , 2012 ) . A separate analysis indicates that the remainder are at best active in 5–20 lineages ( Awasaki et al . , 2014 ) . The routine use of lineage-cell type intersections for neural circuit mapping will thus require more refined methods of isolating neuronal lineages or sub-lineages . To achieve such lineage refinement , we introduce here a combinatorial method analogous to the Split Gal4 technique used to restrict neuronal targeting to the intersection of two TEG expression patterns ( Luan et al . , 2006 ) . We restrict reconstitution of a Split Cre recombinase to the expression patterns of two independent NB-active enhancers ( i . e . NBEs ) . Only NBs in which both enhancers are active thus make full-length Cre . Cre is then used to selectively promote activity of the Gal4 transcription factor—expressed under the control of a TEG enhancer—in the mature progeny of these NBs , thus implementing a second intersection . Our method ( i . e . ‘Split Cre-assisted Restriction of Cell Class-Lineage Intersections , ’ or SpaRCLIn ) generalizes the capabilities of the CLIn technique introduced by Ren et al . ( 2016 ) by expanding the range of possible intersections to most Drosophila lineages while maintaining compatibility with all existing Drosophila Gal4 driver lines . To facilitate SpaRCLIn’s use , we have generated a variety of tools , including two libraries of transgenic fly lines , each of which expresses distinct Split Cre components under the control of 134 different NBEs . We characterize the efficacy of these SpaRCLIn reagents and provide examples of their use in restricted neuronal targeting and circuit-mapping . SpaRCLIn was developed to refine the expression pattern of a Gal4 driver using the basic strategy shown in Figure 1 . In common with other existing methodologies , SpaRCLIn uses a recombinase ( i . e . Cre ) to excise an otherwise ubiquitously expressed construct encoding Gal80 , a suppressor of the Gal4 transcription factor ( Figure 1A–B ) . As in the CLIn technique , recombinase expression—and thus the excision of Gal80—occurs only in targeted NBs , rendering the progeny of these NBs permissive to Gal4 activity ( Figure 1C ) . Those progeny that lie within the expression pattern of the Gal4 driver will be competent to drive UAS-reporters and effectors , such as UAS-GFP . In the SpaRCLIn technique , distinct NBEs are used to express components of a bipartite Split Cre molecule in restricted subsets of NBs . In lineages of these NBs that contain mature neurons within the Gal4 expression pattern , Gal4 will be active . This population of neurons can be additionally parsed using a tripartite Split Cre to further restrict the subset of NBs that make active Cre ( Figure 1D ) . Although most recombinase-based expression systems in Drosophila , such as MARCM ( Lee and Luo , 1999 ) , Flp-out Gal80 ( Gordon and Scott , 2009 ) , and FINGR ( Bohm et al . , 2010 ) have preferentially used the Flp recombinase for Gal80 excision , we selected Cre for use in SpaRCLIn because of its demonstrated ability to retain high activity in a variety of bipartite forms ( Hirrlinger et al . , 2009; Jullien , 2003; Kawano et al . , 2016; Kennedy et al . , 2010; Rajaee and Ow , 2017 ) . Although Cre activity has been reported to be toxic in Drosophila when chronically expressed at high levels ( Heidmann and Lehner , 2001; Nern et al . , 2011 ) , it has previously been used in NBs without apparent adverse effects ( Awasaki et al . , 2014; Hampel et al . , 2011; Ren et al . , 2016 ) . Because our system requires use of a tripartite Cre to achieve the most refined targeting it was also desirable to use a method of splitting Cre that would permit reconstitution of the intact molecule to obtain the highest activity levels . Split inteins , which are capable of autocatalytically joining two proteins to which they are fused , are well-suited to this purpose and distinct split inteins have been previously shown to support reconstitution of recombinase activity from complementary Cre fragments fused to them ( Ge et al . , 2016; Han et al . , 2013; Hermann et al . , 2014; Wang et al . , 2012 ) . Figure 1E shows the primary structure of Cre , indicating the location of the breakpoints ( green highlight ) at which we introduced split intein moieties into the molecule . These breakpoints separate the amino acid residues in the primary structure that form the DNA-binding sites ( blue ) and the active site ( yellow highlight ) , thus insuring that none of the fragments retains catalytic activity . Two Split Cre fragments , CreAB and CreC , were generated by the breakpoint between amino acids P250 and S251 to implement the bipartite Split Cre system ( Figure 1F , G ) , while dividing the CreAB fragment at the breakpoint between amino acids D109 and S110 was used to create two further fragments ( i . e . CreA and CreB ) which together with CreC form the basis of the tripartite Split Cre system ( Figure 1H , I ) . The split intein pairs used to generate these fragments , gp41-1 and NrdJ-1 , were chosen based on their trans-splicing efficiency and their lack of cross-reactivity ( Carvajal-Vallejos et al . , 2012 ) . The latter criterion was critical for avoiding the generation of unproductive fusion products of the Cre fragments . After confirming the ability of the bi- and tripartite constructs to reconstitute Cre activity when co-expressed in transfected S2 cells ( data not shown ) , we used them to generate transgenic fly lines in which they were expressed in patterns dictated by individual enhancers that exhibited activity in neuroblasts . Most of the NBEs selected for this purpose were taken from the large collection of enhancer fragments with fully defined sequences created by the Rubin lab ( Pfeiffer et al . , 2008 #43 ) . Most of the NBEs selected were from a previously characterized collection of embryonically active NB enhancers ( Manning et al . , 2012 ) , with the remainder characterized as indicated in the Materials and Methods and Supplementary file 1 . A total of 134 NBEs were used to make two libraries of transgenic fly lines , one expressing the CreB fragment under the control of each of the 134 NBEs and the other similarly expressing the CreC fragment . These lines thus collectively express CreB and CreC in a large number of distinct and often overlapping subsets of NBs ( Figure 1—figure supplement 1 ) . However , because the 134 enhancers are also typically active in mature neurons , the production of full-length Cre is not necessarily restricted to NBs ( Jenett et al . , 2012 ) . To ensure NB-specific reconstitution of Cre activity , we placed the CreA and CreAB fragments under the control of a compound enhancer formed by fusing individual enhancer elements of the NB-specific genes , deadpan ( dpn ) and nervous fingers-1 ( nerfin-1; see Materials and methods ) . This synthetic dpn-nerfin-1 enhancer ( i . e . DNE ) combines the complementary temporal characteristics of both component enhancers , maintaining strong , broad , and specific activity throughout embryonic neurogenesis ( Figure 1—figure supplement 2A , B ) . Use of the DNE thus ensured that full-length , active Cre would be generated only in NBs where expression of the Cre fragments overlapped , and not in fully-differentiated neurons ( Figure 1G , I ) . This enhancer also expresses in most of the NBs that give rise to the Drosophila CNS with the exception of those found in the late-developing optic lobes , and thus guarantees substantial coverage of the mature neurons found within the expression patterns of Gal4 lines . To detect activity of the Split Cre constructs in vivo , we created transgenic flies carrying a reporter construct in which the floxed Gal80 gene , the expression of which is driven by a ubiquitously active Actin 5C promoter , is followed by the gene encoding the red fluorescent protein , tdTomato ( Figure 1J ) . Expression of tdTomato from this construct , which we call Cre80Tom , thus identifies neurons in which Gal80 has been excised . Gal80 excision , identified by the appearance of tdTomato expression , is identifiable as early as embryonic stage eight when the CreAB and CreC fragments are driven by the DNE and it is widespread in the developing CNS by stage 13 ( Figure 1—figure supplement 2C ) . The onset of Cre activity is sufficiently early to label many of the progeny of a defined NB lineage targetable by the R59E09-Gal4 line , previously identified by Lacin and Truman ( 2016 ) ( Figure 1—figure supplement 2D ) . These data suggest that Gal80 excision occurs relatively early . The bipartite system , using the DNE-CreAB and CreC fragments expressed under the control of two different neuroblast enhancers ( NBE43H02 and NBE44F03 ) , also generated expression patterns in neuronal lineages of third instar larvae ( Figure 1K and L ) . The expression patterns include not only the NBs in which Cre activity is reconstituted , but also the progeny of these NBs , since tdTomato expression is activated in all cells born within these lineages after Gal80 is excised . Although the expression patterns differ in the two cases , they share a small number of common NB lineages as is revealed by application of the tripartite Cre system using the NBE44F03 and NBE43H02 enhancers to drive CreB and CreC , respectively , together with DNE-CreA ( Figure 1M ) . Expression in this case is limited to approximately three bilateral lineages in the ventral nerve cord ( VNC ) and two in the brain . These examples illustrate how the bi- and tripartite Split Cre constructs selectively reconstitute Cre activity in NBs targeted by individual NBEs , and demonstrate that the tripartite Split Cre system can be used to restrict Cre activity to only those NBs in which two distinct NBEs are active . As this example illustrates , the tripartite system generates intersections of two NBE-CreC expression patterns by substituting one CreC fragment with a CreB fragment driven by the identical NBE . To facilitate such substitutions , all NBE-CreC insertions were made on Chromosome III and all NBE-CreB insertions were made on Chromosome II . To evaluate the reproducibility of expression driven by individual NBEs , we examined NBE-CreC∩DNE-CreAB >CreTom crosses for all 134 NBEs in multiple preparations ( on average four per NBE ) and compared these patterns with expression observed in all NBE-CreB∩DNE-CreA∩ DNE-CreC >CreTom crosses . The large size of the patterns and the inability to reliably identify identical neurons and lineages across preparations prevented a systematic analysis , but in general the patterns were similar across preparations for any given cross ( Figure 1—figure supplement 3 ) In most cases , patterns obtained with a given NBE-CreB line also resembled the pattern obtained with the NBE-CreC line made with the same enhancer ( compare panels A and B in Figure 1—figure supplement 3 ) . However , for 24 NBEs overt differences were observed ( Figure 1—figure supplement 3C vs 3D ) . Apart from these NBEs , which have been marked with an asterisk in Supplementary file 1 along with guidance as to their use , the tripartite system represents a reliable intersectional method for restricting Cre activity to subsets of NBs . The progeny of these NBs that are generated after Cre activation will not only express the reporter tdTomato , but will also fail to express the Gal80 transgene , thus permitting Gal4 to function . The selective disinhibition of Gal4 activity in targeted lineages permits UAS-transgenes to be expressed in cells of those lineages whenever they lie within the expression pattern of a Gal4 driver . This allows targeted lineages to be parsed according to the properties of the mature neurons to which they give rise using cell-type specific Gal4 drivers . Such so-called ‘cell class-lineage intersections’ have been previously performed to identify subsets of neurons generated by Type II NBs of the Drosophila brain , which can be selectively targeted using a Type II-specific enhancer ( Ren et al . , 2016; Ren et al . , 2018; Ren et al . , 2017 ) . Among the neurons generated by Type II NBs are several populations of dopaminergic neurons , identified by a Tyrosine Hydroxylase-specific Gal4 driver ( TH-Gal4 ) . Dopaminergic neurons are of considerable interest because of their roles in a variety of important neurobiological processes , including learning , sleep , and locomotion ( for review see Kasture et al . , 2018 ) . The approximately 120–130 dopaminergic neurons in the Drosophila CNS are produced by diverse NBs and numerous reagents have been generated to selectively target them ( Aso et al . , 2014; Friggi-Grelin et al . , 2003; Xie et al . , 2018 ) . As a first test of the SpaRCLIn system , we therefore asked whether it could restrict expression of the TH-Gal4 driver ( Figure 2A ) to small numbers of distinct dopaminergic neurons based on their different lineages of origin . Using a small subset of the NBE-CreC lines in combination with DNE-CreAB , we examined the expression patterns produced by intersection with TH-Gal4 . The expression patterns produced by these intersections were noticeably reduced compared with the full pattern of the TH-Gal4 driver , but they typically still contained 10’s of dopaminergic neurons distributed broadly across the neuraxis ( Figure 2B–C ) . In cases where the expression patterns produced by the bipartite crosses shared a neuron ( Figure 2B–C , arrows ) , combining the relevant NBEs using the tripartite system succeeded in isolating these neurons from most others in the two original crosses ( Figure 2D ) . In general , restricting NB expression using the tripartite system—by pairing the NBE-CreC constructs with NBE-CreB constructs made with different enhancers—produced significantly reduced expression patterns , sometimes consisting of one to two cells or bilateral cell pairs ( Figure 2D–H ) . The expression patterns from 14 NBE-CreC∩NBE-CreB intersections—produced by combining 15 distinct NBEs—were analyzed in detail to quantify both the average number of dopaminergic neurons and the stereotypy of expression for each intersection ( Figure 2I ) . We found that the average number of labeled neurons per preparation did not exceed 8 . 5 ( ±3 . 8 , n = 16 ) for any intersection and was less than 4 . 3 ( ±2 . 3 , n = 17 ) for two-thirds of them . This sparseness of expression suggests that the NBEs tested do not overlap extensively in their NB expression patterns . Stereotypy of expression was also generally present despite considerable variability . Only in one extreme case , did there appear to be a complete absence of stereotypy , with all CNS preparations that had expression displaying a distinct pattern ( Figure 2—figure supplement 1 ) . For all other intersections , at least one principal neuron was found that was shared by multiple preparations , based on cell position and morphology ( Figure 2I , black bars ) . For over half of the intersections , this principal common neuron was shared by 50% or more of preparations . In most cases , other neurons were also found , though preparations containing only such neurons typically occurred at lower frequency ( Figure 2I , gray bars ) . Consistent with this variability of expression , neurons that recurred across preparations were not necessarily found in the same combinations ( Figure 2—figure supplement 2 ) . The sparseness of labeling combined with the variability of expression likely accounts for why half of the intersections yielded at least one preparation without any expression . Interestingly , four of the seven intersections that yielded preparations devoid of expression shared an enhancer ( R14E10 ) , suggesting that particular enhancers may strongly influence the extent of labeling . Variability of labeling also appeared to be enhancer-dependent in that use of the same enhancer ( i . e . R17A10 ) to drive both CreB and CreC components did not necessarily reduce stochasticity . Indeed , although all preparations that had expression shared a common identifiable neuron in this case ( Figure 2I ) , their expression in other neurons varied considerably . A possible source of this variability of expression is weak NBE activity that results in lowered expression of Cre components and consequently more sporadic reconstitution of Cre activity . More work will be required to examine this hypothesis . Regardless , our results demonstrate SpaRCLIn’s ability to substantially restrict expression of a Gal4 driver with sufficient stereotypy in single neurons to be useful for the neuronal manipulations employed in neural circuit mapping . To examine SpaRCLIn’s efficacy for circuit mapping , we used it to identify neural substrates of proboscis extension ( PE ) , a motor pattern normally elicited by gustatory stimuli , but also by the hormone Bursicon in newly eclosed flies ( Peabody et al . , 2009 ) . Robust PE can be readily induced even in older flies using a driver ( rkpan-Gal4 ) that selectively expresses in Bursicon-responsive neurons ( Video 1 , Figure 3A , B; Diao and White , 2012 ) . Expressing the heat-sensitive ion channel UAS-dTrpA1 under the control of this driver , we performed an initial ( ‘Step 1’ ) screen of the CreC library using the bipartite SpaRCLIn system ( Figure 3—figure supplement 1A ) . In this screen , crosses were conducted between each NBE-CreC line and a line that combined all other components , including Cre80Tom , DNE-CreAB , rkpan-Gal4 , and UAS-dTrpA1 . To facilitate visualization of neurons within the resulting expression pattern without requiring additional genomic insertions , we used a dual expression construct ( Cre80Tom- GFP ) that contained actin^Gal80^myr-tdTomato and a 10XUAS-mCD8GFP reporter ( Figure 3—figure supplement 2 ) . Progeny were videorecorded in small chambers on a temperature-controlled plate and assayed for heat-induced PE . Interestingly , several different PE phenotypes were apparent , but only those that involved full extension of the proboscis could be reliably scored under our assay conditions and we therefore focused on the latter . Applying this criterion , we identified 23 NBE-CreC∩DNE-CreAB intersections for which UAS-dTrpA1 activation reliably induced robust PE in greater than 50% of the progeny . The expression patterns resulting from these CreAB∩C∩rkpan-Gal4 ( i . e . Step 1 ) intersections , examined using a UAS-GFP reporter , were clearly restricted relative to rkpan-Gal4 expression ( Figure 3C–D ) , but they were insufficiently sparse to readily identify the neurons—or population of neurons—responsible for inducing the PE motor pattern . Taking advantage of SpaRCLIn’s ability to further restrict expression , we used the tripartite system to carry out a second ( ‘Step 2’ ) screen in which the 23 identified NBE-CreC components were combined pairwise with NBE-CreB components made using the same 23 enhancers ( Figure 3—figure supplement 1B ) . The latter were selected from the NBE-CreB library and crosses were made that combined distinct NBE-CreB and NBE-CreC components with DNE-CreA , rkpan-Gal4 , and Cre80Tom-GFP . These Step two crosses resulted in CreA∩B∩C ∩ rkpan-Gal4 intersections that were assayed for PE as before . Approximately 70 intersections were tested before screening was discontinued because 11 intersections had already yielded PE phenotypes in greater than 50% of flies . The phenotype observed was typically less sustained than that produced by activation of the full rkpan-Gal4 expression pattern in that activation typically caused rhythmic , rather than tonic , extension of the proboscis , which after prolonged heating often transitioned to lifting of the rostrum rather than full extension ( Video 2; Figure 3E ) . The rkpan-Gal4 expression patterns in flies exhibiting this phenotype were substantially reduced for many of the intersections tested and they consistently included particular neurons in the subesophageal zone ( SEZ ) that were characterized by somata near the saddle , broad arbors along the superior gnathal ganglion ( GNG ) , and axons that extended medially before turning , with one branch coursing down each side of the midline and then turning laterally along the medial-inferior edges of the GNG ( Video 3 ) . Two closely apposed neurons of this type were observed , sometimes as bilateral pairs ( Figure 3F ) , and sometimes on only one side ( Figure 3G ) . These neurons , which we call the PErk neurons , were notably prominent in the 16H11-CreB∩44F09-CreC intersection , where they constituted the entire expression pattern of 16 animals ( n = 78 total ) , all of which exhibited PE induction upon heating . Indeed , all 36 animals from this intersection that tested positive for the PE phenotype and were successfully dissected showed expression in the PErk neurons , while none of the animals ( n = 38 ) that tested negative had such expression ( Figure 3I ) . Most of the latter , in fact , had little to no expression . Similar results were obtained with a second intersection ( 44F09-CreB∩10G07-CreC ) . All 19 animals that exhibited induced PE in this intersection had expression in the PErk neurons , and in three animals these were the only neurons present . A third intersection that yielded the PE phenotype in all animals likewise showed consistent expression in the PErk neurons , but the correlation between the PE phenotype and expression in these neurons was somewhat less readily established because of expression in other neurons ( 5 . 6 ± 1 . 8; n = 14 preparations; Figure 3—figure supplement 3 ) . The above examples demonstrate that SpaRCLIn can be used to rationally parse the expression patterns of Gal4 drivers using the workflow shown in Figure 3—figure supplement 1 . One challenge to using this system , however , is the large number of transgenes required to implement it . This is especially true for Step two screening with the tripartite system . To mitigate this burden , we have created several reagents that will facilitate use of the system . In addition to the Cre80Tom-GFP construct described above , we have developed other dicistronic constructs to facilitate manipulating neuronal activity in SpaRCLIn screens ( see Key Resources Table ) . These include constructs and fly lines for Cre80-Kir2 . 1 and Cre80-dTrpA1 ( Figure 3—figure supplement 2 ) . In addition , we have developed an alternate Step one strategy that may avert the need for Step two screening in favorable cases . The alternate strategy uses a transiently expressed DNE-CreAB designed to be active only during early stages of neurogenesis . This construct , which we call ‘FRTerminator , ’ is self-excising in that it is flanked by Flp Recombination Target ( FRT ) sites and encodes a Flp recombinase gene that is co-expressed with CreAB ( Figure 4A ) . Upon expression under control of the DNE enhancer , this construct will remove the CreAB gene and thus limit its expression to early ( embryonic ) neuroblasts ( Figure 4B ) . CreAB will thus be available to reconstitute Cre activity only with complimentary CreC fragments that are also expressed at this time . CreCs whose expression is driven by NBEs that become active only after the elimination of CreAB from neuroblasts , will not lead to the generation of Gal4-competent neurons . Expression patterns resulting from the combination of FRTerminator with NBE-CreCs will thus , in general , be reduced relative to those produced by DNE-CreAB ( Figure 4C , D ) . To determine whether the FRTerminator might therefore expedite parsing of Gal4 expression using the SpaRCLIn system , we repeated selected crosses from the rkpan-Gal4 Step one screen described above . We focused on the 23 NBE-CreC lines that yielded flies with PE phenotypes , combining each with the FRTerminator , rkpan-Gal4 and Cre80-GFP . Progeny were tested for PE upon dTrpA1 activation . We found that three NBE-CreC lines ( 44F09 , 57B09 , and 14E10 ) produced progeny with PE phenotypes at frequencies ranging from 9–17% . Although these frequencies were considerably lower than those obtained using DNE-CreAB , the resulting expression patterns were substantially sparser compared with those of progeny from DNE-CreAB crosses ( Figure 4E , F ) . All animals examined that had PE phenotypes also included in their expression patterns the PErk neurons ( n = 40 ) . In contrast , only one of the animals examined that lacked the phenotype had these neurons ( n = 39 ) . A strong correlation between PE and the presence of the PErk neurons was thus observed , again permitting the conclusion that these neurons are substrates for the behavioral phenotype . We conclude that FRTerminator-based Step one screens may serve as a useful shortcut to serial Step one and Step two screens for restricting Gal4 expression and identifying functionally important neuronal subsets . Our use of SpaRCLIn to identify the RK-expressing neurons that trigger robust proboscis extension demonstrates SpaRCLIn’s ability to systematically parse a neuronal group and identify the functionally relevant subset . Just over 200 crosses—134 crosses for the Step 1 screen of NBE-CreC lines and 70 NBE-CreB∩C Step two crosses—were required to identify two pairs of command-like neurons capable of inducing PE upon activation ( i . e . the PErk neurons ) . Importantly , we discontinued the Step two screen after testing 70 of the 253 possible intersections because of evident redundancy of the command-like neurons . The latter were prominent in the expression patterns of numerous independent Step two intersections and were readily correlated with PE induction in three that produced particularly reduced expression patterns . In the intersection with the sparsest expression , the two pairs of PE-inducing neurons often comprised the entire observable pattern in flies that had the PE phenotype , illustrating the extreme reduction in expression achievable with SpaRCLIn . The demonstration that the PErk neurons can be isolated in single crosses using the FRTerminator indicates that this reduction in expression can be attained without the labor of Step two screening . However , the lower frequency of the PE phenotype in FRTerminator crosses in our example also suggests that FRTerminator-based screens may require testing more animals for each intersection than a standard Step one screen in order to reliably identify positives . Activation of the PErk neurons elicits rhythmic proboscis extension , rather than the tonic PE elicited by activation of all rkpan-Gal4 neurons . This suggests that additional RK-expressing neurons—perhaps lacking command capability—modulate the effects of activating the PErk neurons . Based on their induction of rhythmic extension and their apparent lack of a projection to the proboscis muscles , we conjecture that the PErk neurons identified here are not motor neurons , the activation of which results in tonic and often partial PE ( Gordon and Scott , 2009; Schwarz et al . , 2017 ) . Similarly , the anatomy of the PErk neurons differs from that of other identified neurons that can drive PE when activated , including second-order projection neurons ( Kain and Dahanukar , 2015 ) , modulatory neurons ( Marella et al . , 2012 ) , and a local SEZ interneuron called the Fdg-neuron ( Flood et al . , 2013 ) . Like the Fdg-neuron , however , the neurons identified here seem to function in a premotor capacity , perhaps as part of the central pattern generator for PE that regulates fly feeding ( Itskov et al . , 2014 ) . Further work will be required to determine the precise role of the PErk neurons in the feeding circuitry and their relationship to other identified neurons involved in PE . It also remains to be determined whether activation of both PErk neurons is required to induce the PE phenotype . Indeed , from the standpoint of the efficacy of the SpaRCLIn system it is important to ask why SpaRCLIn failed to separate these two pairs of neurons . The similarity of the two PErk neurons in both soma position and projection pattern is consistent with their being part of the same lineage . Such neurons will necessarily be more difficult to parse using SpaRCLIn , which can separate neurons within the same lineage only based on their birth order . What would be required to do so is having two NBEs that are active in the same lineage but at different times so that they separate earlier- from later-born neurons . Such NBEs , by generating Cre only in older neuroblasts , will generate sublineages of Gal4-competent neurons . Although many of the NBE’s used to make our CreB and CreC libraries clearly generate such sublineages—based on the patterns shown in Figure 1—figure supplement 1—it is doubtful that they cover more than a fraction of all temporal windows of neurogenesis in all neuronal lineages . A method for systematically isolating sublineages of later born neurons using SpaRCLIn may become possible if neuroblast-specific enhancers can be found that are selectively active at later stages of neurogenesis . These could then be used in lieu of the DNE used here . Candidates for such enhancers are those that determine expression of the so-called ‘temporal transcription factors’ that regulate the progressive divisions of many neuroblasts ( Doe , 2017 ) . Although stochasticity is not an uncommon feature of many expression systems ( Bohm et al . , 2010; Tastekin and Louis , 2017 ) , the variability of expression generated by SpaRCLIn was notable . Even for intersections that reliably produce very similar expression across animals , it is not common to get exactly the same pattern twice . The infidelity of expression may derive , at least in part , from intrinsic stochasticity of NBE activity . However , our results indicate that individual NBEs drive expression in broadly reproducible patterns . Other factors that may contribute to expression variability include the strength and/or temporal properties of NBE activity . If a Cre component is only weakly expressed , or expressed late during neurogenesis , a limited amount of active , full-length Cre may be produced and excision of Gal80 may be sporadic in the expressing neuroblasts . Also , the success of Cre reconstitution may vary , and may be particularly low when the religation of three fragments is required . Although the two split inteins used in the SpaRCLIn system were chosen based on their favorable reaction kinetics as determined in vitro , the speed , efficiency , and variability with which they react in different types of cells remain unknown . Finally , because the efficacy of Cre-mediated excision depends on the distance between the loxP sites flanking the excised fragment , it may prove possible to increase the efficacy of excision—and thus reduce variability of expression—using strategies that decrease the distance between the loxP sites flanking the fragment to be excised . While further work will be required to identify the sources of variable expression within the system , the observed stochasticity is not necessarily a disadvantage for circuit-mapping applications , as illustrated by the examples presented here . By providing partially ‘randomized’ expression patterns , SpaRCLIn permits causative relationships to be inferred between groups of manipulated neurons and the effects produced by their manipulation ( Jazayeri and Afraz , 2017 ) . Such randomization has been commonly exploited in so-called ‘Flp-out’ methods that rely on stochastically induced recombinase activity to remove an FRT-flanked gene or transcription stop cassette ( Flood et al . , 2013; Gordon and Scott , 2009; Kain and Dahanukar , 2015 ) . This logic is naturally implemented in SpaRCLIn , but because randomness of expression is considerably more constrained than that observed in systems that rely on strictly stochastic methods , and because the size of the expression patterns is typically small , correlations can be readily established . One consequence of SpaRCLIn’s stochasticity that must be considered in circuit mapping applications , however , is the lowered frequency of bilateral labeling . Most neurons occur as members of bilateral pairs and we observed numerous instances in which SpaRCLIn-derived expression patterns contained only a single member of each pair in a given preparation—presumably due to the variable success of Gal80 excision in both NBs giving rise to the pair . The reduced bilateral representation of neurons may likewise reduce the frequency of phenotypes observed as a consequence of a particular manipulation if , for example , both neurons in a pair must be affected to produce a phenotype . This is often the case for suppression of function , where both neurons in the pair must be inhibited . It is therefore possible that SpaRCLIn will be most effective in applications that involve neuronal activation where unilateral manipulations are often sufficient to generate an effect as they are for proboscis extension . The ability of SpaRCLIn to isolate a given set of neurons of interest in a Gal4 pattern depends critically on the extent to which the various Split Cre components are expressed in the neuroblast lineages of the fly . This will be determined both by the breadth of NB expression of the DNE enhancer used here to delimit Cre activity and by the collective coverage of NB expression provided by the NBEs represented in the libraries of CreB and CreC lines . Our analysis of 3rd instar larval expression in DNE∩NBE intersections ( Figure 1—figure supplement 1 and data not shown ) indicates that many , if not most , NB lineages of the ventral nerve cord and central brain are likely represented within the libraries . Indeed , many lineages are clearly represented multiple times in that different intersections repeatedly isolated the same neurons ( e . g . the PErk neurons ) for both the rkpan-Gal4 and TH-Gal4 drivers . It is less clear , however , that all members of each lineage are represented as not all NBE’s are active during early NB divisions . This is evident from the restriction in NB expression observed when the FRTerminator construct is used , since this construct acts by eliminating lineages or sublineages in which Cre activity is initiated sometime after neurogenesis has begun . It is also clear that the DNE does not express efficiently in NB lineages in the optic lobe ( data not shown ) . To extend the capability of the system to include these lineages will require either the development of a more general neuroblast-specific enhancer or augmenting the system to include an enhancer that specifically targets optic lobe NBs . The effectiveness of SpaRCLIn also depends critically on the success of Cre reconstitution by the system , which is effected by two pairs of split inteins ( Shah and Muir , 2011; Shah and Muir , 2014 ) . These trans-splicing protein fragments function naturally in protein religation and are an emerging technology for use in transgenic animals ( Hermann et al . , 2014; Wang et al . , 2018; Wang et al . , 2012 ) . Their advantages are that they lend themselves readily to intersectional methods , are genetically encoded , and in numerous cases display rapid reaction kinetics and low cross-reactivity . A disadvantage , on which some recent progress has been made ( Stevens et al . , 2017 ) , is that most split inteins require specific flanking amino acid residues in the proteins to which they are fused , in particular a cysteine or serine residue immediately downstream of the N-intein . We were able to create self-ligating split Cre fragments capable of reconstituting full-length , active Cre enzyme in Drosophila NBs by choosing breakpoints in the Cre sequence preceded by a serine residue—the native condition of the NrdJ-1 and gp41-1 split inteins used here ( Carvajal-Vallejos et al . , 2012 ) . Orthogonal ( i . e . non-interacting ) split inteins thus represent attractive tools for reconstituting the function of multiply split proteins , a methodology that should be applicable in other model organisms . Although sophisticated methods for neuronal targeting have been a hallmark of neurobiological studies on the fly , and single cell manipulations are being leveraged in a growing number of cases to elucidate Drosophila brain circuits , targeting every cell in the fly CNS remains an aspirational goal . Recent progress towards this goal has been made using the Split Gal4 system ( Dionne et al . , 2018; Tirian et al . , 2017 ) , and innovative methods continue to be developed using emerging tools ( Garcia-Marques et al . , 2019 ) . An advantage of SpaRCLIn is that it represents a relatively small set of stand-alone reagents for high-specificity neuronal targeting that can be used with the many existing components of the Gal4-UAS system . Importantly , SpaRCLIn also represents an open resource that can readily be augmented by end-users . As methods improve for rationally identifying NB lineages based on gene expression and enhancer activity , the existing SpaRCLIn libraries can be supplemented with lines that together permit the selective targeting of an increasing number of neuroblast lineages . The NB-specific enhancers recently identified by Lacin and Truman ( 2016 ) and used here to characterize our split Cre components ( Figure 1—figure supplement 2D ) provide good examples of reagents that can be used to improve the SpaRCLIn libraries . By combining these libraries with an optimized set of Gal4 drivers that express in distinct subsets of brain cells ( distinguished , for example , by transcription factor expression ) , one can imagine having a set of 3 libraries that in combination can selectively target most neurons in CNS . Vinegar flies of the species Drosophila melanogaster were used in this study . Unless otherwise noted , all flies were grown on BDSC Cornmeal Food and maintained at 25°C in a constant 12 hr light–dark cycle . Both male and female progeny of the genotypes indicated in Supplementary file 2 were used in this study . Previously described fly stocks and their sources are listed in the Key Resources Table . Fly lines generated for this study were made using the DNA constructs described below . Injection of these constructs to produce transgenic flies was carried out by Rainbow Transgenic Flies , Inc ( Camarillo , CA ) . All transgene insertions except the insertion of the DNE-Gal4 were mediated by ΦC31 integrase and placed in the defined attP landing sites indicated in Key Resources Table . Flies made with the DNE-Gal4 were generated by p-element mediated transgenesis . Transgenic flies of the NBE-CreB library have transgene insertions on the 2nd chromosome at attP40 , while all flies in the NBE-CreC library have insertions on the 3rd chromosome at either VK00033 or VK00027 . All oligonucleotide and gBlock synthesis was carried out by Integrated DNA Technologies , Inc ( Coralville , Iowa ) , and all final constructs were verified by sequencing ( Eurofins Scientific , Louisville , KY or Macrogen Corp , Rockville MD ) . For routine molecular biology , the following reagents were used according to the manufacturers’ supplied protocols: PCR amplification: Q5 High-Fidelity 2X Master Mix #M0492S ( New England Biolabs , Ipswich , MA ) ; DNA ligation: Quick Ligation Kit #M2200L ( New England Biolabs , Ipswich , MA ) ; Cloning: Gateway LR Clonase II Enzyme mix #11791100 ( Thermofisher Scientific , Waltham , MA ) , and In-Fusion HD Cloning Plus #638911 ( Takara Bio USA , Inc , Mountain View , CA ) . gBlocks were used to generate most of the final and intermediate constructs described below , including the DNA fragments encoding the NrdJ-1 and gp41-1 split inteins and the Cre fragments described in the manuscript . DNA sequences of the split inteins were back-translated from the published protein sequences ( Carvajal-Vallejos et al . , 2012 ) and all sequences were codon biased for Drosophila . Sequences of all gBlock fragments and PCR primers are listed in Supplementary file 3 . The following reagents , which were used to make several constructs as indicated below , are all described in Pfeiffer et al . ( 2010 ) : pBPGal80Uw-5 , pBPLexA::P65 , 10XUAS-IVS-myr::tdTomato , 10XUAS-mCD8::GFP , and pBPGAL80Uw-6 . The indicated Cre80Tom constructs were made stepwise using the described procedures . All split Cre constructs were made by Gateway cloning ( LR reaction ) . Two sets of destination vectors with split Cre components were made: one for use with entry clones containing promoters , and another for entry clones containing enhancers . The 134 NBE entry clones were combined with the latter to make the expression clones used to generate the CreB and CreC libraries . To make the CreA ( HJP-176 ) destination vectors for use with promoter entry clones , a KpnI-IVS-NheI fragment made from annealed oligonucleotides , a NheI-gBlock012-AgeI gBlocks fragment and an AgeI-PmeI-WPRE-HindIII PCR fragment ( amplified from pBPGAL80Uw-6 using PrimerS472 and PrimerS473 ) were placed between the NheI and HindIII restriction sites of the pBPGw vector ( Addgene Plasmid #17574 Pfeiffer et al . , 2008 ) . Other split Cre destination vectors ( i . e . HJP177 ~HJP180; see the Key Resources Table ) were made by replacing the NheI-CreA-gp41-1N-AgeI fragment in CreA ( HJP-176 ) with fragments consisting of: NheI-gBlock010-SphI + SphI-gBlock011-AgeI ( HJP177 ) , NheI-gBlock008-BsaI+BsaI-gBlock009-AgeI ( HJP178 ) , NheI-gBlock007-AgeI ( HJP179 ) , or NheI-gBlock010-SphI+SphI-gBlock015-AgeI ( HJP180 ) . To create a set of destination vectors for use with enhancer entry clones ( ‘the U-series’ ) , an FseI-DSCP-KpnI synthetic core promoter ( Pfeiffer et al . , 2008 ) was made from annealed oligos and inserted between the FseI and KpnI restriction sites of each of the destination vectors made for use with promoter entry clones . This produced constructs HJP194 ~196 , HJP-207 and HJP-208 ( See Key Resources Table ) . Prior to the production of transgenic fly lines , the functionality of all Cre constructs was validated in cultured S2 cells by placing the constructs under the control of the Actin5C promoter and testing in appropriate combinations for expression and activity using a floxed reporter construct . This construct ( HJP-473 ) was made as follows: an AvrII and PmeI flanked DNA fragment ( including partial nerfin-1 enhancer , FRT and Syn21-flipase-T2A-CreA-gp41-1N-AgeI-FRT ) were synthesized ( Epoch Life Science , Inc , Missouri City , TX ) and put between the AvrII and PmeI restriction sites of DNE-CreA-gp41-1N . The resulting construct can be used in place of DNE-CreA in Step 2 SpaRCLIn screens . It was tested , but its use is not described in this manuscript . This construct was used as an intermediary to make the final FRTerminator construct by inserting gBlock-043 ( part of the CreAB sequence and Nrdj-1N ) into its SbfI and AgeI restriction sites using the In-Fusion HD cloning technique . Two constructs were used to pre-screen candidate enhancers driving Gal4 expression . These included CreStop ( HJP225 ) and UAS-CreC ( HJP266 ) . The CreStop construct was made using a NgoMIV-loxP-hsp70 terminator-MluI gBlock to replace the loxP-Gal80 in Cre80Tom ( HJP223 ) by In-Fusion HD cloning . UAS-CreC was made by cloning a NotI-NrdJ-1C-CreC-XbaI PCR fragment ( Primer116 and Primer117; CreC as template ) between the NotI and XbaI sites of pJFRC1-10XUAS-mCD8::GFP using the In-Fusion HD cloning technique . Embryos were prepared and immunostained as described by Lécuyer et al . ( 2008 ) . Excised nervous system whole mounts were prepared from wandering third-instar larvae or adults after dissection into PBS and fixation in 4% paraformaldehyde in PBS for 20–30 min . Immunostaining was done with the antibodies listed in the Key Resources Table at the indicated dilutions . For confocal imaging , all tissues were attached to poly-L-lysine coated cover glass and mounted in Vectashield ( Vector Laboratories , Burlingame , CA ) prior to imaging with a Nikon C-2 confocal microscope . Z-series were acquired in 0 . 85 μm increments using a 20X objective using 488 nm , 543 nm or 633 nm laser emission lines for fluorophore excitation . The images shown are maximal projections of volume rendered z-stacks of confocal sections taken through the entire nervous system . NB expression of Gal4 driven by the DNE enhancer was examined in embryonic fillets by in situ hybridizations as previously described ( Ross et al . , 2015 ) . Flies assayed for proboscis extension were raised at 25°C until the white prepupa stage and then transferred to 18°C until the time of testing . For neuronal activation using dTrpA1 , the chambers were placed on the surface of the Echotherm Chilling/Heating Dry Bath IC25 ( Torrey Pines Scientific , Inc , Carlsbad , CA ) at 31°C . For the Step 1 SpaRCLIn screen , approximately a dozen adult flies ( 3–10 d old ) of each genotype were placed in glass TriKinetics tubes ( 3 mm inner diameter; TriKinetics Inc , Waltham , MA ) and videorecorded at 31°C for 3 min using a Sony NEX-VG10 videocamera . Proboscis extension behavior was analyzed from these recordings . If two or more flies exhibited robust , full-length extension , the cross was scored as positive . For the Step two tripartite SpaRCLIn screen , two flies at a time ( one male and one female ) were videorecorded together in glass minichambers ( 0 . 3 cm diameter X 0 . 7 cm length ) for 3 min at 18°C followed by 3 min at 31°C . Flies were subjected to these temperature transitions twice and proboscis extension behavior was analyzed following the recording . The criteria for positive proboscis extension was three or more bouts of full proboscis extension in both tests . For the FRTerminator behavior experiments flies were subjected to only one test . Flies used to make the videos included in the manuscript were back-mounted on a 200 uL pipette tip with 5-Minute-Rapid-Curing , General Purpose Adhesive Epoxy ( ITW polymers Adhesive , Danvers , MA ) and placed just above the heating plate , which was adjusted to apply temperature changes .	In humans – as well as flies and most other animals – the brain controls how we move and behave , and regulates heartbeat , breathing and other core processes . To perform these different roles , cells known as neurons form large networks that quickly carry messages around the brain and to other parts of the body . In order to fully understand how the brain works , it is important to first understand how individual neurons connect to each other and operate within these networks . Fruit flies and other animals with small brains are often used as models to study how the brain works . There are several methods currently available that allow researchers to manipulate small groups of fruit fly neurons for study , and in some cases it is even possible to target individual neurons . However , it remains an aspirational goal to be able to target every neuron in the fly brain individually . The Gal4-UAS system is a way of manipulating gene activity widely used to study neurons in fruit flies . The system consists of two parts: a protein that can bind DNA and control the activity of genes ( Gal4 ) ; and a genetic sequence ( the UAS ) that tells Gal4 where to bind and therefore which genes to activate . Fruit flies can be genetically engineered so that only specific cells make Gal4 . This makes it possible , for example , to limit the activity of a gene under the control of the UAS to a specific set of neurons and therefore to identify or target these neurons . Luan et al . developed a new technique named SpaRCLIn that allows the targeting of a subset of neurons within a group already identified with the Gal4-UAS system . During embryonic development , all neurons originate from a small pool of cells called neuroblasts , and it is possible to target the descendants of particular neuroblasts . SpaRCLIn exploits this strategy to limit the activity of Gal4 to smaller and smaller numbers of neuroblast descendants . In this way , Luan et al . found that SpaRCLIn was routinely capable of limiting patterns of Gal4 activity to one , or a few , neurons at a time . Further experiments used SpaRCLIn to identify two pairs of neurons that trigger a well-known feeding behavior in fruit flies . Luan et al . also developed a SpaRCLIn toolkit that will form the basis of a community resource other researchers can use to study neurons in fruit flies . These findings could also benefit researchers developing similar tools in mice and other animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2020	Cre-assisted fine-mapping of neural circuits using orthogonal split inteins
Direct lineage conversion of adult cells is a promising approach for regenerative medicine . A major challenge of lineage conversion is to generate specific cell subtypes . The pancreatic islets contain three major hormone-secreting endocrine subtypes: insulin+ β-cells , glucagon+ α-cells , and somatostatin+ δ-cells . We previously reported that a combination of three transcription factors , Ngn3 , Mafa , and Pdx1 , directly reprograms pancreatic acinar cells to β-cells . We now show that acinar cells can be converted to δ-like and α-like cells by Ngn3 and Ngn3+Mafa respectively . Thus , three major islet endocrine subtypes can be derived by acinar reprogramming . Ngn3 promotes establishment of a generic endocrine state in acinar cells , and also promotes δ-specification in the absence of other factors . δ-specification is in turn suppressed by Mafa and Pdx1 during α- and β-cell induction . These studies identify a set of defined factors whose combinatorial actions reprogram acinar cells to distinct islet endocrine subtypes in vivo . Cellular reprogramming is a rapidly expanding area of regenerative medicine . With suitable reprogramming factors , adult cells can be instructively converted to induced pluripotent stem cells ( pluripotent reprogramming ) or other types of adult cells ( lineage reprogramming ) ( Gurdon and Melton , 2008; Graf and Enver , 2009 ) . Induced pluripotent stem cells ( iPS ) can be differentiated into many cell types in the body . However , the generation of iPS cells and their subsequent differentiation is a lengthy and technically demanding process . Lineage conversion between adult cell types offers a promising alternative , directly producing defined cell types in vitro or even in vivo that may be used for disease modeling and cellular therapies ( Zhou and Melton , 2008; Vierbuchen and Wernig , 2011 ) . Recent examples of lineage reprogramming include the conversion of pre-B cells to macrophages , pancreatic acinar , α-cells , and gut cells to insulin-secreting β-cells , cardiac fibroblasts to cardiomyocyte-like cells , amniotic cells to endothelial cells , and skin fibroblasts to neurons , oligodendrocytes , neural precursors , or blood progenitors ( Xie et al . , 2004; Zhou et al . , 2008; Ieda et al . , 2010; Szabo et al . , 2010; Thorel et al . , 2010; Vierbuchen et al . , 2010; Caiazzo et al . , 2011; Yang et al . , 2011; Ginsberg et al . , 2012; Han et al . , 2012; Song et al . , 2012; Talchai et al . , 2012; Thier et al . , 2012; Najm et al . , 2013; Yang et al . , 2013 ) . Despite increasing success of lineage conversion , a major challenge of this approach is to direct the formation of specific cell types: there is a great diversity of cell types in the adult body , and many of them are further differentiated into closely related subtypes . The most extensive subtype diversification can be found in the mammalian central nervous system , where hundreds of neuronal subtypes exist . A few mammalian neuronal subtypes , including dopaminergic-like and motoneuron-like cells , have been produced from fibroblast conversion ( Caiazzo et al . , 2011; Son et al . , 2011 ) ; methods to generate many others remain to be defined . To study subtype specification in lineage reprogramming , it is necessary to first establish models , where a defined set of factors promote formation of distinct subtypes . A recent study in Caenorhabditis elegans provided such an example , where removal of a chromatin factor confers neurogenic competence to germ cells , which can be subsequently converted to different neuronal subtypes by neuron selector genes ( Tursun et al . , 2011 ) . To establish models of mammalian subtype specification in lineage reprogramming , we focused our studies in a relatively simple system , the adult pancreas , where the endocrine islets are surrounded by acinar cells , a type of exocrine cells that secret digestive enzymes . The islets contain three major endocrine subtypes: insulin+ β-cells , glucagon+ α-cells , and somatostatin+ δ-cells . β-cells secret insulin and play a key role in blood glucose regulation , whereas α- and δ-cells secrete glucagon and somatostatin to support β-cell function ( Edlund , 2001; Jensen , 2004 ) . We reported previously that pancreatic acinar cells can be directly converted to insulin+ β-cells in adult mouse pancreas by combined actions of three transcription factors , Ngn3 , Pdx1 , and Mafa ( referred to as M3 factors ) ( Zhou et al . , 2008 ) . We now report that acinar cells can also be converted to the other endocrine subtypes , namely , somatostatin+ δ-like cells and glucagon+ α-like cells , by Ngn3 and Ngn3+Mafa respectively . A defined set of factors can therefore reprogram acinar cells to the three major islet endocrine subtypes . Further studies indicate that Ngn3 , but not Mafa and Pdx1 , promotes establishment of a generic endocrine state in acinar cells at the onset of reprogramming by suppressing acinar fate-regulators and activating pan-endocrine genes . Ngn3 also promotes δ-subtype specification in the absence of other factors . Mafa and Ngn3 in turn suppress δ-specification in α- and β-cell formation , thus ensuring creation of singular endocrine subtypes . Our studies establish a series of models where combinatorial functions of defined factors convert pancreatic acinar cells to three distinct endocrine subtypes in vivo . These models provide a powerful system to gain mechanistic understanding of the lineage reprogramming process . We have previously reported that pancreatic acinar cells can be converted to insulin+ β-like cells by the combined activity of three reprogramming factors: Ngn3 , Mafa , and Pdx1 , referred to as M3 factors ( Zhou et al . , 2008 ) . Employing the same experimental system of adenoviral expression in adult mouse pancreas , which specifically targets acinar cells ( Figure 1A , Figure 1—figure supplement 1 ) , we examined the role of individual M3 factors in endocrine reprogramming . Surprisingly , Ngn3 alone induced formation of somatostatin+ ( Sst ) cells in approximately 40% of infected cells ( Figure 1B–D ) , whereas Mafa or Pdx1 alone did not induce any hormone positive cells ( Figure 1—figure supplement 2 ) . In addition , co-infection of Ngn3- and Mafa-induced formation of both glucagon+ ( Gcg ) and somatostatin+ cells , which are distinct from each other ( Figures 1E , F ) . The other two-factor combinations , Ngn3 with Pdx1 and Pdx1 with Mafa , did not yield hormone positive cells ( Figure 1—figure supplement 2 ) . Somatostatin and glucagon are the principle hormones of endocrine δ- and α-cells . These data suggest that different combinations of three reprogramming factors could convert pancreatic acinar cells in vivo to the three major islet endocrine cell types: δ- , α- and β-cells . The expression of reprogramming factors in δ- and α-cell induction is transient ( Figure 1—figure supplement 3 ) , similar to β-cell induction using the same experimental approach ( Zhou et al . , 2008 ) . To confirm the identity of the induced Sst+ and Gcg+ cells , we examined whether the induced cells have key features of endogenous δ- and α-cells . 10 . 7554/eLife . 01846 . 003Figure 1 . Induction of somatostatin+ , glucagon+ , and insulin+ cells with defined factors in adult mouse pancreas in vivo . ( A ) Schematic diagram of experimental strategy . Adenoviruses co-expressing reprogramming factor ( R . F . ) and mCherry ( cherry ) were used to directly induce conversion of acinar cells in adult pancreas . 2A peptide that mediates polycistronic expression . Phenotypes were analyzed 10 days after induction . ( B–D ) Expression of Ngn3 alone induced 40 ± 3% of the infected mCherry+ cells to become somatostatin+ ( Sst ) . ( E–G ) Co-infection of two separate viruses carrying Ngn3 and Mafa resulted in the formation of both glucagon+ ( Gcg ) and somatostatin+ cells in 11 ± 6% and 9 ± 5% of infected cells , respectively . ( H–I ) Co-expression of Ngn3 , Mafa , Pdx1 , and mCherry from a single polycistronic construct led to exclusive formation of insulin+ cells in 47 ± 8% of the mCherry+ cells . ( K ) Summary of pancreatic acinar cell conversion to endocrine subtypes with different combinations of factors . A , acinar cells . Quantifications are shown in mean ± s . d . At least 1000 cherry+ cells counted from three different animals . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 00310 . 7554/eLife . 01846 . 004Figure 1—figure supplement 1 . Adenoviral constructs used in the experiments and polycistronic factor expression . ( A ) . Diagrams of the constructs used . CMV: cytomegaloviral promoter . Dark gray bar: 2A peptide that mediates polycistronic protein expression . Cherry: monomeric cherry fluorescent protein . ( B ) . Immunostaining of HEK293 cells infected with pAd-M3 polycistronic expression virus . The majority of cherry+ cells express the M3 factors ( Ngn3 , Mafa , Pdx1 ) , indicating excellent co-expression from this construct . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 00410 . 7554/eLife . 01846 . 005Figure 1—figure supplement 2 . Mafa alone , Pdx1 alone , and combinations of Pdx1 . Mafa and Ngn3 . Pdx1 do not induce endocrine cells in pancreas . ( A–L ) Mafa alone , Pdx1 alone , Pdx1 . Mafa ( polycistronic coexpression ) , Ngn3 . Pdx1 ( polycistronic coexpression ) do not induce the three principle hormones of pancreatic islets . Sst , somatostatin; Gcg , glucagon . ( M–O ) Control staining of hormones in pancreatic islets . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 00510 . 7554/eLife . 01846 . 006Figure 1—figure supplement 3 . Transgene expression mediated by adenoviral infection in adult pancreas is transient . We performed qPCR analyses at four different time points after viral infection ( day 2 , 10 , 30 , and 60 ) in Ngn3cherry mediated delta cell induction ( A ) or Ngn3cherry+Mafacherry mediated alpha cell induction ( B ) . Tissues were harvested from cherry+ regions under a dissecting fluorescent microscope . Transgene expression peaked at day 2 , declined by day 10 , and returned to baseline by day 30 . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 006 Among the major islet endocrine cell types , relatively little is known about δ-cell biology and genes important for δ-cell development and function . Among the few δ-cell-specific genes identified are somatostatin and cholecystokinin receptor B ( Cckbr ) ( Morisset et al . , 2000 ) . Our analysis revealed that the majority of induced δ-like cells co-express Sst and Cckbr 30 days after induction ( 87 ±7% by immunohistochemistry , Figure 2A ) . The Sst+ cells also express the endocrine factors Pax6 and synaptophysin ( Figure 2B , C ) . The Sst+ induced δ-cells were present in adult pancreas 2 months after induction ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 01846 . 007Figure 2 . δ-like cell induction by Ngn3 . ( A–C ) Induced δ-cells co-express somatostatin ( SST ) and cholecystokinin receptor B ( Cckbr ) ( A and A′ ) . They also co-express the endocrine markers Pax6 ( B and B′ ) and Synaptophysin ( Syn , C and C′ ) . Scale bar: 50 µm . ( D and E ) Ultrastructure of endogenous and induced δ-cells in electron micrographs . D′ and E′ are magnified view of the boxed areas in D and E , showing the characteristic morphology of δ-cell granules . White arrow indicates a neighboring acinar cell with dense ER ( endoplasmic reticulum ) assembly . Induced δ-cells were found intermingled among acinar cells . In comparison , endogenous δ-cells reside exclusively in islets . ( F and G ) Transcriptional profiling identified 1283 genes enriched in induced δ-cells 30 days after induction . 632 of the induced genes are present in a whole-islet gene signature ( F ) . Many of the top 30 induced δ-cell genes show medium to low expression in whole islet samples , which contain mostly β-cells . β- and α-specific genes , including Ins1 ( insulin1 ) , Ins2 ( insulin2 ) , NKX6 . 1 , and Gcg ( glucagon ) , are absent from the induced δ-cell samples . ( H ) DNA methylation analysis of the proximal promoters of Amylase 2a and Insulin2 genes in acinar cells , islet δ-cells , and induced δ-cells ( 20 days after induction ) . Methylation status of the induced and endogenous δ-cells is similar , indicating appropriate methylation changes during acinar to δ-cell conversion . ( I ) Induced δ-cells released somatostatin in response to the secretagogue Arginine ( 20 mM ) in an in vitro assay . Acinar cells and islets were used as controls . Data were normalized as fold increase over baseline ( no Arginine ) . Quantifications are shown in mean ± SD , n = 3 animals . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 00710 . 7554/eLife . 01846 . 008Figure 2—figure supplement 1 . Induced δ-cells persist in adult pancreas . The induced δ-cells are detectable in adult pancreas 2 month after induction and they continue to express Pax6 . Ecad: E-cadherin . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 00810 . 7554/eLife . 01846 . 009Figure 2—figure supplement 2 . Ultrastructure comparison of induced and endogenous endocrine subtypes . Representative images of endogenous α- , δ- , and β-cells are presented in A , C , E , whereas that of induced α- and δ-cells are presented in B , D . Images in A′–E′ are magnified views of the boxed areas in A–E . Note that α-cell granule has a thin halo around the matrix , the δ-cell granule has no separation between the membrane and matrix , whereas β-cell granule has a large halo that surrounds the dense-core matrix . Arrows in B and D indicate the neighboring acinar cells . Note that A , B are identical to Figure 3E , F whereas Figure 3C , D are identical to Figure 2D , E . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 00910 . 7554/eLife . 01846 . 010Figure 2—figure supplement 3 . Genomic maps of CpG sites in the promoter region of mouse insulin2 ( ins2 ) and amylase2a2 ( Amy2a2 ) genes . The genomic region around the transcriptional start site ( TSS ) is shown ( 1 kb upstream and 1 kb downstream ) . CpG are represented as triangles . CpG analyzed in this study are shown as solid triangles . Black bars represent exons . Mammalian conservation is shown from the USCS genome viewer . Note that the CpGs analyzed fall within short genomic regions , which have been shown to be sufficient to direct cell type-specific expression of Ins2 and Amy2a2 in pancreas . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 01010 . 7554/eLife . 01846 . 011Figure 2—figure supplement 4 . Purification of endogenous δ- and α-cells , and induced δ-cells by intracellular FACS for DNA methylation studies . Endogenous δ- and α-cells were purified by staining wide-type islet cells and intracellular FACS ( A ) , yield 10 . 1% glucagon+ cells and 5 . 5% somatostatin+ cells from islets . Induced δ-cells were isolated by harvesting the acinar fraction of infected pancreatic samples 20 days after infection , followed by intracellular FACS . The acinar fraction contains very few endogenous endocrine cells ( B , first and second panels , showing no-virus control and cherry alone control ) . After induction , 0 . 36% of total isolated cells were somatostatin+ in Ngn3cherry samples ( B , last panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 01110 . 7554/eLife . 01846 . 012Figure 2—figure supplement 5 . Somatostatin promoter DNA methylation analysis . ( A ) The genomic region around the transcriptional start site ( TSS ) of somatostatin gene is shown ( 1 kb upstream and 1 kb downstream ) . CpG are represented as triangles . CpG analyzed in this study are shown as solid triangles . Black bars represent exons . Mammalian conservation is shown from the USCS genome viewer . We analyzed seven CpGs that fall within a stretch of highly conversed promoter region . ( B ) No methylation was detected at the somatostatin promoter in acinar cells , endogenous δ-/α-cells , and induced δ-/α-cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 01210 . 7554/eLife . 01846 . 013Figure 2—figure supplement 6 . COBRA analysis of Amylase promoter . Combined bisulfite restriction analysis ( COBRA ) confirmed that the Amylase 2 promoter is lightly methylated in acinar cells ( middle lane ) , but heavily methylated in both islet δ-cells and induced δ-cells , consistent with sequencing results presented in Figure 2H . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 013 A major method to recognize and distinguish the different islet endocrine subtypes is by ultrastructural analysis . In particular , the secretory granules of each islet subtype have characteristic morphology ( Larsson et al . , 1976; Leiter et al . , 1979 ) . Electron microscopy analysis revealed that the secretory granules of induced δ-cells are spherical or ellipsoidal with matrix filling the entire granule space ( Figure 2E , E′ ) . This morphology is typical of endogenous δ-cells ( Figure 2D , D′ ) and distinct from that of α-or β-granules ( Figure 2—figure supplement 2 ) . In addition , we observed that the induced δ-cells were embedded among acinar cells ( Figure 2E , arrow points to dense assembly of endoplasmic reticulum in a neighboring acinar cell ) , consistent with their origins from acinar cells . In contrast , endogenous δ-cells reside exclusively within islets ( Figure 2D ) . To further characterize the induced δ-cells , we generated gene expression data from FACS purified induced δ-cells ( day 10 after infection ) using mCherry , a fluorescent marker coexpressed with Ngn3 ( Figure 1A ) . Cherry+ cells contain approximately 40% induced δ-cells . Gene profiling with illumina arrays yielded 1283 genes enriched in the induced δ-cells ( 30 days after induction ) relative to acinar cells ( GEO: GSE52522 ) . Because there is currently no method available that allows purification of endogenous δ-cells , we compared their expression profile with that of whole islets , which are comprised largely of β-cells . 632 of the induced genes ( 49% ) overlapped with the islet-enriched gene set that we previously reported ( Figure 2F ) ( Zhou et al . , 2008 ) ( GEO: GSE12025 ) . Given that δ-cells represent a minor fraction of total mouse islet cells ( 5 . 5% as determined by FACS , see Figure 2—figure supplement 4 ) , the non-overlapping genes ( 651 ) may contain δ-cell-enriched factors that are under-represented in the whole-islet samples . Indeed , among the top 30 most highly induced genes in the δ-like cells , many show low expression in whole islet samples ( Figure 2G ) . We detected up-regulation of Hhex ( Figure 2G ) , a gene recently implicated in δ-cell biology ( 72nd ADA abstract , Klaus Kaestner lab ) . In contrast , glucagon , insulin , and the β-cell marker Nkx6 . 1 are abundantly expressed in islets but absent from induced δ-cells ( Figure 2G ) . We analyzed the DNA methylation status of several gene promoters to assess epigenetic changes in the conversion of acinar to δ-like cells . These genes included Somatostatin ( δ-cells ) , Amylase2 ( acinar cells ) , and Insulin2 ( β-cells ) . Studies have shown that insulin2 gene expression is regulated by DNA sequences located within approximately 400 bp upstream of the transcription start site ( TSS ) ( Hay and Docherty , 2006 ) . Similarly , approximately 200 bp of the Amylase 2 promoter is sufficient to direct acinar-specific expression ( Minami et al . , 2005 ) . We therefore assayed CpGs located in these critical promoter regions ( Figure 2—figure supplement 3 ) . There are very limited studies on the Somatostatin promoter so we assayed seven CpGs that fall within a highly conversed promoter region ( Figure 2—figure supplement 5 ) . We adapted an intracellular FACS protocol to purify endogenous and induced δ-cells after staining with somatostatin ( Figure 2—figure supplement 4 ) ( Pechhold et al . , 2009 ) . We note that this protocol allows isolation of genomic DNA but not intact mRNA from the pancreatic tissues . We have not been successful at generating gene-profiling data from the induced δ-cells using the intracellular FACS method . None of the CpG sites assayed in the somatostatin promoters was methylated in all samples tested ( Figure 2—figure supplement 5 ) , indicating that this promoter is not subject to regulation by DNA methylation . Amylase2 , a gene exclusively expressed in pancreatic acinar cells , was largely unmethylated in acinar cells ( Figure 2H ) . In contrast , the induced δ-cells showed strong methylation in this promoter similar to islet δ-cells ( Figure 2H and Figure 2—figure supplement 6 ) . The Insulin2 promoter was fully methylated in acinar cells but partially demethylated in both endogenous and induced δ-cells ( Figure 2H ) . For both amylase and Insulin promoters , the methylation differences of acinar/islet δ-cell and acinar/induced δ-cell are statistically highly significant ( p<0 . 001 ) whereas there is no significant difference between islet δ-cells and induced δ-cells ( p=0 . 23 and 0 . 30 for amylase and insulin promoter respectively ) . These studies suggest that substantial DNA methylation changes occurred during acinar to δ-cell conversion in the promoters we studied . It is notable that not all cell fate conversion events are associated with DNA methylation changes . For example , no significant DNA methylation was observed in the conversion of pre-B cells to macrophages ( Rodriguez-Ubreva et al . , 2012 ) . We evaluated the ability of induced δ-cells to secret hormones in an in vitro secretion assay . The acinar fraction that contains induced δ-cells was isolated 30 days after induction and stimulated with the secretagogue Arginine . The induced δ-cells responded to Arginine and released somatostatin , in a manner similar to isolated islets which contain endogenous δ-cells ( Figure 2I ) . In contrast , control acinar cells did not respond to Arginine stimulation ( Figure 2I ) , consistent with the fact that they do not express endocrine hormones . These data suggest that induced δ-cells possess cellular machineries necessary for hormone production , storage , and release . The data described above collectively indicate that δ-like cells induced by Ngn3 expression in adult pancreas possess key features of endogenous δ-cells . Co-infection of adult mouse pancreas with two different adenoviruses carrying Ngn3- and Mafa-induced formation of glucagon+ cells ( Figure 3A ) . This co-infection also induced somatostatin+ cells , which are distinct from the glucagon+ cells ( Figure 3A′ ) . We tested different ratios of Ngn3/Mafa viruses and observed that a 1:1 ratio yielded the most number of Gcg+ cells ( Figures 1F and 9 ± 5% at 1:1 ratio , and data not shown ) . Due to the random nature of co-infection , cells that received predominately Ngn3 infection likely become the Sst+ cells . In contrast to co-infection by two separate viruses , polycistronic co-expression of Ngn3 and Mafa from a single construct yielded substantially reduced number of glucagon+ cells ( less than 1% , data not shown ) . We therefore used co-infection to induce glucagon+ cells in all subsequent experiments . 10 . 7554/eLife . 01846 . 014Figure 3 . α-like cell induction by Ngn3 and Mafa . ( A ) Co-infection of two separate viruses carrying Ngn3 and Mafa led to the induction of Glucagon ( Gcg+ ) cells . Somatostatin ( Sst+ ) cells were also induced as a separate population ( A′ ) . ( B–D ) Induced Gcg+ cells express α-cell fate regulator Arx ( B and B′ ) and endocrine factors Pax6 ( C and C′ ) and synaptophysin ( Syn , D and D′ ) . Arrows indicate double positive cells . Scale bar: 50 µm . Syn is expressed in both Gcg+ and Sst+ cells . ( E and F ) Electron micrographs of endogenous and induced α-cells . E′ and F′ are magnified view of the boxed areas in E and F , showing the characteristic morphology of α-cell granules . Arrows indicate zymogen granules of a neighboring acinar cell . Endogenous α-cells reside within islets , whereas induced α-cells reside among acinar cells . ( G ) DNA methylation analysis of the proximal promoters of Glucagon , Amylase 2a , and Insulin2 genes in acinar cells , islet α-cells , and induced α-cells ( 20 days after induction ) . Methylation status of the induced and endogenous α-cells is similar , indicating appropriate methylation changes during acinar to α-cell conversion . ( H ) Induced α-cells responded to stimulation by the secretagogue Arginine ( 20 mM ) and released glucagon . Acinar and islets were used as controls . Data were normalized as fold increase over baseline ( no Arginine ) . Quantifications are shown in mean ± s . d . , n = 3 animals . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 01410 . 7554/eLife . 01846 . 015Figure 3—figure supplement 1 . Induced α-cells persist in adult pancreas . The induced α-cells are detectable in adult pancreas 2 month after induction . They are Pax6+ and Ecadherin+ . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 01510 . 7554/eLife . 01846 . 016Figure 3—figure supplement 2 . Purification of induced α-cells by intracellular FACS for DNA methylation studies . Induced α-cells were isolated by harvesting the acinar fraction of infected pancreatic samples 20 days after infection ( co-infection with Ngn3cherry and Mafacherry ) , following by intracellular FACS . The acinar fraction contains very few endogenous α-cells ( first and second panels , showing no-virus control and cherry alone control ) . After induction , 0 . 5% of total isolated cells were glucagon+ in induced samples ( last panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 01610 . 7554/eLife . 01846 . 017Figure 3—figure supplement 3 . Genomic map of CpG sites in the promoter region of mouse glucagon gene . The genomic region around the transcriptional start site ( TSS ) is shown ( 1 kb upstream and 1 kb downstream ) . CpG are represented as triangles . CpG analyzed in this study are shown as solid triangles . Black bars represent exons . Mammalian conservation is shown from the USCS genome viewer . We analyzed nine CpGs that fall within a short stretch of highly conversed promoter region . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 01710 . 7554/eLife . 01846 . 018Figure 3—figure supplement 4 . A small number of Gcg+ cells are partially reprogrammed . A small number of induced Gcg+ α-cells express the acinar factor Amylase 30 days after induction . White arrows indicate properly converted Gcg+Amylase− cells . Yellow arrows indicate partially converted Gcg+Amylase+ cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 018 The most important gene that controls endogenous α-cell development and identity is the transcription factor Arx . Genetic deletion of Arx during embryonic development results in the absence of α-cells , whereas ectopic expression of Arx in β-cells causes their phenotypic drift towards α-cells ( Collombat et al . , 2003; Dhawan et al . , 2011 ) . We observed strong Arx expression in all induced Gcg+ cells , although some Gcg− cells also expressed Arx ( Figures 3B , B′ ) . In addition , all Gcg+ cells expressed the endocrine genes Pax6 ( Figure 3C , C′ ) and synaptophysin ( Figure 3D , D′ ) . Due to the intermingling of induced Gcg+ cells and Sst+ cells , we were unable to purify these cells for detailed transcriptome analysis . The induced α-cells are readily observable 2 month after induction ( Figure 3—figure supplement 1 ) . Electron microscopy analysis of the induced α-cells indicates that they share a similar ultrastructure as the endogenous α-cells ( Figure 3E , F ) . The secretory granules have a narrow halo between the core and membrane ( Figure 3E′ , F′ ) that is typical of α-granules and distinct from δ- and β-granules ( Figure 2—figure supplement 2 ) . The induced α-like cells were intermingled among acinar cells ( Figure 3F , arrow points to zymogene granules of a neighboring acinar cells ) , whereas endogenous α-cells were found exclusively within islets ( Figure 3E ) . In an in vitro hormone secretion assay , the acinar fraction that contains induced α-cells isolated from the adult pancreas 30 days after induction responded to the secretagogue Arginine and released glucagon ( Figure 3H ) , indicating their ability to produce and secrete hormone . We used intracellular FACS to isolate endogenous and induced α-cells ( Figure 2—figure supplement 4 and Figure 3—figure supplement 2 ) . DNA methylation analysis at the Glucagon , Amylase2a , and Insulin2 promoters showed a general similarity of induced α-cells with endogenous α-cells ( Figure 3G , Figure 3—figure supplement 3 ) . Statistical analysis showed no significant difference between endogenous and induced cells at glucagon and insulin promoters ( p=0 . 26 , 0 . 50 , respectively ) . However , a difference was detected at the amylase promoter ( p=0 . 02 ) . Amylase protein expression was absent in the majority of Gcg+ cells ( over 95% ) . However , a small fraction ( <5% of all Gcg+ cells ) were Amylase+ ( Figure 3—figure supplement 4 ) , indicating incomplete reprogramming in a small subset of glucagon+ cells . Taken together , these results indicate that induced α-cells possess key features of endogenous α-cells . We previously reported that the acinar cell is the cell-of-origin for most of the induced β-cells in M3 factor-mediated reprogramming of adult pancreas ( Zhou et al . , 2008 ) . This is in part due to preferential infection of adenovirus for acinar cells but not the other pancreatic cell types ( Zhou et al . , 2008 ) . To test whether the induced δ-and α-cells also derive from acinar cells , we used a genetic lineage tracing strategy similar to the previous study . A Ptf1aCreER mouse line , which drives CreER expression exclusively in acinar cells of the adult pancreas , was crossed with a Rosa-floxed-Stop-YFP ( RosaYFP ) reporter line to create bigenic Ptf1aCreER::RosaYFP animals . Tamoxifen induction resulted in the labeling of 20–30% of mature acinar cells ( Figure 4A′ , B′ ) , consistent with published report of this line ( Pan et al . , 2013 ) . After adenoviral delivery of Ngn3 or Ngn3+Mafa , which targets acinar cells , we confirmed that acinar cells can give rise to both δ-like and α-like cells ( Figure 4A , B ) . The induced endocrine cells are also smaller than acinar cells ( Figure 4A , B , arrows ) , consistent with previous report ( Zhou et al . , 2008 ) . 10 . 7554/eLife . 01846 . 019Figure 4 . Induced δ- and α-like cells are converted from acinar cells in the absence of cell proliferation . ( A–B ) Genetic lineage tracing of induced δ- and α-like cells . Tamoxifen induction of bigenic Ptf1aCreER::RosaYFP animals led to specific and indelible labeling of approximately 20% of adult pancreatic acinar cells . Delivery of Ngn3+Mafa or Ngn3 by adenovirus in the pancreas resulted in formation of Gcg+YFP+Cherry+ ( A–A′′ , arrows ) and Sst+YFP+Cherry+ ( B–B′′ , arrows ) cells , indicating that the induced cells derive from adult acinar cells . Note that both endogenous and induced endocrine cells are smaller than acinar cells . ( C–E ) Continued BrdU labeling during the first 10 days of δ- and α-induction showed that few induced cells incorporated BrdU , indicating a lack of proliferation during this period . Arrows indicate BrdU+ cells . A total of 1000 Sst+ or Gcg+ cells were quantified from three animals . i-δ: induced δ-cells . i-α: induced α-cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 019 Continuous BrdU labeling showed that the conversion of acinar cells to induced δ- and α-cells occurred largely in the absence of cell proliferation ( Figure 4C , D ) . Only about 1% of the induced δ- and α-cells incorporated BrdU in a 10-day reprogramming period ( Figure 4E ) . These studies confirmed that induced δ- and α-cells derive from adult acinar cells and that the acinar conversion occurred largely in the absence of cell proliferation . In principle , the conversion of one cell fate to another involves two major components: suppression of the original cell fate and activation of a new one within the same cell . We examined the ability of the three reprogramming factors to suppress acinar fate-regulators and activate pan-endocrine genes , two key events necessary for establishing an endocrine fate in acinar cells . Ptf1a , Nr5a2 , and Mist1 are key acinar cell-fate regulators . They are expressed in adult acinar; their genetic deletion results in abnormalities of acinar development and function ( Pin et al . , 2001; Kawaguchi et al . , 2002; Lin et al . , 2004; Beres et al . , 2006; Holmstrom et al . , 2011 ) . We observed that Ngn3 and Mafa , but not Pdx1 , strongly suppressed Ptf1a , Mist1 , and Nr5a2 expression 4 days after gene delivery in pancreas ( Figure 5A , B , E , F , I , J , M ) . 10 . 7554/eLife . 01846 . 020Figure 5 . Ngn3 can simultaneously suppress acinar fate-regulators and activate pan-endocrine genes to establish an endocrine state . Immunohistochemistry showed that 4 days after expression of the three reprogramming factors individually in the pancreas , Ngn3 and Mafa , but not Pdx1 , strongly suppressed the expression of the acinar fate-regulators Mist1 , Ptf1a , and Nr5a2 ( A , B , E , F , I , J , M ) . Ngn3 also activated expression of the pan-endocrine genes Pax6 and Islet1 ( C , D , M ) , whereas Mafa and Pdx1 did not ( G , H , K , L , M ) . Ngn3 alone can therefore establish an endocrine state in acinar cells by simultaneous suppression of acinar factors and activation of pan-endocrine genes . Infection with Cherry was used as control ( M ) . Quantifications are shown as mean ± s . d . At least 1000 cherry+ cells counted from three different animals . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 020 Pax6 and Islet1 are endocrine genes expressed in all islet endocrine cells . Genetic studies have established their important role in the development of all islet endocrine subtypes ( Ahlgren et al . , 1997; Sander et al . , 1997; Ashery-Padan et al . , 2004 ) . Ngn3 , but not Mafa or Pdx1 , activated the expression of these pan-endocrine genes at day 4 after pancreas infection ( Figure 5C , D , G , H , K , L , M ) . The data discussed above together indicate that among the three reprogramming factors , Ngn3 alone possesses the ability to initiate two key events in endocrine reprogramming , namely , acinar suppression and pan-endocrine activation , thereby establishing a generic endocrine state in acinar cells . Given the important role played by key acinar factors in maintaining acinar cell fate , we hypothesized that these factors could act as molecular barriers in endocrine conversion . We tested this hypothesis by co-infecting pancreas with two viruses carrying Nr5a2 and Ngn3 . 10 days after the infection , analyses revealed that persistent expression of Nr5a2 strongly blocked induction of Pax6 and Sst ( Figure 6B , E , G ) , compared with Ngn3 alone controls ( Figure 6A , D , G ) . Another acinar factor Ptf1a similarly suppressed Pax6 and Sst induction ( Figure 6C , F , G ) . These results demonstrate that key acinar factors are potent molecular barriers; their down-regulation is a prerequisite for endocrine reprogramming . 10 . 7554/eLife . 01846 . 021Figure 6 . Acinar factors are molecular barriers of endocrine reprogramming . Compared with the robust induction of Pax6 and Sst by Ngn3 alone ( A , D , G ) , co-expression of Nr5a2 and Ngn3 ( by co-infection of two separate viruses ) strongly inhibited the activation of both endocrine genes ( B , E , G ) . A similar suppression was observed when Ngn3 was co-expressed with Ptf1a ( C , F , G ) . Samples were analyzed 10 days after infection . Quantifications are shown as mean ± s . d . At least 1000 cherry+ cells counted from three different animals . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 021 To further understand the temporal sequence of acinar to endocrine subtype conversion , we examined three key events: acinar suppression , pan-endocrine activation , and subtype-specific gene activation . In Ngn3-induced acinar to δ-cell conversion , we observed strong suppression of the acinar factor Mist1 at day 2 after Ngn3 delivery ( Figure 7A ) . The pan-endocrine gene Pax6 was also robustly activated at day 2 ( Figure 7B ) . In contrast , expression of δ-specific genes somatostatin and CCKbr was not detected until day 4 ( Figure 7G , H ) and became strongly expressed at day 10 ( Figure 7K , L ) . A similar temporal sequence of acinar suppression ( Mist1 ) and pan-endocrine activation ( Pax6 ) followed by β-specific gene activation ( Nkx6 . 1 and insulin ) was also observed in acinar to β-cell conversion ( Figure 8 ) . These data collectively suggest that a generic endocrine state was established in acinar cells at the onset of reprogramming , temporally preceding endocrine subtype specification . 10 . 7554/eLife . 01846 . 022Figure 7 . Acinar suppression and pan-endocrine activation precedes subtype-specific gene activation in acinar to δ-cell conversion . In acinar to δ-cell conversion induced by Ngn3 , strong suppression of the acinar factor Mist1 was observed in the Cherry+-infected cells at day 2 ( A ) . The pan-endocrine factor Pax6 was also induced at day 2 ( B ) . The Mist1-Pax6+ state was maintained in the majority of Cherry+ cells at later time points ( E , F , I , J ) . In contrast , δ-subtype specific factors Sst and CCkbr were not induced until day 4 ( G , H ) and became robustly expressed at day 10 ( K , L ) . In control samples expressing cherry alone , the majority of cherry+ acinar cells had Mist1 expression ( M ) , and none had induced endocrine gene expression ( N , O , P ) . Quantifications are shown as mean ± s . d ( Q ) . At least 1000 cherry+ cells counted from three different animals . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 02210 . 7554/eLife . 01846 . 023Figure 8 . Acinar suppression and pan-endocrine activation precedes subtype-specific gene activation in acinar to β-cell conversion . In acinar to β-cell conversion induced by M3 ( Ngn3+Mafa+Pdx1 ) , near complete suppression of the acinar factor Mist1 was observed in the Cherry+-infected cells at day 2 ( A ) . The pan-endocrine factor Pax6 was also robustly induced at day 2 ( B ) . The Mist1−Pax6+ state was maintained in the majority of Cherry+ cells at later time points ( E , F , I , J ) . In contrast , β-subtype specific factors insulin and Nkx6 . 1 were not induced until day 5 ( G , H ) and became more robustly expressed at day 10 ( K , L ) . In control samples expressing cherry alone , the majority of cherry+ acinar cells had Mist1 expression ( M ) , and none had induced endocrine gene expression ( N , O , P ) . Quantifications are shown as mean ± s . d ( Q ) . At least 1000 cherry+ cells counted from three different animals . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 023 The studies discussed above indicate that Ngn3 has two main functions in acinar to endocrine conversion: establishment of an endocrine state , and δ-subtype specification in the absence of other factors . Ngn3 is also part of the reprogramming factor combination in the induction of α-cells ( Ngn3+Mafa ) and β-cells ( Ngn3+Mafa + Pdx1 ) , which raises the question of how a singular α- or β-cell fate is established . We tested the possibility that δ-specification may be suppressed by Pdx1 and/or Mafa . Polycistronic co-expression of Mafa with Ngn3 reduced the efficiency of somatostatin induction from 40 ± 3% in control Ngn3 animals ( Figure 9A ) to 24 ± 7% in Ngn3 . Mafa animals ( Figure 9B , D ) . Further increasing the amount of Mafa in the mixture led to a stronger suppression of δ-cell induction ( data not shown ) . Polycistronic co-expression of Pdx1 with Ngn3 led to near complete suppression of somatostatin+ cells ( 2 ± 1% , Figure 9C , D ) . Immunostaining with pan-endocrine factor synaptophysin revealed that Cherry+somatostatin− cells were synaptophysin+ ( Figure 9C , inset ) , suggesting that Ngn3 converted acinar cells to an endocrine state , but that δ-subtype specification was blocked . These data indicate that Pdx1 and Mafa can suppress δ-subtype specification , which is likely part of the mechanism to ensure formation of distinct α- and β-subtypes upon coexpression of multiple reprogramming factors . 10 . 7554/eLife . 01846 . 024Figure 9 . Pdx1 and Mafa can suppress δ-subtype specification . Compared with robust induction of Sst+ cells by Ngn3 alone ( A ) , polycistronic co-expression of Mafa and Ngn3 led to strong reduction of Sst+ cells ( B ) . Polycistronic co-expression of Pdx1 and Ngn3 nearly completely suppressed Sst+ cell induction ( C ) . The Sst-cherry+ cells expressed synaptophysin ( Syn ) ( C , inset ) , suggesting that these cells acquired an endocrine identity but δ-specification was blocked . Quantifications are shown as mean ± s . d ( D ) . At least 1000 cherry+ cells counted from three different animals . p<0 . 01 , *p<0 . 001 . Mann–Whitney test . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 024 Our studies indicate that pancreatic acinar cells can be directly converted to endocrine δ- and α-like cells by in vivo expression of Ngn3 or Ngn3+Mafa respectively . Together with our previous report of β-cell reprogramming with Ngn3+Mafa+Pdx1 , these studies provide a set of reprogramming models where combinatorial actions of three factors lead to conversion of acinar cells to the three major islet endocrine subtypes in vivo ( Figure 10A ) . 10 . 7554/eLife . 01846 . 025Figure 10 . Direct in vivo conversion of pancreatic acinar cells to three islet endocrine subtypes by combinatorial actions of three factors . ( A ) Summary of acinar to islet endocrine conversion with defined factors . ( B ) Our studies suggest that there are two main processes in pancreatic acinar to endocrine reprogramming . Ngn3 plays a critical role in establishing a generic endocrine state in acinar cells by suppressing acinar fate regulators ( Ptf1a , Nr5a2 , Mist1 ) and activating pan-endocrine factors ( Pax6 , Islet1 , etc ) ( upper panel ) . Down-regulation of acinar regulators is critical as they can block reprogramming . In endocrine subtype-specification ( lower panel ) , Ngn3 promotes δ-fate in the absence of other factors . Mafa and Pdx1 act in concert with Ngn3 to promote α- and β-specification . Both Mafa and Pdx1 can suppress δ-subtype specification , whereas α-specification is also suppressed in β-induction . Combinatorial actions of the three reprogramming factors therefore led to formation of distinct endocrine subtypes . ( C ) A summary table of reprogramming factor functions . Asterisks: combinatorial actions of multiple factors are required to specify α- and β-cells from acinar cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 025 Our studies indicate that acinar to endocrine reprogramming includes two main processes: establishment of a generic endocrine state and endocrine subtype specification ( Figure 10B ) . Ngn3 plays a central role in promoting the endocrine state at the onset of reprogramming by suppressing acinar fate regulators and activating pan-endocrine genes ( Figure 10B , upper panel ) . This early step is critical as continued expression of key acinar factors will block endocrine conversion . Compared with Ngn3 , Mafa suppresses acinar regulators but does not activate pan-endocrine genes , whereas Pdx1 lacks the ability to initiate either of these two events . It is likely that the main function of Mafa and Pdx1 is to collaborate with Ngn3 to activate subtype-specific genes such as Arx in α-induction ( this study ) and Nkx6 . 1 in β-induction ( Zhou et al . , 2008 ) . Another important function of Mafa and Pdx1 is to suppress δ-specification , thus ensuring formation of singular α- and β-subtypes , and preventing hybrid cells ( Figure 10B , lower panel ) . Our data together suggest that each of the three factors possesses both activator and suppressor functions ( Figure 10C ) . It will be important to fully elucidate the molecular details of these functions in future studies . During pancreatic development , numerous studies have highlighted the importance of Ngn3 in controlling endocrine fate specification ( Jensen , 2004 ) . A recent study showed that ectopic expression of Ngn3 in embryonic pancreas suppressed exocrine fate specification ( Qu et al . , 2013 ) . Thus , Ngn3 can suppress exocrine fate and promote endocrine fate during both embryogenesis in progenitor cells and during adulthood in differentiated acinar cells . The function of Pdx1 has also been extensively studied and demonstrated to be critical in early pancreatic fate determination , as well as a later role in beta cell biology ( Murtaugh , 2007 ) . It is not clear , however , exactly what role Pdx1 plays in δ- and α-cell specification in embryogenesis . Mafa is expressed in pancreatic β-cells during development and plays a role in controlling β-cell gene transcription ( Hang and Stein , 2011 ) . The ability of Mafa to induce α-cell formation in collaboration with Ngn3 from acinar is nevertheless not entirely surprising . Its close homologue Mafb is expressed during pancreas development in a subset of endocrine progenitor cells ( Nishimura et al . , 2006 ) . Mafb deletion leads to decreased α- and β-cell numbers , suggesting a role for Maf factor in α-cell formation ( Artner et al . , 2007; Nishimura et al . , 2008 ) . Taken together , there are clear similarities in the function of Ngn3 , Mafa , and Pdx1 in normal endocrine development and endocrine reprogramming from acinar cells . Nevertheless , the epigenetic landscape of pancreatic progenitors and adult acinar cells in which these factors operate is presumably very different . Further , the ‘generic endocrine state’ established by Ngn3 in acinar cells is not equivalent to an endocrine progenitor in pancreas development . For example , expression of pancreatic progenitor genes such as Sox9 and Hnf6 was not detected during the reprogramming process ( data not shown ) . One surprising finding from our study is that Ngn3 alone promotes δ-cell formation from acinar cells . Among the pancreatic endocrine subtypes , relatively little is known about the δ-cells . No genetic factors have been discovered that specify δ-cell fate in pancreas development . The molecular details of how Ngn3 functions to specify δ-subtype remains to be determined . Conversion of acinar cells to α-cells requires both Ngn3 and Mafa . The optimal method to induce Gcg+ α-like cells is by co-expression of Ngn3 and Mafa from two separate viruses at a 1:1 ratio . Varying the ratio of the two viruses or polycistronic co-expression of both factors led to reduced α-cell formation . These results are reminiscent of induced pluripotent reprogramming from fibroblasts where polycistronic co-expression of OSKM factors led to substantially reduced efficiency of iPS formation compared with random infection by OSKM factors ( Carey et al . , 2009; Chang et al . , 2009 ) . This difference has been suggested to result from a need for appropriate reprogramming factor stoichiometry ( Carey et al . , 2011 ) . We speculate that the induction of α-like cells may similarly require optimized Ngn3/Mafa stoichiometry along with other conditions such as specific expression level and dynamics of the factors , which can be met at the more ‘flexible’ co-injection system with each cell expressing variable levels of Ngn3/Mafa , compared with the more ‘fixed’ polycistronic expression system . Future experiments will be required to elucidate mechanisms of α-cell induction . Our studies suggest that a suitable strategy to produce specific subtypes of cells by lineage conversion is to combine factors that confer broad competence with factors that confer subtype specificity . An elegant example was demonstrated in C . elegans , where removal of a chromatin factor and employment of neuron selector genes allows conversion of germ cells to different neuronal subtypes ( Tursun et al . , 2011 ) . The mammalian system is far more complex , but similar principles may well apply . It is hoped that insight from lineage reprogramming studies will lead to informed design and improved technology , thereby helping to unlock the tremendous therapeutic potential of this approach . Genes of interest were first cloned into a shuttle vector containing a 2A-Cherry , then into the pAd/CMV/V5-DEST adenoviral vector ( Invitrogen , Grand Island , NY ) . High titer virus ( 2–10 × 1010 pfu/ml ) was obtained by purification with the Vivapure Adenopack ( Sartorius , Bohemia , NY ) . Viral tittering was performed with direct immuno-staining of inserted genes ( Pdx1 , Mafa , Ngn3 ) 2 days after infection in HEK293A cells . Viral preparations that did not reach at least 2 × 1010 pfu/ml in one round of purification had poor induction efficiency in vivo , and were not used . Rag1−/− animals were obtained from Jackson Labs ( Bar Harbor , ME ) . Adult animals ( 2–3 month ) were injected with 100 µl of purified adenovirus ( typically 1–2 × 109 pfu , dilution with saline of high titer stocks ) directly into the splenic lobe of the dorsal pancreas with a 3/10 cc Insulin Syringe ( Becton Dickinson , East Rutherford , NJ ) . All experiments were performed under approved institutional regulations . Adult mouse pancreata were processed as previously described ( Zhou et al . , 2008 ) . The following primary antibodies were used: goat anti-Ngn3 ( Santa Cruz ) , guinea pig anti-Insulin ( Dako , Carpinteria , CA ) , guinea pig anti-Glucagon ( Linco , Charles , MO ) , rabbit anti-somatostatin ( Dako ) , goat anti-Somatostatin ( Santa Cruz ) , goat anti-Pdx1 ( Santa Cruz , Dallas , TX ) , rabbit anti-mafA ( Bethyl , Montgomery , TX ) , goat anti-Glut2 ( Santa Cruz ) , rabbit anti-synaptophysin ( Abcam , Cambridge , MA ) , rabbit anti-Ptf1a ( BCBC ) , mouse anti-Mist1 ( Santa Cruz ) , rabbit anti-Nkx6 . 1 ( BCBC , Nashville , TN ) , rabbit anti-Sox9 ( Santa Cruz ) , mouse anti-Pax6 ( DSHB , Iowa City , IA ) , and mouse anti-islet ( DSHB ) . Secondary antibodies were obtained from the Jackson Immunoresearch laboratories ( West Grove , PA ) and Life Technologies . Pictures were taken with a Zeiss LSM 510 META confocal microscope . Standard procedures were used to separate adult mouse pancreas into islet fraction and exocrine fraction after intra-ductal perfusion and digestion with liberase ( Roche , Indianapolis , IN ) . After Dithizone staining , the acinar fraction is manually picked to eliminate all visible islets . Islets and acinar clusters are further dissociated into single cells by EGTA treatment . Cherry+ cells were subsequently isolated by fluorescent activated sorting ( FACS ) with FACSaria ( BD Bioscience , San Jose , CA ) . RNA was extracted ( Qiagene RNeasy kit , Germantown , MD ) , cRNA synthesized ( Ambion Amplification kit , Grand Island , NY ) , and genome-wide gene profiling performed with Illumina arrays . Intracytoplasmic staining of pancreatic cells was performed as previously described ( Pechhold et al . , 2009 ) with minor modifications . Cells were fixed with 4% paraformaldehyde in PBS for 5 min on ice , diluted in wash buffer ( WB ) ( 1:10 ) , centrifuged at 250×g for 5 min , and permeabilized with detergent wash buffer ( WB ( d ) ) for 30 min on ice . Primary antibodies and final concentration used for the intracytoplasmic staining are mouse monoclonal anti-Glucagon ( K79bB10 , Sigma St . Louis , MO; 1/1000 ) and Goat anti-Somatostatin ( Santa Cruz; sc-7819; 1/500 ) . All primary antibodies were pre-labeled with Alexa Fluor 594 , Alexa Fluor 488 , or Alexa Fluor 647 , using Zenon antibody labeling kits according to the manufacture’s protocol . Intracellular FACS was carried out with FACSaria ( BD Bioscience ) . Genomic DNA was purified using RecoverAll Total Nucleic Acid Isolation Kit ( Invitrogen ) and treated with EpiTect Bisulfite Kit ( QIAGEN ) according to the manufacture’s protocols . The bisulfited genomic DNA was amplified using a touch-down PCR protocol ( Amylase2a2 and Ins2 promoters , detailed PCR parameters available upon request ) or using a nested PCR protocol ( Sst and Glucagon promoters ) . All PCR reactions were performed using HotStart Taq DNA polymerase ( QIAGEN ) . For nested PCR: 95°C for 15 min followed by 45 cycles of 95°C/30 s , 52°C/30 s , 72°C/1 min , and last elongation at 72°C for 10 min . The final PCR product was purified using MinElute PCR Purification Kit ( Qiagen ) , cloned , and sequenced . The sequences were analyzed using BiQ Analyzer software ( Bock et al . 2005 ) . For COBRA analysis ( Combined bisulfite restriction analysis ) of Amylase 2a promoter , the PCR product was generated and cut with Taqα1 enzyme . All primers used are listed in Table 1 . 10 . 7554/eLife . 01846 . 026Table 1 . PCR primers for DNA methylation assaysDOI: http://dx . doi . org/10 . 7554/eLife . 01846 . 026GenesRound #Primer sequences ( 5’ to 3’ ) ( forward; reverse ) Amylase 2aTouch-down PCRTTTTATTTTTATTTGGAATGGTG; TCATATTAAACCCAACAAAACCInsulin2Touch-down PCRTTTAAGTGGGATATGGAAAGAGAGATA; ACTACAATTTCCAAACACTTCCCTAATAGlucagonNested 1TTATATAATGTGGATGAGTGGG; TCTACCCTTCTACACCAAAATACGlucagonNested 2TTTGTTTGTTTAGATGAATGATT; TCTACCCTTCTACACCAAAATAGlucagonNested 3AAGGGATAAGATTTTTAAATGAGA; TCTACCCTTCTACACCAAAATACGlucagonNested 4AAGGGATAAGATTTTTAAATGAGA; ACTCTCCAAACTATTTAACCTTACASomatostatinNested 1ATTGTTTGGTTTTTGTGGTATG; TCTTCCTTACCTCAAACAACCSomatostatinNested 2TGGGTGTAGGTTTTTTTTTTTT; TCTTCCTTACCTCAAACAACC Ptf1aCreER;RosaYFP double heterozygous animals were generated by mating homozygous Cpa1CreER males with RosaYFP homozygous females ( Jackson lab ) . 2-month-old Cpa1CreER;RosaYFP adults were injected with Tamoxifen at 6 mg per animal every third day for four times to label acinar cells . Exocrine tissues that contain the induced endocrine cells were harvested after pancreas dissociation and islets removal . After washing with Preculture Medium ( 1 and 5 mM glucose in PBS for Sst and Gcg RIA , respectively ) , exocrine tissue ( approximately 20 mg per sample ) and islets ( 20-30 islets per sample ) were incubated in 1 ml pre-warmed Preculture Medium at 37° C for 1 hr . After removing Preculture Medium , the exocrine tissue and islet samples were re-suspended in 500 ml PBS in the absence or presence of 20 mM Arginine and incubated at 37° C for 1 hr . Supernatants were harvested . Somatostatin EURIA kit ( Euro Diagnostica , Malmö , Sweden ) and Glucagon RIA kit ( Millipore , Darmstadt , Germany ) were used to assess Somatostatin and Glucagon content of the incubation medium .	In mammals , the pancreas is responsible for controlling blood sugar by secreting insulin from specialized β-cells . Other cells in the pancreas , called δ-cells and α-cells , secrete other hormones to assist the β-cells . Diabetes is caused when this system breaks down: either the body attacks its own β-cells ( type I diabetes ) , or the body stops responding properly to insulin ( type II ) . Type I diabetes is usually treated with insulin injections , but there is increasing interest in the possibility of replacing the defective β-cells instead . Building on previous work in which a fourth type of pancreatic cell , called an acinar cell , was reprogrammed to become a β-cell , Li et al . have now shown that the same technique can be used to produce α- and δ-cells as well . Just as the reprogrammed β-cells secreted insulin , like real β-cells , the reprogrammed α- and δ-cells also behaved like real α- and δ-cells . The reprogramming technique relies on using a combination of three transcription factors—which are called Ngn3 , Pdx1 and Mafa—to treat the acinar cells from mice . Previously , it was shown that using a combination of all three transcription factors reprogrammed the acinar cells to become β-cells . Now , Li et al . show that the Ngn3 transcription factor on its own appears to suppress certain genes that are usually expressed in acinar cells , and goes on to cause the acinar cells to become δ-cells . However , a combination of Ngn3 and Mafa produces a mixture of α- and δ-cells . The next challenge is to adapt this reprogramming technique to generate different types of hormone secreting cells from human tissue sources in order to explore its therapeutic potential .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2014	In vivo reprogramming of pancreatic acinar cells to three islet endocrine subtypes
Apoptosis is a prominent metazoan cell death form . Yet , mutations in apoptosis regulators cause only minor defects in vertebrate development , suggesting that another developmental cell death mechanism exists . While some non-apoptotic programs have been molecularly characterized , none appear to control developmental cell culling . Linker-cell-type death ( LCD ) is a morphologically conserved non-apoptotic cell death process operating in Caenorhabditis elegans and vertebrate development , and is therefore a compelling candidate process complementing apoptosis . However , the details of LCD execution are not known . Here we delineate a molecular-genetic pathway governing LCD in C . elegans . Redundant activities of antagonistic Wnt signals , a temporal control pathway , and mitogen-activated protein kinase kinase signaling control heat shock factor 1 ( HSF-1 ) , a conserved stress-activated transcription factor . Rather than protecting cells , HSF-1 promotes their demise by activating components of the ubiquitin proteasome system , including the E2 ligase LET-70/UBE2D2 functioning with E3 components CUL-3 , RBX-1 , BTBD-2 , and SIAH-1 . Our studies uncover design similarities between LCD and developmental apoptosis , and provide testable predictions for analyzing LCD in vertebrates . Animal development and homeostasis are carefully tuned to balance cell proliferation and death . Cell death not only counters cell production , but also supports morphogenesis and tissue sculpting , and destroys cells that could be harmful , such as autoreactive cells in the immune system or genetically abnormal cells that may promote tumorigenesis . Developmental and homeostatic cell elimination are not passive processes but rather follow a highly coordinated , genetically encoded program ( Green , 2011 ) . A major goal has been to identify the molecular basis of the programs controlling regulated cell demise in development . One such program , apoptosis , has been studied extensively over the past four decades . Some proteins that promote apoptosis , such as BCL2 and FAS , are mutated in human disease ( Fisher et al . , 1995; Rieux-Laucat et al . , 1995; Tsujimoto et al . , 1985 ) , indicating that apoptosis contributes to normal human physiology . Nonetheless , caspase-dependent apoptosis does not account for many cell death events that take place during normal animal development . For example , in the moth Manduca sexta , salivary gland and body muscle remodeling is apparently caspase-independent and the ultrastructural morphology acquired by dying cells is non-apoptotic ( Haas et al . , 1995 ) . Similarly , mice homozygous for knockout alleles of key apoptotic genes , including caspase-3 , caspase-9 , Apaf-1 , or Bax and Bak , can survive to adulthood ( Honarpour et al . , 2000; Kuida et al . , 1998; Lindsten et al . , 2000 ) , a surprising observation given the prevalence of cell death in murine development . Indeed , nearly half of spinal cord motor neurons generated during vertebrate development are normally deleted , and this process occurs unabated in the absence of caspase-3 or caspase-9 ( Oppenheim et al . , 2001 ) . While caspase-independent non-apoptotic processes may play key roles in developmental cell death , little is known about their molecular underpinnings . To date , none of the non-apoptotic cell death pathways that have been described have a role in normal development ( Zhou and Yuan , 2014 ) . The Caenorhabditis elegans linker cell provides direct evidence that caspase-independent non-apoptotic cell death pathways operate during animal development . This male-specific gonadal leader cell guides the elongation of the gonad and vas deferens during development , and then dies near the cloaca , presumably to facilitate fusion of the vas deferens with the cloacal sperm-exit channel ( Kimble and Hirsh , 1979 ) . Linker cell death still occurs in the absence of the main apoptotic caspase , CED-3 , and even in animals lacking all four C . elegans caspase-related genes ( Abraham et al . , 2007; Denning et al . , 2013 ) . Other canonical apoptosis genes are also not required , nor are genes implicated in autophagy or necrosis ( Abraham et al . , 2007 ) . Consistent with these genetic observations , the morphology of a dying linker cell , characterized by lack of chromatin condensation , a crenellated nucleus , and swelling of cytoplasmic organelles , differs from the morphology of apoptotic cells ( Abraham et al . , 2007 ) . Intriguingly , cell death with similar features ( linker cell-type death [LCD]; Blum et al . , 2012 ) has been documented in a number of developmental settings in vertebrates ( Pilar and Landmesser , 1976 ) and is characteristic of neuronal degeneration in patients with and mouse models of polyglutamine disease ( Friedman et al . , 2007 ) . A molecular understanding of LCD is necessary to determine the prevalence and importance of this process in development . Genetic studies of C . elegans linker cell death have identified genes that promote this process , including pqn-41 , which encodes a glutamine-rich protein of unknown function , and tir-1/TIR-domain and sek-1/MAPKK , which may function in the same pathway ( Blum et al . , 2012 ) . Intriguingly , the Drosophila and vertebrate homologs of TIR-1 promote distal axon degeneration following axotomy ( Osterloh et al . , 2012 ) , supporting a conserved role for this protein in cell and process culling . The let-7 microRNA and its indirect target , the Zn-finger transcription factor LIN-29 , also promote LCD , and may act early in the process ( Abraham et al . , 2007; Blum et al . , 2012 ) . Nonetheless , the molecular logic of LCD is not understood . Here , we describe a molecular-genetic framework governing LCD in C . elegans . Our studies represent the first such framework for a non-apoptotic cell death program regulating developmental physiology . We demonstrate that LCD is controlled by two Wnt signals , one pro-death and one pro-survival , that function in parallel , and partially redundantly with the LIN-29 , and SEK-1/MAPKK pathways to control non-canonical activity of HSF-1 , a conserved transcription factor that mediates heat-shock and other stress responses . Our functional , genetic , and molecular studies demonstrate that HSF-1 adopts a specific role , distinct from its well-described protective role in the heat-shock response , to promote LCD . We show that let-70 , encoding a conserved E2 ubiquitin-conjugating enzyme , is an important transcriptional target of this pro-death developmental activity of HSF-1 , but not of the HSF-1 stress-response function . LET-70 expression , as well as expression of ubiquitin and some proteasome components , increases just before LCD onset , and this increase requires the Wnt , LIN-29 , SEK-1/MAPKK pathways , and HSF-1 . CUL-3/cullin , RBX-1 , BTBD-2 , and SIAH-1 E3-ubiquitin ligase components function in the same pathway as LET-70 and promote LCD . Our studies reveal design similarities between LCD and apoptosis . In C . elegans , cell lineage specifies the initiation of developmental apoptosis by transcriptional induction of the egl-1 gene ( Thellmann et al . , 2003 ) , encoding a pro-apoptotic BH3-only protein , or the ced-3 gene , encoding the key executioner caspase ( Maurer et al . , 2007 ) . Pathways linking cell lineage specification to transcriptional initiation of apoptosis have been described for some cells and appear to consist of multiple coordinated inputs . Thus , in both LCD and apoptosis diverse signals control specific transcriptional inputs that , in turn , control protein degradation machinery . The molecular conservation of all the elements comprising the LCD program , together with the characteristic cell death ultrastructure , suggest that this program may be broadly conserved and provide an opportunity for probing the process in other settings . To determine how LCD is initiated , we noted that mutations in the gene him-4 , encoding the secreted protein hemicentin , prevent posterior migration of the linker cell , and result in low-level ( ~15% ) linker cell survival ( Abraham et al . , 2007 ) . Thus , linker cell position might , in part , dictate cell death onset . We considered the possibility that secreted ligands of the Wnt pathway , which are expressed in restricted spatial domains in C . elegans , contribute to LCD . We examined animals carrying lesions in each of the five C . elegans Wnt genes and found that in egl-20/Wnt mutants , the linker cell survives inappropriately ( Figure 1A , B ) , and surviving cells are not engulfed ( Figure 1—figure supplement 1 ) . Importantly , linker cell migration , a complex multi-step process dependent on many genes ( Schwarz et al . , 2012 ) , is unaffected in egl-20 single mutants . Likewise , expression of reporter genes , including lag-2 promoter::GFP ( Figure 1B , Figure 1—figure supplement 1A , B ) , appears unaffected . Thus , egl-20 mutations do not generally perturb linker cell fate , suggesting that the gene has a specific role in LCD control . 10 . 7554/eLife . 12821 . 003Figure 1 . An egl-20/Wnt pathway promotes Llnker cell death . ( A ) Linker cell survival in indicated genotypes . Strains contain qIs56[lag-2p::GFP] linker cell reporter transgene and him-5 ( e1490 ) for males . gsk-3 ( nr2047 ) is linked to unc-101 ( sy216 ) . p<10–4 , no . animals scored is inside bars . ( B ) Adult egl-20 ( n585 ) male expressing lag-2p::GFP . ( C ) Linker cell survival in mig-5 ( rh147 ) animals with indicated transgenes . p<10–4 , *p< . 002 . ( D ) bar-1 ( ga80 ) rescue with hsp-16 . 2p::ΔN-BAR-1 . p<10–4 . ( E ) egl-20p::EGL-20::GFP expression in L4 male . ( F ) bar-1p::GFP expression in L4 male . In ( B ) , ( E ) , ( F ) , white caret , linker cell; arrow , Ul/r . p cells; scale bar , 10 μm . ( G ) EM of surviving linker cell in bar-1 ( ga80 ) adult . Arrow , mitochondria . Arrowheads , nuclear envelope . Carets , healthy ER . Scale bar , 1 μm . ( H ) Linker cell survival in indicated genotypes . p<10–4 from the single mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 00310 . 7554/eLife . 12821 . 004Figure 1—figure supplement 1 . Surviving linker cells in egl-20 mutants are not engulfed , but dying ones are . ( A , B ) 2h-old egl-20 ( n585 ) adult male with a surviving linker cell . lag-2p::GFP marks the linker cell ( white carets ) . lin-48::mCherry marks the U . l/rp cells ( arrowheads ) . Note that in ( A ) , the U . l/rp cells abut the surviving linker cell without surrounding it completely , whereas in ( B ) , the U . l/rp cells have entirely engulfed the linker cell . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 00410 . 7554/eLife . 12821 . 005Figure 1—figure supplement 2 . Expression of receptive Wnt components in the linker cell . ( A , B ) Shown are typical L4 males harboring reporters for ( A ) mig-5; ( B ) lin-17 . White carets and dashed circles , linker cell . Scale bars , 10 μm . ( C ) Expression of lin-44p::GFP reporter in an L4 male . Intestinal expression is an artifact of the vector . Dashed circle , linker cell . Scale bars , 10 μm . ( D ) Expression of wrm-1p::GFP in an L4 male . Dashed circle , linker cell . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 005 To determine whether EGL-20 promotes LCD in the context of Wnt signaling , we examined mutants defective in other pathway components . Animals carrying mutations in the mig-14/Wntless gene , which is required for Wnt secretion , also exhibit surviving linker cells at the cloaca ( Figure 1A ) . Similarly , mig-5/Dishevelled and bar-1/β-catenin mutants , as well as lin-17/Frizzled; mom-5/Frizzled double mutants , exhibit linker cell survival without defects in migration or reporter expression ( Figure 1A ) . Other Wnt mutants or mutant combinations do not block LCD ( Supplementary file 1A ) . The kinase GSK3β curtails Wnt signaling by promoting degradation of β-catenin , and a gsk-3 mutation restores LCD to egl-20 mutants ( Figure 1A ) . Furthermore , a heat-shock-inducible promoter driving a cDNA encoding a stabilized N-terminally-truncated BAR-1/β-catenin protein ( hsp-16 . 2 promoter::ΔN-bar-1 ) displays heat-shock-dependent restoration of LCD not only to bar-1/β-catenin mutants , but also to mig-5/Dishevelled mutants ( Figure 1C , D ) . These data support involvement of a canonical Wnt pathway in promoting LCD . Cells surrounding the hermaphrodite cloaca have been previously shown to express EGL-20 ( Whangbo and Kenyon , 1999 ) . These cells , including the U . l/rp cells that engulf the linker cell , but not the linker cell , also express EGL-20 in males at the time of LCD ( Figure 1E ) , consistent with a specific role in LCD . To determine whether receptive Wnt components function in the linker cell to promote its demise , we examined their expression patterns . An 11-kb regulatory region upstream of the bar-1/β-catenin gene fused to GFP is not expressed in cloacal cells or in the trailing gonad but is strongly expressed in the linker cell ( Figure 1F ) . Likewise , mig-5/Dishevelled::GFP and lin-17/Frizzled::GFP reporters are expressed in the linker cell ( Figure 1—figure supplement 2A , B ) . Consistent with these data , expression of a mig-5/Dishevelled cDNA using a linker-cell-specific mig-24 promoter ( Tamai and Nishiwaki , 2007 ) restores cell death to mig-5 mutant males ( Figure 1C ) , indicating a cell-autonomous role for this gene . To examine when Wnt signaling is required for LCD , we heat shocked bar-1/β-catenin mutants carrying a heat-inducible hsp-16 . 2 promoter::ΔN-bar-1 transgene at different time points during larval development , and assessed restoration of cell death . We found that induction as late as the early L4 stage rescued inappropriate linker cell survival ( Figure 1D ) , suggesting that bar-1 activity just before cell death onset is likely sufficient to drive cell death . This observation also supports the notion that EGL-20/Wnt signaling specifically controls LCD and not identity . Unlike surviving cells in pqn-41 or sek-1 mutants , in which organelle changes accompanying cell death are evident ( Blum et al . , 2012 ) , surviving linker cells in bar-1/β-catenin mutants do not exhibit death-associated ultrastructural features ( Abraham et al . , 2007 ) ( Figure 1G ) , supporting a role for the Wnt pathway in cell death initiation . Taken together , our data suggest that the linker cell responds to an EGL-20/Wnt signal emanating from surrounding cells just prior to its death , using redundant activities of the receptors LIN-17 and MOM-5 and the signal transduction components MIG-5/Dishevelled and BAR-1/β-catenin . Unexpectedly , mutations in pop-1 , the sole C . elegans homolog of the canonical Wnt signaling transcription factor Tcf , cause no or weak defects in LCD ( Figure 2—figure supplement 1A ) . Furthermore , while RNAi against pop-1/Tcf promotes highly penetrant defects in other contexts in C . elegans ( Siegfried and Kimble , 2002 ) , only low-level linker cell survival is evident even in RNAi-sensitized backgrounds ( Figure 2—figure supplement 1A ) . pop-1/Tcf lesions also do not enhance or suppress linker cell survival in egl-20/Wnt mutants ( Figures 2A , Figure 2—figure supplement 1A ) , and a pop-1/Tcf activity reporter is not expressed in the linker cell before or during death ( Figure 2—figure supplement 1B–D ) . Likewise , while BAR-1/β-catenin physically and functionally interacts with the transcription factor DAF-16/FOXO ( Essers et al . , 2005 ) , we found that a daf-16 mutation does not block LCD ( Supplementary file 1A ) . 10 . 7554/eLife . 12821 . 006Figure 2 . A lin-44/Wnt pathway promotes linker cell survival . ( A ) Linker cell survival in indicated genotypes . In ( A-C ) strains also contain qIs56 and him-5 ( e1490 ) . p<10–3; *p <10–4; ns , not significant; Fisher’s exact test . lit-1 ( t512 ) is linked to unc-32 ( e189 ) . ( B ) Linker cell survival in egl-20 ( n585 ) and mig-1 ( e1787 ) ; egl-20 ( n585 ) animals harboring a mig-24p::mig-1 transgene . p<0 . 001 . ( C ) Linker cell survival in indicated genotypes . ns , not significant; Fisher’s exact test . ( D ) Model for Wnt pathway interactions in LCD . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 00610 . 7554/eLife . 12821 . 007Figure 2—figure supplement 1 . pop-1 does not play a significant role in linker cell death . ( A ) Linker cell survival in 0-2h adults of the indicated genotypes . All strains also contain the qIs56 reporter transgene to visualize the linker cell and him-5 ( e1490 ) to increase the incidence of males . pop-1 ( RNAi ) performed with RNAi-sensitizing rrf-3 ( pk1426 ) allele . ( B-D ) All panels are images of strain unc-119 ( ed4 ) ; him-5 ( e1490 ) ; syIs187[POPTOP::HIS-24-mCherry] . Linker cell outlined in dashed white . ( B ) Late L3/early L4 male . ( C ) Mid-L4 male . Note the already-apparent linker cell cytoplasmic changes in the DIC image . ( D ) Late L4 male . mCherry-staining nucleus at the top right of the linker cell in ( D ) , belongs to a neighboring overlying cell . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 007 While testing genetic interactions between egl-20/Wnt mutants and other C . elegans Wnt mutants , we found , surprisingly , that mutations in lin-44/Wnt strongly suppressed inappropriate linker cell survival in egl-20 mutants ( Figure 2A ) . These data suggest that two opposing Wnt pathways control LCD: an EGL-20/Wnt pathway promotes , and a LIN-44/Wnt pathway prevents cell death . To test this idea , we examined genetic interactions between EGL-20/Wnt pathway components and other related genes . LCD is also restored to egl-20/Wnt mutants by mutations in mig-1/Frizzled , cfz-2/Frizzled , lit-1/NLK or wrm-1/β-catenin . lin-44/Wnt mutations also suppress inappropriate linker cell survival in bar-1 mutants ( Figure 2A ) . lin-44 is expressed in the C . elegans male tail ( Figure 1—figure supplement 2C ) ( Herman et al . , 1995 ) , consistent with a role in LCD . wrm-1/β-catenin is expressed in the linker cell , as well as other cells ( Figure 1—figure supplement 2D ) . Furthermore , expression of a mig-1/Frizzled cDNA specifically in the linker cell restores inappropriate linker cell survival to mig-1/Frizzled; egl-20/Wnt double mutants ( Figure 2B ) . These results suggest that a tail-derived LIN-44/Wnt signal impinges on the MIG-1/Frizzled and CFZ-2/Frizzled receptors . These receptors function together in the linker cell , through lit-1/NLK and wrm-1/β-catenin , to promote its survival ( Figure 2D ) . While we were unable to score wrm-1; bar-1 double mutants , as these have a fully penetrant , early block in linker cell migration ( 100% , n>100 ) as well as other defects in larval development , our results suggest that the EGL-20/Wnt pathway antagonizes the LIN-44/Wnt pathway at or downstream of WRM-1/β-catenin . Null alleles of egl-20/Wnt block LCD in about 60% of animals ( Figure 1A ) , and early expression of ΔN-BAR-1/β-catenin fails to promote premature onset of LCD ( Figure 1D ) , suggesting that additional cues promote LCD initiation . The linker cell dies at a specific place and time during C . elegans male development , and previous studies showed that a developmental timing cue transduced by the Zn-finger transcription factor LIN-29 partially controls LCD ( Abraham et al . , 2007 ) ( Figure 1H ) . In lin-29; egl-20/Wnt and lin-29; bar-1/β-catenin double mutants , nearly all linker cells survive inappropriately ( Figure 1H ) , suggesting that the LIN-29 timing cue and the EGL-20/Wnt cue function in parallel to control LCD initiation . We previously showed that the MAPKK gene sek-1 also functions in parallel to lin-29 ( Blum et al . , 2012 ) . In egl-20; sek-1 double mutants , nearly all linker cells also survive ( Figure 1H ) . Furthermore , although lin-44/Wnt mutations suppress ectopic linker cell survival in egl-20/Wnt and bar-1/β-catenin mutants , they do not restore LCD to lin-29 , sek-1 , or pqn-41 mutants ( Figure 2C ) . Thus , EGL-20/Wnt and LIN-44/Wnt , LIN-29 , and SEK-1/MAPKK define three parallel , partially redundant pathways initiating LCD . Heat-shock factors are transcriptional regulators , activated in response to certain stresses including heat shock , whose targets include chaperones and other effectors that maintain cell viability during stress . While exploring roles for stress response genes in LCD , we found that a hypomorphic allele ( sy441 ) of the single C . elegans heat-shock factor gene , hsf-1 , causes inappropriate linker cell survival ( Figure 3A ) . The hsf-1 ( sy441 ) allele is a loss-of-function allele that truncates the region encoding the HSF-1 transcriptional transactivation domain ( Hajdu-Cronin et al . , 2004 ) . This observation suggests , surprisingly , that rather than protecting the linker cell , HSF-1 promotes its demise . Indeed , either a single copy hsf-1 promoter::hsf-1::GFP transgene , expressed at roughly the same level as the native hsf-1 locus ( Morton and Lamitina , 2013 ) , or a wild-type hsf-1 cDNA expressed specifically in the linker cell using the mig-24 promoter , rescue the hsf-1 ( sy441 ) LCD defect ( Figure 3A ) , showing that hsf-1 can function cell-autonomously to promote LCD . 10 . 7554/eLife . 12821 . 008Figure 3 . HSF-1 promotes linker cell death . ( A ) Linker cell survival in indicated genotypes . In ( A-C ) , strains also contain qIs56 and him-5 ( e1490 ) . p<10–2;p<10–3; Fisher’s exact test . hsf-1p::hsf-1 ( WT/R145A ) transgenes are fused to GFP . WT: animals raised at indicated temperature after hatching . +HS: WT animals heat shocked at 37°C for 15 min at 6 hr or 4 hr before the L4-adult molt . hsf-1 ( sy441 ) : mig-24p::hsf-1 bar is average of three independent extrachromosomal array lines . hsf-1p::hsf-1 ( R145A ) bar is average of two independent single-copy integrated lines . hsf-1p::hsf-1 ( R145A ) : animals were raised at the indicated temperature after hatching . ( B ) Linker cell survival in indicated genotypes . HSF-1 ( R145 ) , hsf-1p::hsf-1 ( R145A ) . The drSi28[hsf-1p::hsf-1 ( R145A ) ] transgene was used . For hsf-1p::hsf-1 ( R145A ) ; bar-1 ( ga80 ) , two other independent single-copy integrated lines gave similar results . ( C ) Linker cell survival in indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 00810 . 7554/eLife . 12821 . 009Figure 3—figure supplement 1 . Markers of the heat-shock response are not induced during linker cell death . ( A-D ) Shown are L4 males harboring reporters for ( A ) hsp-1; ( B ) hsp-16 . 2; ( C ) hsp-16 . 41; ( D ) hsp-4 . At least 20 animals were examined for each reporter . hsp-4 is not a typical heat-shock hsf-1 target but harbors cryptic heat-shock elements in its proximal promoter . ( E ) DIC ( left ) and fluorescence ( right ) images of an L4 male treated with NaN3 to induce HSF-1 nuclear stress granules . Dashed square magnified 1 . 5x in inset . White carets , LC . White arrowheads in inset , nuclear stress granules in the LC . Black arrowhead , stress granule in another cell . Scale bars , 10 μm . ( F ) Same as ( E ) except animal treated with tetramisole , which does not induce HSF-1 granules . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 009 Compromised stress responses do not generally block LCD , as neither unfolded protein response mutants ( Blum et al . , 2012 ) , nor daf-16/FOXO or daf-21/HSP90 mutants ( Supplementary file 1A ) show LCD defects ( Blum et al . , 2012 ) . This raises the possibility that the role of HSF-1 in LCD may be different from its role in the canonical heat-shock response . To test this idea directly , we examined expression of GFP reporters for HSF-1 target genes normally induced during heat shock and found that they are not induced in the linker cell during LCD ( Figure 3—figure supplement 1A–D ) . Supporting this conclusion , the hsp-16 . 2 gene is a target of HSF-1 in the heat-shock response , and LCD is restored to bar-1 mutants by the hsp-16 . 2 promoter::ΔN-bar-1 transgene following a heat shock . However , no rescue is evident without heat exposure ( Figure 1D ) , suggesting that the hsp-16 . 2 promoter is not normally induced during LCD . Similarly , while nuclear-cytoplasmic shuttling does not control HSF-1 activity in C . elegans ( Morton and Lamitina , 2013 ) , HSF-1 does form nuclear aggregates in all cells following stress exposure ( Morton and Lamitina , 2013 ) . While aggregates can be seen in dying linker cells in stressed animals ( Figure 3—figure supplement 1E ) , no aggregates are evident in the dying linker cell , or the surrounding cells , of unstressed animals ( Figure 3—figure supplement 1F ) , supporting a novel role for HSF-1 in LCD . In addition to functional differences between the role of HSF-1 in the heat-shock response and LCD , we also found distinct genetic requirements . The HSF-1 ( R145A ) protein contains a mutation in the putative HSF-1 DNA binding domain . Previous studies demonstrated that expression of this protein restores the heat-shock response to hsf-1 ( sy441 ) mutants lacking the distal portion of the HSF-1 transactivation domain ( Morton and Lamitina , 2013 ) . Trans-complementation in the active HSF-1 trimer likely explains how these two loss-of-function lesions can , together , promote a normal heat-shock response . However , instead of rescuing the LCD defect of hsf-1 ( sy441 ) mutants , we found that a single copy hsf-1 ( R145A ) transgene enhanced inappropriate linker cell survival from 29% to 61% ( Figure 3A ) . Taken together , our results show that the role of HSF-1 in LCD is functionally and genetically distinct from its role in the heat-shock response . A prediction arising from these data is that the LCD and heat-shock functions of HSF-1 might compete with each other . To test this , we first observed that while the linker cell of wild-type males raised at 20°C always dies , some wild-type adult males raised at 25°C harbor a surviving linker cell ( Figure 3A ) , suggesting that the heat-shock role of HSF-1 might compete with its LCD role . To test this more directly , we subjected males to a 37°C heat shock 4 hr prior to LCD onset and found that these animals also exhibit a surviving linker cell . Importantly , males heat-shocked 6 hr before LCD onset exhibit fewer surviving linker cells ( Figure 3A ) . These results are consistent with the idea that heat-shock functionally sequesters HSF-1 away from its LCD role , and that activity of HSF-1 just before the cell dies is required to promote LCD . These results also explain why we were able to rescue bar-1 mutants with the hsp-16 . 2 promoter::ΔN-bar-1 transgene , as heat shock was performed 10 hr before LCD onset , well before the activity of HSF-1 is required . An examination of males carrying the hsf-1 ( R145A ) transgene in an otherwise wild-type background revealed that LCD progressed to completion in all animals even at 25°C ( Figure 3A ) . This result suggests that in a wild-type hsf-1 background , hsf-1 ( R145A ) functions as a gain-of-function allele , promoting LCD . One possibility for how this might occur is that the allele preferentially disrupts HSF-1 complexes promoting the heat-shock response , thereby promoting LCD . Regardless of the precise mode of action , our serendipitous discovery of the gain-of-function nature of the R145A protein allowed us to dissect the functional relationships between HSF-1 and the parallel pathways controlling LCD onset . Strikingly , we found that three independent hsf-1 ( R145A ) single-copy transgene isolates restored LCD to egl-20/Wnt and bar-1/β-catenin mutants ( Figure 3B ) . Importantly , a lin-44/Wnt mutation could not restore cell death to hsf-1 ( sy441 ) animals ( Figure 2C ) . Likewise , hsf-1 ( R145A ) transgenes also restored LCD to lin-29 , sek-1/MAPKK , and pqn-41/Q-rich mutants ( Figure 3B ) . These results suggest that the Wnt , LIN-29 , and SEK-1/PQN-41 pathways all require HSF-1 function to promote LCD . Consistent with these observations , we found a synergistic increase in linker cell survival in animals carrying mutations in egl-20 , lin-29 , sek-1 , or pqn-41 and the hsf-1 ( sy441 ) partial loss-of-function mutation ( Figure 3C ) , as might be predicted if HSF-1 functions downstream of all LCD initiation pathways we described . To understand the mechanism by which HSF-1 promotes LCD , we sought genes that function downstream . We previously performed a genome-wide RNA interference ( RNAi ) screen identifying genes required for LCD ( described in Blum et al . , 2012 ) . From this screen , we found that males fed bacteria expressing dsRNA targeted against the gene let-70 , encoding a putative E2 ubiquitin-conjugating enzyme , exhibit robust linker cell survival , indicating that the gene is required for LCD ( Figure 4A-C , 4E ) . As RNAi can induce off-target effects , we confirmed our results by examining two non-overlapping RNAi targeting fragments and obtained similar results ( Figure 4A , E ) . let-70 ( RNAi ) animals exhibit surviving linker cells with normal ultrastructure ( Figure 4C ) , indicating that let-70 likely acts in promoting LCD and not in corpse degradation . Consistent with this observation , surviving linker cells are unengulfed ( Figure 4—figure supplement 1A ) . 10 . 7554/eLife . 12821 . 010Figure 4 . let-70 promotes linker cell death . ( A ) let-70 gene structure and mutations/RNAi clones used in our studies . Black boxes , exons; white boxes , 5’ or 3’ untranslated regions . Scale bar , 200 bp . ( B ) Combined DIC and fluorescent images of let-70 ( RNAi ) adult male . lag-2p::GFP marks the linker cell . Arrow , linker cell . White line , cloaca . Scale bar , 10 μm . ( C ) EM of surviving let-70 ( RNAi ) linker cell . Scale bar , 2 μm . Asterisk , mitochondria , Arrowhead , ER , N , nucleus , nl , nucleolus . ( D ) Purified 6xHis-LET-70 , Drosophila UBA1 , DIAP1 and ubiquitin . causes DIAP1 auto-ubiquitination . ( E-H ) Linker cell survival in indicated genotypes . No . animals scored , inside bars . Error bars , SEM . p<0 . 001; *p<0 . 0001; Fisher’s Exact Test; ns , not significant . Animals contained qIs56 and him-5 ( e1490 ) . In ( F ) animals also contained rrf-3 ( pk1426 ) for increased RNAi efficiency . In LC-only experiments , mig-24p was used to drive rde-1 cDNA in rde-1 ( ne219 ) ; him-8 ( e1489 ) ; qIs56 mutants . ND , not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 01010 . 7554/eLife . 12821 . 011Figure 4—figure supplement 1 . let-70 ( RNAi ) animals have unengulfed linker cells . ( A ) Surviving let-70 ( RNAi ) linker cell fail is not engulfed . Scalebar , 5 μm . Arrow/red cell , linker cell . Green cells , engulfing U . l/rp cells . ( B ) Cell-specific linker cell RNAi by restoring rde-1 expression in rde-1 mutants only to the linker cell using a mig-24 promoter::rde-1 cDNA transgene . lag-2::GFP is used to mark the linker cell . Top left: GFP is expression in animals with RNAi against GFP without rde-1 rescue in the linker cell . Bottom left: GFP expression is reduced in animals with rde-1 rescued in the linker cell subjected to GFP RNAi . Right: Quantification of fluorescence intensity . n=16 for each genotype . Error bars , SD . p<0 . 0001 , Student’s t-test . Scalebar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 01110 . 7554/eLife . 12821 . 012Figure 4—figure supplement 2 . Expression of siah-1 , rbx-1 , and cul-3 in migrating and dying linker cells . ( A ) siah-1p::GFP . Left , merged DIC/fluorescent image of a dying linker cell at the cloaca; middle , fluorescent image of the linker cell at left; right , merged DIC/fluoresent image of a migrating linker cell . Scalebar , 10 μm . Arrow , linker cell . ( B ) Same as ( A ) except rbx-1p::GFP . ( C ) Same as ( A ) except cul-3p::GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 012 To confirm the let-70 RNAi results , we sought animals carrying inactivating mutations in the gene . Animals homozygous for a previously isolated allele , tm2777 , or a CRISPR/Cas9-induced deletion we generated , ns636 , exhibit larval lethality and adult sterility as previously reported for the s689 allele ( Zhen et al . , 1993 ) , precluding studies of LCD . However , ns770 , a CRISPR/Cas9-induced C-to-T point mutation we made that is predicted to generate a P61S alteration in the LET-70 protein , is viable ( Figure 4A ) . A similar lesion confers instability to the S . cerevisiae UBC4 E2 enzyme at 39°C ( Tongaonkar et al . , 1999 ) , suggesting that ns770 may be a partial loss-of-function allele . Indeed , we found that 57% of let-70 ( ns770 ) animals possess surviving linker cells ( Figure 4E ) . This defect is not temperature dependent in the growth range of C . elegans ( 15°C: 62% , n=78; 25°C: 57% , n=87 ) , perhaps because these temperatures are much lower than those abolishing function in yeast . To determine where LET-70 acts to promote death , we generated let-70 ( ns770 ) animals carrying a mig-24 promoter::let-70 cDNA construct expressed specifically in the linker cell . As shown in Figure 4E , LCD is restored in these animals , suggesting that let-70 acts within the linker cell to promote its demise ( Figure 4E ) . To confirm this idea , we carried out linker-cell-specific let-70 RNAi . RDE-1 is an argonaute protein required for RNAi , and in rde-1 ( ne219 ) ; mig-24 promoter::rde-1 cDNA animals , RNAi is only functional in the linker cell ( Figure 4—figure supplement 1B ) . RNAi against let-70 in these animals also prevents LCD ( Figure 4E ) . We conclude that LET-70 acts cell autonomously to kill the linker cell . let-70 is predicted to encode a protein 94% identical to the mammalian E2 ubiquitin-conjugating enzyme UBE2D2 ( Zhen et al . , 1993 ) . To confirm that LET-70 indeed functions as an E2 enzyme , we assayed its ability to mediate ubiquitin transfer in an in vitro ubiquitylation assay . Incubation of LET-70 with the Drosophila UBA1 E1-activating enzyme and the Drosophila E3 ligase DIAP1 results in DIAP autoubiquitylation in the presence of ATP , magnesium ions , and ubiquitin ( Figure 4D ) . A similar reaction without LET-70 does not yield DIAP ubiquitylation , suggesting that LET-70 functions as an E2 . To determine whether LET-70 functions as an E2 enzyme in vivo , we first examined let-70 ( ns770 ) animals expressing a mig-24 promoter::let-70 ( C85S ) cDNA transgene predicted to change the LET-70 catalytic cysteine 85 to serine . However , transgenic animals exhibited hermaphrodite sterility ( presumably due to expression in the hermaphrodite distal tip cell required for gonad development ) , suggesting that LET-70 ( C85S ) is a dominant negative protein . However , in another set of experiments , we found that while a silently-mutated RNAi-resistant let-70 cDNA partially rescues the LCD defect of let-70 ( RNAi ) animals , a similar cDNA encoding the C85S mutation does not ( Figure 4E ) . Taken together , our studies suggest that the ubiquitin-conjugating activity of LET-70 is required in vivo for LCD . To determine whether other E2 enzymes are also required for LCD , we performed RNAi against 13/22 E2-encoding genes with available dsRNA bacterial clones and found no evidence of inappropriate linker cell survival , indicating that LET-70 likely acts specifically to promote LCD ( Supplementary file 1B ) . To determine whether LET-70 functions as part of the ubiquitin proteasome system ( UPS ) for LCD , we first tested if UBA-1 , the sole E1 activating enzyme in C . elegans , is also required . While systemic RNAi against uba-1 is early-larval lethal , linker-cell-specific RNAi against uba-1 produces a robust defect in LCD ( Figure 4F ) . Similarly , weak uba-1 ( it129 ) mutant animals survive to adulthood and display weak but significant linker cell survival ( 17 ± 2% survival , n=209 ) . Furthermore , RNAi against the gene encoding ubiquitin , ubq-1 , also blocks LCD ( Figure 4F ) . Thus , LCD requires canonical components of the ubiquitin-mediated protein degradation pathway . We also examined the effects of inhibiting components of the 19S proteasome regulatory subunit on LCD and found that systemic RNAi against the rpn-3 , rpn-8 , or rpn-11 genes results in linker cell survival in about one third of animals ( Figure 4F ) , and linker cell-specific RNAi against these genes results in similar inhibition ( Figure 4F ) . Taken together , our results strongly suggest that LET-70 functions in the linker cell as a component of the UPS . E2 enzymes such as LET-70 function through E3 proteins to mediate protein degradation ( Hershko et al . , 1983 ) . We therefore sought to identify E3 ubiquitin ligase components that mediate LET-70 activity . Cullin proteins are subunits of many E3 enzymes , and the C . elegans genome encodes six such proteins ( CUL-1-6 ) . We tested whether any of these is involved in LCD and found that RNAi against the cul-3 gene results in inappropriate linker cell survival ( Figure 4G , Supplementary file 1C ) . Linker-cell-specific RNAi against cul-3 also yields surviving linker cells , supporting a cell autonomous function for this gene . Strikingly , cul-3 ( RNAi ) ; let-70 ( ns770 ) animals exhibit a synergistic increase in linker cell survival well above each single mutant , indicating that these genes likely function together , in sequence or in parallel , to promote LCD ( Figure 4G ) . Previous studies had demonstrated interactions between CUL-3 and the RING protein RBX-1 in C . elegans ( Pintard et al . , 2003 ) . While many RING-finger encoding genes we tested by RNAi do not appear to have roles in LCD ( Supplementary file 1B ) , we found that RNAi against the rbx-1 gene does promote modest linker cell survival ( Figure 4G ) , suggesting a role in LCD . Supporting this notion , cul-3 and rbx-1 are both expressed in the linker cell ( Figure 4—figure supplement 2 ) . CUL-3 E3 complexes often contain BTB-domain substrate binding proteins . We screened 23 BTB proteins by RNAi and/or mutation , and identified two that block LCD when inactivated ( Supplementary file 1C ) . One of these , EOR-1 , will be described elsewhere . The other , BTBD-2 , is homologous to human BTBD2 , and its inactivation results in linker cell survival in roughly half of animals examined ( Figure 4G ) . Expression of BTBD-2 using the mig-24 linker-cell-specific promoter restored linker cell death to btbd-2 ( gk474281 ) mutants ( 47 ± 3% survival in btbd-2 ( gk474281 ) mutants ( N=90 ) vs . 31 ± 4% survival in transgenic lines , 2 lines examined ( N=81 ) ) . We also examined 55 genes encoding protein domains commonly found in E3 enzymes ( Supplementary file 1B , C ) . We found that RNAi against the seven-in-absentia homolog siah-1 causes a modest but significant linker cell survival defect ( Figure 4G ) . To confirm this observation , we examined animals defective for the siah-1 ( tm1968 ) mutation , which deletes most of exon 4 of the gene and is likely a molecular null , and found a similar survival defect . Interestingly , both siah-1 ( tm1968 ) ; cul-3 ( RNAi ) and siah-1 ( tm1968 ) ; rbx-1 ( RNAi ) double mutants exhibit greater linker cell survival than either single mutant ( Figure 4G ) . We conclude that CUL-3 , RBX-1 , BTBD-2 , and SIAH-1 , all function to promote LCD and likely do so downstream of LET-70 . To study the expression and localization of LET-70 , we generated animals carrying let-70 promoter::let-70::GFP or let-70 promoter::GFP transgenes . We found that both reporters are expressed in the linker cell and that the translational fusion reporter is evenly distributed between the nucleus and cytoplasm ( Figure 5B , data not shown ) . Importantly , we found that expression of neither reporter is constitutive . Rather , while GFP fluorescence is not detected during migration of the linker cell , it is induced 1–2 hr before obvious morphological features of cell death appear ( Figure 5A–C ) . Similar induction is seen with a fosmid recombineered to contain 18 . 9 kb surrounding the genomic let-70 locus fused to GFP ( n>25 ) . We wondered whether the expression of other components of the UPS might also be induced in the linker cell with similar kinetics . Although some reporter genes we tested are not induced ( Figure 4—figure supplement 2 ) , expression of GFP reporter fusions to the ubq-1 gene , encoding C . elegans ubiquitin , and to the proteasome component gene rpn-3 is induced ( Figures 5D–I ) . These results suggest that expression of some UPS components is upregulated in the linker cell just prior to cell death onset . 10 . 7554/eLife . 12821 . 013Figure 5 . let-70 , ubq-1 , and rpn-3 expression is induced just before linker cell death onset . ( A-C ) let-70p::let-70::GFP expression in migrating ( A ) or dying ( B ) linker cell . Scale bar , 10 μm . ( C ) Expression quantification in ( A , B ) . Error bars , SEM . Number inside bar , no . animals scored . ( D-F ) Same as ( A-C ) for ubq-1p::ubq-1::GFP . ( G-I ) Same as ( A-C ) for rpn-3p::GFP . ( J ) Expression of indicated GFP reporters in surviving linker cells in him-8 ( e1489 ) animals of indicated genotype . ( K ) All animals contained qIs56 and him-5 ( e1490 ) . let-70 ( RNAi ) instead of let-70 ( ns770 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 013 To understand how the induction of UPS genes is regulated , we looked at the expression of let-70 promoter::let-70::GFP and ubq-1 promoter::ubq-1::GFP reporter transgenes in surviving cells in mutants of the Wnt , LIN-29 , and SEK-1/MAPKK pathways we identified as LCD regulators . Wild-type rpn-3 promoter::GFP expression decreases in all cells in the first hours of adulthood and was not bright enough to reliably score in mutant backgrounds . We found that surviving linker cells in mutants of all three pathways often failed to express either GFP reporter ( Figure 5J ) , but the effects were more pronounced for the let-70 reporter . Double mutants between mutant components of each of the three regulatory pathways and let-70 ( ns770 or RNAi ) demonstrate additive interactions ( Figure 5K ) , as would be expected with combinations of partial loss-of-function mutants functioning in the same pathway . Taken together , our results are consistent with a model in which LET-70 functions downstream of the Wnt , LIN-29 , and SEK-1/MAPKK signals . To examine the relationship between let-70 and hsf-1 , we looked at the expression of the let-70 promoter::let-70::GFP and ubq-1 promoter::ubq-1::GFP reporter transgenes in an hsf-1 ( sy441 ) partial loss-of-function mutant . As shown in Figure 6A , GFP expression was significantly reduced for both , with a more pronounced effect for the let-70 reporter . These studies indicate that wild-type HSF-1 activity is required to induce let-70 expression , and , therefore , that LET-70 functions downstream of HSF-1 . let-70 promoter::let-70::GFP expression is not induced by heat shock , consistent with previous Northern blot studies ( Figure 6B ) ( Zhen et al . , 1993 ) . Therefore , consistent with our characterization of HSF-1 , HSF-1 must be acting in a manner distinct from the heat-shock response to induce let-70 expression and cell death in the linker cell . 10 . 7554/eLife . 12821 . 014Figure 6 . HSF-1 controls LET-70 expression . ( A ) Expression of indicated GFP reporter in surviving linker cells in him-8 ( e1489 ) animals of indicated genotype . *p<0 . 0001 , p<0 . 005 , Fisher’s exact test . Error bars , SEM . ( B ) let-70p::let-70::gfp expression in head region after heat shock . Error bars , SD . ns , not significant , Student’s t-test . ( C ) let-70 promoter sequence alignment across indicated nematodes . Red , conserved nucleotides . ( D ) Same as ( C ) but for rpn-3 . ( E ) let-70p::GFP and let-70∆HSE::GFP expression . Error bar , SEM . *p<0 . 0001 , Fisher’s exact test . ( F ) let-70 and btbd-2 interactions with hsf-1 . Error bars , SEM . Number within bars , no . of animals scored . Animals contained qIs56 and him-5 ( e1490 ) . ND= not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 01410 . 7554/eLife . 12821 . 015Figure 6—figure supplement 1 . ∆HSE reduces let-70 promoter::let-70::GFP expression in the linker cell . ( A ) Male containing an integrated wild-type let-70p::let-70::GFP transgene; inset: higher magnification image of linker cell . Scale bar , 10 μm . Arrow , linker cell . ( B ) Same as ( A ) except with ∆HSE . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 015 The DNA motif TTCTAGAA is enriched in regulatory regions of genes induced in C . elegans in response to heat shock ( GuhaThakurta et al . , 2002 ) , and the motif TTCnnGAAnnTTC has been defined as an HSF binding element from yeast to mammals . A comparison of let-70 genomic sequences upstream of the ATG start codon revealed a region highly conserved between C . elegans and at least three other related nematode species ( Figure 6C ) . Within this region we identified two conserved motifs . The upstream motif ( motif 1 ) is identical to the HSF consensus binding site , whereas the downstream motif ( motif 2 ) contains two potential HSF monomer binding sites ( TTC and GAA ) . We also identified a highly conserved heat-shock element ( HSE ) upstream of the rpn-3 gene ( Figures 5G-I , 6D ) . A consensus HSF binding site was not found within the regulatory sequences used for the ubq-1 reporter studies; however , a number of one-off sites were found , perhaps explaining the weaker regulation of our ubq-1 reporter by HSF-1 . To test the functional relevance of the let-70 heat-shock element homology region for let-70 expression , we generated animals harboring a let-70 promoter::GFP reporter transgene in which a 97 nucleotide region including conserved motif 1 and 2 were deleted ( let-70∆HSE::GFP ) . As shown in Figure 6E , transgenic animals failed to express the reporter in the dying linker cell in about 40% of animals , consistent with the similar defect we observed in let-70::GFP expression in hsf-1 ( sy441 ) loss-of-function mutants . Importantly , GFP expression in other cells of let-70∆HSE::GFP animals was not perturbed ( Figure 6E , Figure 6—figure supplement 1 ) , suggesting a specific role for this DNA element in controlling linker cell expression of let-70 . Our results demonstrate that let-70 expression is under the control of HSF-1 , likely acting through a consensus heat-shock element in the let-70 5’ control region , but not through the canonical heat-shock response pathway . To functionally probe this model , we tested the genetic relationship between let-70 and hsf-1 . hsf-1 ( sy441 ) ; let-70 ( ns770 ) animals harboring partial loss-of-function perturbations of each gene , have increased linker survival compared to either single mutant alone ( Figure 6F ) . More importantly , while the hsf-1 ( R145A ) gain-of-function transgene restores cell death to all previously tested LCD mutants ( see above ) , it fails to rescue inappropriate linker cell survival in let-70 ( ns770 ) mutants ( Figure 6F ) . Likewise , the hsf-1 ( R145A ) gain-of-function transgene fails to restore cell death to btbd-2 ( gk474281 ) mutants ( Figure 6F ) . Taken together , our results suggest that LET-70 and BTBD-2 function downstream of a linker-cell-specific non-canonical function of HSF-1 to promote LCD . Our data also suggest that other HSF-1 targets are likely relevant , and that let-70 may be under the control of additional regulators . The results presented here allow us to construct a model for the initiation and execution of LCD in C . elegans ( Figure 7 ) . The logic of the LCD pathway may be similar to that of developmental apoptotic pathways . In C . elegans and Drosophila , where the control of specific cell deaths has been primarily examined , cell lineage or fate determinants control the expression of specific transcription factors that then impinge on proteins regulating caspase activation ( Fuchs and Steller , 2011 ) . Likewise , LCD is initiated by redundant determinants that require a transcription factor to activate protein degradation genes . 10 . 7554/eLife . 12821 . 016Figure 7 . Model for linker cell death . Green , upstream regulators . Orange , HSF-1 . Purple , proteolytic components . DOI: http://dx . doi . org/10 . 7554/eLife . 12821 . 016 Our data suggest that three partially redundant signals control LCD initiation . The antagonistic Wnt pathways we describe may provide positional information to the linker cell , as the relevant ligands are expressed only near the region where the linker cell dies . The LIN-29 pathway , which controls timing decisions during the L4-adult molt , may ensure that LCD takes place only at the right time . Finally , while the TIR-1/SEK-1 pathway could act constitutively in the linker cell , it may also respond to specific cues from neighboring cells . Indeed , MAPK pathways are often induced by extracellular ligands . We propose that these three pathways , together , trigger activation of HSF-1 . Our data support a model in which HSF-1 is present in two forms , HSF-1LC , promoting LCD , and HSF-1HS , protecting cells from stresses , including heat shock . We postulate that the redundant LCD initiation pathways tip the balance in favor of HSF-1LC , allowing this activity to bind to promoters and induce transcription of key LCD effectors , including LET-70/UBE2D2 and other components of the ubiquitin proteasome system ( UPS ) , functioning through E3 ligase complexes consisting of CUL-3 , RBX-1 , BTBD-2 , and SIAH-1 . Importantly , the molecular identification of LCD components and their interactions opens the door to testing the impact of this cell death pathway on vertebrate development . For example , monitoring UBE2D2 expression during development could reveal upregulation in dying cells . Likewise , genetic lesions in pathway components we identified may lead to a block in cell death . Double mutants in apoptotic and LCD genes would allow testing of the combined contributions of these processes . As is the case with caspase proteases that mediate apoptosis ( Pop and Salvesen , 2009 ) , how the UPS induces LCD is not clear , and remains an exciting area of future work . That loss of BTBD-2 , a specific E3 ligase component , causes extensive linker cell survival suggests that a limited set of targets may be required for LCD . Previous work demonstrated that BTBD2 , the vertebrate homolog of BTBD-2 , interacts with topoisomerase I ( Khurana et al . , 2010; Xu et al . , 2002 ) , raising the possibility that this enzyme may be a relevant target , although other targets may exist . The UPS has been implicated in a number of cell death processes in which it appears to play a general role in cell dismantling , most notably , perhaps , in intersegmental muscle remodeling during metamorphosis in moths ( Haas et al . , 1995 ) . However , other studies suggest that the UPS can have specific regulatory functions , as with caspase inhibition by IAP E3 ligases ( Ditzel et al . , 2008 ) . During Drosophila sperm development , caspase activity is induced by the UPS to promote sperm individualization , a process that resembles cytoplasm-specific activation of apoptosis ( Arama et al . , 2007 ) . While C . elegans caspases are dispensible for LCD , it remains possible that they participate in linker cell dismantling or serve as a backup in case the LCD program fails . Finally , the proteasome contains catalytic domains with target cleavage specificity reminiscent of caspases; however , inactivation of the caspase-like sites does not , alone , result in overt cellular defects ( Britton et al . , 2009 ) , suggesting that this activity may be needed to degrade only specific substrates . Although the proteasome generally promotes proteolysis to short peptides , site-specific cleavage of proteins by the proteasome has been described ( Chen et al . , 1999 ) . It is intriguing to speculate , therefore , that caspases and the proteasome may have common , and specific , targets in apoptosis and LCD . Our discovery that C . elegans heat-shock factor , HSF-1 , promotes cell death is surprising . Heat-shock factors are thought to be protective proteins , orchestrating the response to protein misfolding induced by a variety of stressors , including elevated temperature . Although a role for HSF1 has been proposed in promoting apoptosis of mouse spermatocytes following elevated temperatures ( Nakai et al . , 2000 ) , it is not clear whether this function is physiological . In this context , HSF1 induces expression of the gene Tdag51 ( Hayashida et al . , 2006 ) . Both pro- and anti-apoptotic activities have been attributed to Tdag51 ( Toyoshima et al . , 2004 ) , and which is activated in sperm is not clear . Recently , pathological roles for HSF1 in cancer have been detailed ( e . g . Mendillo et al . , 2012 ) , but in these capacities HSF1 still supports cell survival . Developmental functions for HSF1 have been suggested in which HSF1 appears to act through transcriptional targets different from those of the heat-shock response ( Jedlicka et al . , 1997 ) , although target identity remains obscure . Here , we have shown that HSF-1 has at least partially non-overlapping sets of stress-induced and developmental targets . Indeed , typical stress targets of HSF-1 , such as the small heat-shock gene hsp-16 . 49 as well as genes encoding larger chaperones , like hsp-1 , are not expressed during LCD , whereas let-70 , a direct transcriptional target for LCD , is not induced by heat shock . Interestingly , the yeast let-70 homologs ubc4 and ubc5 are induced by heat shock ( Seufert and Jentsch , 1990 ) , supporting a conserved connection between HSF and UBE2D2-family proteins . However , the distinction between developmental and stress functions is clearly absent in this single-celled organism , raising the possibility that this separation of function may be a metazoan innovation . What distinguishes the stress-related and developmental forms of HSF-1 ? One possibility is that whereas the stress response appears to be mediated by HSF-1 trimerization , HSF-1 monomers or dimers might promote LCD roles . Although this model would nicely account for the differential activities in stress responses and LCD of the HSF-1 ( R145A ) transgenic protein , which would be predicted to favor inactivation of a larger proportion of higher order HSF-1 complexes , the identification of conserved tripartite HSEs in the let-70 and rpn-3 regulatory regions argues against this possibility . Alternatively , selective post-translational modification of HSF-1 could account for these differences . In mammals , HSF1 undergoes a variety of modifications including phosphorylation , acetylation , ubiquitination , and sumoylation ( Xu et al . , 2012 ) , which , depending on the site and modification , stimulate or repress HSF1 activity . In this context , it is of note that p38/MAPK-mediated phosphorylation of HSF1 represses its stress-related activity ( Chu et al . , 1996 ) , and the LCD regulator SEK-1 encodes a MAPKK . However , no single MAPK has been identified that promotes LCD ( E . S . B . , M . J . K . unpublished results ) , suggesting that other mechanisms may be at play . Our finding that POP-1/TCF does not play a significant role in LCD raises the possibility that Wnt signaling exerts direct control over HSF-1 through interactions with β-catenin . However , we have not been able to demonstrate physical interactions between these proteins to date ( M . J . K , unpublished results ) . Finally , a recent paper ( Labbadia and Morimoto , 2015 ) demonstrated that in young adult C . elegans , around the time of LCD , global binding of HSF-1 to its stress-induced targets is reduced through changes in chromatin modification . Remarkably , we showed that chromatin regulators play a key role in let-70 induction and LCD ( J . A . M . , M . J . K and S . S . , manuscript in preparation ) , suggesting , perhaps , that differences in HSF-1 access to different loci may play a role in distinguishing its two functions . Previous studies from our lab raised the possibility that LCD may be related to degenerative processes that promote vertebrate neuronal death . Nuclear crenellation is evident in dying linker cells and in degenerating cells in polyQ disease ( Abraham et al . , 2007 ) and the TIR-1/Sarm adapter protein promotes LCD in C . elegans as well as degeneration of distal axonal segments following axotomy in Drosophila and vertebrates ( Osterloh et al . , 2012 ) . The studies we present here , implicating the UPS and heat-shock factor in LCD , also support a connection with neurodegeneration . Indeed , protein aggregates found in cells of patients with polyQ diseases are heavily ubiquitylated ( Kalchman et al . , 1996 ) . Chaperones also colocalize with protein aggregates in brain slices from SCA patients , and HSF1 has been shown to alleviate polyQ aggregation and cellular demise in both polyQ-overexpressing flies and in neuronal precursor cells ( Neef et al . , 2010 ) . While the failure of proteostatic mechanisms in neurodegenerative diseases is generally thought to be a secondary event in their pathogenesis , it is possible that this failure reflects the involvement of a LCD-like process , in which attempts to engage protective measures instead result in activation of a specific cell death program . C . elegans strains were cultured using standard methods ( Brenner , 1974 ) and were grown at 20°C unless otherwise indicated . Wild-type animals were the Bristol N2 subspecies . Most strains harbor one of two mutations that generate a high percentage of male progeny , him-8 ( e1489 ) IV or him-5 ( e1490 ) V , as well as one of two integrated linker cell markers , qIs56[lag-2p::GFP]V or nsIs65[mig-24p::Venus] X . Other alleles and transgenes used are as follows: LGI: hsf-1 ( sy441 ) , lin-44 ( n1792 ) , mig-1 ( e1787 ) , lin-17 ( n3091 ) , unc-101 ( sy216 ) , gsk-3 ( nr2047 ) , pop-1 ( q624 ) , pop-1 ( q645 ) , pop-1 ( hu9 ) , daf-16 ( mu86 ) , unc-13 ( e1091 ) . LGII: mig-14 ( ga62 ) , lin-29 ( n333 ) , lin-29 ( n546 ) , mig-5 ( rh147 ) , cam-1 ( gm122 ) , cwn-1 ( ok546 ) , rrf-3 ( pk1426 ) , drSi13[hsf-1p::hsf-1-gfp] , drSi28[hsf-1p::hsf-1 ( R145A ) -GFP] , nsSi2[hsf-1p::hsf-1 ( R145A ) -GFP] , nsSi3[hsf-1p::hsf-1 ( R145A ) -GFP] . LGIII: pqn-41 ( ns294 ) , wrm-1 ( ne1982 ) , lit-1 ( t512 ) , unc-32 ( e189 ) , mom-4 ( ne1539 ) , mom-4 ( or39 ) , unc-119 ( ed4 ) . LGIV: siah-1 ( tm1968 ) , egl-20 ( n585 ) , cwn-2 ( ky756 ) , cwn-2 ( ok895 ) , let-70 ( ns770 ) , uba-1 ( it129 ) , btbd-2 ( gk474281 ) . LGV: rde-1 ( ne219 ) , cfz-2 ( ok1201 ) , mom-2 ( ne834 ) , daf-21 ( p673 ) . LGX: bar-1 ( ga80 ) , sek-1 ( ag1 ) , lin-18 ( e620 ) . See Supplemental file 2A . Two additional lines of hsf-1p::hsf-1 ( R145A ) -GFP were generated from pOG124 ( a gift of T . Lamitina ) , by the ‘direct’ method , as previously described ( Frøkjaer-Jensen et al . , 2014 ) . One line failed to exhibit bar-1 mutant suppression , but also did not enhance hsf-1 ( sy441 ) survival , suggesting it was inactive , and was therefore not used in analysis . Inserts were verified by PCR and expression of HSF-1::GFP . let-70 ( ns770 ) was generated using co-CRIPSR-based CRISPR/Cas9-mediated genome editing as previously described ( Arribere et al . , 2014 ) . pJA42 ( Addgene , Cambridge , MA ) was edited using PCR mutagenesis with a ‘universal’ forward primer ( 5’- GTTTTAGAGCTAGAAATAGCAAGTTAAAATAAGGCTAG -3’ ) and a let-70 specific reverse primer ( TTTCTAGCTCTAAAACATGGATAGTCTGTTGGGAAG CAAGACATCTCGCAATAG ) to generate the let-70 targeting vector . Single-stranded oligodeoxynucleotide ‘repair’ templates were ordered from Sigma for let-70 ( ns770 ) ( 5’ TTAAATTTATTTTTTTCCAATTTCGATCAATACCTTTGGTGGTTTAAATGAATAGTCTGTTGGGAAGTGGATAGTGAGGAAGAAGACACCTCCCTGATAGG 3’ ) and dpy-10 ( cn64 ) ( Arribere et al . , 2014 ) . N2 animals were injected with the following mix: 50 ng pDD162 ( Addgene ) , 25 ng pJA58 ( dpy-10 sgRNA , Addgene ) , 25 ng let-70 targeting vector , 20 ng dpy-10 ( cn64 ) repair oligo , 20 ng let-70 ( ns770 ) repair oligo in 1x injection buffer ( 20mM potassium phosphate , 3mM potassium citrate , 2% PEG , pH 7 . 5 ) . F1 generation was screened for animals with a roller or dumpy-roller phenotype , indicating successful repair of the dpy-10 break using the provided dpy-10 oligonucleotide template , which were picked to individual plates . These animals were allowed to lay eggs and then genotyped for successful co-conversion of the let-70 locus by PCR and Sanger sequencing . Non-roller F2 animals were then picked from successfully let-70-converted F1s and homozygosed for let-70 ( ns770 ) before outcrossing twice . See Supplemental file 2B . RNAi was performed by feeding on the strains indicated ( Blum et al . , 2012 ) . Bleached embryos from gravid hermaphrodites were synchronized at the L1 stage by leaving them overnight in M9 . L1s animals ( 30–50% of which were male ) were added to each RNAi plate and grown for approximately 48 hr at 20–22°C . 0–2 hr adults were scored using a fluorescent dissecting scope ( Leica ) . Clones were either newly created by cloning into the L4440 vector , or were already published clones from the Ahringer feeding library . Total RNA was extracted using TRIzol ( Theromfisher , Waltham , MA ) using standard protocols . cDNAs were amplified from day one adult Caenorhabditis briggsae using the SuperScript II Reverse Transcriptase ( Thermofisher ) . C . briggsae let-70 cDNA with silent mutations was generated using GeneArt Gene Synthesis ( Thermofisher ) and cloned into plasmid using standard conditions . C85S point mutation was generated using Pfu turbo polymerase ( Agilent , Santa Clara , CA ) and DpnI digest ( NEB , Ipswich , MA ) using standard Quikchange protocol ( Agilent ) . Germline transformation was carried out as previously described ( Mello et al . , 1991 ) . For GFP expression analysis , all plasmids were injected into unc-119 ( ed3 ) III; him-8 ( e1489 ) IV hermaphrodites with unc-119 ( + ) ( Maduro and Pilgrim , 1995 ) as a transformation marker . All plasmids were injected at between 1–50 ng/ul . pBluescript ( Agilent ) was used to adjust the DNA concentration of injection mixtures if necessary . For rescue studies , animals were picked under a fluorescent dissecting microscope ( Leica ) the previous night as L3s with YFP- or mCherry-expressing linker cells to a new RNAi plate and scored the following day . Throughout , only correctly-migrated linker cells were used in determining survival percentages . Linker cell death was scored as previously described ( Blum et al . , 2012 ) . Briefly , worms were synchronized by treating gravid hermaphrodites with alkaline bleach and allowing the eggs to hatch in M9 medium overnight . Synchronized L1s were released onto fresh NGM plates seeded with OP50 or HT115 E . coli containing the RNAi clone of interest , and maintained at 20°C . Animals were picked to a new plate as late L4s with a fully retracted tail tip with rays visible under the unshed L4 cuticle . Two hours later , newly molted adults were mounted on slides on 2% agarose-water pads , anaesthetized in 30 mM sodium azide or 5 mM tetramisole , and examined on a Zeiss Axioplan 2 or AxioScope A1 under Nomarski optics and widefield fluorescence at 40x or 63x . Images were acquired through a Zeiss AxioCam and the Axiovision software . The linker cell was identified by green fluorescence ( from reporter transgenes ) as well as by its location and morphology . A linker cell was scored as surviving if its nucleus was circular with an intact nucleolus , if the cell shape was not rounded , and if the cell had not shed any large blebs . All other cells were scored as dead or dying . Rescuing extrachromosomal arrays contained a lag-2p::mCherry construct , and , in an effort to prevent selection bias towards survival , males with mCherry-expressing linker cells were picked as L3s for scoring the following day as young adults , as above . Some Wnt pathway mutants exhibited two linker cells . For these strains , animals with only one visible linker cell were picked as L3s to score the following day . Throughout , only correctly migrated cells that had reached the cloaca were used in determining survival percentages . For GFP expression assays , 0–2 hr adults containing the let-70p::let-70::GFP ( nsIs241 ) or ubq-1p::let-70p::GFP ( nsIs386 ) transgenes were scored for the presence or absence of GFP expression in the linker cell . The fraction of animals expressing GFP was divided by the fraction of animals with surviving linker cells in order to obtain an accurate measure of linker cell expression . This method was verified by looking at GFP expression of reporters with a lag-2p::mCherry coinjection marker; results using the two different methodologies showed similar expression patterns . Just-molted ( 0–2 hr ) qIs56 him-5 ( e1490 ) ; bar-1 ( ga80 ) or let-70 ( RNAi ) adult males with surviving linker cells were imaged using a Zeiss Axioplan 2 compound microscope to measure the relative location of the linker cell within the worm using the AxioVision software ( Zeiss ) . Animals were then fixed , stained , embedded in resin , and sectioned using standard methods ( Lundquist et al . , 2001 ) . Images were acquired on an FEI TECNAI G2 Spirit BioTwin Transmission Electron Microscope with a Gatan 4K x 4K digital camera at The Rockefeller University EM Resource Center . An unpaired t-test was used for GFP quantification in rde-1 knockdown animals and in let-70p::let-70::GFP animals following heat shock . Fisher’s Exact Tests were used for quantification of LCD experiments as well as quantification of GFP+ linker cells . let-70 cDNA cloned into the vector pET28b ( + ) ( Novagen , ) was transformed into BL21 ( DE3 ) cells using heat shock . Cells were induced overnight with 500 mM IPTG at 25°C . Purification was performed using a previously described protocol ( Sandu et al . , 2010 ) . In vitro ubiquitination assay: A 40 μL reaction containing 3 μg each of purified Drosophila Uba1 , Diap1 , and ubiquitin ( Gift from C . Sandu ) were incubated with C . elegans His-LET-70 and reaction buffer ( 25 mM Tris , pH 7 . 5 , 50 mM NaCl , 250 μM DTT , 4 mM ATP and 4 mM MgCl2 ) for 30 min at 25°C ( Sandu et al . , 2010 ) . One half of the reaction was run on an SDS-PAGE gel and stained with Coomassie Blue to visualize proteins . Animals were cultured on 4 cm NGM agar plates seeded with E . coli OP50 . These plates were sealed with parafilm , placed in a water bath at the indicated temperature for the indicated time , agar face down , and subsequently returned to the 20°C incubator , until animals were picked for scoring as above .	Embryos make numerous new cells as they develop , but also destroy many cells to remove the faulty ones and to ensure that tissues grow to the right size and shape . This deliberate form of cell death must be precisely regulated to prevent too many cells or healthy cells , from being destroyed . Understanding the molecular mechanisms that govern cell death is therefore important for understanding normal development and also human disease . One well-studied process that leads to cell death is called apoptosis . This process carefully dismantles and breaks down the components of a cell , but does not seem to account for all cell death that occurs during animal development . Recently another developmental cell-death pathway , called the linker-cell-type death , was discovered in a small roundworm called Caenorhabditis elegans . This pathway appears to work in mammalian cells as well , and may help to break down nerve fibers that are not needed . However , many of this pathway’s component parts remained unknown . Kinet , Malin et al . have now used a combination of genetics and cell biology in C . elegans to uncover the components of linker-cell-type death and to investigate how they interact . The results of these studies revealed a hierarchy of genetic interactions that governs this pathway in C . elegans . One protein called HSF-1 plays a particularly important role . This protein is a transcription factor and it binds to , and regulates , the activities of various genes . HSF-1 usually works in cells to protect them from stress , but Kinet , Malin et al . showed that it instead promotes linker-cell-type death by activating a molecular machine , called the proteasome , that breaks down proteins . The experiments also revealed two proteins ( called BTBD-2 and SIAH-1 ) that may be important for shuttling specific proteins for degradation by the proteasome . Three signalling pathways that regulate important developmental processes also regulate the activation of linker-cell-type death . Kinet , Malin et al . propose that these signalling pathways do so by working together to activate HSF-1 , which in turn activates the genes that lead to the destruction of cells by the proteasome . A future challenge is to understand in more detail how the more recently discovered cell death pathway actually kills cells . Further work could also explore how HSF-1 , a protein that normally protects cells , is transformed into a cell-killing protein .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	HSF-1 activates the ubiquitin proteasome system to promote non-apoptotic developmental cell death in C. elegans
Engineered allosteric regulation of protein activity provides significant advantages for the development of robust and broadly applicable tools . However , the application of allosteric switches in optogenetics has been scarce and suffers from critical limitations . Here , we report an optogenetic approach that utilizes an engineered Light-Regulated ( LightR ) allosteric switch module to achieve tight spatiotemporal control of enzymatic activity . Using the tyrosine kinase Src as a model , we demonstrate efficient regulation of the kinase and identify temporally distinct signaling responses ranging from seconds to minutes . LightR-Src off-kinetics can be tuned by modulating the LightR photoconversion cycle . A fast cycling variant enables the stimulation of transient pulses and local regulation of activity in a selected region of a cell . The design of the LightR module ensures broad applicability of the tool , as we demonstrate by achieving light-mediated regulation of Abl and bRaf kinases as well as Cre recombinase . Dissection of biological processes is greatly aided by optogenetic methods that allow researchers to mimic and manipulate the function of individual proteins . Multiple broadly applicable approaches have been developed to enable tight regulation of protein localization and protein interactions ( Guntas et al . , 2015; Wang et al . , 2016; Strickland et al . , 2012; Kawano et al . , 2015 ) ; however , the direct control of enzymatic activity remains a challenge . Only a few existing strategies achieve direct regulation of enzymatic activity , but they either lack the ability to activate enzymes locally within a cell , or they are difficult to apply to multiple enzyme classes ( Wang et al . , 2019; Dagliyan et al . , 2016; Zhou et al . , 2017; Diaz et al . , 2017; Wu et al . , 2009; Hongdusit et al . , 2020 ) , thus limiting their use for addressing mechanistic questions in cell biology . This highlights the need for an optogenetic approach which is broadly applicable , easy to implement and provides all the advantages of optogenetics in one tool including tight temporal and local subcellular regulation . An attractive strategy that can be harnessed for this purpose is the application of a rationally designed light-sensitive domain that can allosterically control protein activity . This protein engineering approach offers several important advantages . It enables targeted regulation of one domain within a multidomain protein ( Karginov et al . , 2010 ) . Regulation is achieved ‘remotely’ without steric interference with the catalytic pocket of the enzyme and substrate binding . Also , allosteric switch domains are genetically encoded into the targeted protein simplifying the application of the tool . Only four optogenetic methods for allosteric regulation of activity have been reported so far and they suffer from critical limitations ( Dagliyan et al . , 2016; Wu et al . , 2009; Hongdusit et al . , 2020; Winkler et al . , 2015; Reynolds et al . , 2011 ) . Two of these strategies achieve subcellular control of certain targeted proteins but they cannot be applied to many other enzymes due to their design ( Wu et al . , 2009; Hongdusit et al . , 2020 ) . One approach is broadly applicable but it does not achieve local control and triggers inactivation rather than activation of the protein ( Dagliyan et al . , 2016 ) . A similar approach described by Reynolds et al achieves regulation in both directions for dihydrofolate reductase ( DHFR ) and PDZ domain but efficiency of this approach for applications in mammalian cells and its ability to regulate proteins at subcellular level has not been demonstrated ( Reynolds et al . , 2011 ) . Furthermore , existing allosteric switches lack the ability to tune the activation/inactivation kinetics , an important feature required for mimicking different temporal signaling modes of enzymes . Among the various potential targets of value for cell biology applications , protein kinases constitute an important family of enzymes that regulate key physiological functions and thus their activity is tightly controlled . Aberrant kinase regulation is the underpinning of many diseases , including the development and progression of malignant tumors ( Lee and Yaffe , 2016 ) . Many kinases are therefore important therapeutic targets and significant research effort is directed towards uncovering their function in cells . Such functional studies are challenging because a single kinase often induces drastically different responses depending on the location , level and timing of its activation ( Marshall , 1995; Rauch et al . , 2016; Cohen-Saidon et al . , 2009; Bugaj et al . , 2018; Toettcher et al . , 2013 ) . Furthermore , transient , sustained , or oscillatory kinase activation can result in distinct outcomes . Therefore , interrogating kinase-mediated signaling requires approaches capable of mimicking complex spatiotemporal control of kinase activity down to the subcellular resolution . Traditional methods using genetic manipulation and small molecule inhibitors lack this level of regulation . Optogenetics , on the other hand , overcomes these limitations by leveraging the precision of light-mediated activation ( Toettcher et al . , 2011 ) . Some existing optogenetic tools regulate the dimerization or localization of kinases ( Toettcher et al . , 2013; Wend et al . , 2014; Grusch et al . , 2014; Chang et al . , 2014; Kim et al . , 2014 ) . These approaches control the catalytic activity only indirectly and can only be applied to a narrow group of kinases . Other approaches employ light-sensitive caging or allosteric switches for the regulation of kinase catalytic domains ( Wang et al . , 2019; Dagliyan et al . , 2016; Zhou et al . , 2017 ) . However , these methods do not achieve subcellular control of activity and some of these approaches are also limited in their broader applicability due to structural requirements ( Wang et al . , 2019; Dagliyan et al . , 2016; Zhou et al . , 2017 ) . Thus , the development of a broadly applicable approach to enable optogenetic regulation of protein kinases remains challenging and highly desirable . Here , we report the development of a protein engineering approach that enables light-mediated allosteric control of enzymatic activity and combines critical advantages of optogenetics in one tool: it provides tight temporal regulation of activity with tunable kinetics , enables local control that can be achieved with subcellular resolution , and allows for broad applicability to different types of enzymes . We achieve this by employing an engineered light sensitive switch domain that is genetically encoded and enables the allosteric regulation of kinase catalytic domains . It provides fast , specific , and tunable control of kinase activity in living cells using low intensity blue light . Furthermore , this method enables reversible and repetitive activation of a kinase and provides local control of kinase activity at a subcellular level . Application of this tool allowed us to uncover temporal dynamics of phosphorylation-mediated signaling following short term ( 10 s ) as well as prolonged activation of a kinase . Importantly , we show that this approach can be applied not only to different kinases but also to other enzymes . To modulate kinase activity with light , we engineered a Light-Regulated ( LightR ) domain that can potentially function as an allosteric switch when inserted into a catalytic domain of a kinase . LightR is comprised of two tandemly connected Vivid ( VVD ) photoreceptor domains from Neurospora crassa ( Nihongaki et al . , 2014; Vaidya et al . , 2011; Zoltowski and Crane , 2008; Zoltowski et al . , 2007 ) . VVD is a monomer in the dark , and it forms an antiparallel homodimer upon illumination with blue light ( Nihongaki et al . , 2014; Vaidya et al . , 2011; Zoltowski and Crane , 2008; Zoltowski et al . , 2007; Wang et al . , 2012 ) . Dimerization is accompanied by a major flip of the N-terminal tail , bringing it close to the C-terminus of the other VVD in the dimer ( Figure 1A; Vaidya et al . , 2011; Zoltowski and Crane , 2008; Zoltowski et al . , 2007 ) . Therefore , we surmised that a tandem connection of two VVDs via a flexible linker would generate a clamp-like switch of 335 amino acid total size that opens in the dark and closes in response to blue light . To connect two VVD molecules , we designed a flexible 22 amino acid linker ( GGS ) 4G ( GGS ) 3 which provides sufficient flexibility and length ( approximately 25–30 Å when extended in the dark state ) to accommodate the association and dissociation of the VVD monomers . We hypothesized that inserting this LightR clamp domain into a small flexible loop within the catalytic domain of an enzyme would enable light-mediated regulation of its activity . In the dark , the opening of the LightR clamp could increase the distance between its N- and C- termini up to approximately 25 Å , which should distort the native structure of the catalytic domain and thereby inactivate the enzyme . Illumination with blue light would close the clamp and bring the N- and C-termini of LightR together resulting in restoration of the native structure of the catalytic domain and recovery of the enzyme activity ( Figure 1B ) . We first applied this optogenetic approach to regulate the tyrosine kinase Src which is a key regulator of cell migration , angiogenesis and tumor progression ( Sen and Johnson , 2011 ) , and generated light-regulated tyrosine kinase Src ( LightR-Src ) . Our previous studies demonstrate that the insertion of a regulatory domain at position Gly288 in the Src catalytic domain can be employed to achieve allosteric control of activity Karginov et al . , 2010; thus we focused on this site for the insertion of LightR domain ( Figure 1C ) . We first performed molecular dynamics simulations of the designed LightR-Src catalytic domain to predict a possible mechanism for the regulation of kinase activity by LightR domain . To build the initial LightR domain model for the dark and lit state , we modified the crystal structures of VVD in the light ( PDB: 3RH8 ) and dark ( PDB: 2PD7 ) by connecting the VVD proteins with a 22-residue flexible linker , ( GGS ) 4G ( GGS ) 3 . The GPGGSGG and GSGGPG linkers were then added to the N- and C-termini of the LightR domain , respectively ( Figure 1C ) . All of these flexible linkers were refined in Modeller . The LightR-Src structures were then constructed by inserting LightR domain into crystal structures of c-Src catalytic domain ( PDB:1Y57 ) replacing G288 . To ensure the best possible approximation of a starting structure , the engineered molecules were also subjected to energy minimization to allow for resolution of potential steric clashes , unusual torsion angles , or other energetically unfavorable aspects potentially introduced in the construction of the model . The geometry optimized molecules were equilibrated at 37°C to reach an equilibrium state structure at the desired temperature prior to starting the simulations . The analysis of 100 ns simulations demonstrated stabilization of fluctuations in root mean square deviation ( RMSD ) of the backbone ( Figure 1—figure supplement 1A ) . This is indicative of a structure occupying an energetic minima and thus represents a reasonable structure for further analysis . Analysis of LightR-Src in the dark shows increased RMSD of individual residues within the catalytic domain relative to the active wild-type Src ( Figure 1D , E; Figure 1—figure supplement 1B ) and a strong correlation between the motion of the residues in the LightR clamp and in the Src catalytic domain ( Figure 1—figure supplement 2 ) suggesting a possible allosteric mechanism . In contrast , the lit state of LightR-Src exhibits much lower RMSD , closer to wild-type Src ( Figure 1D , E; Figure 1—figure supplement 1B ) , and loses the motion correlation between LightR domain and Src residues ( Figure 1—figure supplement 2 ) . This indicates that the open conformation of LightR clamp in the dark can cause changes in the catalytic domain of Src that can be reversed upon illumination with blue light . Interestingly , LightR-Src in the dark shows substantial deviation of the G-loop ( Figure 1D , E; Figure 1—figure supplements 1B and 3 ) , a critical functional element of most protein kinases ( Cowan-Jacob , 2006; Krupa et al . , 2004 ) . This suggests a possible allosteric mechanism by which LightR domain regulates kinase activity via disruption of the G-loop . To test whether LightR insertion enables light-mediated regulation of Src , we evaluated Src catalytic activity by an in vitro kinases assay using purified N-terminal fragment of paxillin as a substrate ( Karginov et al . , 2010; Cai et al . , 2008; Klomp et al . , 2016 ) . In this and all following experiments , we used a LightR-Src construct bearing a Y527F mutation ( avian cSrc position ) that disrupts Src autoinhibition and prevents its negative regulation by endogenous mechanisms ( Karginov et al . , 2010; Zheng et al . , 2000 ) . We employed this to ensure that LightR-Src is only regulated by light . We compared LightR-Src activity to the activity of the catalytically inactive mutant of LightR-Src ( D388R in mouse cSrc , sequence ID: NP_001020566 . 1 ) and the constitutively active mutant of cSrc ( Y527F in avian cSrc ) . LightR-Src , as well as the control constructs , were transiently overexpressed in and immunoprecipitated from LinXE cells . Our results show that LightR insertion in Src completely disrupts its activity in the dark which is recovered in response to blue light to levels comparable to that of the constitutively active Src ( Figure 2A ) . Next , we tested the regulation of LightR-Src in living cells by evaluating changes in phosphorylation of known endogenous Src substrates , p130Cas ( Y249 ) and paxillin ( Y118 ) ( Cunningham-Edmondson and Hanks , 2009; Sachdev et al . , 2009 ) . Our results show that illumination of LinXE cells , transiently expressing LightR-Src , induces robust phosphorylation of the endogenous Src targets paxillin and p130Cas ( Figure 2B , Figure 2—figure supplement 1A ) . Importantly , cells expressing the catalytically inactive mutant of LightR-Src ( D388R ) that were also exposed to blue light did not show any increase in phosphorylation of Src substrates ( Figure 2B ) . This demonstrates that the increase in phosphorylation is a direct consequence of LightR-Src activation and that blue light by itself had no effects on the Src targets . The activation time-course demonstrates noticeable phosphorylation of Src targets after only two minutes of light irradiation ( Figure 2C ) . The level of activity of LightR-Src can be regulated by attenuation of light intensity ( Figure 2—figure supplement 1B ) . These data support our model for regulation of kinase activity using LightR clamp domain and demonstrate efficient and specific regulation of LightR-Src in living cells . To further assess the function of LightR-Src and the kinetics of its activation in cells , we evaluated the phosphorylation of a broad panel of Src substrates at different time points of activation using quantitative LC-MS/MS analysis . For this analysis , we used HeLa cell line stably expressing LightR-Src-mCherry-myc construct at levels comparable to that of endogenous Src ( Figure 2—figure supplement 1C ) . Phosphorylation of known Src targets , including caveolin1 ( Y14 ) , p130Cas ( Y12 , Y128 , Y249 , and Y287 ) , paxillin ( Y88 and Y118 ) , p120 catenin ( Y217 and Y228 ) , and cortactin ( Y421 ) , was significantly increased in LightR-Src expressing cells , but not in control cells , exposed to blue light ( Figure 2—source data 1 ) . Principal component analysis ( PCA ) revealed that LightR-Src-expressing cells exhibited a broadly distinct phosphoproteome dynamics from control cells following light exposure ( Figure 2—figure supplement 2A ) . Protein interaction network analysis along principal component one revealed that LightR-Src-induced phosphorylation events were enriched for cell migration , cell adherens junctions , and focal adhesions , all of which are processes known to be driven by Src activation ( Figure 2—figure supplement 2B ) . To assess the kinetics of LightR-Src signaling , we analyzed changes in the phosphoproteome at different time points after LightR-Src activation . Several distinct phosphorylation kinetics profiles were identified . ‘Early responders’ exhibited a significant increase in phosphorylation as early as ten to thirty seconds after LightR-Src activation , further demonstrating rapidity of activation achieved by LightR-Src ( Figure 2D ) . ‘Intermediate responders’ show increased phosphorylation at the 1 to 5 min time points ( Figure 2E ) . Interestingly , we detected Src autophosphorylation on Y416 at 1 min , indicating that Src phosphorylates some targets before it even undergoes autophosphorylation . A group of ‘late responders’ were phosphorylated at 1 hr ( Figure 2F ) . A distinct group of proteins comprised of MAP kinases ERK1 and ERK2 exhibited only transient increase in phosphorylation at 5 min ( Figure 2—figure supplement 3 ) . This phosphorylation is mediated by upstream MAP kinase cascades and leads to activation of ERK kinases ( Bugaj et al . , 2018; Toettcher et al . , 2013; Cargnello and Roux , 2011 ) . Thus , our data indicate that Src only transiently activates specific MAP kinase pathways . Importantly , all these phosphorylation changes were not detected in control HeLa cells that were exposed to blue light but did not express LightR-Src ( Figure 2—figure supplement 4A–C ) . Overall , these results demonstrate that LightR-Src phosphorylates known Src substrates , show the fast kinetics of LightR-Src signaling within seconds , and uncover distinct temporal patterns of Src target protein phosphorylation in living cells . Since VVD dimerization is reversible ( Kawano et al . , 2015 ) , we hypothesized that LightR-Src should become inactive when light is switched off . To test this , LinXE cells transiently expressing LightR-Src were illuminated with blue light for 30 min and then placed in the dark for different periods of time . Our results show that incubation in the dark led to a significant decrease in phosphorylation of paxillin ( Figure 3A ) . However , it took up to 2 hr for phosphorylation to return to basal levels , indicating the slow inactivation kinetics of LightR-Src . To achieve faster inactivation of LightR-Src , we introduced an I85V mutation into both VVD domains . This mutation reduces the half-life of VVD dimer in the dark from 18 , 000 s to 780 s and thus should facilitate faster LightR-Src inactivation ( Zoltowski et al . , 2009 ) . Indeed , our results demonstrate that compared to the original LightR-Src , the I85V/I85V variant ( FastLightR-Src ) shows much faster reversibility . Within two minutes after the light was switched off , we observed a significant decrease in paxillin phosphorylation ( Figure 3B ) . By fifteen minutes , phosphorylation reached basal level . However , we noticed that activation of FastLightR-Src leads to a lower p130Cas phosphorylation level when compared to the same activation time point of LightR-Src ( Figure 3—figure supplement 1 ) . This is potentially due to the fast cycling of I85V mutants between lit and dark state ( Zoltowski et al . , 2009 ) . This cycling could happen even when cells are illuminated and thus would reduce the fraction of active LightR-Src molecules at a given time ( Zoltowski and Crane , 2008 ) . Thus , FastLightR-Src may allow researchers to mimic function of Src kinase cycling between activation and inactivation states in living cells ( Kaimachnikov and Kholodenko , 2009 ) . Overall , our results show that the off-kinetics of the LightR switch can be tuned by modifications of the VVD domains . This provides the flexibility required for mimicking different temporal modes of kinase signaling , a capability that existing optogenetics approaches lack ( Wang et al . , 2019; Dagliyan et al . , 2016; Zhou et al . , 2017 ) . Protein kinases are often activated transiently and can undergo repeated cycles of activation/inactivation ( Kholodenko , 2006; Purvis and Lahav , 2013; Conlon et al . , 2016; Li et al . , 2017; Zhang et al . , 2018; Roche et al . , 1995; Hilioti et al . , 2008; Jacquel et al . , 2009; Kholodenko , 2000; Maeda et al . , 2004 ) . These oscillations of kinase activity can drive a specific biological response ( Kholodenko , 2006; Purvis and Lahav , 2013; Conlon et al . , 2016; Li et al . , 2017; Roche et al . , 1995; Hilioti et al . , 2008; Jacquel et al . , 2009; Kholodenko , 2000; Maeda et al . , 2004 ) . Thus , we wanted to determine whether FastLightR construct can be used to mimic oscillations of kinase activity in living cells . To test this , FastLightR-Src was activated for two periods of twenty minutes each , separated by ten minutes of deactivation . Our results reveal successful cycles of activation and inactivation as indicated by changes in phosphorylation of p130Cas ( Figure 3C ) . Previous studies show that activation of Src leads to stimulation of cell spreading ( Klomp et al . , 2016; Karginov et al . , 2014; Kaplan et al . , 1995; Cary et al . , 2002; Fu et al . , 2018 ) . Therefore , we tested whether LightR-Src activation induces a similar response in living cells . Our results show that cells expressing FastLightR-Src start spreading upon irradiation with blue light ( Figure 4A; Video 1 ) and stop immediately when the light is turned off . Repeated irradiation of cells with blue light induced corresponding cycles of cell spreading; demonstrating again that the LightR tool can be used to mimic oscillation of kinase activity in living cells ( Figure 4A ) . Importantly , illumination of cells expressing catalytically inactive mutant of LightR-Src ( D388R ) did not induce any cell-spreading ( Figure 4B ) . Also , we observed that inactive FastLightR-Src localizes in the perinuclear region and translocates to focal adhesions and cell membrane upon activation ( Figure 4C; Video 1 ) . This translocation is reversible and correlates with activation/inactivation cycles . Notably , this change in localization mimics that of wild type Src ( Kaplan et al . , 1995; Chu et al . , 2014 ) . Overall , our results demonstrate that LightR-Src activation can mediate cell morphodynamic changes and functions similarly to what has been observed for native Src kinase . Current optogenetic tools enable localized kinase signaling only by re-localizing and sequestering a constitutively active kinase to an organelle or to specific areas in the cell ( Kerjouan , 2019; O'Banion et al . , 2018; Kakumoto and Nakata , 2013 ) . Light-mediated regulation of kinase catalytic activity per se has not been achieved at a subcellular level . We hypothesized that FastLightR-Src activation/inactivation kinetics should enable local activation of the kinase only in a defined subcellular location . Localized activation of protein kinases and Src is a critical determinant in the regulation of cell function . Previous studies have suggested that local activation of Src at the cell periphery should stimulate the formation of local membrane protrusions ( Cary et al . , 2002; Baumgartner et al . , 2008 ) . However , this hypothesis has only been indirectly supported and has not been rigorously evaluated due to the limitations of existing methods . The LightR-Src approach would allow us to define the effects of local activation of Src in living cells . Indeed , we observed that local illumination of HeLa cells transiently expressing FastLightR-Src induced the formation of membrane protrusions within the illuminated area and caused polarization of the cell towards the light ( Figure 5A–C; Figure 5—figure supplement 1A; Video 2 ) . This effect was reversed as soon as the light was switched off ( Figure 5A; Video 2 ) . Cells expressing the catalytically inactive LightR-Src ( D388R ) did not polarize in response to local light irradiation ( Figure 5B , C ) . Notably , FastLightR-Src translocated to focal adhesions only in the area illuminated with blue light ( Figure 5D; Figure 5—figure supplement 1B; Video 2 ) and relocated back once the light was switched off ( Figure 5D; Video 2 ) . This is again consistent with known activation-dependent changes in localization of wild type Src ( Dhar and Shukla , 1991; Weernink and Rijksen , 1995 ) . These results reveal that local activation of Src is sufficient to induce local protrusions and demonstrate that LightR approach can be used for the regulation of kinase activity at a subcellular level . Local regulation of protein kinase allows us to assess the dynamics of local morphological changes . A previous study using Src family kinases ( SFK ) biosensor suggested strong correlation between SFK activity at the cell periphery and cell-edge velocity ( Gulyani et al . , 2011 ) . LightR-Src allows us to determine the direct effect of local Src activity on protrusion dynamics . To achieve this , we evaluated temporal changes in cell edge velocity upon continuous local activation of LightR-Src in HeLa cells . Our analysis revealed that Src induced local waves of increased cell-edge velocity ( Figure 5E , H; Figure 5—figure supplement 2A ) ; suggesting the induction of recurrent local contractions that slow down membrane protrusion . Inhibition of Rho associate protein kinase ( ROCK ) and myosin light chain kinase ( MLCK ) , major regulators of contractile machinery ( Parri et al . , 2007; Totsukawa et al . , 2000 ) , perturbed these waves and reduced the average velocity of protrusions ( Figure 5E–J; Figure 5—figure supplement 2B ) . Interestingly , inhibition of ROCK caused significantly smaller average velocity reduction than inhibition of MLCK ( Figure 5—figure supplement 2B ) , consistent with its limited role in phosphorylating MLC at the cell periphery ( Totsukawa et al . , 2004 ) . Thus , by applying LightR-Src , we demonstrated that sustained local Src activity induces waves of protrusions that are mediated by ROCK and MLCK . To verify that the LightR approach is applicable to other kinases , we set out to engineer LightR variants of tyrosine kinase Abl and a dual specificity kinase bRaf . Since the majority of kinases share a conserved catalytic domain structure ( Fabbro et al . , 2015 ) , we inserted LightR switch into Abl and bRaf at a position analogous to the insertion site used in Src ( Figure 6A ) , replacing K282 residue in Abl and H477 in bRaf . Our results show that illumination of LinXE cells transiently expressing LightR-Abl-GFP construct induced the phosphorylation of known Abl substrate p130Cas ( Figure 6B ) . Stimulation of LightR-bRaf-Venus construct transiently expressed in LinXE cells induced phosphorylation of its direct target , MEK1 , and led to downstream activation of ERK1 and ERK2 kinases at levels comparable with the constitutively active mutant of bRaf ( V600E ) ( Figure 6C ) . To demonstrate that LightR-bRaf off-kinetics are tunable , we generated a FastLightR-bRaf variant following the same strategy used to generate FastLightR-Src . This variant exhibited significantly faster deactivation kinetics , with a half-life time around 15 min compared to approximately 3 hr half-life of LightR-bRaf ( Figure 6—figure supplement 1 ) . We also assessed whether FastLightR-bRaf can undergo cyclic activation and deactivation by monitoring ERK2 kinase translocation into the nucleus , a known outcome of bRaf activation ( Burack and Shaw , 2005 ) . Indeed , ERK2 shuttles in and out of nucleus upon activation/inactivation cycles of FastLightR-bRaf ( Figure 6D; Video 3 ) . Overall , our data show that the LightR approach can be applied to achieve light-mediated regulation of different protein kinases . To demonstrate the broad applicability of the LightR tool to other types of enzymes , beyond kinases , we engineered LightR-Cre recombinase . Cre-recombinase has become an essential tool in biomedical research because it allows for genetic recombination and induced activation or deletion of genes ( Abremski and Hoess , 1984; Abremski et al . , 1983 ) . We created four variants of Light-Cre fused to miRFP fluorescent protein that differ in their insertion site of LightR domain . The insertion sites were selected in four loops within Cre that are distant from the DNA binding site but connected to critical catalytic residues through an α-helix ( Figure 6E ) . To generate these four constructs , we replaced selected amino acids ( N60 , D153 , D189 and D278 ) with the LightR domain . The variant with LightR domain inserted at D153 residue in Cre showed activation in response to blue light ( Figure 6F; Figure 6—figure supplement 2 ) . This demonstrates that the LightR approach can be used broadly for the precise regulation of several types of enzymes . Our study describes an optogenetic approach that provides several advantages for the interrogation of signaling pathways and demonstrates its broad applicability to address important biological questions . The key features of this method include: ( 1 ) allosteric regulation of the enzymatic activity , ( 2 ) tight temporal control of activity with tunable kinetics , ( 3 ) local regulation of activity at a subcellular level , and ( 4 ) broad applicability to different enzymes . Importantly , unlike other approaches , LightR combines all these advantages in one tool , thus , simplifying application of optogenetics in biological research . We achieved direct regulation of enzymatic activity via an allosteric control by inserting the LightR switch into small loops within the protein structure . This modular design provides significant flexibility in the selection of the insertion site and allows for specific regulation of catalytic activity without compromising key functions of the protein such as its interactions with binding partners or its native localization in the cell ( Karginov et al . , 2010 ) . Furthermore , variation of the flexible linkers , both between the kinase and VVD as well as between the two VVDs , or changing the insertion site for LightR in the targeted enzyme could result in reversing or altering the regulatory mechanism of LightR , as was previously demonstrated for insertion of LOV2 domain in DHFR ( Reynolds et al . , 2011 ) . While we implemented only one linker type and only one insertion site in LightR-Src , this is a venue for future work to generate more modular tools . Application of LightR approach to different classes of enzymes suggests its broad applicability for the regulation of a wide variety of protein functions in living cells . Thus , unlike other light-regulated allosteric switches ( Wu et al . , 2009; Hongdusit et al . , 2020 ) , this method combines multiple important advantages of optogenetic regulation with its broad applicability to a wide range of enzymes . We showed that regulation by LightR domain is tunable , enabling different modes of regulation . LightR switch with slow off kinetics will be useful for long-term activation of LightR-enzymes; since brief periodic pulses of light will be sufficient to maintain activity while avoiding phototoxicity caused by long exposure to blue light . The FastLightR switch , on the other hand , is more suitable for studies that mimic transient , oscillatory or localized activation of a protein . It could also be used to study the kinetics of negative regulators of signaling pathways immediately after a signaling input is turned off . Activation of a LightR-enzyme requires low intensity light ( Figure 2—figure supplement 1B , 0 . 6 mW/cm2 ) , minimizing the phototoxic effects of blue light illumination . This level is lower than the intensity used in other optogenetic studies using light-sensitive switches to regulate enzymes ( Dagliyan et al . , 2016; Zhou et al . , 2017; Wang et al . , 2012 ) . The activation of LightR is limited to the blue light spectrum , thus enabling its multiplexing with other red-shifted optogenetic tools or FRET biosensors . Several optogenetic approaches for the regulation of protein kinases have been described previously; however , all had specific shortcomings . Several methods regulate the localizationof a kinase rather than its catalytic activity ( Kawano et al . , 2015; Graziano et al . , 2017; Mühlhäuser et al . , 2019; Moitrier et al . , 2019; Katsura et al . , 2015; Zhang et al . , 2014 ) . The method described by Zhou et al provides efficient control of kinases by light-regulated steric hindrance but it does not enable regulation on a subcellular level ( Zhou et al . , 2017 ) . Furthermore , one of the regulatory domains has to be inserted in the FG loop , a substrate-interacting region in several kinases including Src family kinases ( Zhou et al . , 2017; Shah et al . , 2016 ) . Thus , this approach may not be applicable to some critical kinases . The insertion site that we selected for LightR domain is intentionally positioned away from any substrate binding elements , which should minimize any steric hindrance by this domain . Dagliyan et al . , achieved regulation of Src through insertion of LOV domain at the same site that we used for the LightR approach ( Dagliyan et al . , 2016 ) . However , this method also did not achieve local regulation at a subcellular level . Furthermore , this approach only enabled inactivation of a kinase by illumination with light . To activate a kinase at a specific time , researches have to first inhibit the kinase by keeping cells under blue light for a significant period of time , then release the inhibition by switching the light off . This requirement increases the possibility of potential cytotoxic effect of light and makes the implementation of this tool problematic for studies where activation of a kinase is desired . While the insertion of LOV domain at a different site may result in light-induced activation , as was previously demonstrated for DHFR ( Reynolds et al . , 2011 ) , this has not been achieved for protein kinases at the moment . Thus , combined advantages of LightR enable broader application of this tool for interrogation of kinase signaling . Tight and efficient control of LightR allowed us to mimic fast signaling dynamics of Src kinase in cells , and to establish stimulation of different signaling patterns over time . Several identified targets link early effects of Src activation to focal adhesion regulation and membrane protrusion formation . Our phosphoproteomics analysis revealed early Src-dependent phosphorylation of talin-1 at Y26 and Y70 residues ( Figure 2D ) . These residues are located in the N-terminal F0 FERM domain of talin-1 ( Bouaouina et al . , 2008; Goult et al . , 2010 ) . When talin-1 is in the inactive closed conformation , the F0 domain remains exposed and is proposed to act as an early detector of changes in the environment ( Bouaouina et al . , 2008 ) . Thus , phosphorylation of talin-1 on Y26 and Y70 by Src may represent an early step in the inside-out integrin activation mechanism performed by talin ( Vinogradova et al . , 2002; Vinogradova et al . , 2004 ) . Later phosphorylation ( 1–5 min ) of proteins involved in formation of new focal adhesions and membrane protrusions ( such as lamellipodin , p130Cas , palladin , tensin-3 , cortactin , GIT1 , and paxillin ) correlates with the time at which we observed cell spreading and increased protrusion velocity ( Figures 4A and 5H ) . Elevated phosphorylation of many proteins at 1 hr is indicative of oncogenic transformation processes mediated by prolonged activation of Src ( Hunter and Sefton , 1980; Sefton et al . , 1980 ) . Our phosphoproteomics studies also revealed that continuous Src activity induces only transient activation of MAP kinases ERK1 and ERK2 ( Figure 2—figure supplement 3 ) . This suggests that short term Src activation induces activation of ERK1/2 whereas prolonged activity triggers inactivation of this pathway . Previous studies demonstrated that stimulation of cells with EGF results in transient activation of ERK1 and ERK2 ( Olsen et al . , 2006; Sasagawa et al . , 2005 ) . In the future studies , it will be interesting to study whether Src also plays a role in negative regulation of ERK1/2 downstream of EGF . Importantly , our approach allowed us to identify temporal categories of Src targets; distinguishing between targets that are phosphorylated at early , intermediate , or delayed time points . Such temporal compartmentalization can be elucidated by the optogenetic approach and provides unprecedented insights into the complexity and diversity of signaling networks . Periodic activation of kinases occurs during different cellular processes ( Kaimachnikov and Kholodenko , 2009; Roche et al . , 1995; Hilioti et al . , 2008; Jacquel et al . , 2009; Kholodenko , 2000; Maeda et al . , 2004 ) . Depending on the kinase and the physiologic context , periodicity of these cycles ranges from minutes to hours , which often determines the functional outcome ( Hilioti et al . , 2008; Jacquel et al . , 2009; Kholodenko , 2000; Maeda et al . , 2004; Karginov et al . , 2014 ) . Using FastLightR , we demonstrated that we can modulate activation cycles of protein kinases , Src ( Figures 3C and 4A , C; Video 1 ) and bRaf , ( Figure 6D; Video 3 ) with light . Tight regulation of FastLightR allows us to control the periodicity and the amplitude of these activation pulses . Therefore , FastLightR can be potentially applied to any enzyme of interest to directly reconstruct its oscillatory signaling and identify its functional role . The ability to regulate proteins on a subcellular level is one of the main capabilities desired in optogenetic tools . However , this has been challenging to achieve for methods employing steric hindrance and allosteric control as a mechanism of regulation ( Dagliyan et al . , 2016; Zhou et al . , 2017 ) . Efficiency and fast reversibility of FastLightR allowed us to control Src activity at a specific location in the cell and demonstrate that its activity is sufficient to induce local protrusion . One of the interesting observations is the fact that continuous local activation of Src induces waves of protrusions instead of causing persistent protrusions ( Figure 5E , H ) . Several groups have previously described periodic protrusions at the cell edge ( Tsai and Meyer , 2012; Giannone et al . , 2004; Machacek and Danuser , 2006 ) , and correlated this phenotype with periodic contractions due to phosphorylation of myosin light chain ( MLC ) ( Tsai and Meyer , 2012; Giannone et al . , 2004; Machacek and Danuser , 2006 ) . Our results demonstrate that both kinases that phosphorylate MLC ( ROCK and MLCK ) participate in the formation of the recurrent waves of protrusion in response to local Src activation . However , ROCK appears to be less important for maintaining the overall elevated protrusion velocity downstream of Src activation ( Figure 5—figure supplement 2B ) . Inhibition of MLCK resulted in much more noticeable reduction in average protrusion velocity ( Figure 5—figure supplement 2B ) . Previous studies suggested that MLCK predominantly acts at the cell periphery where it mediates formation of stable protrusions and persistent cell migration ( Totsukawa et al . , 2004 ) . Our data is consistent with these prior studies and suggest that MLCK is a predominant factor regulating cell protrusions downstream of Src . Several previous studies described approaches for the regulation of Cre recombinase activity by light ( Wu et al . , 2020; Morikawa et al . , 2020; Kawano et al . , 2016; Taslimi et al . , 2016; Sheets et al . , 2020; Kennedy et al . , 2010 ) . However , these methods are designed based on the split protein reassembly strategy , which has its limitations . Split proteins tend to reassemble spontaneously , leading to undesirable basal activity of engineered split Cre ( Dagliyan et al . , 2018; Liu and Tucker , 2017 ) . Moreover , when activated , not all split halves of the protein reassemble , which often leads to suboptimal activation levels of the engineered enzyme ( Dagliyan et al . , 2018 ) . Furthermore , this method requires co-expression of two proteins and , to achieve optimal results , both split halves must be co-expressed in equimolar ratios in a cell . LightR overcomes some of these limitations , because LightR-Cre is , by design , a single construct that expresses a complete Cre enzyme fused with LightR domain . Although high levels of LightR-Cre expression sometimes lead to low basal activity ( Figure 6—figure supplement 2 ) , this could be solved by optimizing the expression level of LightR-Cre and/or by modifying the linkers connecting LightR domain to the Cre recombinase . We believe that such optimization is feasible because we did not detect any basal activity for the LightR-kinases regardless of their expression level in cells . In summary , the LightR-approach to regulate enzyme activity demonstrates versatility and broad applicability to a variety of enzymes and can thus help unravel signaling pathways and networks as well as study defined biological processes that are direct consequences of the regulated enzymes . The following antibodies were used: anti-phospho-p130Cas ( Y249 ) ( BD Pharmigen cat . no . 558401 ) , anti-p130Cas ( BD Pharmigen cat . no . 610271 ) , anti-GFP ( Clontech , cat . no . 632381 ) , anti-paxillin ( Fisher Scientific , cat . no . BDB612405 ) , anti-phospho-paxillin ( Y118 ) ( Invitrogen , cat . no . 44–722G ) , anti-Src ( Santa Cruz , cat . no . 8056 ) , anti-GAPDH ( Ambion , cat . no . AM4300 ) , anti-Myc ( Millipore , cat . no . 05–724 ) , anti-phospho-MEK1/2 ( Ser17/221 ) ( Cell Signaling , cat . no . 9121 ) , anti-MEK1/2 ( Cell Signaling , cat . no . 9122 ) , anti-p44/42 ( ERK1/2 ) ( Cell Signaling , cat . no . 9102 ) , anti-p44/42 ( ERK1/2 ) ( Thr202/Tyr204 ) ( Cell Signaling , cat . no . 9101 ) . The reagents used were: IgG-coupled agarose beads ( Millipore , cat . no . IP04-1 . 5ML ) , Leupeptin hemisulfate ( Gold Biotechnology , cat . no . L-010–5 ) , Aprotinin ( Gold Biotechnology , cat . no . A-655–25 ) , Y-27632 dihydrochloride ( Millipore-Sigma , cat . no . Y0503 ) , MLCK Inhibitor Peptide 18 ( Cayman , cat . no . 224579-74-2 ) , 2X Laemmli Sample buffer ( BIO-RAD , cat . no . 161–0737 ) , 2-Mercaptoethanol ( Fisher Chemical , cat . no . 60-24-2 ) , Trypsin ( Promega , cat . no . V5113 ) , Trypsin-digested BSA for coating tubes ( Sigma , cat . no . A7906 ) , Fetal Bovine Serum ( Omega Scientific , cat . no . FB-01 ) . Coverslips for live imaging were purchased from ( ThermoFisher , cat . no . 25CIR-1 . 5 ) . The materials used were: C18 cartridges ( Waters , cat . no . WAT02350 ) , TMT11plex ( ThermoFisher , cat . no . A34807 ) , TMT10plex ( ThermoFisher , cat . no . 90406 ) , Fe-NTA spin columns ( ThermoFisher , cat . no . A32992 ) , glutathione sepharose ( GE Healthcare , cat . no . GE17-0756-01 ) . The following cell lines were used: HeLa cells ( ATCC , cat . No . CCL- 2 ) , Human embryonic kidney HEK293T cells ( ATCC , cat . no . CRL-3216 ) , and human embryonic kidney LinXE cell line ( derived from HEK 293 cells , gift from Klaus Hahn Lab , UNC ) ( Bravo-Cordero et al . , 2013 ) . All cell lines were cultured at 37°C and 5% carbon dioxide in DMEM medium supplemented with 10% FBS and 1 mM L-glutamine . All experiments were performed with cells grown for less than 20 passages after thawing . All cells were tested negative for Mycoplasma contamination . Cell lines identity was confirmed by the supplier using STR analysis . DH5α bacteria cells ( NEB , cat . No . C2987H ) were used for GST-paxillinN-C3 production . All LightR-Src structures used for molecular dynamics simulations were created using the Src catalytic domain from PDB: 1Y57 combined with dark and lit state VVD from PDB: 2PD7 and 3RH8 , respectively . Using UCSF Chimera ( version 1 . 12 ) software ( Pettersen et al . , 2004 ) , the asymmetric flexible linkers , GPGGSGG and GSGGPG , were respectively added to the N- and C- termini of VVD and the monomers were linked with a 22 amino acid flexible linker ( GGS ) 4G ( GGS ) 3 . These VVD constructs were then inserted into the active Src catalytic domain using Chimera’s structure editor . The flexible linkers were then refined using Modeller ( Sali and Blundell , 1993 ) . All structures were energy minimized in NAMD ( version 2 . 13 ) using a steepest descent gradient for 1000 steps ( Phillips et al . , 2005 ) . These structures were then equilibrated for 0 . 5 ns at increasing temperature from 60 K to 310 K , then for an additional 1 ns at 310 K , with generalized Born implicit solvent with an ion concentration of 0 . 15 M . All MD simulations were then carried out under these same conditions over 100 ns . Simulations for the LightR constructs were conducted in triplicate for a total simulation time of 300 ns per structure . Time points for the pre-equilibrated structure , before RMSD convergence , were discarded for analysis to ensure structure used was reasonable . Starting structures were qualitatively validated by evaluating RMSD convergence ( Figure 1—figure supplement 1A ) . The RMSD analysis and visualization was carried out using VMD ( version 1 . 9 . 3 ) and principal component analysis ( PCA ) was conducted using the NMWiz GUI for ProDY ( Bakan et al . , 2011 ) . Distances between the N and C-termini of the LightR domain as well as the length of the flexible linker when extended in the dark state were determined using the average position of these residues from the MD simulations . The averaged structure of the dark and lit state LightR-Src were opened in Chimera to obtain these distances with Chimera’s measurement tool . Amino acid sequence of LightR domain is: GPGGSGGHTLYAPGGYDIMGYLIQIMNRPNPQVELGPVDTSCALIL CDLKQKDTPIVYASEAFLYMTGYSNAEVLGRNCRFLQSPDGMVKP KSTRKYVDSNTINTMRKAIDRNAEVQVEVVNFKKNGQRFVNFLTM IPVRDETGEYRYSMGFQCETEGGSGGSGGSGGSGGGSGGSGGSHTLYAPGGYDIMGYLIQIMNRPNPQVELGPVDTSCALILCDLKQK DTPIVYASEAFLYMTGYSNAEVLGRNCRFLQSPDGMVKPKSTRK YVDSNTINTMRKAIDRNAEVQVEVVNFKKNGQRFVNFLTMIPVR DETGEYRYSMGFQCETEGSGGPG . Linkers are italicized . Bolded and underlined sequences are the first and second VVD sequences , respectively . LightR gene was codon optimized so that the two tandem VVD DNA sequences are as different as possible to make cloning using PCR easier . LightR sequence was designed so that the two VVD proteins were connected with a flexible twenty-two amino acid linker ( GGS ) 4G ( GGS ) 3 . LightR DNA sequence was ordered as a gBlock from Integrated DNA Technologies . The gBlock was amplified using PCR with forward primer encoding a GPGGSGG linker and a 24–28 nucleotide sequence that anneals upstream of the insertion site of interest . The reverse primer encodes a GSGGPG linker and 24–28 nucleotide sequence that anneals downstream of the insertion site of interest . The resulting PCR product of this reaction acts as a megaprimer that we use to insert LightR domain at a site of interest in a gene using a modification of QuickChange site-directed mutagenesis ( Karginov and Hahn , 2011 ) . LightR-Src-mCherry-myc construct was generated by using RapR-Src-mCherry-myc construct ( Klomp et al . , 2016 ) and replacing iFKBP insert with the LightR . In the resulting construct , LightR is replacing G288 in cSrc ( position in avian Src ) . Constitutively active CA-Src-mCherry-myc construct was generated by replacing Cerulean in CA-Src-Cerulean-myc construct ( Klomp et al . , 2016 ) with mCherry using a modification of QuickChange site-directed mutagenesis ( Karginov and Hahn , 2011 ) . Stargazin-iRFP670 was generated from the previously described stargazin-mCherry construct ( Karginov et al . , 2014 ) , by replacing mCherry with iRFP670 using a modification of QuickChange site-directed mutagenesis ( Karginov and Hahn , 2011 ) . pmiRFP670-N1 was a gift from Vladislav Verkhusha ( Addgene plasmid # 79987; http://n2t . net/addgene:79987; RRID:Addgene_79987 ) ( Shcherbakova et al . , 2016 ) . pCAG-iCre was a gift from Wilson Wong ( Addgene plasmid # 89573; http://n2t . net/addgene:89573; RRID:Addgene_89573 ) , and pcDNA3 . 1_Floxed-STOP-mCherry was a gift from Moritoshi Sato ( Addgene plasmid # 122963; http://n2t . net/addgene:122963; RRID:Addgene_122963 ) . Using a modification of QuickChange site-directed mutagenesis ( Karginov and Hahn , 2011 ) , iCre gene was cloned into the pmiRFP670-N1 backbone to obtain iCre-miRFP670 plasmid . Similarly , ERK2 gene from pCEFL-ERK2 ( a gift from Dr . Channing Der’s lab , UNC ) was cloned into mCherry-C1 backbone to obtain mCherry-ERK2 plasmid . We obtained GFP-Abl ( Homo sapiens ) construct ( gift from Dr . Steven Dudek , UIC ) and introduced P242E/P249E mutations to make it constitutively active using site-directed mutagenesis , bRaf-Venus construct bearing V600E mutation was a gift from Dr . John O’Bryan ( MUSC ) , LeGO-iV2 was a gift from Boris Fehse ( Addgene plasmid # 27344; http://n2t . net/addgene:27344; RRID:Addgene_27344 ) ( Weber et al . , 2008 ) , pMD2 . G was a gift from Didier Trono ( Addgene plasmid # 12259; http://n2t . net/addgene:12259; RRID:Addgene_12259 ) , psPAX2 was a gift from Didier Trono ( Addgene plasmid # 12260; http://n2t . net/addgene:12260; RRID:Addgene_12260 ) , GST-paxillinN-C3 construct was a gift from Dr . Michael Schaller , WVU ( Lyons et al . , 2001 ) . All DNA constructs were transfected transiently using Fugene 6 ( Promega Corporation ) transfection reagent according to the manufacturer protocol . N-terminal fragment of paxillin was purified following previously described procedure ( Lyons et al . , 2001 ) . Briefly , GST-paxillinN-C3 construct was expressed in DH5α bacteria cells following induction with 0 . 1 mM Isopropyl β-D-1-thiogalactopyranoside for 4 hr . Bacterial pellet was resuspended in 30 ml of TETN buffer ( 20 mM TRIS pH 8 , 100 mM NaCl , 1 mM EDTA , 0 . 5% Triton X100 ) and lysed by sonication . GST-paxillinN-C3 was purified from the cleared lysates by affinity chromatography using Glutathione Sepharose following previously described protocol ( Lyons et al . , 2001 ) . A detailed protocol for this experiment was previously described ( Cai et al . , 2008 ) . Briefly , Src kinase constructs bearing an mCherry and a myc tandem tags at the C-terminus were transiently overexpressed in LinXE cells . Cells were exposed to continuous blue light ( 3 mW/cm2 , 465 nm wavelength ) for the indicated times or kept in the dark . Lysates were then collected under red light illumination ( to prevent activation of LightR ) using the lysis buffer ( 20 mM HEPES-KOH , pH 7 . 8 , 50 mM KCl , 1 mM EGTA , 1% NP40 , 1 mM NaF , 0 . 2 mM Na3VO4 , aprotinin 16 µg/ml , and Leupeptin hemisulfate 3 . 2 µg/mL ) . Lysates were centrifuged at 4000 rpm , 4°C , for 10 min , and the cleared lysates were incubated with ProteinG sepharose beads conjugated with the anti-myc antibody ( 4A6 from Millipore-Sigma ) for 1 . 5 hr at 4°C . Beads were then washed with wash buffer ( 20 mM Hepes-KOH , pH 7 . 8 , 100 mM NaCl , 50 mM KCl , 1 nM EGTA , 1% NP40 ) and then with kinase reaction buffer ( 25 mM HEPES , pH 7 . 5 , 5 M MgCl2 , 0 . 5 mM EGTA , 0 . 005% BRIJ-35 ) . The beads were resuspended in kinase reaction buffer and incubated with 0 . 1 mM ATP and 0 . 05 mg/ml purified N-terminal fragment of paxillin ( GST-paxillinN-C3 ) at 37°C for 10 min . The reaction was terminated by adding 2X Laemmli sample buffer with 5% v/v 2-Mercaptoethanol then incubating at 110°C for 5 min . The phosphorylation of paxillin was examined by western blotting . LinXE cells grown in 3 cm plates to 60–80% confluency were transfected with the LightR construct of interest using Fugene six transfection reagent as recommended by the manufacturer . Transfected cells were incubated overnight at 37°C in the dark to prevent unwanted activation . The following day , the cells were exposed to blue light by placing them 10 cm above an HQRP LED Plant Grow Panel Lamp System ( 3 mW/cm2 , 465 nm wavelength ) . Temperature was maintained during incubation times by placing the setup inside a tissue culture incubator ( Figure 2—figure supplement 1A ) . Light was shining continuously for the desired time . In experiments that required cycles of activation and inactivation , we manually unplugged the LED panel during inactivation times and plugged it back in during activation times before we lysed the cells . At the end of the experiment time course , media was aspirated and cells were lysed under safe red-lights with 700 µl of 2X Laemmli sample buffer containing 5% v/v 2-Mercaptoethanol . Cell lysates were collected and incubated at 110°C for 5 min . 10–15 µL of the cell lysates were analyzed by a western blot to probe for phosphorylation of endogenous proteins . To assess LightR-Cre activity we used the previously described Floxed-STOP-mCherry reporter system ( Kawano et al . , 2016 ) . LinXE cells were co-transfected overnight with LightR-Cre-iRFP670 and Floxed-STOP-mCherry DNA constructs ( 1:9 ratio ) and kept in the dark . The next day , cells were pulsed with light using the HQRP LED Plant Grow Panel Lamp System ( 3 mW/cm2 , 465 nm wavelength ) . A microcontroller ( Arduino Uno ) and power relay ( IoT Relay , Digital Data Loggers INC . ) were used to turn on the LED panel for 2 s every 10 s . Cells were then imaged using EVOS Auto 2 Invitrogen fluorescence microscope using 20 × air objective to analyze the expression of mCherry reporter . Using LighR-Src-mCherry-myc plasmid as a template , LightR-Src-mCherry-myc gene was PCR-amplified with primers that introduced restriction sites NotI and BsiWI on the 5’ and 3’ end , respectively . This PCR product and the LeGO-iV2 plasmid were digested with NotI and BsrGI and then ligated to generate a LightR-Src-mCherry-myc lentiviral construct . This construct was then co-transfected with pMD2 . G , a plasmid expressing VSV-G lentivirus envelope protein , and psPAX2 , a second-generation lentiviral packaging plasmid , into HEK 293 T cells ( ATCC CRL-3216 ) . After 1–3 days , the conditioned media was centrifuged ( 1000 × g , 10 min , 25°C ) , and virus-containing supernatant was used to directly infect HeLa cells . Transduced HeLa cells were sorted via FACS by selecting the brightest 20 percentile of mCherry-expressing cells . In order to quantify the response of cell edge to the local kinase activation with light , we performed the following steps of the analysis: See Supplementary file 1 .	Cells need to sense and respond to their environment . To do this , they have dedicated proteins that interpret outside signals and convert them into appropriate responses that are only active at a specific time and location within the cell . However , in many diseases , including cancer , these signaling proteins are switched on for too long or are active in the wrong place . To better understand why this is the case , researchers manipulate proteins to identify the processes they regulate . One way to do this is to engineer proteins so that they can be controlled by light , turning them either on or off . Ideally , a light-controlled tool can activate proteins at defined times , control proteins in specific locations within the cell and regulate any protein of interest . However , current methods do not combine all of these requirements in one tool , and scientists often have to use different methods , depending on the topic they are researching . Now , Shaaya et al . set out to develop a single tool that combines all required features . The researchers engineered a light-sensitive ‘switch’ that allowed them to activate a specific protein by illuminating it with blue light and to deactivate it by turning the light off . Unlike other methods , the new tool uses a light-sensitive switch that works like a clamp . In the dark , the clamp is open , which ‘stretches’ and distorts the protein , rendering it inactive . In light , however , the clamp closes and the structure of the protein and its activity are restored . Moreover , it can activate proteins multiple times , control proteins in specific locations within the cell and it can be applied to a variety of proteins . This specific design makes it possible to combine multiple features in one tool that will both simplify and broaden its use to investigate specific proteins and signaling pathways in a broad range of diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology", "tools", "and", "resources" ]	2020	Light-regulated allosteric switch enables temporal and subcellular control of enzyme activity
Tryparedoxin peroxidases , distant relatives of glutathione peroxidase 4 in higher eukaryotes , are responsible for the detoxification of lipid-derived hydroperoxides in African trypanosomes . The lethal phenotype of procyclic Trypanosoma brucei that lack the enzymes fulfils all criteria defining a form of regulated cell death termed ferroptosis . Viability of the parasites is preserved by α-tocopherol , ferrostatin-1 , liproxstatin-1 and deferoxamine . Without protecting agent , the cells display , primarily mitochondrial , lipid peroxidation , loss of the mitochondrial membrane potential and ATP depletion . Sensors for mitochondrial oxidants and chelatable iron as well as overexpression of a mitochondrial iron-superoxide dismutase attenuate the cell death . Electron microscopy revealed mitochondrial matrix condensation and enlarged cristae . The peroxidase-deficient parasites are subject to lethal iron-induced lipid peroxidation that probably originates at the inner mitochondrial membrane . Taken together , ferroptosis is an ancient cell death program that can occur at individual subcellular membranes and is counterbalanced by evolutionary distant thiol peroxidases . Ferroptosis is characterized by the iron-dependent accumulation of cellular lipid hydroperoxides to lethal levels . This form of regulated cell death has been implicated in the pathology of degenerative diseases ( e . g . Alzheimer’s , Huntington’s and Parkinson’s diseases ) , cancer , ischemia-reperfusion injury and kidney degeneration ( for recent reviews see [Doll and Conrad , 2017; Stockwell et al . , 2017; Galluzzi et al . , 2018] ) . The term ferroptosis was coined in 2012 to describe a form of death in cancer cells induced by the small molecule erastin ( Dixon et al . , 2012; Dixon et al . , 2014; Cao and Dixon , 2016 ) . Consequence is an inhibition of glutathione biosynthesis and thus of glutathione peroxidase 4 ( GPx4 ) activity . Of the eight glutathione peroxidases in mammalian cells , GPx4 is the only one that is able to detoxify lipid hydroperoxides even within membranes ( Thomas et al . , 1990; Seiler et al . , 2008; Brigelius-Flohé and Maiorino , 2013 ) . Inactivation of GPx4 by chemicals like for example RSL3 induces ferroptosis ( Yang et al . , 2014 ) ; and inducible GPx4-deletion in mice revealed that ferroptosis is a pervasive and dynamic form of cell death also in non-transformed cells ( Friedmann Angeli et al . , 2014 ) . Heat-stressed plants can also undergo a ferroptosis-like cell death ( Distéfano et al . , 2017 ) . A hallmark of a cell death occurring exclusively by ferroptosis is its suppression by iron chelators ( Dixon et al . , 2012 ) or lipophilic antioxidants ( Seiler et al . , 2008; Krainz et al . , 2016 ) as well as by the depletion of polyunsaturated fatty acids in membrane lipids ( Doll et al . , 2017; Kagan et al . , 2017 ) . African trypanosomes ( Trypanosoma brucei species ) are the causative agents of human sleeping sickness and Nagana cattle disease . The obligate free living protozoan parasites multiply as bloodstream ( BS ) form in the mammalian host and as procyclic ( PC ) form in the tsetse fly vector . BS T . brucei rely exclusively on glycolysis for energy supply and have an only rudimentary mitochondrion , whereas in the PC stage , the single mitochondrion is fully elaborated and the parasites gain ATP via oxidative phosphorylation . Trypanosomes have an unusual thiol redox metabolism that is based on trypanothione [N1 , N8-bis ( glutathionyl ) spermidine , T ( SH ) 2 ) ] and the flavoenzyme trypanothione reductase ( TR ) ( Krauth-Siegel and Leroux , 2012; Manta et al . , 2013; Manta et al . , 2018 ) . The trypanothione system delivers the reducing equivalents for a variety of crucial pathways . Most of the reactions are mediated by tryparedoxin ( Tpx ) , an essential distant member of the thioredoxin protein family ( Comini et al . , 2007 ) . Trypanosomes lack catalase . Hydroperoxides are detoxified by 2-Cys-peroxiredoxins ( Tetaud et al . , 2001; Budde et al . , 2003; Wilkinson et al . , 2003 ) and non-selenium glutathione peroxidase-type ( Px ) enzymes ( Hillebrand et al . , 2003; Wilkinson et al . , 2003; Schlecker et al . , 2005 ) . Whereas the peroxiredoxins use hydrogen peroxide and peroxynitrite as main substrates ( Thomson et al . , 2003; Trujillo et al . , 2004 ) , the Px-type enzymes preferably detoxify lipid-derived hydroperoxides ( Diechtierow and Krauth-Siegel , 2011 ) . With NADPH as ultimate electron donor , the reducing equivalents flow via TR , T ( SH ) 2 , and tryparedoxin onto the peroxidases which therefore have been named tryparedoxin peroxidases ( Castro and Tomás , 2008; Krauth-Siegel and Comini , 2008; Krauth-Siegel and Leroux , 2012; Manta et al . , 2013 ) . Tandemly arranged genes encode three virtually identical Px-type enzymes ( Px I , II , and III ) in T . brucei ( Hillebrand et al . , 2003 ) . RNA-interference against the Px-type enzymes results in a severe growth defect in both BS and PC T . brucei ( Wilkinson et al . , 2003; Schlecker et al . , 2005 ) . Proliferation of the Px-depleted BS parasites can , however , be restored by supplementing the medium with the vitamin E-analog Trolox [ ( ± ) −6-hydroxy-2 , 5 , 7 , 8-tetramethylchromane-2-carboxylic acid] ( Diechtierow and Krauth-Siegel , 2011 ) . The same is true for cells lacking GPx4 , the closest related enzyme in mammals ( Seiler et al . , 2008 ) . Selective knockout ( KO ) of the gene encoding the mitochondrial Px III has only a mild and transient effect on the in vitro proliferation of BS T . brucei and the mutant parasites are fully infectious in the mouse model . In contrast , cells that lack the cytosolic peroxidases Px I and II die after transfer into Trolox-free medium ( Diechtierow and Krauth-Siegel , 2011; Hiller et al . , 2014 ) . The lethal phenotype of the BS parasites starts at the lysosome and is closely linked to the endocytosis of iron-loaded host transferrin ( Hiller et al . , 2014 ) . In the PC insect stage of T . brucei , the presence of either the cytosolic or the mitochondrial peroxidases is sufficient for viability . However , deletion of the complete peroxidase ( px I-III ) locus is lethal . The cell death can be reverted by exogenous Trolox , deferoxamine ( Dfx ) or MitoQ which suggested an iron-induced lipid peroxidation that may start at the mitochondrion ( Schaffroth et al . , 2016 ) . Here we show that the death program in the peroxidase-deficient PC T . brucei closely resembles ferroptosis in mammalian cells . Integrity and viability of the parasites are preserved by known ferroptosis inhibitors . Strikingly , sensors for mitochondrial oxidants and chelatable iron as well as overexpression of an iron-dependent superoxide dismutase in the mitochondrion attenuate cell lysis . Transmission electron microscopy of the Px I-III knockout ( KO ) cells reveals mitochondrial matrix condensation and formation of enlarged cristae . Our data show that oxidants and iron in the mitochondrial matrix play a critical role in the process and that ferroptosis is an evolutionary ancient cell death program that is prevented by distant thiol peroxidases . The Px-type tryparedoxin peroxidases protect African trypanosomes against fatal membrane damages ( Diechtierow and Krauth-Siegel , 2011; Hiller et al . , 2014; Schaffroth et al . , 2016 ) . In PC T . brucei , deletion of the px I-III locus is lethal but the mutant cells proliferate like wildtype parasites if the medium is supplemented with either 100 µM of Trolox or Dfx ( Schaffroth et al . , 2016 ) . The minimum concentration to ensure full viability of the Px I-III KO cells is 50 µM Trolox and 100 µM Dfx , respectively . 10 µM Trolox and 20 µM Dfx alone are not sufficient but in combination , restore cell viability ( Figure 1—figure supplement 1 ) . Evidently , partial complexation of iron by Dfx lowers the concentration of Trolox that is required to protect the cells . The concerted action of Dfx and Trolox is probably related to the fact that free iron ions react with hydroperoxides generating highly deleterious alkoxy radicals . The protective effect of radical trapping agents is usually associated with their reactivity towards peroxy radicals ( Shah et al . , 2018 ) . Yet , this reaction would regenerate hydroperoxides and thus , in peroxidase-deficient cells maintain the vicious cycle . As shown recently , Trolox is a more powerful scavenger of alkoxy radicals compared to peroxy radicals ( Alberto et al . , 2013 ) . The reaction converts alkoxy radicals into the respective alcohols which more likely explains the protecting effect of Trolox and probably other radical trapping agents as well . At 1 µM , α-Tocopherol ( α-Toc ) fully reverts the lethal phenotype of mammalian cells lacking GPx4 ( Seiler et al . , 2008 ) . Therefore , we asked if the biologically most active form of vitamin E is also more efficient than Trolox in protecting the Px I-III KO cells . Unexpectedly , even 100 µM α-Toc was not sufficient to fully prevent cell lysis . In a second approach , the cells were pre-loaded with 10 µM α-Toc in the presence of 100 µM Trolox and then transferred into Trolox-free medium ± α-Toc . In this case , the Px I-III KO cells even slightly proliferated within the 6 hr observation time , remarkably also in the absence of further exogenous α-Toc . Apparently , uptake of α-Toc by the parasites is slow compared to Trolox , but after proper incorporation , indeed α-Toc is the superior protecting agent . If this is due to the ability of α-Toc to integrate in the phospholipid bilayer of biological membranes is not clear . As reported recently , an analog in which the isoprenoid side chain is truncated to a methyl group proved to be an even better antioxidant than α-Toc ( Zilka et al . , 2017 ) . The finding that lysis of the Px I-III KO T . brucei was prevented by either a lipophilic antioxidant or an iron chelator was reminiscent of ferroptosis ( Dixon et al . , 2012; Friedmann Angeli et al . , 2014; Doll and Conrad , 2017 ) . Most potent inhibitors of this regulated cell death described so far are ferrostatin-1 ( Fer-1 ) and liproxstatin-1 ( Lpx-1 ) which protect mammalian cells at nanomolar concentrations . Indeed also in trypanosomes , 100 nM Fer-1 or 200 nM Lpx-1 completely abolished the lethality of the Px I-III KO cells ( Figure 1A ) . Mechanistically , the compounds act as radical-trapping agents which in phospholipid bilayers are significantly more potent antioxidants than α-Toc ( Zilka et al . , 2017 ) . Mammalian ferroptotic cells show a time-dependent increase of both lipid and soluble reactive oxygen species ( ROS ) production ( Dixon et al . , 2012; Yang et al . , 2014 ) . Thus , the Px I-III KO cells were incubated in medium ± Trolox and treated with H2DCFDA , a broadly used dye for detection of cellular oxidants ( Kuznetsov et al . , 2011 ) . After 2 hr in Trolox-free medium , DCF fluorescence was strongly increased indicating that the parasites are subject to general oxidative stress ( Figure 1B ) . Next we measured lipid peroxidation using BODIPY 581/591 C11 , a redox-sensitive dye that integrates into membranes and shifts its fluorescence from red to green upon oxidation ( Pap et al . , 1999 ) . Px I-III KO parasites were incubated with BODIPY in medium that was supplemented with Trolox , Dfx , Fer-1 and Lpx-1 , respectively , or lacked any protecting agent and analyzed by flow cytometry . Cells that were kept for 2 hr in non-supplemented medium showed strong green fluorescence whereas cells incubated in the presence of the radical scavengers or iron chelator were protected from lipid peroxidation ( Figure 1C ) . Taken together , in trypanosomes , a cell death that is due to impaired removal of lipid-derived hydroperoxides can be prevented by known ferroptosis inhibitors . RSL3 ( RAS-selective lethal ) is a ferroptosis-inducing agent . The ( 1S , 3R ) -RSL3 isomer kills RAS-transformed tumorigenic fibroblast cell lines with EC50-values of 10 nM and irreversibly inactivates GPx4 . The other three diastereomers display EC50-values of 2 . 5 to 5 µM ( Yang et al . , 2014 ) . BS T . brucei were cultured for up to 72 hr in the presence of different concentrations of ( 1S , 3R ) -RSL3 or the racemic mixture and then subjected to ATPlite measurements . After 24 and 48 hr , EC50-values of 2 and 4 µM were obtained which increased about two-fold after 72 hr probably because of long-term instability of the compounds in the medium . Thus , RSL3 was trypanocidal but selectivity for the ( 1S , 3R ) -isomer could not be detected ( Supplementary file 1 ) . A lack of stereo-specificity and EC-values in the low micromolar range were reported for non-transformed fibroblast cell lines as well ( Yang et al . , 2014 ) . To further assess a putative interaction of RSL3 with the peroxidases , the analysis was conducted in the presence of Trolox , Lpx-1 or Fer-1 . Under these conditions , viability and proliferation of T . brucei is independent of the Px-type enzymes ( Figure 1A ) . The EC50-values obtained in the presence or absence of the radical-trapping antioxidants were virtually the same which implies that the trypanocidal activity of RSL3 is not related to inhibition of the peroxidases . This may , at least partially , be due to the fact that the parasite peroxidases have an active site cysteine instead of the selenocysteine in GPx4 . Mammalian mouse embryonic fibroblasts ( MEF and PFa1 cells ) in which authentic GPx4 is replaced by a Cys-mutant are much less sensitive towards ( 1S , 3R ) -RSL3 than the respective wildtype cells ( Ingold et al . , 2018 ) . RSL3 carries a chloroacetamido group . In a large scale screen against the peroxidase cascade of T . brucei , several compounds with this substituent proved to be trypanocidal and to inactivate Tpx ( Fueller et al . , 2012 ) . All protein components of the parasite peroxidase system , TR , Tpx and the Px-type enzymes , are essential and possess reactive cysteine residues . Firstly , RSL3 was studied as putative covalent inhibitor of TR . 1 µM reduced TR was incubated for up to 2 hr with 100 µM RSL3 racemate as described in Materials and methods . The activity of TR remained constant ruling out any inactivation of the reductase . Next , the effect of RSL3 on the peroxidase cascade was studied . The mixture of NADPH , T ( SH ) 2 , TR , Tpx and Px was treated with RSL3 . After different times , H2O2 was added and NADPH consumption was followed . Finally , a mixture containing all components except the peroxidase was incubated with RSL3 and the reaction started by adding Px and H2O2 . In both approaches , RSL3 caused a time-dependent decrease of the activity but the degree of inactivation was identical irrespective of the presence or absence of the peroxidase in the pre-incubation mixture ( Supplementary file 1 ) . This strongly suggested that RSL3 is a time-dependent inhibitor of Tpx . The parasite-specific essential oxidoreductase is a distant relative of thioredoxins and glutaredoxins ( Comini et al . , 2007 ) . Tpx transfers reducing equivalents from trypanothione not only to the peroxidases but also methionine-sulfoxide reductase and , probably most importantly , ribonucleotide reductase ( Dormeyer et al . , 2001 ) . Thus , inhibition of Tpx likely affects the synthesis of DNA precursors which may be the main reason for the trypanocidal action of RSL3 . The PC Px I-III KO cells were incubated in medium ± Trolox , treated with the mitochondrial membrane potential-sensitive MitoTracker Red or propidium iodide ( PI ) , as indicator of plasma membrane integrity , and subjected to flow cytometry . When kept for 4 hr in the presence of Trolox or directly after transfer into standard medium ( 0 hr – Trolox ) , the majority of cells displayed normal forward scatter ( FSC ) and side scatter ( SSC ) ( Figure 2—figure supplement 1 ) . After 1 or 2 hr in the absence of a protecting agent , most of the cells had reduced SSC and slightly increased FSC probably due to their reduced motility and altered morphology . From 2 hr onwards , a third population arose that comprised severely damaged or dead cells . In accordance with previous fluorescence microscopy studies ( Schaffroth et al . , 2016 ) , already 1 hr after Trolox-withdrawal , the MitoTracker Red signal was reduced and after 2 hr had reached the minimal value . In contrast , the PI fluorescence remained at the basal level . Only when the cells were kept for ≥3 hr without the antioxidant the PI staining strongly increased . Taken together , in the Px I-III-deficient cells , loss of the mitochondrial membrane potential clearly precedes plasma membrane disintegration . To elucidate if/how the loss of the mitochondrial membrane potential affects the morphology of the organelle , the Px I-III KO cells were studied by immunofluorescence microscopy . After 1 hr in Trolox-free medium , many cells still displayed a proper MitoTracker signal but surprisingly weak immune staining for the mitochondrial matrix 2-Cys-peroxiredoxin ( mtTXNPx ) ( Figure 2A ) . After 2 hr in the absence of Trolox , the majority of cells had lost the MitoTracker signal and the mtTXNPx antibodies visualized some bright spots in addition to the faint mitochondrial staining . Respective observations were made when antibodies against two other matrix proteins , lipoamide dehydrogenase and acetate-succinate-CoA-transferase , were used ( not shown ) . The reasons for the impaired and later punctuated staining of mitochondrial matrix proteins in the fixed and permeabilized cells are not clear . One may speculate that the antibody penetration was hampered and later , upon progressive condensation of the matrix , the proteins became concentrated within distinct areas . When antibodies against the voltage-dependent anion channel ( VDAC ) were used , this phenomenon was not observed . Instead , cells that after 1 hr in the absence of Trolox had already lost their MitoTracker Red signal , retained the VDAC staining ( Figure 2B ) . Even after 2 hr , when the majority of cells no longer showed MitoTracker staining and had an overall swollen cell body , the antibodies against the outer mitochondrial membrane protein still visualized the tubular mitochondrion . To get a deeper insight in the morphological changes , the Px I-III KO cells were subjected to transmission electron microscopy . Cells kept in the presence of Trolox or for only 30 min in Trolox-free medium were indistinguishable from wildtype parasites . These cells are characterized by a highly elongated cell body with densely packed cytoplasm and distinguishable subcellular structures such as the nucleus , mitochondrion , Golgi apparatus , ER , acidocalcisomes , glycosomes and lipid droplets . Staining of the mitochondrion and cristae was comparable or even lighter than that of the cytoplasm ( Figure 3A ) . When the Px I-III KO cells were incubated for ≥1 hr in Trolox-free medium , cells appeared that still had an elongated shape , but , compared to the controls , lighter cytoplasm and a more electron-dense mitochondrion ( Figure 3B and C , Figure 3—figure supplement 1 ) . Both morphological changes appeared to be linked as cells with darkened mitochondrion but normal cytosol or , vice versa , with normal mitochondrion but lighter cytosol were hardly or not detectable ( Figure 3D ) . Other organelles were unaffected ( Figure 3B , Figure 3—figure supplement 1 ) . The lighter cytosol is probably due to the gradual increase of the cell volume . Indeed , upon prolonged incubation in the absence of the antioxidant , light microscopy of the Px I-III KO cells revealed an increase of swollen cells ( not shown ) . The dark mitochondria contained a growing number of less electron-dense area ( Figure 3C ) . These white structures appeared within the organelle but were never found to protrude out into the cytosol . Notably , high magnification revealed three membranes that surrounded the white bleb ( Figure 3C , insert ) . This strongly suggested that the outer mitochondrial membrane remained intact and the blebs were dilated cristae . The phenotype was reminiscent of a condensed matrix and enlarged intermembrane space as observed in isolated mitochondria after transfer into high osmotic medium ( Cortese et al . , 1991 ) . After 2 hr in Trolox-free medium , single cells were found that showed the release of vesicles from the plasma membrane ( Figure 3—figure supplement 1 ) . MitoSOX Red is widely used as indicator for mitochondrial superoxide production . Oxidation of the compound and binding of the oxidation products to nucleic acids is associated with strong red fluorescence . To assess the cellular localization of the sensor , Px I-III KO cells were treated with MitoSOX in Trolox-supplemented medium , incubated in medium ± Trolox , stained with MitoTracker Green and Hoechst 33342 and subjected to fluorescence microscopy . Cells in Trolox-supplemented medium showed a very small single red dot which co-localized with the DAPI signal for the kinetoplast and thus binding of some probably photo-oxidized sensor to the mitochondrial DNA ( Figure 4A , upper panel ) . When kept for 2 hr in Trolox-free medium , most of the MitoSOX-treated Px I-III KO cells had still a normal morphology but many of them displayed a more intense kinetoplast fluorescence ( Figure 4A , lower panels ) . A nuclear staining was not observed . This strongly suggests that at the beginning of the lethal process MitoSOX senses oxidants that are generated within the mitochondrial matrix . Swollen cells that appear upon prolonged incubation without protecting agents revealed an overall week red fluorescence and intense staining of the kinetoplast or nucleus or both structures . Evidently , when the mitochondrial membrane potential is abrogated , MitoSOX leaks out and loses its specificity for the mitochondrial matrix . As proof of principle , Px I-III KO cells in Trolox-supplemented medium were loaded for 10 min with MitoSOX , treated with 2 µM antimycin and subjected to flow cytometry . In agreement with data published for other cells ( Mukhopadhyay et al . , 2007 ) , the treatment resulted in a 5-fold fluorescence increase ( not shown ) . The Px I-III KO cells were then treated ± MitoSOX in Trolox-supplemented medium , transferred into medium ± Trolox , incubated for up to 5 hr and treated with DAPI . When kept in the presence of Trolox , the MitoSOX-loaded cells displayed a basal fluorescence comparable to the auto-fluorescence of entirely untreated cells ( -Trolox , - MitoSOX ) . Incubation in Trolox-free medium resulted in increased red fluorescence ( Figure 4B , left ) . After 3 or 5 hr , the cells displayed a 3-fold higher MitoSOX fluorescence than the Trolox-supplemented control ( Figure 4C ) . Strikingly , after 5 hr in the absence of Trolox but presence of MitoSOX , just a minute fraction of cells showed increased DAPI fluorescence . Only when incubated in the absence of both Trolox and MitoSOX , the cells incorporated DAPI ( Figure 4B , right ) . To further evaluate the putative protecting effect of MitoSOX , light scattering of the Px I-III KO cells was followed . As expected , cells kept in the presence of Trolox had normal FSC and SSC ( P1 population ) whereas those in standard medium showed low FSC and SSC ( P3 ) due to severe damage . When pre-loaded with MitoSOX , indeed , a considerable portion of the Px I-III KO cells in Trolox-free medium remained in the P1 or P2 fractions even after 5 hr of incubation ( Figure 4D ) . Next , we wanted to dissect which cell fraction was mainly responsible for the oxidant production . After 5 hr in Trolox-free medium , all three sub-populations had increased MitoSOX fluorescence but only the P3 fraction showed high DAPI fluorescence in accordance with the presence of severely damaged or dead cells ( Figure 4—figure supplement 1 ) . The strongest increase in MitoSOX fluorescence was observed for the P2 population which suggests that oxidant production was highest in cells with impaired motility and/or morphological alterations but still intact plasma membrane ( DAPI-negative ) . The data indicate that in the peroxidase-deficient parasites production of soluble oxidants starts within the mitochondrial matrix and then rapidly spreads all over the cell and MitoSOX acted as both oxidant sensor and protecting antioxidant . Finally , putative changes in the cellular ATP level were measured . The first analysis was carried out in SDM-79 medium . This could be the reason why lysis of the Px I-III KO cells in the absence of a protecting agent was slightly delayed when compared to studies in MEM-Pros medium . After 2 hr in the absence of Trolox , the ATP level had dropped by more than 50% , but virtually all cells retained normal morphology ( Figure 5A ) . When only highly motile cells were considered , the loss in ATP essentially mirrored the decline in the cell number . Since flagellar movement requires high levels of ATP , motility is expected to be rapidly affected upon ATP depletion ( Langousis and Hill , 2014 ) . Thus , a substantial fraction of parasites ceased to move but retained a virtually normal shape even though the ATP level was already significantly decreased . On the other hand , a small number of cells appeared to be swollen but still had a beating flagellum . This may be due to the degree to which the ATP level drops in different areas of the highly elongated cell . In an individual parasite , ATP depletion may primarily affect flagellum beating , which originates at its tip , or the ion pumps in the plasma membrane . To rule out any bias upon cell counting , the experiment was repeated using MEM-Pros medium and PI staining to follow cell viability . After 1 hr in Trolox-free medium , the cellular ATP had dropped by 50% but no PI staining was observed ( Figure 5B ) . After 3 hr , the cells incorporated PI . At this time point the remaining ATP level was diminished to 20% of the control . Taken together , in the absence of a protecting agent , PC Px I-III KO cells rapidly lose ATP . This correlates with an impaired motility and altered morphology of the cells but not an immediate plasma membrane leakage . As shown in Figure 1 , in the absence of a lipophilic antioxidant or iron chelator , Px I-III KO cells undergo lipid peroxidation . This analysis did , however , not allow to identify the subcellular membrane primarily affected . MitoPerOx is a derivative of BODIPY 581/591 C11 that is taken up into mitochondria of living cells where it is targeted to the interior surface of the inner membrane ( Prime et al . , 2012 ) . To evaluate the subcellular localization of the sensor , Px I-III KO cells in Trolox-supplemented medium were treated with MitoPerOx and stained with MitoTracker Green . Life cell fluorescence microscopy revealed a perfect overlay of both signals confirming a strong enrichment of MitoPerOx in the mitochondrion ( Figure 6A ) . The cells were then incubated in medium ± Trolox that was supplemented with either BODIPY or MitoPerOx . Both sensors revealed a time-dependent fluorescence increase in accordance with lipid peroxidation ( Figure 6B ) . In the MitoPerOx-treated sample , after 30 min in Trolox-free medium , cells with increased fluorescence appeared whereas in the BODIPY-treated cells , the fluorescence started to increase after 60 min ( Figure 6C ) . Differences in the sensitivity of the probes can be ruled out . The use of 2 µM BODIPY and 100 nM MitoPerOx ensured a very similar overall fluorescence . Thus , the mitochondria-targeted probe was slightly faster oxidized than the untargeted sensor reporting on all cellular membranes which suggests that lipid peroxidation starts within the matrix-facing leaflet of the inner mitochondrial membrane . As shown in Figure 4 , the PC Px I-III KO cells generate mitochondrial oxidants but the precise nature of the products is not known . The red fluorescence measured could arise from reaction of MitoSOX with superoxide but also other oxidants such as iron/H2O2 ( Kalyanaraman et al . , 2012 ) . African trypanosomes express four iron-superoxide dismutases of which two ( SODA and SODC ) are mitochondrial proteins . BS T . brucei in which the mRNA of SODA is down-regulated exhibit increased sensitivity towards the superoxide-inducing agent paraquat ( Wilkinson et al . , 2006 ) . The coding region of SODA was cloned into a vector that allows the tetracycline ( Tet ) -inducible expression of the protein with C-terminal myc2-tag . PC Px I-III KO cells were transfected with the pHD1700/sodA-c-myc2 construct and five cell lines obtained by serial dilutions ( for details see Materials and methods ) . Western blot analysis revealed for the induced Px I-III KO/SODA-myc cells expression of SODA-myc but , to a low level , also in the non-induced cells , indicating leaky expression ( Figure 7A ) . Since antibodies against the authentic T . brucei SODA were not available and those against the T . cruzi ortholog did not detect the protein in total lysates of T . brucei , the cellular level of overexpression could not be assessed . Immunofluorescence microscopy of the Px I-III KO/SODA-myc cells with myc antibodies revealed a perfect overlay with the MitoTracker Red signal ( Figure 7B ) . In comparison , antibodies against cytosolic peroxiredoxin evenly stained the whole cell body . The apparent partial co-localization of SODA-myc with the cytosolic marker is explained by the fact that the mitochondrion occupies 25% of the total cell volume ( Böhringer and Hecker , 1975 ) . In medium lacking any protecting agent , the non-induced Px I-III KO/SODA-myc cells died within 4 hr as did the parental Px I-III KO cells . The level of SODA-myc in the non-induced cells was not sufficient for protection . The induced Px I-III KO/SODA-myc cells displayed a slightly delayed lysis ( Figure 7C ) . In the presence of 25 µM Dfx , a concentration that protected the non-induced and parental cell lines only partially , the induced Px I-III KO/SODA-myc cells remained viable . To evaluate the effect of SODA-overexpression in more detail , induced and non-induced Px I-III KO/SODA-myc cells were incubated in medium ± 25 µM Dfx , treated with PI and subjected to flow cytometry . Overexpression of SODA or the presence of 25 µM Dfx partially prevented PI incorporation whereas in combination both treatments caused strong protection , in accordance with the cell counting results . Notably , in the absence and presence of Dfx , expression of SODA had a protecting effect ( Figure 7D ) . Overexpression of the mitochondrial SOD rendered the peroxidase-deficient parasites less sensitive to an iron-induced cell death . Superoxide oxidizes [4Fe4S]-cluster proteins , a process that generates hydrogen peroxide and releases iron ( Winterbourn , 2008 ) . The ectopic expression of SODA may lower the concentration of free iron either by diminishing the damage of iron sulfur clusters or because of the overexpression of an iron-containing protein . In any case , the data indicate a crucial role of mitochondrial matrix iron in the cell death of the Px I-III-deficient parasites . The involvement of iron is the primary characteristic of cells undergoing ferroptosis , however the subcellular site of iron that triggers the death program is not clear ( Doll and Conrad , 2017 ) . RPA is a mitochondria-targeted iron sensor . It is composed of a hydrophobic cationic rhodamine B moiety that mediates the uptake into mitochondria and a phenanthroline part that reacts with Fe ( II ) . Quenching of the RPA fluorescence is a selective indicator of mitochondrial chelatable iron in intact cells ( Petrat et al . , 2002; Rauen et al . , 2007 ) . RPAC has the same fluorophore and linker but lacks iron-chelating capacity and can serve as control . Preliminary experiments revealed that RPAC-treated parasites had a much higher fluorescence than cells loaded with RPA . Therefore , in experiments following the fluorescence , the cells were treated with different concentrations of the fluorophores . Px I-III KO cells were loaded with 150 nM RPA or 10 nM RPAC in PBS , followed by 15 min incubation in Trolox-containing medium to ensure the selective uptake of the sensor into the mitochondrion . Afterwards , the cells were stained with MitoTracker Green and subjected to fluorescence microscopy . The RPA and RPAC fluorescence coincided with the MitoTracker signal confirming a mitochondrial localization ( Figure 8A ) . Next , the cells were loaded with 150 nM RPA or 10 nM RPAC and then incubated in medium ± Trolox ( Figure 8B ) . When the cells were kept in the presence of Trolox , both sensors displayed clear mitochondrial localization . After 2 hr in the absence of Trolox , the RPA-treated cells were virtually unaffected ( not shown ) whereas the RPAC-treated sample displayed swollen cells with comparably faint and unspecific fluorescence . When kept for 3 hr in Trolox-free medium , most of the RPA-loaded cells appeared swollen and highly fluorescent while the majority of RPAC-treated parasites were rounded up and had lost their fluorescence . This time-dependent increase of RPA fluorescence and decline of RPAC fluorescence probably reflects leakage of the sensors to the extra-mitochondrial space caused by the loss of the mitochondrial membrane potential . The observation that after 3 hr in Trolox-free medium most of the RPA-loaded cells were only swollen whereas the RPAC-loaded control cells were highly damaged suggested a putative protecting effect by the iron-chelating sensor . To analyze this in more detail , the Px I-III KO cells were pre-loaded with different concentrations of RPA , transferred into Trolox-free medium and cell viability was followed for 6 hr by counting living cells . The presence of 150 nM RPA slowed down cell lysis; and 1 . 5 µM RPA exerted a protecting effect comparable to that observed with 50 µM Dfx ( Figure 8C and Figure 1—figure supplement 1 ) . The parasites were then stained with 150 nM RPA or RPAC , incubated for 3 hr ± Trolox and inspected by light scatter analysis . Indeed , after loading with RPA , but not with RPAC , a large portion of cells retained normal FSC and SSC ( Figure 8D ) indicating that chelation of mitochondrial iron by RPA protected the cells . Next the cells were loaded with 50 nM RPA or 1 nM RPAC and then incubated for different times in medium ± Trolox and subjected to flow cytometry ( Figure 8E ) . At this concentration , RPA exerted an only minor protecting effect . When the Px I-III KO cells were kept for ≥2 hr in the absence of protecting agent , the RPA fluorescence strongly increased and the RPAC signal decreased . That both 3 hr + Trolox probes had a higher fluorescence compared to the respective 0 hr – Trolox time points is probably a technical artifact as the 3 hr samples were treated with the dyes immediately after thawing the fluorescent compounds whereas the other samples were loaded later depending on the respective incubation times . Clearly , when comparing the fluorescence of the 3 hr ± Trolox samples , the RPA fluorescence was shifted to higher and the RPAC signal to lower values . Taken together , the fluorescence imaging and flow cytometry data indicate that upon loss of the mitochondrial membrane potential , the sensors leak out into the cytosol where the RPA fluorescence is de-quenched due to the lower iron concentration compared to the mitochondrion and that of RPAC is low due to dilution or loss . To determine the fluorescence of the maximally quenched and de-quenched probe , respectively , Px I-III KO cells were pre-incubated in medium with Trolox ± Fe ( III ) /HQ complex , Fer-1 or Dfx , washed and loaded with 50 nM RPA . In the presence of Trolox or Fer-1 , the cells displayed the same low fluorescence suggesting that the sensor was largely quenched by the mitochondrial chelatable iron ( Figure 8F ) . When the cells were pre-treated with Fe ( III ) /HQ , a highly membrane-permeable complex that intracellularly is rapidly reduced ( Rauen et al . , 2007 ) , the RPA fluorescence was shifted to even lower values which probably represents the maximally quenched signal . When instead of the lipophilic antioxidants , Dfx was used as protecting agent , the RPA fluorescence was strongly increased . This indicates that Dfx causes a decline of the mitochondrial iron available for reaction with RPA . Hoyes and Porter showed a time-dependent strong accumulation of radiolabeled Dfx in the soluble/cytosol fraction of the cell ( Hoyes and Porter , 1993 ) . This is further supported by our own data that in contrast to free Dfx , starch-coupled Dfx , which is restricted to the endosomal/lysosomal compartments ( Zhang and Lemasters , 2013 ) , is unable to protect the PC Px I-III KO cells ( Schaffroth et al . , 2016 ) . Finally , we studied if overexpression of mitochondrial SODA affects the mitochondrial iron level . Px I-III KO/SODA-myc cells , that were cultured for 20 hr in the presence or absence of Tet , were loaded with 50 nM RPA . As shown in Figure 8G , the fluorescence of the induced cells was slightly higher when compared to the non-induced cells . The only minor shift is probably due to the fact that the cultures are not homogenous but contain cells that express SODA to various levels ( as seen in the immunofluorescence analysis ) . Even so , the reproducibly observed shift to higher fluorescence suggests that overexpression of the iron-protein in the mitochondrial matrix lowers the iron available for chelation by RPA . Taken together , the data show that mitochondrial iron plays a crucial role in the death program of the peroxidase-deficient cells . Here we report that trypanosomes that lack the lipid hydroperoxide-detoxifying tryparedoxin peroxidases undergo a cell death which is suppressed by iron chelation ( Dfx ) or lipophilic antioxidants ( Trolox , α-Toc , Lpx-1 , Fer-1 ) and involves accumulation of lipid hydroperoxides and thus fulfils all criteria defining ferroptosis ( Stockwell et al . , 2017; Galluzzi et al . , 2018 ) . The first alterations observed after transfer of Px I-III-deficient PC parasites into medium without protecting agent were mitochondrial lipid peroxidation , loss of the mitochondrial membrane potential , drop of cellular ATP and production of mitochondrial oxidants indicating that in these cells the ferroptosis-type death started at the mitochondrion . The implication of mitochondria in ferroptosis is still highly controversial and debated in the mammalian system . Human cancer cells undergoing erastin-induced ferroptosis retain normal ATP levels; and cells that lack a functional electron transport chain or are depleted of mitochondria can experience ferroptosis ( Dixon et al . , 2012; Gaschler et al . , 2018 ) . Clearly , an active mitochondrial electron transport chain is not a prerequisite for a cell to undergo ferroptosis . Yet , even in these cells , morphological changes such as the formation of smaller mitochondria with increased membrane density were observed ( Dixon et al . , 2012 ) . GPx4-deficient mouse embryonic fibroblasts ( Pfa1 cells ) show a breakdown of the mitochondrial membrane potential ( Seiler et al . , 2008 ) , but initial lipid peroxidation occurs outside the mitochondrial matrix and involves a disruption of the outer mitochondrial membrane ( Friedmann Angeli et al . , 2014 ) . In redox phospholipidomics , GPx4-inactivated Pfa 1 cells revealed increased levels of oxygenated derivatives for all major classes of phospholipids , with the exception of cardiolipin and lipid peroxidation was found to take place predominantly in endoplasmic reticulum ( ER ) -associated sites ( Kagan et al . , 2017 ) . On the other hand , serum-induced ferroptosis in MEFs under amino acid starvation causes ATP depletion ( Gao et al . , 2015 ) and erastin-induced ferroptosis in neuronal cells is accompanied by a loss of the mitochondrial membrane potential and cellular ATP ( Neitemeier et al . , 2017 ) . Strikingly , GPx4-deficient kidneys display a time-dependent formation of peroxidized cardiolipin ( Friedmann Angeli et al . , 2014 ) and mitochondria-targeted nitroxides are able to inhibit ferroptosis in a variety of tissue types and across multiple growth conditions demonstrating that mitochondrial lipid oxidation is critical in promoting the ferroptotic cell death ( Krainz et al . , 2016 ) . The mitochondrial phenotype of the peroxidase-deficient PC T . brucei is not a specialized property of these protozoa and their single mitochondrion . In the mammalian bloodstream form – which relies exclusively on glycolysis for energy supply and acquires iron by endocytosis of host transferrin – lack of the cytosolic peroxidases triggers an iron-dependent lipid peroxidation and cell death that originates at the lysosome ( Hiller et al . , 2014 ) . A respective phenotype has been reported for human fibrosarcoma HT 1080 cells or lung cancer Calu-1 cells which are protected from erastin toxicity by inhibitors of lysosomal activities ( Torii et al . , 2016 ) . Our results strongly suggest that ferroptosis can be initiated at distinct subcellular sites . The role of mitochondria in ferroptosis execution appears to depend on the individual cell type and even the culture conditions such as the use of high glucose medium . A crucial factor could be that rapidly proliferating cells such as many cancer and embryonic cells rely more on glycolysis whereas differentiated cells primarily use oxidative phosphorylation for ATP production . In addition , ferroptosis starting at one cellular compartment may rapidly spread to other sites . Lipid transport proteins disseminate oxidative stress between compartments by preferably trafficking peroxidized lipids to mitochondria ( Kriska et al . , 2010; Vila et al . , 2004; Friedmann Angeli et al . , 2014 ) . After loss of the mitochondrial membrane potential , Px I-III-deficient cells still displayed a tubular staining for VDAC , an integral protein of the outer mitochondrial membrane . The most prominent alterations revealed by electron microscopy were a condensation of the mitochondrial matrix and appearance of white blebs which probably reflect enlarged cristae . A rupture of the outer membrane , described for mammalian cells after inactivation of GPx4 ( Friedmann Angeli et al . , 2014; Doll et al . , 2017 ) , was not observed . In accordance with the situation in ferroptotic mammalian cells , other subcellular structures , and in particular the nucleus , remained unaffected ( Stockwell et al . , 2017 ) . In the absence of a protecting agent , the cytosol of the Px I-III-deficient parasites became progressively lighter due to swelling of the cell body which finally resulted in cell death . That mitochondrial membrane peroxidation is an early event in the death of the peroxidase-deficient cells is supported by the finding that MitoPerOx presented a faster oxidation kinetic than the untargeted BODIPY C11 sensor . The inner mitochondrial membrane is particularly susceptible to oxidative damage because of its very large surface area and the proximity to the superoxide-producing respiratory chain and the high content of peroxidation-sensitive unsaturated fatty acids in its phospholipids , notably cardiolipin ( Prime et al . , 2012 ) . A comparative study on different phospholipids revealed cardiolipin with its two phosphate groups and thus most anionic phospholipid , as strongest binder of GPx4 ( Cozza et al . , 2017 ) . T . brucei is rich in polyunsaturated fatty acids , with C22:4–6 and C20:2–5 as the most abundant species ( Richmond et al . , 2010 ) and also one of the cardiolipin species identified in isolated mitochondria contains a C22:6 fatty acid ( Guler et al . , 2008 ) . This clearly creates a vulnerability of trypanosomes for membrane oxidation . As a characteristic feature of cells prone to ferroptosis , the Px I-III-deficient cells retain full viability in the presence of an iron chelator but the subcellular site of the iron involved remained elusive . PC T . brucei lack a transferrin receptor but take up iron from ferric complexes via a 2-step mechanism in which ferric iron is reduced to ferrous iron and is subsequently transported ( Mach et al . , 2013 ) . The parasites possess Mit1 , a homolog of the iron transport protein mitoferrin-1 in the inner mitochondrial membrane of vertebrate cells . Mit1 is essential in PC T . brucei and depletion of its mRNA specifically affects the mitochondrion ( Mittra et al . , 2016 ) . Here we present several lines of evidence that mitochondrial matrix iron triggers the cell death in the PC Px I-III-deficient trypanosomes . RPA , an iron-chelating sensor that is targeted to the mitochondrion , prolonged viability of the parasites . In Dfx-treated cells the sensor was de-quenched indicating that Dfx lowers the iron available for chelation by RPA . Radiolabeled Dfx was demonstrated not to be retained in the endosomal/lysomal compartments but to accumulate in the soluble/cytosolic fraction of mammalian cells ( Hoyes and Porter , 1993 ) . This is supported by our own data . Whereas free and starch-coupled Dfx equally protect BS T . brucei which lack the cytosolic peroxidases and show a lysosomal lethal phenotype , only free , but not starch-coupled Dfx , is able to protect the PC peroxidase-deficient cells ( Hiller et al . , 2014; Schaffroth et al . , 2016 ) . It is very likely that Dfx exerts its iron chelating and thus protecting effects in the cytosol . This could significantly alter our view on the mechanisms of ferroptosis that are partially based on the assumption that Dfx is membrane-impermeable and restricted to the lysosomal compartment ( Cao and Dixon , 2016; Gaschler et al . , 2018 ) . Upon incubation of the Px I-III KO cells without any protecting agent , the RPA fluorescence became de-quenched , most probably because , due to breakdown of the mitochondrial membrane potential , the sensor leaks out into the cytosol where the chelatable iron concentration is expectedly lower than in the mitochondrial matrix . Further support for the role of matrix iron in the process came from the finding that overexpression of the mitochondrial iron-SODA lowered the Dfx concentration required to protect the peroxidase-deficient cells . Three putative protecting mechanisms could be envisaged: an attenuated generation of free iron from FeS clusters by superoxide , a diminished level of reactive iron by affecting the recycling of Fe2+ from Fe3+ and incorporation of iron into the newly expressed mitochondrial protein . Each of these mechanisms would lower the reactive iron level in the mitochondrial matrix . To our knowledge this is the first study in which the subcellular site of iron triggering ferroptosis has been identified . Mechanisms that result in increased levels of poly-unsaturated fatty acids or cellular iron as well as those that decrease the capacity of the cell to detoxify peroxidized lipids such as depletion of GSH or GPx4 are able to trigger ferroptosis . Enzymatic effectors for example lipoxygenases can drive ferroptotic oxidation of poly-unsaturated fatty acids ( Seiler et al . , 2008; Stockwell et al . , 2017 ) . However , the lipoxygenase-mediated lethal effect is observed under GSH depletion conditions but not if ferroptosis is induced by GPx4 inactivation demonstrating that the presence of a lipoxygenase is not essential for the cell death program ( Yang et al . , 2016 ) . As shown recently for diverse types of lipoxygenase inhibitors , there is a correlation between the anti-ferroptotic activity and reactivity as radical-trapping agents ( Shah et al . , 2018 ) . Clearly , ferroptosis can be triggered spontaneously for instance by a rise in peroxidized lipids . There is a close link between the cellular lipid composition and ferroptosis sensitivity . ACSL4 ( acyl-CoA synthetase long-chain family member 4 ) is a critical determinant of the membrane lipid composition and sensitizes cells to ferroptosis whereas cells lacking ACSL4 show marked resistance to ferroptosis induction ( Doll et al . , 2017 ) . The T . brucei genome encodes at least five acyl-CoA synthetases ( ACSs ) . One of the four enzymes studied so far ( ACS1 ) accepts arachidonic acid and 22:6 fatty acids as preferred substrates ( Jiang and Englund , 2001 ) and the so far uncharacterized ACS5 ( Tb927 . 10 . 3260 ) has been annotated as putative long-chain-fatty-acid-CoA-ligase . An interesting aspect of future studies will be the putative role of long-chain ASCs in triggering ferroptosis in these protozoa . As shown here , ferroptosis is not restricted to mammalian cells ( Seiler et al . , 2008; Dixon et al . , 2012; Friedmann Angeli et al . , 2014; Stockwell et al . , 2017 ) and plants ( Distéfano et al . , 2017 ) . The regulated cell death occurs in trypanosomes , one of the earliest branching eukaryotes . The incorporation of polyunsaturated fatty acids into cell membranes was probably highly advantageous during evolution , allowing to modulate the membrane fluidity and functionality of membrane proteins and cells to adapt to different environments and temperatures . This is particularly important for African trypanosomes which during their digenetic life cycle switch between the 37°C bloodstream of a mammalian host and the 27°C midgut of the tsetse fly vector . The presence of poly-unsaturated fatty acids in the membranes , however , requires sophisticated antioxidant systems . In most cells this is achieved by the GSH/GPx4 couple . Trypanosomes developed a defense system that is coupled to the unique trypanothione/tryparedoxin system and replaces the GSH/glutathione peroxidase GPx4 pair . Tetracycline ( Tet ) , Trolox , α-tocopherol ( α-Toc ) , liproxstatin-1 ( Lpx-1 ) , ferrostatin-1 ( Fer-1 ) , deferoxamine mesylate ( Dfx ) , DAPI , 8-hydroxyquinoline ( HQ ) , iron ( III ) chloride x 6 H2O , hemin , penicillin/streptomycin and phleomycin were purchased from Sigma , Munich , Germany . Hygromycin B was from Carl Roth , Karlsruhe , Germany . DCFH-DA and propidium iodide ( PI ) were purchased from ThermoFisher , Schwerte , Germany . BODIPY 581/591 C11 ( BODIPY ) , MitoSOX Red , MitoTracker CMXRos , and MitoTracker Green were from Life Technologies , Darmstadt , Germany . Rhodamine B-[ ( 1 , 10-phenanthroline-5-yl ) -aminocarbonyl]benzyl ester ( RPA ) and rhodamine B-[ ( phenanthren-9-yl ) -aminocarbonyl]-benzylester ( RPAC ) were from Squarix Biotechnology , Marl , Germany and fetal calf serum ( FCS ) from Biochrome , Berlin , Germany . A sample of Hoechst 33342 was kindly provided by Dr . Walter Nickel , Heidelberg , Germany . MitoPerOx was a kind gift from Dr . Mike Murphy , Cambridge , UK . Drs José Pedro Friedmann Angeli and Marcus Conrad , Munich , Germany , are kindly acknowledged for samples of ( 1S , 3R ) -RSL3 and the racemic mixture of RSL3 ( Yang et al . , 2014 ) . Additional ( 1S , 3R ) -RSL3 was purchased from Cayman Chemical , Ann Arbor , Michigan . Trypanothione and trypanothione disulfide ( Comini et al . , 2009 ) as well as tag-free recombinant T . brucei trypanothione reductase ( TR ) ( Persch et al . , 2014 ) , tryparedoxin ( Tpx ) ( Fueller et al . , 2012 ) and peroxidase ( Px ) ( Melchers et al . , 2008 ) were prepared as described . Polyclonal rabbit antibodies against the cytosolic T . brucei 2-Cys-peroxiredoxins ( cTXNPx ) and guinea pig antibodies against the mitochondrial 2-Cys-peroxiredoxin ( mtTXNPx ) were obtained previously ( Ceylan et al . , 2010; Ebersoll et al . , 2018 ) . Polyclonal rabbit antibodies against T . brucei aldolase and VDAC and T . cruzi SODA were kindly provided by Drs Christine Clayton , Heidelberg , Germany , André Schneider , Bern , Switzerland , and Rafael Radi , Montevideo , Uruguay . Monoclonal mouse antibodies against the c-Myc protein as well as horseradish peroxidase-conjugated goat antibodies against mouse and rabbit immunoglobulins G were purchased from Santa Cruz Biotechnology , Heidelberg , Germany . The parasites used in this work were all culture-adapted Trypanosoma brucei brucei of the cell line 449 , descendants from the Lister strain 427 that stably express the tet repressor ( Cunningham and Vickerman , 1962; Biebinger et al . , 1997 ) . The cells were kindly provided by Dr Christine Clayton , Heidelberg , Germany . BS cells were grown in HMI-9 medium at 37°C . If not stated otherwise , PC parasites were cultivated at 27°C in MEM-Pros medium as described previously ( Schlecker et al . , 2005 ) . Distinct experiments were conducted in SDM79-CGGGPPTA medium ( GE Healthcare , Munich , Germany ) . All media were supplemented with 10% heat-inactivated FCS , 50 units/ml penicillin , 50 µg/ml streptomycin and 0 . 2 µg/ml phleomycin . In addition , both media for PC parasites contained 7 . 5 µg/ml hemin but no glucose . PC cells lacking the complete Px I-III locus ( Px I-III KO cells ( Schaffroth et al . , 2016 ) were routinely cultured in the presence of 100 µM Trolox ( 100 mM stock solution in ethanol ) . For phenotypic analyses , the parasites were harvested in the logarithmic growth phase , washed with PBS , resuspended in 5 ml medium adjusted to a density of about 5 × 105 cells/ml and subjected to different treatments . Stock solutions of Dfx ( 10 mM in PBS ) , Fer-1 ( 10 mM in DMSO ) , Lpx-1 ( 10 mM in DMSO ) , and α-Toc ( 10 mM in ethanol ) were prepared and various concentrations added to the cells . After different times , viable cells ( defined as those with normal , elongated shape ) were counted in a Neubauer chamber . The mean and standard deviation ( SD ) of three independent experiments were calculated using GraphPad Prism software ( GraphPad Software , La Jolla , CA ) . The pHD1700 plasmid contains a hygromycin resistance gene and a cassette that allows for Tet-inducible expression of a protein with C-terminal myc2-tag . The coding sequence of sodA without stop codon ( 714 bp ) was amplified from genomic DNA of WT T . brucei by PCR using the primers 5’CGATAAGCTTATGAGGTCTGTCATGATGC3’ and 5’CGATGGATCCCTTCATAGCCTGTTCATAC3’ ( restriction sites underlined ) . The amplicon as well as plasmid pHD1700/grx2-c-myc2 ( Ceylan et al . , 2010 ) were digested with HindIII and BamHI , purified and ligated yielding pHD1700/sodA-c-myc2 . PC Px I-III KO cells were transfected with the NotI-linearized plasmid . After 24 hr cultivation , hygromycin was added and resistant cells were selected by serial dilutions as described in detail ( Musunda et al . , 2015 ) . The plate reader-based assay was carried out as described previously ( Fueller et al . , 2012; Latorre et al . , 2016 ) . Stock solutions of 10 mM ( 1S , 3R ) -RSL3 and RSL3 racemate were prepared in DMSO . Each well contained 90 μl HMI-9 medium with 2500 cells/ml . The compounds were serially 1:3 diluted in 10 steps and 10 μl of the dilutions added to the cells to give final concentrations between 30 µM and 1 nM . 10% DMSO served as positive control . The highest DMSO concentration added with the compounds was 0 . 3% ( which does not affect parasite viability , negative control ) . Since the Px-type enzymes are dispensable if the cells are kept in the presence of Trolox ( Hiller et al . , 2014 ) , a second analysis was done in medium that was supplemented with 100 µM Trolox , 100 nM Fer-1 or 200 nM Lpx-1 . After 24 hr , 48 hr and 72 hr cultivation , 50 μl ATPlite one step solution ( PerkinElmer , Rodgau , Germany ) was added and the luminescence measured in a Victor X4 plate reader ( PerkinElmer ) at room temperature . The luminescence intensities were plotted against the logarithmic compound concentrations and EC50-values calculated using GraphPad Prism . Chlorhexidine ( 37 µM to 11 nM ) , a trypanocidal compound and known TR inhibitor ( Meiering et al . , 2005; Beig et al . , 2015 ) , served as positive control . A putative irreversible inhibition of TR was studied essentially as described previously ( Otero et al . , 2006 ) . In a total volume of 1 ml of 40 mM Hepes , 1 mM EDTA , pH 7 . 5 , 1 µM T . brucei TR was incubated at 25°C with 100 µM RSL3 racemate in the presence of 500 µM NADPH . After different times ( 0–120 min ) , 5 µl of the reaction mixture was removed , and the remaining activity measured in a 1 ml standard TR assay . Because of the dilution , reversible inhibition is not recorded under these conditions . Controls contained either buffer , TR and NADPH or buffer , TR and inhibitor in the pre-incubation mixtures . The effect of RSL3 on the parasite peroxidase system was measured as described previously ( Fueller et al . , 2012 ) . Shortly , in a total volume of 200 µl of 100 mM Tris , 5 mM EDTA , pH 7 . 6 , 150 µM NADPH , 200 mU TR , 100 µM T ( SH ) 2 , 10 µM Tpx , 60 nM Px and 40 µM RSL3 or 5 µl DMSO ( control ) were incubated at 25°C . After 1 , 15 , and 40 min , the reaction was started by adding 100 µM H2O2 and the absorption decrease followed at 340 nm . In a second approach , the pre-incubation mixture contained all components except the peroxidase , and the assay was started by adding both Px and H2O2 . All treatments were performed at 27°C in the dark . Logarithmically growing Px I-III KO cells were harvested , split into samples of about 106 cells , washed with cold PBS and re-suspended in 1 ml medium . For the detection of general cellular ROS , the cells were re-suspended in SDM-79 medium ± Trolox , incubated for 2 hr , washed and stained for 30 min with 10 µM H2DCFDA ( 10 mM stock solution in DMSO ) in medium + Trolox . All following analyses were conducted in MEM-Pros medium . To measure cellular lipid peroxidation , the cells were transferred into medium supplemented with either 100 µM Trolox , 100 µM Dfx , 100 nM Fer-1 , 200 nM Lpx-1 or without any addition , containing 2 µM BODIPY 581/591 C11 ( BODIPY; 10 mM stock solution in DMSO ) and incubated for 2 hr . For the comparison of MitoTracker Red and PI staining , the cells were incubated for 0–4 hr in medium ± Trolox , treated with MitoTracker , as described for the fluorescence microscopy , or for 5 min with 5 µg/ml PI ( 1 mg/ml stock solution in water ) in PBS . For the MitoSOX experiments , the cells were pre-loaded for 10 min with 5 µM of the dye ( 5 mM stock solution in DMSO ) in medium + Trolox and then incubated in medium ± Trolox . As control , the cells were treated for 5 min with 20 ng/ml DAPI ( 50 µg/ml in water ) in PBS . For comparing general and mitochondrial lipid peroxidation , the cells were incubated in medium ± Trolox containing 2 µM BODIPY or 100 nM MitoPerOx ( 2 mM stock solution in DMSO ) . As proof-of-principle for the experiments with RPA , the cells were transferred into medium supplemented with either 100 µM Trolox , 100 µM Dfx , 100 nM Fer-1 or Trolox plus 100 µM Fe ( III ) /HQ ( freshly prepared by mixing equal volumes of 10 mM FeCl3 in water and 10 mM HQ in 50% v/v ethanol in water; [Petrat et al . , 2002] ) and incubated for 2 hr . Thereafter , the cells were washed , re-suspended in PBS , incubated for 15 min with 50 nM RPA ( 1 mM stock solution in DMSO ) , washed and incubated for another 15 min in medium + Trolox to allow optimal enrichment of the dye in the mitochondrion . To follow changes in RPA fluorescence after Trolox withdrawal , starting with the longest time point , the cells were stained with 50 nM RPA or 1 nM RPAC ( 1 mM stock solution in DMSO ) and treated as described above followed by up to 3 hr incubation in medium ± Trolox . To detect possible changes in the mitochondrial iron content due to SODA-overexpression , the Px I-III KO/SODA-myc cells were cultured for about 20 hr ± Tet , and stained with 50 nM RPA . After the respective treatments , the cells were washed with cold PBS , re-suspended in 1 ml cold PBS , transferred into FACS tubes ( Sarstedt ) and immediately subjected to flow cytometry in a BD FACSCanto or FACSCantoII instrument at the Flow Cytometry and FACS Core Facility ( FFCF ) of the Center of Molecular Biology ( ZMBH ) of Heidelberg University . The following excitation lasers and emission filters ( ex:em ) were applied: BODIPY and MitoPerOx 488:530/30 nm , MitoTracker 561:586/15 nm , PI 488:570 nm , DAPI 405:450/50 nm , MitoSOX 488:585/42 nm , RPA and RPAC 561:610/20 . In each experiment 10000 events were recorded . The data were analyzed using FlowJo software ( FlowJo , LLC ) . About 1 . 2 × 106 cells were used per sample . MitoTracker staining , cell fixation , permeabilization and antibody treatment were performed as described previously ( Hiller et al . , 2014; Schaffroth et al . , 2016 ) . Antibodies against mtTXNPx , VDAC , cMyc and cTXNPx were diluted 1:1000 , 1:500 , 1:200 , and 1:1000 , respectively . For visualization , Alexa Fluor 488-conjungated goat anti-guinea pig ( 1:1000 ) , anti-rabbit ( 1:1000 ) , Alexa Fluor 546 goat anti-rabbit ( 1:1000 ) and Alexa Fluor 488-conjungated goat anti-mouse ( 1:250 ) ( Molecular Probes ) antibodies were used . The cells were examined using a Carl Zeiss Axiovert 200 M microscope equipped with an AxioCam MRm digital camera and the AxioVision software ( Zeiss , Jena , Germany ) . Aliquots of 5 × 107 logarithmically growing Px I-III KO cells were transferred into medium ± Trolox and incubated for 0 . 5 , 1 and 2 hr . The cell fixation and embedding procedures were adapted from ( Höög et al . , 2010 ) . Briefly , cells were fixed by 2 . 5% glutaraldehyde and 2% p-formaldehyde in 100 mM cacodylate buffer , pH 7 . 2 at 4°C overnight . The cells were centrifuged in fixative and the pellet was processed in one piece . After rinsing in buffer the samples were further fixed in 1% osmium tetroxide in cacodylate buffer , washed in water , and incubated with 1% uranyl acetate in water overnight . Dehydration was done in 10 min steps in an aceton gradient followed by Spurr resin embedding and polymerization at 60°C . The blocks were cut in 70 nm thin sections using a Leica UC6 ultramicrotome ( Leica Microsystems Vienna , Austria ) and collected on pioloform-coated mesh grids . The post-stained sections were imaged on a JEOL JEM-1400 electron microscope ( JEOL , Tokyo , Japan ) operating at 80 kV and equipped with a 4K TemCam F416 ( Tietz Video and Image Processing Systems GmBH , Gautig , Germany ) . The analysis was performed by the Electron Microscopy Core Facility ( EMCF ) of Heidelberg University . Logarithmically growing Px I-III KO cells in SDM-79 or MEM-Pros medium were divided into six aliquots . Starting with the latest time point , the samples were washed with PBS , transferred into 5 ml medium ± Trolox and incubated at 27°C . After 0 to 4 hr , an aliquot was removed and kept on ice for cell counting , another one was treated with PI and subjected to flow cytometry as described above . The remaining cells were centrifuged and re-suspended in 300 µl medium + Trolox . Three aliquots of 100 µl , each containing 2 × 106 cells , were transferred into a 96-well plate , mixed with 50 µl ATPlite one-step solution ( PerkinElmer ) according to the manufacturer’s protocol and luminescence was measured in a Victor X4 plate reader ( PerkinElmer ) . The luminescence of the cell-free medium was subtracted . All cells that still had an elongated shape or only highly motile cells were counted . All data were analyzed and are presented as percentage of the starting values using GraphPad Prism . PC Px I-III KO/SODA-myc cells , cultivated for 18 hr in the presence or absence of 1 µg/ml tet , were harvested , re-suspended in reducing SDS-sample buffer and boiled . Lysates from 5 × 106 cells were loaded per lane onto a 12% gel and subjected to SDS-PAGE . After electrophoresis , the proteins were transferred onto a PVDF membrane and probed with antibodies against c-Myc ( 1:200 ) and aldolase ( 1:20 , 000 ) , followed by development with HRP-conjugated goat antibodies against mouse ( 1:5000 ) and rabbit ( 1:10 , 000 ) IgGs , respectively . All sample preparations were performed at 27°C in the dark in a total volume of 1 ml . Per sample , about 106 cells were harvested and washed with cold PBS . Treatment of the cells with MitoSOX or MitoPerOx was done as described above for the flow cytometry . MitoSOX-treated cells were then transferred into medium ± Trolox and incubated for 2 hr . Afterwards the cells were treated for 15 min with 120 nM MitoTracker Green in Trolox-supplemented medium , followed by 30 min incubation in Trolox-containing medium for optimal enrichment of the dye in the mitochondrion . For nuclear and kinetoplast DNA staining , the cells were treated for 15 min with 3 µg/ml Hoechst 33342 in PBS . In the case of RPA and RPAC , the cells were incubated for 15 min with 150 nM RPA or 10 nM RPAC in Trolox-supplemented PBS followed by 15 min in Trolox-supplemented medium . Subsequently , the cells were either stained with MitoTracker Green or transferred into medium ± Trolox and incubated for up to 3 hr at 27°C in the dark . After the respective treatments , the cells were washed with PBS , re-suspended in 20 µl PBS , placed on ice and within ≤30 min inspected under a Carl Zeiss Axiovert 200 M microscope equipped with an AxioCam MRm digital camera and the AxioVision software ( Zeiss , Jena ) . Except where stated otherwise , all experiments were performed three times on separate days as independent biological replicates . The data shown represent the mean ± SD of these replicates . The data were evaluated using GraphPad Prism ( GraphPad Software , La Jolla , CA ) .	Plants , animals and fungi all belong to a group of organisms known as eukaryotes . Their cells host a variety of compartments , with each having a specific role . For example , mitochondria are tasked with providing the energy that powers most of the processes that keep the cell alive . Membranes delimit these compartments , as well as the cells themselves . Iron is an element needed for chemical reactions that are essential for the cell to survive . Yet , the byproducts of these reactions can damage – ‘oxidize’ – the lipid molecules that form the cell’s membranes , including the one around mitochondria . Unless enzymes known as peroxidases come to repair the oxidized lipids , the cell dies in a process called ferroptosis . Scientists know that this death mechanism is programmed into the cells of humans and other complex eukaryotes . However , Bogacz and Krauth-Siegel wanted to know if ferroptosis also exists in creatures that appeared early in the evolution of eukaryotes , such as the trypanosome Trypanosoma brucei . This single-cell parasite causes sleeping sickness in humans and a disease called nagana in horses and cattle . Before it infects a mammal , T . brucei goes through an ‘insect stage’ where it lives in the tsetse fly; there , it relies on its mitochondrion to produce energy . Bogacz and Krauth-Siegel now show that if the parasites in the insect stage do not have a specific type of peroxidases , they die within a few hours . In particular , problems in the membranes of the mitochondrion stop the compartment from working properly . These peroxidases-free trypanosomes fare better if they are exposed to molecules that prevent iron from taking part in the reactions that can harm lipids . They also survive more if they are forced to create large amounts of an enzyme that relies on iron to protect the mitochondrion against oxidation . Finally , using drugs that prevent ferroptosis in human cells completely rescues these trypanosomes . Taken together , the results suggest that ferroptosis is an ancient cell death program which exists in T . brucei; and that , in the insect stage of the parasite's life cycle , this process first damages the mitochondrion . This last finding could be particularly relevant because the role of mitochondria in ferroptosis in mammals is highly debated . Yet , most of the research is done in cells that do not rely on this cellular compartment to get their energy . During their life cycle , trypanosomes are either dependent on their mitochondria , or they can find their energy through other sources: this could make them a good organism in which to dissect the precise mechanisms of ferroptosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2018	Tryparedoxin peroxidase-deficiency commits trypanosomes to ferroptosis-type cell death
Hallmark social activities of humans , such as cooperation and cultural learning , involve eye-gaze signaling through joint attentional interaction and ostensive communication . The gaze-signaling and related cooperative-eye hypotheses posit that humans evolved unique external eye morphologies , including uniformly white sclera ( the whites of the eye ) , to enhance the visibility of eye-gaze for conspecifics . However , experimental evidence is still lacking . This study tested the ability of human and chimpanzee participants to discriminate the eye-gaze directions of human and chimpanzee images in computerized tasks . We varied the level of brightness and size in the stimulus images to examine the robustness of the eye-gaze directional signal against simulated shading and distancing . We found that both humans and chimpanzees discriminated eye-gaze directions of humans better than those of chimpanzees , particularly in visually challenging conditions . Also , participants of both species discriminated the eye-gaze directions of chimpanzees better when the contrast polarity of the chimpanzee eye was reversed compared to when it was normal; namely , when the chimpanzee eye has human-like white sclera and a darker iris . Uniform whiteness in the sclera thus facilitates the visibility of eye-gaze direction even across species . Our findings thus support but also critically update the central premises of the gaze-signaling hypothesis . In humans , eye-gaze is employed in critical ways for conspecific communication during social activities such as cooperation , teaching , and language learning ( Csibra and Gergely , 2009; Tomasello et al . , 2005 ) . Humans excel at detecting another’s direct gaze and following another’s gaze directions from early infancy ( Farroni et al . , 2002; Farroni et al . , 2004 ) . Moreover , during social interactions , humans exchange communicative intentions with one another via eye contact ( Senju and Johnson , 2009; Kleinke , 1986 ) , and an experience of being watched by another affects one’s reputational concerns ( Bateson et al . , 2006; Engelmann et al . , 2012 ) . Comparative studies have found that closely related species such as nonhuman great apes also excel at detecting another’s direct gaze ( Myowa-Yamakoshi et al . , 2003 ) and following another’s gaze directions ( Bräuer et al . , 2005; Okamoto et al . , 2002 ) , and adjust their gaze-following behaviors flexibly and cognitively ( Kano and Call , 2014; Okamoto-Barth et al . , 2007; Povinelli and Eddy , 1996; Tomasello et al . , 1999 ) . Moreover , they make eye contact with other individuals at critical moments of social interactions , such as when requesting food from an experimenter ( Gomez , 1996 ) and initiating affiliative social interactions with conspecifics ( Heesen et al . , 2021; Yamagiwa , 1992 ) . However , several potential differences might exist between humans and nonhuman apes in their use and interpretation of gaze behaviors . For example , in the gaze-following/cueing experiments , while humans primarily relied on another’s eye-directional cues , nonhuman apes relied on head-directional rather than eye-directional cues ( Tomasello et al . , 2007; Tomonaga , 2007 , but see Povinelli and Eddy , 1996; Deaner and Platt , 2003 ) . In general , reliance on eye-directional rather than head-directional cues indicates one’s inclination/ability to identify a specific spot that another’s foveae are directing at , rather than a large area that another’s field of view covers . Thus , the former strategy leads to efficient identification of the interaction partners’ focused objects during joint attentional interaction and ostensive communication . Relatedly , a previous study observed that , while both human infants and nonhuman apes alternately looked at both the interaction partner’s face and the focused object during joint-attentional interactions , nonhuman apes made only brief looks to the interaction partner’s face while human infants made more long-lasting looks ( Carpenter and Tomasello , 1995 ) . Furthermore , in a communicative gaze-following task , while human infants responded to an agent’s eye contact as if they interpreted it as an ostensive signal preceding referential information ( Senju and Csibra , 2008; Okumura et al . , 2020; but see Gredebäck et al . , 2018 ) , great apes did not show any evidence supporting this understanding ( Kano et al . , 2018 ) . These potential differences between human and nonhuman apes might be partly related to differences in their sociocognitive skills but also humans’ specialization to eye-gaze signals in conspecific communication . Relatedly , one influential hypothesis , the gaze-signaling hypothesis ( Kobayashi and Kohshima , 2001; Kobayashi and Kohshima , 1997 ) , proposes that humans have evolved special morphological features in the eye , including uniform whiteness in the exposed sclera , to enhance the visibility of eye-gaze directions and thereby help conspecifics communicate via eye-gaze without much attentional effort . This hypothesis is based on comparative analyses showing that ( 1 ) uniformly white sclera is a unique feature of the human eye among primates and that ( 2 ) the human eye is horizontally more elongated , and the human sclera is horizontally more exposed compared to other primates’ eyes ( Kobayashi and Kohshima , 2001; Kobayashi and Kohshima , 1997 ) . The cooperative-eye hypothesis ( Tomasello et al . , 2007 ) extended this gaze-signaling hypothesis based on the results from their gaze-following study , proposing that humans evolved these morphological features as well as a special sensitivity to these features to facilitate joint-attentional and communicative interactions in a cooperative context . Others discussed that humans’ white sclera not only signals gaze direction but also emotional cues in combination with fine musculatures around the eyes ( Whalen et al . , 2004; Baron-Cohen et al . , 2001; Jessen and Grossmann , 2014 ) , and also emotional and health-status cues by its color variations ( Provine et al . , 2013a; Provine et al . , 2013b ) . Moreover , typically developing human adults and children show basic preferences for an animal agent having white sclera ( Segal et al . , 2016 ) . Despite the widespread popularity of the gaze-signaling and related cooperative-eye hypotheses in the literature , these hypotheses have been severely challenged by recent quantitative morphological studies showing that certain eye features of humans are not necessarily unique among great ape species ( Mayhew and Gómez , 2015; Perea-García et al . , 2019; Caspar et al . , 2021; Mearing and Koops , 2021; Kano et al . , 2021 ) . Accordingly , there is currently a heated debate over whether the external eye morphology of humans has any communicative function . Specifically , those studies have shown that ( 1 ) the sclera of humans is not necessarily more exposed than that of other great ape species ( Mayhew and Gómez , 2015; Caspar et al . , 2021; Kano et al . , 2021 ) , that ( 2 ) the color contrast/difference between the iris and the sclera is similar between humans and other great apes ( Perea-García et al . , 2019; Caspar et al . , 2021; Mearing and Koops , 2021; Kano et al . , 2021 ) , and that ( 3 ) there is substantial individual variation in the extent to which sclera is unpigmented among some nonhuman ape species ( Mayhew and Gómez , 2015; Caspar et al . , 2021; Mearing and Koops , 2021; Kano et al . , 2021; Perea García , 2016 ) . Given these new findings , one recent study suggested that one ( and possibly only ) eye feature that both distinguishes humans from other great apes and contributes to the visibility of eye-gaze direction is uniform whiteness in humans’ exposed sclera; i . e . , depigmentation all the way from the iris edge to the eye corners ( Kano et al . , 2021 ) . This previous study used image analysis and computer vision techniques and identified that uniformly white sclera characterizes clear visibilities of both iris and eye-outline edges – the two essential features that contribute to the visibility of eye-gaze directions ( see Appendix 3—figure 1 for the illustration of this aspect ) . More specifically , while the iris is highly visible in all great ape species , the visibility of eye-outline edges is limited in nonhuman apes because most nonhuman ape individuals have darker or more graded/patchy sclera colors compared to humans’ sclera colors , which more easily blend into the adjacent skin/hair colors . On the other hand , nearly all human individuals have uniformly white sclera , which is clearly distinguished from the adjacent skin/hair colors around the eyes ( irrespective of the skin color variations ) even when the eyes are viewed in visually challenging conditions ( e . g . , shading , distancing ) . However , experimental evidence is lacking to support these findings . Previously , several experimental studies demonstrated that humans’ white sclera facilitates gaze perception at least in human participants . Ricciardelli et al . , 2000 tested human participants in a gaze-discrimination task and found that reversing the contrast polarity of human eye images makes the judgment of gaze direction less accurate . Specifically , the gaze directions were more accurately judged in eyes having a positive ( normal ) contrast polarity ( white sclera with a darker iris ) than those having a negative ( reversed ) contrast polarity ( black sclera with a bright iris ) , even though these two eye images were composed of the same colors . Ricciardelli et al . thus suggested that humans have perceptual expertise in positive contrast polarity of eyes ( see Itier et al . , 2006; Yoshizaki and Kato , 2011 for related results ) . Also , Yorzinski and Miller , 2020 tested human participants in a gaze-discrimination task in which the sclera colors of human faces were manipulated to be either conspicuous ( white or lighter than the iris color ) or inconspicuous ( similar or darker than the iris color ) ; human participants detected the faces with conspicuous sclera colors faster and more accurately . Yorzinski and Miller thus suggested that humans’ white sclera facilitates their gaze perception . Importantly , although these two studies found similar results , it remains unclear whether humans’ white sclera facilitates their gaze perception by its perceptual properties per se or by human participants’ perceptual expertise in the positive contrast polarity of human eyes . To answer this question , a comparative approach is helpful . Tomonaga and Imura , 2010 tested whether a chimpanzee could be trained to discriminate eye-gaze directions of human facial images , and in one of their experimental conditions , they replicated the results from Ricciardelli et al . , 2000 . Specifically , they showed that the chimpanzee’s task performance dropped when the contrast polarity of the eyes was reversed in the stimulus human faces . Unfortunately , however , the conclusion from this result is severely limited because the chimpanzee was trained to discriminate the gaze directions of the human positive eyes ( positive contrast polarity ) before being tested in the trials presenting the human negative eyes ( reversed contrast polarity ) . Therefore , it remains unclear whether the chimpanzee’s performance was affected by the change in the eye contrast polarity or merely the change in the trained color pattern . Critically , no previous study adopted a fully crossed design for species comparison to examine the effect of sclera colors on primate gaze perception; namely , a study design that presents the stimuli of both species to the participants of both species . Chimpanzees are a suitable species for this design because their eyes have relatively uniformly dark sclera and a bright iris , namely , having a negative contrast polarity opposite to the positive contrast polarity that human eyes have . Therefore , in a fully crossed design with chimpanzees and humans , one can test whether the uniformly white sclera of human positive ( normal ) eyes and chimpanzee positive ( reversed ) eyes facilitates the gaze-discrimination performances of both chimpanzee and human participants . Such a question would clarify whether the uniformly white sclera has a perceptual advantage independently from perceptual advantages of other eye features or the participants’ perceptual expertise in a certain eye contrast polarity ( as we will detail below ) . Consequently , this study tested both humans and chimpanzees on their ability to discriminate the gaze direction of both chimpanzee and human faces with the eyes manipulated to have both normal and reversed contrast polarities . The contrast polarity of both species’ eyes ( only the eyeball regions ) was reversed by inverting the lightness ( or grayscale ) values of the eyeball images in the faces ( Figure 1A and Appendix 3—figure 3 ) . This manipulation created artificial white sclera in chimpanzee eyes and artificial dark sclera in human eyes , while not affecting the iris-sclera ( or pupil-iris ) color differences in each species’ eyes . The task for human and chimpanzee participants was to discriminate the gaze direction of human and chimpanzee eye images on a computer monitor ( either in a keypress or visual search task; Figure 1B ) . Our experimental procedures followed a previous study adopting training procedures ( Tomonaga and Imura , 2010 ) instead of spontaneous gaze-following/cueing procedures for chimpanzees ( Tomasello et al . , 2007; Tomonaga , 2007 ) because , while the former study successfully trained a chimpanzee to discriminate eye-gaze directions of a stimulus ( human ) face , the latter studies reported that chimpanzees predominantly rely on head-directional cues but not eye-gaze directional cues in the spontaneous tasks . Also , as the strength of the visual signal generally depends on its degradation caused by natural noises ( Endler , 1990 ) , we tested the robustness of eye-gaze signals against shading and distancing ( Figure 1A ) , namely the simulated visual conditions where stimulus eyes were presented darker ( in shadows ) and smaller ( in the distance ) , following a previous study on great ape eye color ( Kano et al . , 2021 ) . We developed five sets of hypotheses and predictions ( Table 1 ) . H1 is our key hypothesis , which posits the perceptual advantage of uniformly white sclera ( Kobayashi and Kohshima , 2001; Kobayashi and Kohshima , 1997; Kano et al . , 2021 ) and thus predicted increased performance for both human and chimpanzee positive eyes in participants of both species . H2 posits the perceptual advantage of the iris-sclera color difference . As the iris-sclera color difference ( note the difference between the contrast and difference measures employed by previous studies; Perea-García et al . , 2019; Caspar et al . , 2021; Mearing and Koops , 2021; Kano et al . , 2021 ) was similar between chimpanzees and humans in general ( Kano et al . , 2021 ) and did not differ between our chimpanzee and human stimuli with both positive and negative eyes ( see Appendix 3—figure 2 for the quantitative evaluation of our stimuli ) , we predicted no performance difference between the stimulus types in participants of both species . H3 posits the perceptual advantage of the horizontally elongated shape . As human eyes were horizontally longer than chimpanzee eyes in general ( Mayhew and Gómez , 2015; Caspar et al . , 2021; Kano et al . , 2021 ) and also in our stimulus set ( Appendix 3—figure 2 ) , we predicted increased performance for the human stimuli ( with both positive and negative eyes ) in participants of both species . H4 posits the participants’ perceptual expertise in the contrast polarity of own-species eyes ( Ricciardelli et al . , 2000 ) and thus predicted increased performance for the positive eyes of both species in human participants and the negative eyes of both species in chimpanzee participants . H5 addressed a general possibility arising from our manipulation of both species’ eye colors , namely , that participants of both species perform more poorly in such artificial conditions , and thus predicted increased performance for the human positive eyes and the chimpanzee negative eyes in participants of both species . Related to this last hypothesis , our chimpanzee participants had extensive experiences in interacting with both conspecifics and humans from youth and thus were familiar with both species’ eyes . We also ensured that our human participants had a minimum of a few months ( to decades ) of experiences in interacting with chimpanzees and thus were familiar with both species’ eyes . In Study 1 , we tested 25 adult human participants ( 14 females , 11 males ) in two experiments . Experiment 1 presented participants with the stimuli of both humans and chimpanzees with normal contrast polarity at four stimulus levels ( L1–4 ) varying in size and brightness . We tested the effect of stimulus level and species on participants’ correct responses ( correct , incorrect ) in a binomial generalized linear mixed model ( GLMM; see Appendix 1—table 1 for the formulas ) and found that human participants performed better in trials presenting the human stimuli than those presenting the chimpanzee stimuli ( χ2 = 37 . 08 , df = 1 , p<10–8 ) . We also found that their performance was worse in trials presenting smaller and more shaded stimuli ( χ2 = 45 . 85 , df = 3 , p<10–9; see Appendix 1—table 2 for the full GLMM results; Figure 2A ) . Experiment 2 presented the same participants with the stimuli of both species with both normal and reversed contrast polarities at stimulus levels L3 and L4 . We tested the effect of stimulus level , species , and contrast polarity on participants’ correct responses in GLMM and found a significant three-way interaction effect between these factors ( Figure 2B; χ2 = 17 . 57 , df = 1 , p<10–4; also see Appendix 1—table 2 ) . We then performed simple effects tests to examine the observed interaction effect further ( Figure 2B and Appendix 1—table 2 ) . Critically , we found that human participants performed better in trials presenting the positive ( reversed ) eyes of chimpanzees ( i . e . , the white sclera and a darker iris ) than those presenting the negative ( normal ) eyes of chimpanzees ( i . e . , the dark sclera and a bright iris ) , while they performed worse in trials presenting the negative ( reversed ) eyes of humans than those presenting the positive ( normal ) eyes of humans . In Study 2 , we began with training 10 chimpanzees . Only three ( Natsuki , Hatsuka , Pendesa ) passed all the required training and subsequently participated in test phases ( see Appendix 2—table 1 for details about participants; also see Appendix 4—table 1 and Figure 1 for the number and performance of training sessions for each chimpanzee ) . Study 2 involved two experiments . Experiment 1 tested those three chimpanzees and presented them with the human and chimpanzee eye images with normal contrast polarity . To test chimpanzees at stimulus levels higher than L1 ( i . e . , smaller and more shaded ) , we gradually incremented the stimulus level ( by 0 . 5 ) across sessions when individuals showed high performance in target trials presenting a given stimulus level ( above 85% in two successive sessions ) . The test phase was defined as the sessions presenting stimulus level higher than ( or equal to ) L2 . 5 , given that we observed clear performance differences between stimulus species at stimulus levels higher than L2 in Study 1 ( see Appendix 4—table 3 for the number of sessions in the pre-test and test phases ) . We tested chimpanzees’ correct responses during the test phase across repeated sessions at the individual level ( with the α level corrected for the number of individuals in the Bonferroni correction , α = 0 . 05/3 ) . Each chimpanzee completed a minimum of 20 test sessions . To avoid the ceiling effect , we incremented the stimulus level also during the test phase when chimpanzees showed high performance in target trials based on the same criteria . We tested the effect of stimulus species in binomial GLMM on each chimpanzee’s correct responses ( correct , incorrect ) during the test phase and found that all three chimpanzees performed significantly better for the human stimuli than the chimpanzee stimuli ( Natsuki: χ2 = 8 . 28 , df = 1 , p=0 . 004; Hatsuka: χ2 = 9 . 50 , df = 1 , p=0 . 002; Pendesa: χ2 = 21 . 94 , df = 1 , p<10–5; also see Appendix 1—table 2 ) . Experiment 2 tested two ( Natsuki and Hatsuka ) out of the three chimpanzees . Pendesa was dropped from this experiment because she took about twice as many training sessions as the other two chimpanzees ( Appendix 4—table 2 and Figure 1 ) . Experiment 2 presented them with eye stimuli having reversed contrast polarity in the first test phase ( Test B ) and then eye stimuli having normal contrast polarity in the next test phase ( Test A2 ) ; thus , together with the results from Experiment 1 ( also called Test A1 phase ) , we tested chimpanzees in the ABA design . The Test A2 phase started from stimulus level L3 , which these two chimpanzees reached during the Test A1 phase . The other procedures were identical with Experiment 1 ( with α = 0 . 05/2 ) . We compared each chimpanzee’s correct responses in target trials across Test A1 and Test B in GLMM and found a significant interaction effect between stimulus species and phase in both chimpanzees ( Natsuki: χ2 = 34 . 61 , df = 1 , p<10–8; Hatsuka; χ2 = 8 . 39 , df = 1 , p=0 . 004; also see Appendix 1—table 2 ) . We then compared each chimpanzee’s performance across Test B and Test A2 and found a significant interaction effect between the two factors in both chimpanzees ( Natsuki: χ2 = 37 . 04 , df = 1 , p<10–8; Hatsuka; χ2 = 33 . 75 , df = 1 , p<10–8 ) . To examine these observed interaction effects further , we performed simple effects tests ( Figure 3 and Appendix 1—table 2 ) . Critically , we found that both chimpanzees’ performance significantly increased from Test A1 ( normal contrast polarity ) to Test B ( reversed contrast polarity ) and then their performance decreased from Test B to Test A2 ( normal contrast polarity ) in trials presenting the chimpanzee stimuli . Natsuki’s performance significantly decreased from Test A1 to Test B and then increased from Test B to Test A2 in trials presenting the human stimuli . Hatsuka’s performance did not significantly decrease from Test A1 to Test B but significantly increased from Test B to Test A2 in those trials . Overall , these results revealed a striking advantage of eyes having positive contrast polarity , namely , the eyes of both species with the uniformly white sclera and a darker iris , in the gaze-discrimination performance of both human and chimpanzee participants . Our results thus supported H1 ( perceptual advantage of uniformly white sclera ) . We also found that , although both human and chimpanzee eye-gaze directions are reliably discernible when those eyes were presented sufficiently large and bright ( i . e . , L1–2 stimuli ) , the human eye-gaze directions were more discernible than the chimpanzee eye-gaze particularly when those eyes were presented smaller and more shaded ( i . e . , L3–4 stimuli ) . Our alternative hypotheses ( H2–5 ) cannot explain the overall patterns of our results . H2 ( perceptual advantage of the iris-sclera color difference ) cannot explain our results likely because it supposes clear visibility of only iris but not that of eye-outline edges , another critical feature that contributes to the visibility of eye-gaze ( Kobayashi and Kohshima , 2001; Kano et al . , 2021 ) ( also see Appendix 3—figure 1 ) . H3 ( perceptual advantage of the horizontally elongated eye shape ) also cannot explain our results likely because small variations in the horizontal eye length do not critically affect the visibility of eye-gaze . However , it should be noted that humans can take more extreme sideway eye positions than chimpanzees due to their horizontally elongated eye shape ( Kobayashi and Kohshima , 2001 ) , and such mechanistic difference may be one advantage of the human eye in eye-gaze signaling . Yet , this fact might be of limited relevance to real-life social interaction because previous eye-tracking studies measuring eye movements of chimpanzees and humans in naturalistic conditions indicated that the majority of eye positions fall within 20° , the eye position adopted in our stimuli , in both species ( Kano and Tomonaga , 2013; Kothari et al . , 2020 ) . H4 and H5 ( perceptual expertise in own-species and normal eye contrast polarity ) also cannot largely explain our results . Finally , one pattern of our results could not be explained by H1 alone , specifically that the negative human eyes and positive chimpanzee eyes affected participants’ task performance similarly in some conditions . More specifically , human participants ( collectively ) performed similarly in trials presenting the human and chimpanzee stimuli in the L4-reversed condition ( i . e . , the negative human eye and positive chimpanzee eye in the darkest and smallest images ) . Also , chimpanzee Hatsuka performed similarly in trials presenting the human and chimpanzee stimuli during the Test B ( reversed contrast polarity ) phase ( unlike Natsuki ) . These partial results are explained by either H2 or the combination of H1 and H3 . Overall , however , our results indicate that the presence of the uniformly white sclera ( and a darker iris ) in our stimuli is the primary factor affecting the similarity of our results between the participants of both species in our study design . Our results thus suggest that the visibility of human eye-gaze is primarily supported by basic color properties of the eyes , and thus by a basic perceptual mechanism shared among human and chimpanzee participants . Relatedly , there are a number of previous studies documenting the similarities in visual perception , including spectrum sensitivity , visual acuity , and contrast sensitivity , between humans and nonhuman apes ( Deeb et al . , 1994; Dulai et al . , 1994; Jacobs et al . , 1996; Matsuno et al . , 2004; Matsuno and Tomonaga , 2006; Matsuzawa , 1990; Bard et al . , 1995; Adams et al . , 2017 ) though small differences might exist ( Jacobs et al . , 1996; Adams et al . , 2017 ) . One notable aspect of our results is that chimpanzees required extensive training to distinguish between different gaze directions of human and chimpanzee eyes , and many of our chimpanzees were unable to pass all the required training phases . Given that many of our chimpanzees were already trained for simple color and form perception tasks ( e . g . , Matsuno et al . , 2004; Matsuzawa , 1990 ) , this difficulty might suggest that chimpanzees may generally find it more difficult attending to detailed perceptual features of the eyes compared to humans , consistent with the previous training ( Tomonaga and Imura , 2010 ) and gaze-following/cueing studies ( Tomasello et al . , 2007; Tomonaga , 2007; but see Povinelli and Eddy , 1996 ) . In this sense , consistent with the previous experimental studies in humans ( Tomasello et al . , 2007; Whalen et al . , 2004; Provine et al . , 2013b; Ricciardelli et al . , 2000 ) , our results also suggest that gaze perception in humans is supported by both the uniformly white sclera of eyes and their perceptual expertise in such detailed eye features . One clear limitation of this study is that only a small number of chimpanzees could participate in our test conditions , which hampers the generalization of our results . Nonetheless , if the performance of the current task is supported by simple color properties of eyes and basic visual perception of great apes as argued above , our results should be replicated in other ( trained ) individuals and likely also in other great ape species . However , further tests are necessary to confirm this prediction . Finally , it remains unanswered whether our results are generalizable to other stimuli with some variations in sclera color , specifically given that some chimpanzee individuals have partly unpigmented ( white ) sclera ( Perea-García et al . , 2019; Caspar et al . , 2021; Mearing and Koops , 2021; Kano et al . , 2021 ) . However , such sclera is typically characterized as more graded or patchy compared to humans’ uniformly white sclera . Moreover , as noted earlier , the previous simulation study demonstrated that the visibility of eye-gaze ( especially that of eye-outline edges ) is limited with the eyes of all nonhuman ape species without the human-like uniformly white sclera , particularly in visually challenging conditions ( Kano et al . , 2021 ) . We thus expect a similar pattern of results even when using eyes with partly unpigmented sclera in our experiments . Again , however , further experimental studies are necessary to confirm this prediction . In conclusion , we demonstrated that uniform whiteness in the exposed sclera enhances eye-gaze signaling . We thus provided experimental support for the gaze-signaling hypothesis despite recent criticisms on this hypothesis ( Perea-García et al . , 2019; Caspar et al . , 2021; Mearing and Koops , 2021 ) . However , we also propose several significant updates on this hypothesis . Specifically , we found that it is the uniformly white sclera but not necessarily other distinguishing features , such as iris-sclera color difference and some variation in horizontal eye elongation , that critically distinguishes the human eye from the chimpanzee eye in terms of the visibility of eye-gaze direction . Moreover , we found that one function of the uniformly white sclera is to equip the eye-gaze signal with robustness against degradation caused by natural noises ( e . g . , shading , distancing ) . These new findings , when combined with the original and related hypotheses , suggest that humans have evolved special external eye morphology for conspecific communication and that it is a vital part of communication for humans to read conspecific eye-gaze cues during their everyday cooperative and cultural activities . Study 1 tested 25 human adults ( 14 females , 11 males; 23 East/South Asians and 1 Caucasian male ) who had moderate to extensive experience in caretaking or studying chimpanzees ( 3 months = 1; 1–5 years = 10; 5–10 years = 4; >10 years = 10 ) . Although our human participants were mostly from similar cultural backgrounds , two related experimental studies tested participants from other cultural backgrounds ( Ricciardelli et al . , 2000; Yorzinski and Miller , 2020 ) , and thus our results are complementary to those previous results . Our participants included 10 individuals who had extensive experience interacting with chimpanzees over a decade . We confirmed the same results when we restricted our analyses to those participants . All human participants were workers or students at Kumamoto Sanctuary ( KS ) or Primate Research Institute ( PRI ) who were directly invited to participate in this experiment . All were naïve to the experimental hypotheses in this study . All reported having normal to corrected-to-normal vision and no color blindness . Written informed consent was obtained from all participants before the study . The experimental protocol was approved by the internal ethical committee for human experiments in PRI ( no . 2020-05 ) . Study 2 trained 10 chimpanzees ( nine females , one male ) . Among them , three chimpanzees ( Natsuki , Hatsuka , Pendesa ) passed all the training stages and participated in Experiment 1 . Two of these three chimpanzees ( Natsuki , Hatsuka ) participated in Experiment 2 . Daily veterinary checks indicated no specific visual problems ( including color blindness ) that may have interfered with the execution of current experiments in our chimpanzee participants ( though some minor visual problems may exist in Pendesa; Kaneko et al . , 2013 ) . Chimpanzees lived in a social group of conspecifics at KS or PRI . All chimpanzees were tested in a dedicated testing room at each facility , and their daily participation was voluntary , in that they could decide whether to enter the testing room on a given testing day . They received regular feedings , daily enrichment , and had ad libitum access to water . Animal husbandry complied with institutional guidelines ( KS: Wildlife Research Center ‘Guide for the Animal Research Ethics’; PRI: 2002 version of ‘The Guidelines for the Care and Use of Laboratory Primates’ ) , and the research protocol was approved by the institutional research committee ( KS: WRC-2020-KS008A/009A; PRI: 2020-193/209 ) . See Appendix 2—table 1 for details about participants . Study 1 tested human participants in a standard office setting either in KS or PRI . Two participants were tested remotely online given the COVID-19 situation at the time of the experiment . They received the same task program online and performed the task on their computer in a standard office . Although slight differences existed in experimental setups between these two and the other participants ( detailed below ) , we confirmed that including or not including them in our analysis yielded the same results . Participants sat in front of a 23-inch monitor ( 52 . 7 × 29 . 6 cm , SE2416H , Dell , Round Rock , TX; for one online participant , 52 . 2 × 29 . 3 cm , 243V5QHABA/11 , Phillips , Amsterdam , the Netherlands; for the other online participant , 50 . 9 × 28 . 6 cm xub2390hs-b3 , Iiyama , Japan; all monitors were in 1920 × 1080 pixels and set at 100% brightness and 50% contrast ) . They placed their second- to fourth-digit fingers on the left , down , and right keys of a standard keyboard connected to the computer . With this setup , the viewing distance was about 60–70 cm . Participants were told to sit in front of the monitor as they normally would and not to move their original head position throughout the experiments . Study 2 tested chimpanzee participants in a testing room equipped with touch panels ( ET1790L-7CWB-1-ST-NPB-G , Touch Panel Systems , Yokohama , Japan , in KS; LCD-AD172F2-T , IO-DATA , Kanazawa , Japan , in PRI; both 34 . 5 × 26 . 0 cm; both 1280 × 1024 pixels; both 100% brightness and 50% contrast ) installed with their centers 45 cm from the floor . With this position , the eye level of the chimpanzees was roughly at the center of the monitor when they sat on the floor . The monitor was installed 15 cm behind transparent polycarbonate panels , and there was a rectangle hole sized 40 × 15 cm on the panel so that chimpanzees could view the stimuli through the panel while making a touch response by inserting their arm through the hole ( Figure 1 ) . With this setup , the viewing distance was about 30–40 cm . This visual distance for chimpanzee participants is shorter than that for human participants , and thus overall task difficulty should be lower for them , consistent with other procedural differences that we made to ease the task difficulty for chimpanzees . We prepared chimpanzee and human facial images with different levels of sizes and brightness . Eyeball regions of these facial images were manipulated to have either normal or reversed contrast polarity ( Appendix 3—figure 3 ) . To create the chimpanzee stimuli , we selected 10 high-resolution facial images of chimpanzees of both sexes from image collections obtained from colleagues at KS . As we sampled images from KS chimpanzees , some of our stimulus chimpanzees were familiar to KS chimpanzees ( half of the stimulus chimpanzees used in both training and test sessions ) , while all stimulus chimpanzees were unfamiliar to PRI chimpanzees . Yet , we confirmed that KS and PRI chimpanzees did not systematically differ in their performance during the training sessions ( see Appendix 4—figure 1 ) . All stimulus chimpanzees were familiar to KS human participants , while most stimulus chimpanzees were unfamiliar to PPI human participants . Yet , we confirmed that KS and PRI human participants performed similarly in trials presenting the chimpanzee stimuli . The selection criteria of stimulus photographs were as follows: ( 1 ) the photographed individual oriented both head and eyes directly to the camera , ( 2 ) all eye features of the individual were clearly visible , ( 3 ) no strong shade was visible on the face , and ( 4 ) no expression was shown in the face . See the author’s online repository for the full set of chimpanzee stimuli ( https://osf . io/2xny3/ ) . The selected chimpanzee individuals included juveniles and adults of both sexes , and their sclera colors were uniformly dark ( from the iris edge to the eye corner ) . Although some chimpanzee individuals have unpigmented spots in their sclera ( Perea-García et al . , 2019; Caspar et al . , 2021; Mearing and Koops , 2021; Kano et al . , 2021 ) , we selected individuals having relatively uniformly dark sclera to simplify our experimental comparisons ( and manipulated them to have dark sclera in eye corners of averted gaze faces; see below for details ) . To create the human stimuli , we selected 10 high-resolution facial images of humans from the image collections published for research use ( Egger et al . , 2011 ) based on the same criteria as above . The selected human individuals were teenagers of both sexes in various ethnicities with various skin and eye colors , and their sclera colors were uniformly white ( from the iris edge to the eye corner ) . All stimulus humans were unfamiliar to both chimpanzee and human participants . We balanced the selection of human stimulus individuals so that we could include a wide variety of skin and eye colors among them . Study 1 used the whole sets of chimpanzee and human stimuli , which consisted of 10 stimulus individuals in each stimulus species , and Study 2 used six stimulus individuals in each stimulus species ( due to procedural differences between studies; see below ) . The facial images of both chimpanzees and humans were then cropped to include only the face and hair and auto-level adjusted to reduce the variations in overall brightness and contrast across images using Photoshop ( Adobe , San Jose , CA ) . The cropped images were then pasted into a uniform 50% gray background sized 400 × 400 pixels . The size of each cropped facial image ( for both chimpanzee and human image ) was adjusted based on its iris diameter , which was set at 16 pixels ( 4 . 2–4 . 4 mm on the monitors used in both studies 1 and 2 ) . This size adjustment was performed to test the effect of eye shape ( horizontal elongation of eye-opening , related to H3 ) independently from its absolute size and also to test the effect of white sclera independently from its exposed area size ( see below for the quantification of these parameters ) . It should be noted that , due to these controls , our chimpanzee stimuli were presented as slightly larger than the size proportional to human stimuli because the eyeball size of humans is generally slightly larger ( about 5–10% ) in that of chimpanzees ( Ross and Kirk , 2007; Bekerman et al . , 2014; Kirk , 2004 ) . To create facial images with averted gaze , the eyeball part of each face ( with direct gaze ) was cropped and then shifted six pixels to the side ( Appendix 3—figure 3; this corresponded to the rotation of the eyeball of about 20° in both stimulus species ) . We then filled the blank areas in the shifted eye by copying the sclera colors of the original image using the ‘stamp’ tool in Photoshop . To create the facial images in which the eyes had reversed contrast polarity , we first cropped the eyeball part of each face ( with both direct and averted gaze ) and then inverted the lightness ( or grayscale ) component of the cropped part while keeping its chromaticity component unchanged ( to avoid unnatural bluish appearance in the eyes ) in a custom-made MATLAB program ( MathWorks , Natick , MA ) . We then evaluated the shape and color of the eyes in our images following a previous method ( Kano et al . , 2021 ) . We first created the region-of-interest ( ROI ) mask respectively for iris and sclera by tracing and filling the edge of each feature in Photoshop and a custom-made MATLAB program . We then calculated the color of each ROI as the mean of CIELAB color in all pixels within that ROI . CIELAB color system is created to simulate a perceptually uniform color space in humans and is also considered applicable to nonhuman primates with human-like trichromatic color vision ( Stevens et al . , 2009 ) . We then calculated the color difference between iris and sclera in each image as a Euclidean difference between the mean values of these ROIs . We confirmed that , as found in a previous study ( Kano et al . , 2021 ) , the iris-sclera color differences did not significantly differ between our human and chimpanzee stimuli ( with both normal and reversed contrast polarities; Appendix 3—figure 2 ) . We also measured the shape of the eyes in each image using the same ROIs . We confirmed that , as found in the previous study ( Kano et al . , 2021 ) , the human eye was horizontally longer than the chimpanzee eye , but the size of the sclera ROI did not differ between the human and chimpanzee eye in our stimuli ( Appendix 3—figure 2 ) . Finally , we converted the facial images to various levels of sizes and brightness . Study 1 used four stimulus levels ( L1–4 ) . L1 stimuli measured 400 pixels in width ( original ) and 100% brightness ( original ) , L2 stimuli measured 200 pixels in width ( 1/2 ) and 50% brightness ( 1/2 ) , L3 stimuli measured 100 pixels in width ( 1/4 ) and 33% brightness ( 1/3 ) , and L4 stimuli measured 50 pixels in width ( 1/8 ) and 25% brightness ( 1/4 ) . These size and brightness levels were determined based on pilot experiments with two human participants ( who did not participate in Study 1 ) so that the gaze direction of L4 stimuli was recognizable to both participants at least in one of the stimulus species with either positive or negative eyes . In Study 2 , we prepared three additional stimulus levels , L1 . 5 , L2 . 5 , and L3 . 5 , which were the intermediate between L1 and 2 , L2 and 3 , and L3 and 4 , respectively , in terms of size and brightness ( i . e . , L1 . 5: 300 pixels in width , 75% brightness; L2 . 5: 150 pixels in width , 42% brightness; L3 . 5: 75 pixels in width , 29% brightness ) so that chimpanzees could move to the next stimulus level without showing substantial drops in their performances ( see details about the test procedures below ) . Studies 1 and 2 used identical stimuli except that Study 1 presented the whole face in a 1:1 square image ( e . g . , 400 × 400 pixels ) , while Study 2 presented only the eye region in a 4:1 rectangle image ( e . g . , 400 × 100 pixels ) to reduce attentional demands on chimpanzees . We made the task procedures of studies 1 and 2 as similar as possible , although several unavoidable differences existed because chimpanzees required extensive training to master the gaze-detection task . In Study 1 , the task for the human participants was to indicate the direction of gaze ( left/front/right ) in the stimulus face presented at the center of the screen by keypress in each trial . They were instructed to answer as accurately and quickly as possible . Study 1 consisted of two experiments . All participants completed experiments 1 and 2 in this order . Experiment 1 presented stimuli with normal eye contrast polarity at L1–4 levels . Experiment 2 presented stimuli with eyes having both normal and reversed contrast polarities at L3–4 levels . Before each experiment , they completed 20 practice trials presenting L1 stimuli ( with the stimuli with normal eye contrast polarity for Experiment 1 and those with both normal and reversed eye contrast polarities for Experiment 2 ) . Each experiment consisted of a total of 96 trials with eight blocks ( 12 trials within each block ) . In both experiments , each block presented the stimuli of the same species , and the eight blocks alternately presented chimpanzee and human stimuli . In Experiment 1 , each block started with three consecutive trials presenting L1 stimuli of either species ( with normal eye contrast polarity ) and then increased the stimulus level every three trials; namely , 1st–3rd trials , 4th–6th trials , 7th–9th trials , and 10th–12th trials respectively presented L1 , L2 , L3 , and L4 stimuli of either species . Thus , in Experiment 1 , 12 trials presented stimuli of either species ( chimpanzee , human ) at each stimulus level ( L1–4 ) . In Experiment 2 , each block ( 12 trials ) started with six consecutive trials presenting L3 stimuli of either species and then six consecutive trials presenting L4 stimuli of the same species . The first two blocks ( first and second blocks ) presented stimuli of the two species with normal eye contrast polarity ( each block presented one species ) , and then the next two blocks ( third and fourth blocks ) presented stimuli of the two species with reversed eye contrast polarity . The fifth and sixth blocks and the seventh and eighth blocks again presented stimuli of the two species with normal eye contrast polarity and then those with reversed eye contrast polarity . Thus , in Experiment 2 , 12 trials presented stimuli of either species ( chimpanzee , human ) at each stimulus level ( L3 , L4 ) in either eye contrast polarity ( normal , reversed ) . Each participant completed all experiments in 25–30 min . All experiments were conducted in November 2020 . The number of times in which each gaze direction ( left/front/right ) was presented was balanced in each participant ( i . e . , each direction was presented in 32 trials per participant ) , and the number of times in which each stimulus individual was presented was also balanced both within each participant and across participants ( i . e . , each stimulus individual was presented on average 4 . 8 times per participant ) . The order of presenting the chimpanzee or human stimuli in the first block was counterbalanced across participants . The orders of gaze directions ( left/front/right ) and stimulus individuals were pseudorandomized so that the same gaze direction was not presented in more than two successive trials , and the same stimulus individual was not presented in any successive trials . In Study 2 , the task for the chimpanzee participants was to indicate the image with averted gaze ( shifted to the right; called the target image ) among two other images with direct gaze ( called the distractor images ) by a touch response in each trial ( i . e . , three-item visual search task , following the task design by Tomonaga and Imura , 2010 ) . The target and distractor images differed only in their eye-gaze direction ( but not in their eye contrast polarity or brightness/size ) . The three images were centered at 220 , 640 ( center ) , and 860 pixels horizontally and 512 pixels vertically on a 1280 × 1024 pixels monitor . Chimpanzees were given a sip of grape juice or a piece of apple ( depending on their preference ) when they answered correctly in each trial ( the same amount of reward was given for each chimpanzee throughout the study ) . Before training , we performed a pilot experiment ( 200–600 trials for each chimpanzee ) to decide the general task design , especially in terms of the number of distractors in each trial , the number of trials in each session , and the features of initial stimuli , so that the chimpanzees could gradually learn the task . As in Study 1 , Study 2 consisted of two experiments . Experiment 1 presented stimuli with normal eye contrast polarity , and Experiment 2 presented those with reversed eye contrast polarity and then those with normal eye contrast polarity . Thus , these two experiments presented stimuli with normal and reversed eye contrast polarities in the ABA design . Throughout Study 2 ( both training and test ) , each session consisted of 48 trials and four blocks . Each block ( 12 trials ) presented stimuli of the same species , and the four blocks alternately presented the chimpanzee and human stimuli . Each chimpanzee performed 1–8 sessions per day depending on their motivation . Each session lasted about 10 min . Study 2 took about 8 months from August 2020 to March 2021 including both training and test periods . Training performed before these two experiments in Study 2 consisted of six training stages , and chimpanzees were trained for the task in a step-by-step manner through these stages . Training stage 1 presented the target image with no iris with the distractor images with irises ( direct gaze; see Appendix 3—figure 4 for the examples ) . Training stages 2–4 presented the target image in which the iris was positioned in 38° , 30° , and then 20° ( the final iris position was 20°; Appendix 3—figure 4 ) , following the training procedure employed by a previous study ( Tomonaga and Imura , 2010 ) . Training stages 1–4 used two stimulus individuals per stimulus species , and training stages 5 and 6 added two new stimulus individuals per stimulus species in each stage; thus , training stages 5 and 6 respectively presented four and six stimulus individuals per stimulus species ( the final stimulus set in training stage 6 ) . The criterion of passing each training stage was either scoring over 90% in one session or 80% in two consecutive sessions both in trials presenting the chimpanzee stimuli and those presenting the human stimuli . We trained chimpanzees in the same number of trials for the human and chimpanzee stimuli to avoid biasing their learning for either stimulus species . Eight chimpanzees learned this visual search task with the most basic stimulus set ( training stage 1 ) and continued to the next training stages ( training stages 2–6 ) . During the latter training stages , there was no consistent bias across individuals in their performances for the trials presenting the human and chimpanzee stimuli . However , some chimpanzees performed notably better in trials presenting the chimpanzee stimuli than those presenting the human stimuli ( Cleo , Iroha , Mizuki ) and some showed the opposite pattern ( Pendesa ) , while others performed similarly in those trials ( Appendix 4—figure 1 ) . These observed individual differences do not appear to be related to the particular backgrounds of each individual ( Appendix 1—table 1 ) . Three chimpanzees ( Natsuki , Hatsuka , Pendesa ) passed all training stages after extensive training ( Appendix 4—table 2 ) and learned to reliably differentiate the eye-gaze directions of both humans and chimpanzee stimuli ( L1 , normal eye contrast polarity; Appendix 4—figure 1 ) . Three chimpanzees passed all the training stages , namely , that they performed reliably in both trials presenting the chimpanzee and human stimuli and then participated in Experiment 1 . See Appendix 4—table 2 for the number of sessions each chimpanzee had in each training stage . Experiment 1 was divided into pre-test and test phases ( pre-Test A1 and Test A1 phases ) . To test chimpanzees at stimulus levels higher than L1 , we incremented the stimulus level by 0 . 5 when the chimpanzee scored above 85% in two successive sessions ( in all trials at stimulus level L1 and test trials at stimulus level higher than L1; see below for details about the test and baseline trials ) . The pre-test phase started from stimulus level L1 and the test phase started from stimulus level L2 . 5 . We used stimulus levels higher than ( or equal to ) L2 . 5 for the test phase because the results from Study 1 ( human participants ) indicated that clear performance differences between the stimulus species ( i . e . , test conditions ) emerged at stimulus levels higher than L2 . To examine individuals’ performance across sessions , each individual completed a minimum of 20 test sessions . To avoid the ceiling effect , we incremented the stimulus level by 0 . 5 when the individual scored above 85% in two successive sessions in test trials during the test phase . Two chimpanzees ( Natsuki and Hatsuka ) participated in Experiment 2 . The other chimpanzee ( Pendesa ) took more than twice as many sessions as the other two chimpanzees for training and thus was dropped from Experiment 2 ( Appendix 4—table 2 and Figure 1 ) . Experiment 2 first presented stimuli with reversed eye contrast polarity ( pre-Test B and Test B phases ) and then presented those with normal eye contrast polarity ( pre-Test A2 and Test A2 phases ) . As in Experiment 1 , the pre-Test B and pre-Test A2 presented L1–2 stimuli , and the Test B and Test A2 phase presented L2 . 5–4 stimuli . Each chimpanzee completed a minimum of 20 test sessions respectively in the Test B and Test A2 phases . The other procedures were identical to those in Experiment 1 . It should be noted that , although the number of these test sessions varied across individuals and test phases , we confirmed that limiting the dataset to 20 sessions in all individuals and sessions yielded the same results . See Appendix 4—table 3 for the number of sessions each chimpanzee had in each pre-test and test stage . In both experiments 1 and 2 , ( pre- ) test sessions presenting stimulus levels higher than or equal to L1 . 5 consisted of 24 baseline and 24 test trials . The baseline trials presented L1 stimuli , and the test trials presented the stimuli at higher levels . Each block ( 12 trials ) presented six baseline trials consecutively and then six test trials . Thus , in each test session , 12 trials presented stimuli of either species ( chimpanzee , human ) at the L1 ( baseline trials ) or higher levels ( test trials ) . Chimpanzees maintained high performances in baseline trials across sessions for both human and chimpanzee stimuli in the Test A1 phase ( L1 stimuli with normal color; Natsuki: 89% ± 12% vs . 90% ± 8%; Hatsuka: 91% ± 10% vs . 89% ± 8%; Pendesa; 97% ± 5% vs . 91% ± 7%; mean ± SD ) , Test B phase ( L1 stimuli with reversed eye contrast polarity; Natsuki: 91% ± 7% vs . 94% ± 7%; Hatsuka: 92% ± 10% vs . 96% ± 6%; mean ± SD ) , and Test A2 phase ( L1 stimuli with normal eye contrast polarity; Natsuki: 91% ± 10% vs . 94% ± 7%; Hatsuka: 92% ± 7% vs . 96% ± 5; mean ± SD ) . As in Study 1 , the number of times in which the target image ( with averted gaze ) was presented on each location ( left/center/right ) was balanced in each session ( i . e . , each gaze direction was presented in 16 trials per session ) , and the number of times in which each stimulus individual was presented was balanced in each session ( i . e . , each stimulus individual was presented four times per session ) . The order of presenting chimpanzee or human images in the first block was counterbalanced across sessions . The locations of the target images and the order of presenting the stimulus individuals were pseudorandomized so that the target image did not appear on the same location in more than two successive trials , and the same stimulus individual was not presented in any successive trials . See Appendix 4—table 1 for the summarized descriptions of each stage at the training and ( pre- ) test phases and Appendix 4—table 2 and Appendix 4—table 3 for the number of sessions in each stage at the training and ( pre- ) test phases , respectively . To test the participants’ performance differences between conditions , we ran a binomial GLMM in R ( version 4 . 0 . 5 ) . In Experiment 1 of Study 1 , the model included participants’ correct response ( correct , incorrect ) as the response variable , stimulus species ( chimpanzee image , human image ) , and stimulus level ( L1–4 ) as test fixed factors , the interaction between those test factors , block and ( within-block ) trial , which was nested in each block , as control fixed factors , and participant and stimulus individual as random factors ( see Appendix 1—table 1 for the formulas ) . In Experiment 2 of Study 1 , we used the same model with eye contrast polarity ( normal , reversed ) as an additional test fixed factor ( and its interaction with the other test factors; Appendix 1—table 1 ) . In Study 2 ( chimpanzees ) , as we evaluated chimpanzees’ performance in repeated sessions and adjusted their performances by incrementing stimulus levels according to their performance , we treated the session as a random factor and did not include stimulus level in the model . Therefore , the model included stimulus species as a test fixed factor , block and ( within-block ) trial as control fixed factors , and stimulus individual and sequence as random factors ( Appendix 1—table 1 ) . Study 2 performed statistical tests for each chimpanzee with the α level adjusted for the number of individuals in the Bonferroni correction; namely , 0 . 05/3 in Experiment 1 and 0 . 05/2 in Experiment 2 . For all models in studies 1 and 2 , we included all possible random slope components , although we removed the correlations between random slopes and intercepts to ease the nonconvergence issues ( Barr et al . , 2013 ) . Overdispersion was checked using the dispersion parameters derived from the R package ‘blmeco’ and did not seem to be an issue in any of our models ( they ranged between 0 . 77 and 1 . 12 ) . The significance of a given term was tested using a likelihood ratio test . Nonsignificant interaction terms were dropped to test the significance of lower-order terms . When the interaction term was significant , post-hoc comparisons were performed to examine simple effects at each factor level . All data and R scripts are available in our online repository ( https://osf . io/2xny3/ ) .	From an early age , we are able to detect the direction others are looking in ( known as eye-gaze ) and make eye contact with each other to communicate . The front of the human eye has a large white area known as the sclera that surrounds the darker colored iris and pupil in the center . Compared to us , chimpanzees and other nonhuman great apes have sclerae that are much darker in color or at least not as uniformly white as human eyes . Some researchers believe that the white sclera of the human eye may have evolved to make it easier for other individuals to detect the direction of our gaze . However , there is a lack of experimental evidence as to whether white sclerae actually helps humans to distinguish the direction of eye-gaze . Here , Kano , Kawaguchi and Yeow presented human and chimpanzee participants with images of other humans and chimpanzees on a computer screen and asked them to indicate the direction of eye-gaze in each image . The experiments found that both humans and chimpanzees were better able to discriminate the directions of eye-gaze from the images of humans than those of chimpanzees , particularly when the images were smaller or more shaded . Moreover , artificially altering the eyes in the chimpanzee images so that they were more human-like – that is , had a light-colored sclera and a darker iris – enabled both humans and chimpanzees to better discriminate the eye-gaze directions of the chimpanzees . Kano , Kawaguchi and Yeow’s findings indicate that white sclerae do indeed help both humans and chimpanzees to discriminate the direction of eye-gaze , even though only humans have white sclerae . This is likely because humans use eye-gaze in key social activities ( including learning languages , coordinating to complete complex tasks and transmitting cultural information ) , indicating that white sclerae may have evolved to enhance human-specific communication . To learn more about this type of communication , future research could focus on finding out when white sclerae first evolved .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2022	Experimental evidence that uniformly white sclera enhances the visibility of eye-gaze direction in humans and chimpanzees
The Zika virus has emerged as a global public health concern . Its rapid geographic expansion is attributed to the success of Aedes mosquito vectors , but local epidemiological drivers are still poorly understood . Feira de Santana played a pivotal role in the Chikungunya epidemic in Brazil and was one of the first urban centres to report Zika infections . Using a climate-driven transmission model and notified Zika case data , we show that a low observation rate and high vectorial capacity translated into a significant attack rate during the 2015 outbreak , with a subsequent decline in 2016 and fade-out in 2017 due to herd-immunity . We find a potential Zika-related , low risk for microcephaly per pregnancy , but with significant public health impact given high attack rates . The balance between the loss of herd-immunity and viral re-importation will dictate future transmission potential of Zika in this urban setting . To model the transmission dynamics of ZIKV infections and estimate relevant epidemiological parameters , we fitted an ento-epidemiological , climate-driven transmission model to ZIKV incidence and climate data of FSA between 2015 and 2017 within a Bayesian framework , similar to our previous work on a dengue outbreak in the Island of Madeira ( Lourenço and Recker , 2014 ) . The model is based on ordinary differential equations ( ODE ) describing the dynamics of viral infections within the human and mosquito populations ( Equations 1-5 and 6-10 , respectively ) . The human population is assumed to be fully susceptible before the introduction of ZIKV and is kept constant in size throughout the period of observation . After an infectious mosquito bite , individuals first enter an incubation phase , after which they become infectious to a mosquito for a limited period of time . Fully recovered individuals are assumed to retain life-long immunity . We assumed that sexual transmission did not significantly contribute to transmission dynamics and therefore ignored its effects ( Yakob et al . , 2016; Moghadas et al . , 2017; Maxian et al . , 2017 ) . For the dynamics of the vector populations we divided mosquitoes into two life-stages: aquatic and adult females . Adult mosquitoes were further divided into the epidemiologically relevant stages for arboviral transmission: susceptible , incubating and infectious . In contrast to human hosts , mosquitoes remain infectious for life . The ODE model comprised 8 climate-dependent entomological parameters ( aquatic to adult transition rate , aquatic mortality rate , adult mortality rate , oviposition rate , incubation period , transmission probability to human , hatching success rate and biting rate ) , whose dependencies on temperature , rainfall and humidity were derived from other studies ( see Table 1 ) . Four parameters ( baseline mosquito biting rate , mosquito sex ratio , probability of transmission from human-to-vector and human lifespan ) were fixed to their expected mean values , taken from the literature ( see Table 2 ) . To estimate the remaining parameters , alongside parameter distributions regarding the date of first infection , the human infectious and incubating periods , and the observation rate of notified ZIKV cases , we fitted the ODE model to weekly notified cases of ZIKV in FSA using a Bayesian Markov-chain Monte Carlo ( MCMC ) approach . The results are presented both in terms of mean dynamic behaviour of the ODE under the MCMC solutions and posterior distributions of key epidemiological parameters . A full description of the fitting approach and the estimated parameters can be found in the section Materials and methods . The reliance on Aedes mosquitoes for transmission implies that the transmission potential of ZIKV is crucially dependent on temporal trends in the local climate . We therefore investigated daily rainfall , humidity and mean temperature data in FSA between 2013 and May 2017 ( Figure 2A ) . The data showed erratic fluctuations in rainfall with sporadic episodes of intense rain but without a clear seasonal trend . Temperature , on the other hand , presented a much clearer seasonal signature with fixed amplitudes between 22 and 27 degree Celsius , peaking between December and May . Humidity showed an intermediate scenario and appeared correlated with periods of intense rainfall but negatively correlated with temperature . By fitting our climate-driven transmission model to the local climate and ZIKV case data ( see Material and Methods and Figure 2B ) we obtained parameter estimates for the mosquito lifespan as well as the viral extrinsic incubation period ( EIP ) for the same period . Mosquito lifespan and EIP are main drivers of vectorial capacity and both showed seasonal oscillations with median values of around 9 and 5 days , respectively ( Figure 2—figure supplement 3 ) , which are in line with ranges found in the literature ( Trpis et al . , 1995; Trpis and Hausermann , 1986; Andraud et al . , 2012; Hugo et al . , 2014 and Table 3 ) . Importantly , there was a strong negative temporal correlation between these two variables , with periods of longer EIP coinciding with shorter lifespans and vice-versa . This negative relationship resulted in large temporal variations in vectorial capacity and thus seasonal oscillations in the daily reproductive numbers , R0 , with a median value of 2 . 7 in the period 2015–2017 ( range 1 . 0–4 . 3 , Figure 2—figure supplement 3 ) , and 2 . 2 before 2015 , peaking in the local summer months between December and April ( Figure 2C ) . Importantly , R0 remained above 1 for the entire period , indicating a high suitability for ZIKV in FSA . It should be noted that R0 in this context is a time-dependent variable , i . e . R0⁢ ( t ) , but out of convenience we simply refer to it as R0 . We also looked at the relationship between each climatic variable and R0 and case counts ( Figure 2D and E , respectively ) . The transmission potential was strongly and positively correlated with temperature ( r2=0 . 728 ) and negatively with humidity ( r2=0 . 26 ) . As expected , from the highly random patterns in the climate series , there was no correlation between R0 and rainfall ( r2=0 . 008 ) . In contrast , there was an opposite trend in the relationship between the climatic variables and case counts , with a positive correlation with humidity ( r2=0 . 28 ) and a negative correlation with temperature ( r2=0 . 23 ) . As with R0 there was only a weak observable trend in the relationship between rainfall and the number of Zika cases . It should be understood that this macroscopic analysis does not take into account the expected temporal lags due to mosquito development , incubation periods etc . , so the purpose here was simply to identify a general qualitative relationship between climate , vectorial capacity and disease incidence . Four parameters of public health importance were estimated by our MCMC framework: the date of introduction , the human infectious period , the human ( intrinsic ) incubation period , and the case observation rate ( Table 4 ) . The posterior for the introduction date showed a strong support for an introduction into FSA in early-mid December 2014 ( estimated median: 10t⁢h of December ) , i . e . around 7–8 weeks before the first notified case ( Figure 3A ) . The estimated human infectious period was ≈6 days ( Figure 3C , median = 5 . 9 , 95% CI [5 . 47–6 . 14] ) , which was very similar to the estimated incubation period ( Figure 3D , median = 5 . 8 , 95% CI [5 . 6–6 . 15] ) and in line with previously estimated ranges for ZIKV ( Table 3 ) . In this context it is important to note that informative priors had been used for these 2 parameters ( Figure 2—figure supplement 2 . ) , and the posterior for the incubation period presented an adjustment of ≈-0 . 5 days relative to the proposed distribution from the literature . Of particular interest here was the very low observation rate ( Figure 3B ) , with a median of just under 0 . 004 ( median = 0 . 0039 , 95% CI [0 . 0038–0 . 0041] ) , which equates to less than 4 in 1000 infections having been notified during the epidemic in FSA . Although lower than other previously reported estimates , this would explain the relatively long period of low viral circulation before the epidemic took off in April 2015 . That is , based on our estimates , there were around 2 , 700 Zika infections during the first 2 months , of which only 10 were notified . More importantly , when applying this rate to the total number of cases we found that by the end of the first epidemic wave around 65% ( 95% CI [57 . 0–72 . 9] ) of the population in FSA had been infected by the virus . This high attack rate is not unusual for Zika , however , and is in general agreement with observations elsewhere ( Table 3 ) . As illustrated by the cumulative attack rate in Figure 4A , and similar to estimates from other regions in the world ( Table 3 ) , nearly 65% of the population got infected by ZIKV by the end of 2015 , which rose to over 75% ( 95% CI [76 . 9–84 . 3] ) by the end of 2016 . During the first wave most cases occurred off-season , here defined by our estimated daily reproductive number ( R0 ) , while the second wave appeared much more synchronized with the period of high transmission potential . Notably , this temporal phenomenon has also been observed for the chikungunya virus ( CHKV ) when it was first introduced into FSA in 2014 ( Faria et al . , 2016b ) . The amassed accumulation of herd-immunity during the first wave resulted in a marked difference between the estimated basic reproductive number , R0 , and the effective reproductive ratio ( Re ) by the end of 2015 ( Figure 4A ) . This in turn might explain the marked reduction in Zika cases in FSA in 2016 , at a time when the virus was infecting large numbers of individuals elsewhere in the country ( Figure 1A , C ) . At the start of 2016 , Re was estimated to be more than 3 times smaller than R0 , which increased to 5 by the beginning of 2017 . Projecting into the future using average climate data for this region showed that the mean effective reproductive number is expected to remain low and close to 1 for the next few years , suggesting a very weak potential for ZIKV endemicity in the near future . In fact , the sporadic nature of Zika cases in 2017 strongly suggest that herd immunity in this region is at a sufficiently high level to prevent sustained transmission . Furthermore , during 2017 , Re was on average less than 1 ( mean: 0 . 62 , range: 0 . 25–1 . 06 ) , and we would therefore argue that the small number of cases ( 1 . 4% of 2015–2017 ) were mostly a result of small transmission chains , either from resonant transmission from the previous year , or from introduction events from nearby locations . Crucially , this would also explain why our ODE model matched both the dynamics and the sizes of the first two epidemic waves in FSA between 2015 and 2016 but failed to capture the small number of cases during 2017 ( Figure 2B ) . Without external introductions of infectious individuals ( human or vector ) our results predicted an epidemic fade-out by 2017 , in accordance with the lack of notified cases after March 2017 ( Figure 4A ) . We therefore projected ZIKV’s epidemic potential over the next two decades ( until 2040 ) using stochastic simulations ( see Material and Methods ) while assuming different rates of viral introduction ( Figure 4B , C ) . Our results showed that the potential for ZIKV to cause another outbreak or to establish itself endemically in FSA is strongly dependent on the frequency of re-introductions , whereby higher rates of external introductions might in fact help to sustain high levels of herd immunity , whereas infrequent introductions are more likely to result in notable outbreaks . That is , semi-endemic behaviour was only observed in simulations with low introduction rates ( Figure 4B–C ) , as these scenarios strike a fine balance between a low number of new cases affecting herd-immunity levels and population turnover . In contrast , high introduction rates quickly exhaust the remaining susceptible pool , resulting in very long periods without epidemic behaviours . In effect , our estimated observation rate entails the proportion of real infections that would have been notified if symptomatic and correctly diagnosed as Zika . Based on the previously reported Yap Island epidemic of 2007 ( Duffy et al . , 2009 ) , the percentage of symptomatic infections can be assumed to be close to 18% . Unfortunately , measures of the proportion of individuals seeking medical attention and being correctly diagnosed do not exist for FSA , although it is well known that correct diagnosis for DENV is imperfect in Brazil ( Silva et al . , 2016 ) . We therefore performed a sensitivity analysis by varying both the proportions of infected symptomatic individuals seeking medical attention and the proportion of those being correctly diagnosed for Zika . Figure 5A shows that if any of these proportions is less than 10% , or both between 15–20% , our observation rate of 3 . 9 per 1000 infections can easily be explained . Finally we investigated the sensitivity of our results with regards to the expected number of newborns presenting microcephaly ( MC ) . Following the observation that virtually all reported MC cases were issued before the summer of 2016 and with a lag of 5–6 months ( Figure 1A ) , we assumed that the vast majority of Zika-associated MC cases would have been a consequence of the first epidemic wave in 2015 . We used the estimated attack rate of approximately 65% from 2015 ( Figure 4A ) and varied the local birth rate and the theoretical risk of MC to obtain an expected number of cases . In agreement with other reports ( de Araújo et al . , 2016; Cauchemez et al . , 2016; Jaenisch et al . , 2016; Johansson et al . , 2016 ) , our model predicted a relatively low risk for MC given ZIKV infection during pregnancy ( Figure 5B , C ) . In particular , using a conservative total of 21 confirmed MC cases in FSA , i . e . rejecting suspected or other complications , we estimate an average risk of approximately 0 . 35% of pregnancies experiencing ZIKV infection . Including the 3 foetal deaths where ZIKV infections were confirmed during pregnancy , i . e . using a total of 24 cases , only increased the risk to 0 . 39% . More generally , based on the results from our fitting approach and using the average birth rates of FSA as guideline , we estimate that on average 3–4 MC cases are expected per 100 k individuals at 65% exposure to the virus . Using an ento-epidemiological transmission model , driven by temporal climate data and fitted to notified case data , we analysed the 2015–2017 Zika outbreak in the city of Feira de Santana ( FSA ) , in the Bahia state of Brazil and determined the conditions that led to the rapid spread of the virus as well as its future endemic and epidemic potential in this region . Given FSA’s high suitability for ZIKV mosquito-vectors and its particular geographical setting as a state commerce and transport hub , our results should have major implications for other urban centres in Brazil and elsewhere . The pattern of reported ZIKV infections in FSA was characterized by a large epidemic in 2015 , in clear contrast to total reports at the country-level , peaking during 2016 . Most notably for FSA was the epidemic decay in 2016 and fadeout in 2017 . In order to resolve whether this was due to a lower transmission potential of ZIKV in 2016/2017 in FSA , we calculated the daily reproductive number ( R0 ) between 2013 and 2017 but found no notable decrease in 2016 . Interestingly , the maximum R0 in that period was observed in the season 2015/2016 , coinciding with El Niño ( Golden Gate Weather Services , 2017 ) and thus in line with the hypothesis that this phenomenon may temporary boost arboviral potential ( Caminade et al . , 2017; van Panhuis et al . , 2015 ) . By fitting our model to weekly case data we also estimated the observation rate , i . e . the fraction of cases that were notified as Zika out of the estimated total number of infections . It has previously been reported that the vast majority of Zika infections go unnoticed ( Table 3 ) , which is in agreement with our estimates of an observation rate below 1% . Based on this , around 65% of the local population were predicted to have been infected by ZIKV during the first wave in 2015 , which is in the same range as the reported Zika outbreaks in French Polynesia ( 66% ) ( Cauchemez et al . , 2016 ) and Yap Island ( 73% ) ( Duffy et al . , 2009 ) . The accumulation of herd-immunity caused a substantial drop in the virus’s effective reproductive number ( Re ) and hence a significantly lower number of cases during the second wave in 2016 and subsequent demise in 2017 . In the context of FSA , it is possible that the high similarity of case definition to DENV , the concurrent CHIKV epidemic , and the low awareness of ZIKV at that time could have resulted in a significant number of ZIKV infections being classified as either dengue or chikungunya . Furthermore , based on our analysis , we would argue that the percentage of correctly diagnosed ZIKV infections and infected individuals seeking medical attention must have been exceptionally low ( both lower than 20% ) . The age structure of notified cases showed a higher than expected incidence risk ratio ( IRR ) for individuals under the age of 4 years and a lower than expected risk for individuals aged + 50 years . This contrasts the observation during the Zika outbreak on Yap Island in 2007 , where all age classes , except the elderly , presented similar attack rates ( Duffy et al . , 2009 ) . We note here , however , that the Yap Island analysis was based on both a retrospective analysis of historical hospital records and prospective surveillance ( serology , surveys ) . It is therefore possible that the signatures amongst the youngest and oldest individuals in FSA may reflect deficiencies and/or biases in local notified data . Such signatures could emerge by both a rush of parents seeking medical services driven by a hyped media coverage or prioritization of child-care due to the emergence of microcephaly during the Zika epidemic and a small proportion of the elderly seeking or having access to medical attention . In fact , the increased risk in young children in 2016 may have been a result of increased awareness as well as the interventions by the WHO in the second year . We also found a small increase in IRR in the 20–34 years age group , particularly during 2016 , which could be indicative of the small contribution of sexual transmission ( Moghadas et al . , 2017; Maxian et al . , 2017 ) . Most of these observations are speculative , however , and more detailed data will be required to fully understand these age-related risk patterns . For instance , initiatives such as the ZiBRA Project ( ZIBRA , 2016; Faria et al . , 2016c; Faria et al . , 2017 ) , which perform mobile and real-time sampling with portable genome sequencing , could prove to be essential for a retrospective and future analysis of the ZIKV epidemic in Brazil , especially in areas where high levels of herd-immunity will prevent large-scale circulation in the coming years ( Ferguson et al . , 2016 ) . The implicit consideration of climate variables as drivers of vector biology allowed us to ascertain the relative roles of temperature , humidity and rainfall for Zika’s basic and effective reproductive potentials ( R0 and Re , respectively ) . Similar to other studies in temperate and tropical settings , we found that temperature , with its direct influence on mosquito lifespan , aquatic development and extrinsic incubation period , was the key driver of seasonal oscillations in the transmission potential ( Lourenço and Recker , 2014; Mordecai et al . , 2016; Mourya et al . , 2004; Feldstein et al . , 2015 ) . Rainfall , on the other hand , only seemed to play a marginal role and we argue that it may be a relevant player for arboviral transmission mainly in tropical regions subject to intense rain seasons , such as areas in South East Asia ( Cuong et al . , 2011; Hii et al . , 2012; Xuan et al . , 2014 ) . We also noted that the correlations between climatic variables and case counts were inverted when addressed against the transmission potential . For instance , while temperature was positively correlated with R0 it was negatively correlated with Zika cases . This implies that the transmission potential is readily responsive to climatic variation but that the Zika epidemics in FSA showed a slight but expected delay in relation to the peak in transmission potential , with case numbers generally increasing after a stable period of maximum R0 , followed by epidemic peaks that tended to coincide with declining R0 . An interesting observation is that the 2015 epidemic peaked approximately 3 months after the estimated peak in the virus’s transmission potential , whereas there was much higher synchrony during the second wave in 2016 . The same behaviour has been described for the CHIKV outbreak in FSA in 2014–2015 and which has been linked to highly discordant spatial distributions between the first two epidemics ( Faria et al . , 2016b ) . It is likely that similar spatial effects ( Kraemer et al . , 2017 ) were present in FSA’s ZIKV outbreaks . Unfortunately we did not have access to sufficiently detailed spatial data to explore this hypothesis further . A phylogenetic analysis has proposed that the introduction of ZIKV into Bahia took place between March and September 2014 , although without direct evidence for its circulation in FSA at that time ( Naccache et al . , 2016 ) . Our estimated date of introduction showed support for a date in early-mid December 2014 , a few months after the proposed introduction into Bahia and just over 7 weeks before the first case of Zika was notified in FSA . Similar periods between the first notification and estimated introduction often represent the time taken to complete one or more full transmission cycles ( human-mosquito-human ) before a cluster of cases is generated of sufficient size for detection by passive surveillance systems ( Lourenço and Recker , 2014 ) . The case data also shows a 2 months period after the first notification during which weekly case numbers remained extremely low . This long period was unexpected as persistent circulation of ZIKV could hardly be justified by the observed total of only 10 cases . Given our estimated observation rate , however , the number of ZIKV infections during this time could have amounted to over 2700 actual cases . In April , the number of cases increased rapidly , coinciding with the Micareta festival , which we argue may have played a role in igniting the exponential phase of the epidemic by facilitating human-vector mixing as well as a more rapid geographical expansion . After calibrating our model to the 2015–2017 epidemic , we projected the transmission of ZIKV beyond 2017 using stochastic simulations and average climatic variables for this region . Without the possibility of externally acquired infections , local extinction was very likely by 2017 due to the high levels of herd-immunity . According to our study , Zika’s reproductive potential ( Re ) reached its lowest point in 2017 , and it is expected to remain low for the next couple of years , given the slow replenishment of susceptibles in the population through births . When explicitly modelling the importation of infectious cases our projections for the next two decades corroborated the conclusions of previous modelling studies that suggest a weak endemic potential for ZIKV after the initial exhaustion of the susceptible pool ( Ferguson et al . , 2016; Kucharski et al . , 2016 ) . However , our simulations also showed that the future epidemic behaviour is strongly dependent on the frequency of re-introductions , where sporadic and unpredictable epidemics could still be in the order of hundreds of cases . Furthermore , given our estimated observation rate for the 2015–2017 epidemic , passive surveillance systems are unlikely to fully detect the scale and occurrence of such small epidemics , missing their actual public health impact , and as such efforts should thus be placed to improve ZIKV detection and diagnosis in order to optimize the local reporting rates and potential for control . Human sexual and vertical transmission of ZIKV is an important public health concern , especially within the context of potential Zika-associated microcephaly ( MC ) and other neurological complications in pre- and neonatals . With a total of over 10 , 000 live births in 2015 in FSA , our crude estimate for the risk of Zika-associated MC per pregnancy was below 4 cases per 100 , 000 individuals in a generalized population under an attack rate of 65% . As discussed elsewhere ( Cauchemez et al . , 2016 ) , this risk is extremely low when compared to other known viral-associated complications , such as those caused by infections by cytomegalovirus ( CMV ) and the rubella virus ( RV ) ( De Santis et al . , 2006; Naing et al . , 2016 ) . It is therefore crucial to reiterate that what makes the ZIKV a public health concern is not necessarily the per pregnancy risk of neurological complications , but rather the combination of low risk with very high attack rates . Other studies have reported that the risk for complications during the 1s⁢t trimester of gestation is higher than the one estimated here . For example , in the French Polynesia ( FP ) outbreak ( Cauchemez et al . , 2016 ) , the risk associated with ZIKV infection during the 1s⁢t trimester was 1% , while the overall , full pregnancy risk was 0 . 42% , similar to our FSA estimates . For the Yap Island epidemic , no microcephaly cases have been reported . With an estimated 24 births per 1000 females ( census 2000 as in ( Duffy et al . , 2009 ) ) and using an overall risk of approximately 0 . 4% per pregnancy , only 0–3 cases per 100 , 000 individuals would have been expected . However , the island’s small population size ( 7391 individuals ( Duffy et al . , 2009 ) ) together with a general baseline of 0–2 microcephaly cases per 100 , 000 in many areas of the world ( Johansson et al . , 2016; Butler , 2016; EUROCAT , 2003 ) would explain the absence of reported cases . It is also important to consider that a variety of birth defects have been found to be statistically associated with Zika virus infection during pregnancy , of which MC is one possible outcome . While the risk for birth defects per pregnancy is consistently reported to be high , estimations for the risk of MC vary considerably . For example , recent clinical trials ( Honein et al . , 2017; Brasil et al . , 2016 ) suggested that the risk of Zika-associated MC could be an order of magnitude higher than the estimate reported in this or other previous studies ( Cauchemez et al . , 2016; Duffy et al . , 2009 ) . At this stage it is not possible to explain these differences , but it is tempting to speculate that other factors must influence either the actual or estimated risk . For example , there could be diagnostic biases or differences between epidemiological and clinical studies . Alternatively , viral or host genetic background , as well as the pre-exposition to other arboviruses may influence the absolute risk experienced by local populations or cohorts . Official notification of Zika infections in Brazil started on the 1s⁢t of January 2016 , although cases were reported in many other regions in Brazil during 2015 . It is therefore plausible that the observation rate changed upon official guidelines and that the capacity to accurately diagnose and report Zika infections could have been lower in 2015 compared subsequent years . To explore this , we reran our fitting approach allowing for a possible change in the observation rate for 2016 and onwards ( Figure 3—figure supplement 2 ) and found a similar observation rate for 2015 ( 0 . 0039 versus 0 . 0034 ) as well as a similar attack rate between the two model variants . However , the estimated observation rate for 2016 and beyond was ≈4 times larger than for 2015 , implying a positive change due to changes in the surveillance system . Nevertheless , only about 13–14 out of 1000 Zika cases were reported after the 1s⁢t of January 2016 . It is hard to discern where the positive changes took place , but we suggest the revised diagnosis guidelines may have increased the proportion correctly diagnosed while the proportion of symptomatic individuals visiting medical facilities did not change . It is also tempting to speculate that the 2015/2016 imbalance in reporting may have been a general phenomenon across Brazil . As described elsewhere , it is thus possible that FSA is a good example of states and urban centres that may have witnessed larger epidemics than reported in 2015 ( de Oliveira et al . , 2017 ) . This , together with our conclusion that low MC risk with very high attack rates makes ZIKV a public health concern , could explain why most MC reports at the level of the country were in 2015 ( de Oliveira et al . , 2017 ) , although for many regions the total reported number of ZIKV cases may have been surprisingly small that year . There are certain limitations to our approach , many of which could be revisited when more detailed data becomes available . For example , we assumed homogeneous mixing between human and mosquito hosts but it is possible that spatio-temporal heterogeneities may have played a role in FSA . Furthermore , we have curated and integrated functional responses of key entomological parameters to temperature , rainfall and humidity variation , which were originally reported for dengue viruses . Our fitting approach is also dependent on notified case data and it is possible that the reported cases are not representative of the initial expansion of the virus , which may have thwarted the obtained posterior of the introduction date . Finally , our future projections for the endemic and epidemic potential of ZIKV are based on average climatic trends of past years and do not capture the occurrence of natural variation between years , in particular for years affected by major Southern American climate events , such as the El Niño ( Caminade et al . , 2017 ) . In this study we have addressed the local determinants of ZIKV epidemiology in the context of a major urban centre of Brazil . Our results imply that control and surveillance of ZIKV should be boosted and focused in periods of high temperature and during major social events . These factors could identify windows of opportunity for local interventions to mitigate ZIKV introduction and transmission and should be transferable to other areas for which both temperature data and community event schedules are available . We further confirm that the high transmission potential of ZIKV in urban centres can lead to the exhaustion of the local susceptible pool , which will in turn dictate the long-term epidemic and endemic behaviour of the virus . Depending on the rate of re-introduction , sporadic outbreaks are to be expected , although these will be unlikely to result in a notable increase in the number of microcephaly cases due to their limited sizes and low risk per pregnancy . Nonetheless , these local sporadic occurrences could still have important public health consequences , and we argue that much better diagnostics and reporting rates are required for local authorities to detect and respond to such events in the near future . Our integrated mathematical framework is capable of deriving key insights into the past and future determinants of ZIKV epidemiology and its findings should be applicable to other major urban centres of Brazil and elsewhere . Feira de Santana ( FSA ) is a major urban centre of Bahia , located within the state’s largest traffic junction , serving as way points to the South , the Southeast and central regions of the country . The city has a population of approximately 620 . 000 individuals ( 2015 ) and serves a greater geographical setting composed of 80 municipalities ( municipios ) summing up to a population of 2 . 5 million . Although major improvements in water supply have been accomplished in recent decades , with about 90% of the population having direct access to piped water , supply is unstable and is common practice to resort to household storage . Together with an ideal ( tropical ) local climate , these are favourable breeding conditions for species of the Aedes genus of mosquitoes , which are the main transmission vectors of ZIKV , CHIKV and the dengue virus ( DENV ) that are all co-circulating in the region ( Kraemer et al . , 2015; Carlson et al . , 2016 ) . FSA’s population is generally young , with approximately 30% of individuals under the age of 20% and 60% under the age of 34 . In the year of 2015 , the female:male sex ratio in FSA was 0 . 53 and the number of registered births was 10352 , leading to a birth rate standard measure of 31 new-borns per 1000 females in the population . Local climatic data ( rainfall , humidity , temperature ) for the period between January 2013 and May 2017 was collected from the Brazilian open repository for education and research ( BDMEP , Banco de Dados Meteorológicos para Ensino e Pesquisa ) ( Brazil BDMEP , 1961 ) . The climate in FSA is defined as semi-arid ( warm but dry ) , with sporadic periods of rain concetrated within the months of April and July . Between 2013 and 2015 , mean yearly temperature was 24 . 6 celsius ( range 22 . 5–26 . 6 ) , total precipitation was 856 mm ( range 571–1141 ) , and mean humidity levels 79 . 5% ( range 70 . 1–88 . 9% ) . Temperature , humidity and precipitation per day is available as Dataset 1 . ZIKV surveillance in Brazil is conducted through the national notifiable diseases information system ( Sistema de Informação de Agravos de Notificação , SINAN ) , which relies on passive case detection . Suspected cases are notified given the presence of pruritic maculopapular rash ( flat , red area on the skin that is covered with small bumps ) together with two or more symptoms among: low fever , or polyarthralgia ( joint pain ) , or periarticular edema ( joint swelling ) , or conjunctival hyperemia ( eye blood vessel dilation ) without secretion and pruritus ( itching ) ( Brazil SINAN , 2016; Brazil , 2016 ) . The main differences to case definition of DENV and CHIKV are the particular type of pruritic maculopapular rash and low fever ( as applied during the Yap Island ZIKV epidemic ( Duffy et al . , 2009 ) ) . The data presented in Figure 1 for both Brazil and FSA represents notified suspected cases and is available as Dataset 3 ( please refer to the Acknowledgement section for sources ) . Here , we use the terms epidemic wave and outbreak interchangeably ( but see ( Perkins et al . , 2016 ) ) . A total of 53 suspected cases with microcephaly ( MC ) or other neurological complications were reported in FSA between January 2015 and February 2017 . Using guidelines for microcephaly diagnosis provided in March 2016 by the WHO ( as in ( Faria et al . , 2016c ) ) , a total of 21 cases were confirmed after birth and follow-up . A total of 3 fetal deaths were reported for mothers with confirmed ZIKV infection during gestation but for which no microcephaly assessment was available . The first confirmed microcephaly case was reported on the 24t⁢h of November 2015 and virtually all subsequent cases were notified before August 2016 ( with the exception of 2 ) . The microcephaly case series can be found in Dataset 4 . The ordinary differential equations ( ODE ) model and the Markov-chain Monte Carlo ( MCMC ) fitting approach herein used are based on the framework previously proposed to study the introduction of dengue into the Island of Madeira in 2012 ( Lourenço and Recker , 2014 ) . We have changed this framework to relax major modelling assumptions on the mosquito sex ratio and success of egg hatching , have included humidity and rainfall as critical climate variables , and have also transformed the original least squares based MCMC into a Bayesian MCMC . The resulting framework is described in the following sections , in which extra figures are added for completeness . The dynamics of infection within the human population are defined in Equations 1-5 . In summary , the human population is assumed to have constant size ( N ) with mean life-expectancy of μh years , and to be fully susceptible before introduction of the virus . Upon challenge with infectious mosquito bites ( λv→h ) , individuals enter the incubation phase ( Eh ) with mean duration of 1/γh days , later becoming infectious ( Ih ) for 1/σh days and finally recovering ( Rh ) with life-long immunity . ( 1 ) dShdt=μhN−λv→h−μhSh ( 2 ) dEhdt=λv→h−γhEh−μhEh ( 3 ) dIhdt=γhEh−σhIh−μhIh ( 4 ) dRhdt=σhIh−μhRh ( 5 ) N=Sh+Eh+Ih+Rh For the dynamics of the mosquito population ( Equations 6-10 ) , individuals are divided into two pertinent life-stages: aquatic ( eggs , larvae and pupae , A ) and adult females ( V ) as in ( Yang et al . , 2009 ) . The adults are further divided into the epidemiologically relevant stages for arboviral transmission: susceptible ( Sv ) , incubating ( Ev ) for 1/γ˙v days and infectious ( Iv ) for life . The ˙ ( dot ) notation is here adopted to distinguish climate-dependent entomological factors ( further details in the following sections ) . ( 6 ) dAdt=c˙vfθ˙v ( 1−AK ( R+1 ) ) V− ( ϵ˙Av+μ˙Av ) A ( 7 ) dSvdt=ϵ˙AvA−λh→v−μ˙VvSv ( 8 ) dEvdt=λh→v−γ˙vEv−μ˙VvEv ( 9 ) dIvdt=γ˙vEv−μ˙VvEv ( 10 ) V=Sv+Ev+Iv Here , the coefficient c˙v is the fraction of eggs hatching to larvae and f the resulting female proportion . For simplicity and lack of quantifications for local mosquito populations , it is assumed that the sex ratio remains at 1:1 ( i . e . f=0 . 5 ) . Moreover , ϵ˙Av denotes the rate of transition from aquatic to adult stages , μ˙Av the aquatic mortality , μ˙Vv the adult mortality , and θ˙v is the success rate of oviposition . The logistic term ( 1-AK⁢ ( R+1 ) ) can be understood as the ecological capacity to receive aquatic individuals ( Tran et al . , 2013 ) , scaled by a carrying capacity term K⁢ ( R+1 ) in which K determines the maximum capacity and R is the local rainfall contribution ( further details on following sections ) . From Equations 6-10 , the mean number of viable female offspring produced by one female adult during its life-time , i . e . the basic offspring number Q , was derived ( Equation 11 ) . Most parameters defining Q are climate-dependent , and for fixed mean values of the climate variables ( ex . mean rainfall R¯ ) , expressions were derived for the expected population sizes of each mosquito life-stage modelled ( A0 , V0 ) which are used to initialize the vector population ( Equations 12-13 ) . ( 11 ) Q=ϵ˙Avϵ˙Av+μ˙Avc˙fθ˙vμ˙Vv ( 12 ) A0=K ( R¯+1 ) ( 1−1Q ) ( 13 ) V0=K ( R¯+1 ) ( 1−1Q ) ϵAv˙μVv˙ In respect to the infected host-type being considered , the vector-to-human ( λv→h ) and human-to-vector ( λh→v ) incidence rates are assumed to be , respectively , density-dependent and frequency-dependent ( Equations 14-15 ) . Here , av˙ is the biting rate and ϕ˙v→h and ϕh→v are the vector-to-human and human-to-vector transmission probabilities per bite . Conceptually , this implies that ( i ) an increase in the density of infectious vectors should directly raise the risk of infection to a single human , while ( ii ) an increase in the frequency of infected humans raises the risk of infection to a mosquito biting at a fixed rate . The basic reproductive number ( R0 ) is defined similarly to previous modelling approaches ( Equation 16 ) ( Wearing and Rohani , 2006; Lourenço and Recker , 2013 ) . We further derived an expression for the effective reproductive ratio ( Re , Equation 17 ) , taking into account the susceptible proportion of the population in real-time . ( 14 ) λv→h= ( av˙ϕ˙v→hIvSh/N ) ∝Iv ( 15 ) λh→v= ( av˙ϕ˙h→vIhSv/N ) ∝Ih/N ( 16 ) R0= ( V/N ) av˙ av˙ ϕ˙v→h ϕh→v γ˙v γhμ˙Vv ( σh+μh ) ( γh+μh ) ( γ˙v+μ˙Vv ) ( 17 ) Re= ( Sh/N ) × ( Sv/N ) ×R0/ ( V/N ) For the fitting process , the MCMC algorithm by Lourenco et al . is here altered to a Bayesian approach by formalising a likelihood and parameter priors ( Lourenço and Recker , 2014 ) . For this , the proposal distributions ( q ) of each parameter were kept as Gaussian ( symmetric ) , effectively retaining a random walk Metropolis kernel . We define our acceptance probability α of a parameter set Θ , given model ODE output y as: ( 18 ) α=min{1 , π ( y\|Θ⋆ ) p ( Θ⋆ ) q ( Θo\|Θ⋆ ) π ( y\|Θo ) p ( Θo ) q ( Θ⋆\|Θo ) } where Θ⋆ and Θo are the proposed and current ( accepted ) parameter sets ( respectively ) ; π ( y\|Θ⋆ ) and π ( y\|Θo ) are the likelihoods of the ODE output representing the epidemic data given each parameter set; p⁢ ( Θo ) and p⁢ ( Θ⋆ ) are the prior-related probabilities given each parameter set . We fit the Zika virus cumulative case counts per week , for which no age-related or geographical data is taken into consideration . For computational reasons and based on a previous approach ( Dorigatti et al . , 2013 ) , the likelihoods π were calculated as the product of the conditional Poisson probabilities of each epidemic data ( di ) and ODE ( yi ) data point: ( 19 ) π ( y\|Θ ) =∏i=1N[Pr{yi=di}] Note , in this case where we have low cases numbers in a large population , the Poisson likelihood represents a reasonable approximation to the Binomial process , which is expected to underlie the observed data . With the MCMC approach described above , all combinations of the open parameters in the ODE system that most likely represent the outbreak are explored ( Table 4 ) . In summary , the MCMC estimates the distributions for: ( 1 ) the carrying capacity K , an indirect estimate of the number of adult mosquitoes per human; ( 2 ) time point of the first case t0 , assumed to be in a human; ( 3 ) a linear coefficient η that scales the effect of temperature on aquatic and adult mortality rates; ( 4 ) a linear coefficient α that scales the effect of temperature on the extrinsic incubation period; ( 5 ) a non-linear coefficient ρ that scales the effects of humidity and rainfall on entoi mological parameters; ( 6 ) the human infectious period 1/σh; and ( 7 ) the human incubation period 1/γh . By introducing the linear coefficients η and α , the relative effect of temperature variation on mortality and incubation is not changed per se , but instead the baselines are allowed to be different from the laboratory conditions used by Yang et al . ( Yang et al . , 2009 ) . For solutions in which η , α→1 , the laboratory-based relationships are kept . For a discussion on possible biological factors that may justify η and α please refer to the original description of the method in ( Lourenço and Recker , 2014 ) and ( Brady et al . , 2013 ) . Finally , the introduction of ρ allows the MCMC to vary the strength by which entomological parameters react to deviations from local humidity and rainfall means . In practice , the effect of rainfall and humidity can be switched off when ρ→0 and made stronger when ρ→+∞ ( details below ) . Initial analysis of the MCMC output raised an identifiability issue between the human infectious period ( 1/σh ) and the linear coefficient ( η ) that scales the effect of temperature on vector mortality ( η scales the baseline mortality without changes to the response of mortality to temperature ) . Hence , changes in both η and 1/σh result in similar scaling effects on the transmission potential R0 ( Equation 16 ) and thus unstable MCMC chains for η and 1/σh , with the resulting posteriors appearing to be bimodal ( for which there was no biological support ) . We addressed this issue by using informative priors for four parameters for which biological support exists in the literature: η , 1/σh , 1/γh , and α . Gaussian priors were used with means and standard deviations taken from the literature ( see Figure 2—figure supplement 2 ) . The framework described above has only 4 fixed parameters that are neither climate-dependent nor estimated in the MCMC approach ( Table 2 ) . Amongst these , ϕh→v is the per bite probability of transmission from human-to-mosquito , which we assume to be 0 . 5 ( Lounibos and Escher , 2008; Mohammed and Chadee , 2011 ) ; the sex ratio of the adult mosquito population f is assumed to be 1:1 ( Lounibos and Escher , 2008; Mohammed and Chadee , 2011 ) ; the life-expectancy of the human population is assumed to be an average of 75 years ( WHO , 2016c ) ; and the biting rate is taken to be on average 0 . 25 although with the potential to vary dependent on humidity levels ( details below ) ( Trpis and Hausermann , 1986; Yasuno and Tonn , 1970 ) . For each of the temperature-dependent entomological parameters , polynomial expressions are found de novo or taken from previous studies fitting laboratory entomological data with temperature ( T ) values used in Celsius . For rainfall ( R ) and humidity ( U ) , positive or negative relationships to entomological parameters are introduced using simple expressions , with values used after normalization to [0 , 1] . We assume that some parameters are affected by a combination of temperature with either rainfal or humidity , but take their effects to be independent . A list of climate-dependent parameters and references is found in Table 1 . Polynomials of 4th degree for the mortality ( μAv , μVv ) and success ovipositon ( θv ) rates are taken from the study by Yang and colleagues under temperature-controlled experiments on populations of Aedes aegypti ( Equations 19-21 ) ( Yang et al . , 2009 ) . For aquatic to adult ( ϵAv ) rate we use the 7t⁢h degree polynomial of the same study ( Equation 20 ) . For the relationship between the extrinsic incubation period ( 1/γv ) and temperature we apply the formulation by Focks et al . which assumes that replication is determined by a single rate-controlling enzyme ( Focks et al . , 1995; Schoolfield et al . , 1981; Otero et al . , 2006 ) ( Equation 24 ) . The probability of transmission per mosquito bite ( ϕv→h ) is here modelled ( Equation 25 ) as estimated by Lambrechts and colleagues ( Lambrechts et al . , 2011 ) . Finally , the relationship between temperature and the fraction of eggs that successfully hatch ( cv ) is estimated de novo ( Equation 26 ) by fitting a 3r⁢d degree polynomial to Aedes aegypti and albopictus empirical data described by Dickerson et al . ( see Figure 2—figure supplement 1 ) ( Dickerson , 2007; Mohammed and Chadee , 2011 ) . ( 20 ) ϵAv ( T ) =0 . 131−0 . 05723T+0 . 01164T2−0 . 001341T3+0 . 00008723T4−0 . 000003017T5+5 . 153×10−8T6−3 . 42×10−10T7 ( 21 ) μAv ( T ) =2 . 13−0 . 3797T+0 . 02457T2−0 . 0006778T3+0 . 000006794T4 ( 22 ) μVv ( T ) =0 . 8692−0 . 1599T+0 . 01116T2−0 . 0003408T3+0 . 000003809T4 ( 23 ) θv ( T ) =−5 . 4+1 . 8T−0 . 2124T2+0 . 01015T3−0 . 0001515T4 ( 24 ) γv ( T ) =0 . 003359Tk298×exp⁡ ( 15000R ( 1298−1Tk ) ) 1+exp⁡ ( 6 . 203×1021R ( 1−2 . 176×1030−1Tk ) ) ( 25 ) ϕv→h ( T ) =0 . 001044T× ( T−12 . 286 ) × ( 32 . 461−T ) 1/2 ( 26 ) cv ( T ) = ( −184 . 8+27 . 94T−0 . 9254T2+0 . 009226T3 ) /100 . 0 We normalise the time series of rainfall ( R ) and humidity ( U ) , further using the mean normalised values ( R¯ , U¯ ) as reference for extreme deviations from the expected local tendencies ( Bicout and Sabatier , 2004; Tran et al . , 2013 ) . Rainfall is assumed to affect positively the fraction of eggs that successfully hatch ( cv ) ( Alto and Juliano , 2001; Rossi et al . , 2015; Tran et al . , 2013; Madeira et al . , 2002 ) . A similar positive relationship is taken for the vector biting rate ( av ) and humidity levels ( Yasuno and Tonn , 1970 ) , in contrast to a negative effect on the adult mosquito mortality rate ( μVv ) ( Alto and Juliano , 2001 ) . ( 27 ) cv ( R ) = ( R−R¯ ) /1+ ( R−R¯ ) 2 ( 28 ) av ( U ) = ( U−U¯ ) /1+ ( U−U¯ ) 2 ( 29 ) μVv ( U ) =U¯− ( U−U¯ ) /1+ ( U−U¯ ) 2 Below is the complete formulation for each entomological parameter in time ( t ) , depending on the climatic variables for which relationships are assumed to exist , including the MCMC fitted linear ( α , η ) and non-linear ( ρ ) factors described above . ( 30 ) ϵAv ( t ) =ϵAv ( T ) ( 31 ) μAv ( t ) =ημAv ( T ) ( 32 ) μVv ( t ) =ημVv ( T ) [1+μVv ( U ) ]ρ ( 33 ) θv ( t ) =θv ( T ) ( 34 ) γv ( t ) =αγv ( T ) ( 35 ) ϕv→h ( t ) =ϕv→h ( T ) ( 36 ) cv ( t ) =cv ( T ) [1+cv ( R ) ]ρ ( 37 ) av ( t ) =av[1+av ( U ) ]ρ A stochastic version of the ento-epidemiological framework was developed by introducing demographic stochasticity in the transitions of the dynamic system . This followed the original strategy described in ( Lourenço and Recker , 2014 ) , in which multinomial distributions are used to sample the effective number of individuals transitioning between classes per time step . Multinomial distributions are generalized binomials - B⁢i⁢n⁢o⁢m⁢i⁢a⁢l⁢ ( n , p ) - where n equals the number of individuals in each class and p the probability of the transition event ( equal to the deterministic transition rate ) . This approach has also been demonstrated elsewhere ( Lampoudi et al . , 2009 ) . The approach used in this study uses code in C/C++ , bash and R scripts and is available at https://github . com/lourencoj/ArboWeD2/tree/ArboWeD2_V1 . A copy is archived at https://github . com/elifesciences-publications/ArboWeD2 .	Mosquitoes can transmit viruses that cause Zika , dengue and several other tropical diseases that affect humans . Zika virus usually causes mild symptoms , but it is thought that infection during pregnancy can lead to brain abnormalities , including microcephaly , where babies are born with an abnormally small head . Recent studies have shed light on how the Zika virus spread from Africa to reach South America , the Caribbean and North America . However , much less is known about the ecological factors that contribute to the spread of the virus within towns , cities and other local areas . In 2015 , Brazil was struck by an outbreak of the Zika virus that led to an international public health emergency . Lourenço et al . used a mathematical model to explore the local conditions within Feira de Santana ( a major urban center in Brazil ) that contributed to the outbreak . The model took into account numerous factors , including temperature , humidity , rainfall and the mosquito life-cycle , which made it possible to reconstruct the history of the virus over the past three years and to make projections for the next decades . It revealed that most of the infections occured during 2015 , with approximately 65% of the population infected . The incidences of new infections declined in 2016 , as increasing numbers of local people had already been exposed to the virus and became immune . Temperature and humidity appeared to have played a critical role in sustaining the mosquito population carrying the Zika virus . Further analysis suggests that the risk of Zika virus causing microcephaly is very low – only 0 . 3–0 . 5% of the pregnant women in Feira de Santana who were infected with Zika gave birth to a baby with the condition . What therefore makes Zika a public health concern is the combination of a low risk with very high infection rates , which can affect a large number of pregnancies . This study will help researchers and policy makers to predict how the Zika virus will behave in the coming years . It also highlights the limitations and successes of the current system of surveillance . Moreover , it will help to identify critical time periods in the year when mosquito control strategies should be implemented to limit the spread of this virus . In future , this could help shape new local strategies to control Zika virus , dengue and other diseases carried by mosquitoes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "microbiology", "and", "infectious", "disease" ]	2017	Epidemiological and ecological determinants of Zika virus transmission in an urban setting
Spiroplasma poulsonii is a maternally transmitted bacterial endosymbiont that is naturally associated with Drosophila melanogaster . S . poulsonii resides extracellularly in the hemolymph , where it must acquire metabolites to sustain proliferation . In this study , we find that Spiroplasma proliferation specifically depletes host hemolymph diacylglyceride , the major lipid class transported by the lipoprotein , Lpp . RNAi-mediated knockdown of Lpp expression , which reduces the amount of circulating lipids , inhibits Spiroplasma proliferation demonstrating that bacterial proliferation requires hemolymph-lipids . Altogether , our study shows that an insect endosymbiont acquires specific lipidic metabolites from the transport lipoproteins in the hemolymph of its host . In addition , we show that the proliferation of this endosymbiont is limited by the availability of hemolymph lipids . This feature could limit endosymbiont over-proliferation under conditions of host nutrient limitation as lipid availability is strongly influenced by the nutritional state . Many insects harbor facultative bacterial endosymbionts , which despite not being required for host survival have important implications for host biology ( Wernegreen , 2012 ) . Two of the most prevalent and well-characterized facultative insect endosymbionts are Wolbachia and Spiroplasma , which are estimated to infect ∼40% and 5–10% of all insects species , respectively ( Hackett and Clark , 1979; Duron et al . , 2008; Hilgenboecker et al . , 2008 ) . While Wolbachia principally resides intracellularly ( Dobson et al . , 1999; Albertson et al . , 2009 ) , Spiroplasma occupies an extracellular niche , proliferating mainly in the hemolymph that fills the body cavity of arthropods ( Sakaguchi and Poulson , 1961; Anbutsu and Fukatsu , 2006 ) . Spiroplasma and Wolbachia are both maternally transmitted and have developed unique strategies to colonize the germline of their female hosts for transmission to the next generation ( Frydman et al . , 2006; Serbus and Sullivan , 2007; Herren et al . , 2013 ) . Facultative endosymbionts with strict maternal transmission , including Wolbachia and Spiroplasma , increase their prevalence in host populations by virtue of two strategies: ( i ) manipulating host reproduction to increase the fitness of infected hosts ( Werren and O'Neill , 1997 ) ; ( ii ) inducing a direct increase in host fitness in a manner that is usually condition dependent , for example protecting hosts against different classes of parasites ( Hedges et al . , 2008; Jaenike et al . , 2010; Teixeira et al . , 2008 ) . Protective endosymbionts of disease vectors may be useful for the control of vector borne disease , and they are increasingly being studied in this context ( Moreira et al . , 2009 ) . While these interactions are clearly of importance , more fundamental features of facultative endosymbioses are poorly understood and frequently overlooked , including metabolic exchanges and the mitigation of host fitness costs . Genome sequencing has indicated that endosymbiotic bacteria have highly reduced metabolic capacities and depend heavily on their hosts to provide them with a diversity of compounds needed for their sustained proliferation ( Klein et al . , 2012; Moran et al . , 2008 ) . However , the direct study of the metabolism of endosymbiotic bacteria is challenging , due to the high level of integration and interdependence between endosymbionts and their hosts . Therefore , despite a general , genome-centric understanding of the metabolic capacities of numerous endosymbionts , little is known about the nature of specific metabolites required for endosymbiont proliferation and the implications of metabolite acquisition by endosymbionts on host physiology and fitness . Strict maternal transmission is expected to result in the evolution of endosymbionts that have minimized host fitness costs ( Werren and O'Neill , 1997 ) . Experimental studies are generally in line with this prediction , for example Wolbachia and Spiroplasma have relatively minor effects on host fitness ( Martins et al . , 2010; Unckless and Jaenike , 2012 ) , however , fitness costs usually become apparent as hosts age ( Ebbert , 1991; Min and Benzer , 1997; Fry et al . , 2004 ) . For endosymbionts that colonize the germline from the adult soma , it has been demonstrated that endosymbiont titers are positively correlated with transmission fidelity ( Dyer et al . , 2005; Unckless et al . , 2009 ) . In general , endosymbiont titers ( and hence proliferation rates ) will therefore be determined by a compromise between the need to minimize host fitness costs and maximize transmission fidelity ( Jaenike , 2009 ) . Since host fitness costs are most likely minimized by the limiting excessive endosymbiont proliferation , the factors that limit endosymbiont proliferation are of central importance for the biology of endosymbionts; however , few mechanisms that are capable of limiting endosymbiont proliferation have been identified ( Login et al . , 2011 ) . The proliferation of bacteria is often controlled by host immune systems; however , it is notable certain endosymbionts , including Spiroplasma , are not susceptible to host immune responses ( Herren and Lemaitre , 2011 ) suggesting that other factors are likely to be of importance for limiting their proliferation . In this study , we used the genetically tractable insect , Drosophila melanogaster and its endosymbiotic Spiroplasma ( MSRO strain ) to analyze the mechanisms that govern Spiroplasma proliferation and the effects of endosymbiont proliferation on host physiology . We find that under normal rearing conditions MSRO Spiroplasma ( henceforth referred to as Spiroplasma ) shortens the life span of its host , D . melanogaster . Interestingly , under nutrient limitation , where increased competition between Spiroplasma and its host could be expected , Spiroplasma proliferation is compromised with minimal effect on host fitness . We noted that under nutrient limitation , host hemolymph lipids decline significantly and that under normal rearing conditions , the first observable effect of Spiroplasma-infection on host physiology is a depletion of host lipids . We then used RNAi-based strategies to reduce the hemolymph lipid concentration and find that this inhibits Spiroplasma proliferation and extends the life span of Spiroplasma-harboring flies . We therefore demonstrate that: ( i ) specific hemolymph lipids are utilized by Spiroplasma and ( ii ) the availability of hemolymph-lipids limits the proliferation of Spiroplasma . We investigated the impact of harboring Spiroplasma on its host's fitness by measuring survival and egg production in both virgin and mated Spiroplasma-infected and uninfected female flies . When maintained on a rich Drosophila diet , flies-harboring Spiroplasma have a significantly shortened life span compared to flies that do not harbor Spiroplasma ( Figure 1A , Figure 1—figure supplement 1A ) , which is in agreement with a previous study on Drosophila willistoni and WSRO Spiroplasma ( Ebbert , 1991 ) . Notably , the presence of Spiroplasma did not significantly affect the death rate until flies were about 21–25 days old but flies began to exhibit signs of Spiroplasma-induced pathology between 14 and 21 days , as demonstrated by decreased performance in climbing assays ( Figure 1B ) . Prior to death , aged Spiroplasma-harboring flies exhibit an apparent lack of coordination and tremors . The increase in Spiroplasma-induced lethality and pathology in old flies correlates with the increase of Spiroplasma titers , observed by qPCR of whole flies ( Figure 1C ) and fluorescence microscopy of hemolymph ( Figure 1D ) . Consistent with previous studies on other Drosophila-endosymbiotic Spiroplasma strains ( Anbutsu and Fukatsu , 2003; Haselkorn et al . , 2013 ) , we noticed that Spiroplasma titers reached a plateau in old flies ( age >28 days ) suggesting that a factor might limit Spiroplasma proliferation at this stage . In light of the finding that Spiroplama reduces fly life span and the apparent trade-off between life span and reproductive output in Drosophila ( Partridge et al . , 1987; Sgrò and Partridge , 1999 ) , we also compared the rate of egg production between Spiroplasma-infected and uninfected females . There was a twofold increase in the number of eggs laid by Spiroplasma-infected virgin flies compared to Spiroplasma-uninfected virgin flies over 14 days ( Figure 1E ) . The number of eggs laid by Spiroplasma-infected mated flies over a 14-day period was similar to Spiroplasma-uninfected mated flies , however , Spiroplasma-infected mated flies laid an increased number of eggs in the first 2 days post-eclosion ( Figure 1—figure supplement 1B ) . These results indicate that the presence of Spiroplasma stimulates egg production in virgin flies and also in mated flies over the first 2 days post eclosion , while causing a minor decline in egg production of mated flies at later time points . Collectively , these experiments demonstrate that ( i ) Spiroplasma has a low impact on the general fitness of its host , negatively affecting survival and egg laying only in old flies and ( ii ) that this decrease in host fitness correlates with higher Spiroplasma titers . 10 . 7554/eLife . 02964 . 003Figure 1 . Spiroplasma proliferation is associated with life span reduction . ( A ) Life span of virgin flies-harboring Spiroplasma ( Sp ( + ) ) relative to control flies that do not harbor Spiroplasma ( Sp ( − ) ) when kept on a rich Drosophila diet . *p<0 . 0001 , N = 50 flies per condition . Shown is one representative experiment out of three independent experiments . ( B ) The climbing activity of virgin flies-harboring Spiroplasma ( Sp ( + ) ) relative to uninfected flies ( Sp ( − ) ) over time . p<0 . 0001 , N = 20 flies per condition . Shown is one representative experiment out of three independent experiments . ( C ) qPCR quantification of the titers of Spiroplasma in virgin flies over aging . Values for each timepoint have at least three samples ( five flies pooled per sample ) . Shown is one representative experiment out of three independent experiments . ( D ) Fluorescent microscopy images depicting SYTO-9 stained Spiroplasma in Drosophila hemolymph at 7 days ( D1 ) and 21 days ( D2 ) of fly age . ( E ) The number of eggs laid by virgin flies-harboring Spiroplasma ( Sp ( + ) ) relative to control flies that do not harbor Spiroplasma ( Sp ( − ) ) , in total over 14 days ( left panel ) and in 2-day period over 14 days ( right panel ) . In total , Spiroplasma-infected virgin flies laid significantly more eggs . p=0 . 02 . Shown is the mean ± SEM of data pooled from four independent experiments , N = 20 flies per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 00310 . 7554/eLife . 02964 . 004Figure 1—figure supplement 1 . The impact of Spiroplasma infection on survival and egg production by mated females on rich media . ( A ) Life span of mated flies-harboring Spiroplasma ( Sp ( + ) ) relative to control flies that do not harbor Spiroplasma ( Sp ( − ) ) when kept on a rich Drosophila diet . **p<0 . 0001 , N = 20 flies per condition . Shown is one representative experiment out of three independent experiments . ( B ) The number of eggs laid by mated flies-harboring Spiroplasma ( Sp ( + ) ) relative to control flies that do not harbor Spiroplasma ( Sp ( − ) ) , in total over 14 days ( left panel ) and in a 2-day period over 14 days ( right panel ) . In total , there was no significant difference between the number of eggs laid by Spiroplasma-infected and Spiroplasma-uninfected mated flies . p=0 . 02 and NS ( p=0 . 16 ) . Over the first 2 days , Spiroplasma-infected mated females produced significantly more eggs than their uninfected counterparts . **p=0 . 0005 . Shown is the mean ± SEM of data pooled from four independent experiments , N = 20 flies per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 004 We had initially speculated that the competition for resources between Spiroplasma and its host would be more conspicuous upon nutrient scarcity and that under these conditions Spiroplasma might have a more detrimental effect on host fitness . To test this hypothesis , we maintained adult flies on a nutrient poor diet . Under these nutrient-limiting conditions , Drosophila survival is significantly compromised , however it is important to note that this diet still contains sufficient nutritional content to support an entire Drosophila life-cycle ( Vijendravarma et al . , 2012 ) . It is also worth mentioning that we only examined the effects of nutrient deprivation in adult flies and that for all the experiments conducted Drosophila larvae were raised under normal conditions . Surprisingly , Spiroplasma-infected and uninfected virgin flies had a similar life span ( Figure 2A ) and produced similar numbers of eggs under nutrient-limiting conditions ( Figure 2B ) . For mated flies under nutrient limitation , harboring Spiroplasma resulted in a minor reduction in survival relative to flies that did not harbor Spiroplasma ( Figure 2—figure supplement 1A ) . Spiroplasma-infected mated flies produced more eggs than Spiroplasma-uninfected mated flies under nutrient-limiting conditions ( Figure 2—figure supplement 1B ) . These findings suggest that , in contrast to our initial hypothesis , harboring Spiroplasma has a rather limited fitness cost under nutrient-limiting conditions . Importantly , we observed that Spiroplasma proliferation in whole flies ( Figure 2B ) and hemolymph ( Figure 2—figure supplement 2 ) was also significantly inhibited when flies are maintained on a nutrient poor diet . Thus , the inhibition of Spiroplasma proliferation in flies that are maintained under nutrient-limiting conditions could explain why Spiroplasma has limited fitness costs under these conditions . For consistency , and to facilitate the maintenance of flies under identical conditions , subsequent experiments were carried out on virgin females ( unless otherwise specified ) . 10 . 7554/eLife . 02964 . 005Figure 2 . The implications of harboring Spiroplasma under host nutritional depravation . ( A ) Survival of virgin flies on a nutritionally poor diet . Flies-harboring Spiroplasma ( Sp ( + ) ) do not have significantly different mortality from flies that do not harbor Spiroplasma ( Sp ( − ) ) . NS ( p=0 . 9378 ) . N = 50 flies per condition . Shown is one representative experiment out of three independent experiments . ( B ) The number of eggs laid by virgin flies-harboring Spiroplasma ( Sp ( + ) ) relative to control flies that do not harbor Spiroplasma ( Sp ( − ) ) , in total over 14 days ( left panel ) and in 2-day period over 14 days ( right panel ) . Overall , there is no significant difference in the number of eggs laid between Spiroplasma-infected and uninfected virgin flies under nutrient deprivation . NS ( p=0 . 77 ) . Shown is the mean ± SEM of data pooled from four independent experiments , N = 20 flies per experiment . ( C ) Quantification of Spiroplasma titers by qPCR reveals that virgin female flies maintained on the same nutritionally poor diet as in panel A have significantly lower Spiroplasma titers after 8 and 12 days than flies maintained on a rich diet . p=0 . 015 and *p=0 . 0002 , respectively . Values for each time-point have at least three samples ( five flies pooled per sample ) . Shown is the mean ± SEM of one representative experiment out of the three independent experiments that were conducted . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 00510 . 7554/eLife . 02964 . 006Figure 2—figure supplement 1 . The effects of nutrient deprivation on survival and egg production of Spiroplasma-infected mated females . ( A ) Survival of mated flies on a nutritionally poor diet . Mated flies-harboring Spiroplasma ( Sp ( + ) ) do have significantly higher mortality relative to mated flies that do not harbor Spiroplasma ( Sp ( − ) ) on a nutritionally poor diet . p<0 . 0001 . N = 20 flies per condition . Shown is one representative experiment out of three independent experiments . ( B ) The number of eggs laid by mated flies-harboring Spiroplasma ( Sp ( + ) ) relative to control flies that do not harbor Spiroplasma ( Sp ( − ) ) , on a nutrient poor diet , in total over 14 days ( left panel ) and in 2-day period over 14 days ( right panel ) . In total , Spiroplasma-infected mated flies laid significantly more eggs . p=0 . 02 . Shown is the mean ± SEM of data pooled from four independent experiments , N = 20 flies per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 00610 . 7554/eLife . 02964 . 007Figure 2—figure supplement 2 . Spiroplasma titers in fly hemolymph under nutrient deprivation . Quantification by qPCR reveals that Spiroplasma titers are significantly lower in the hemolymph of flies maintained on a nutrient poor diet for 12 days than flies maintained on a rich diet . p=0 . 04 . Values are the mean ± SEM of at least three samples ( hemolymph from 12 flies pooled per sample ) . Shown is one representative experiment out of the three independent experiments that were conducted . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 007 The results above suggest that Spiroplasma proliferation might be dependent on the availability of host factors that are nutritional state-dependent . To identify these factors , we examined the effect that maintaining uninfected flies for 12 days on a nutrient poor diet had on the concentration of metabolites in the hemolymph ( where Spiroplasma reside ) . We found that raising flies on a nutrient poor diet resulted in significantly lower concentrations of protein , sterol and diacylglyceride ( DAG , the main transport lipid in Drosophila ) ( Figure 3A–C ) , while levels of glucose and trehalose were not changed significantly ( Figure 3—figure supplement 1A , B ) and L-amino acids increased ( Figure 3—figure supplement 1C ) . Thus , nutrient limitation led to a specific decline in hemolymphatic protein and lipid concentrations . We then complemented the nutrient poor diet with either inactivated yeast ( rich in protein and lipids ) or sucrose , we found that only inactivated yeast extract was able to recover the Spiroplasma proliferation rates observed on rich media ( Figure 4 ) . This indicates that Spiroplasma proliferation is not only affected by the caloric content of the food but by the composition of the diet . Taken together , these results lend support to the hypothesis that Spiroplasma proliferation is heavily dependent on hemolymph metabolite composition and that certain metabolites ( e . g . , lipids and protein ) could play a more important role than others ( e . g . , sugars ) . 10 . 7554/eLife . 02964 . 008Figure 3 . Nutrient deprivation depletes host lipids . ( A–C ) The protein ( A ) , sterol ( B ) , and DAG ( C ) concentration of the hemolymph of flies maintained on nutritionally poor diets for 12 days is significantly lowered relative to flies maintained on a nutritionally rich diet . Mean ± SEM of three independent experiments is shown , p=0 . 038 , p=0 . 023 , and p=0 . 0018 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 00810 . 7554/eLife . 02964 . 009Figure 3—figure supplement 1 . The impact of nutrient deprivation on hemolymph metabolite concentrations . Quantification of glucose ( A ) , trehalose ( B ) , and L-amino acid ( C ) concentration of the hemolymph of flies maintained on nutritionally poor diets for 12 days relative to flies maintained on a rich diet . L-amino acid concentration of the hemolymph is significantly higher in nutrient poor media relative to flies maintained under nutrient rich conditions while glucose and trehalose concentrations remain unchanged . Mean ± SEM of three independent experiments is shown , NS ( p=0 . 1319 and p=0 . 357 ) and p=0 . 0027 . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 00910 . 7554/eLife . 02964 . 010Figure 4 . Spiroplasma proliferation is influenced by the nutrient composition of the diet . Quantification of Spiroplasma titers by qPCR reveals that complementing nutrient poor media with inactivated yeast results in a significant increase in Spiroplasma titers after 12 days . p=0 . 0003 . In contrast , complementing nutrient poor media with sucrose does not significantly increase Spiroplasma titers . NS ( p=0 . 5 ) . Values are the mean ± SEM of at least four samples ( five flies per sample ) . Shown is one representative experiment out of the three independent experiments that were conducted . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 010 To gain a better insight into the relationship between Spiroplasma and host physiology , we monitored the effects of harboring Spiroplasma on the concentration of metabolites in the hemolymph of flies raised on a rich diet . We observed that the concentration of measured sugars was not changed and that the L-amino acid concentration was even increased in the presence of Spiroplasma ( Figure 5A–C ) . We also found that the hemolymph of flies-harboring Spiroplasma had an increased concentration of proteins and sterol ( Figure 5D , E ) . However , after centrifugation to remove bacterial cells , the protein and sterol concentration was no longer significantly different between Spiroplasma-infected and Spiroplasma-uninfected hemolymph samples . This suggests that Spiroplasma cells contain a substantial amount of proteins and sterol , but that the presence of this endosymbiont does not deplete either of these factors in the hemolymph . Notably , we observed that Spiroplasma-infected flies experienced a significant drop in the concentration of DAG within the hemolymph ( Figure 5F ) . Importantly , this difference is not caused by the previously noted differences in the rate of egg production between Spiroplasma-infected and Spiroplasma-uninfected virgin flies , as a decrease in DAG was also observed in mated flies ( Figure 5—figure supplement 1A ) , where Spiroplasma did not affect the number of eggs laid over 14 days ( Figure 1—figure supplement 1B ) . We also observed that there was a more marked decline in DAG levels in 28-day-old Spiroplasma-infected relative to uninfected flies ( Figure 5—figure supplement 1B ) , where Spiroplasma titers have nearly reached their maximum levels ( Figure 1C ) and egg production declines ( David et al . , 1975; Partridge et al . , 1987 ) . We then investigated the impact of Spiroplasma on metabolite storage in the fat body by quantifying the amount of triacylglyceride ( TAG , the main storage lipid in Drosophila ) and glycogen ( the main storage carbohydrate in Drosophila ) . Quantifications of whole female flies ( reflecting mainly insect fat body energy storage but also the energy contents of the ovaries ) showed a decrease in the amount of TAG in Spiroplasma-infected flies compared to Spiroplasma-uninfected flies ( Figure 5G ) , while the amount of glycogen was not affected by the presence of Spiroplasma ( Figure 5—figure supplement 2 ) . Consistent with the decrease in TAG reserves , 12-day-old Spiroplasma-infected flies succumb more rapidly to acute starvation , in which flies are only given a source of water but no source of nutrition ( Figure 5H ) . Since TAG stored in the fat body largely derived from hemolymph DAG ( Canavoso et al . , 2001 ) , the decrease of TAG is likely to be the outcome of the depletion of hemolymphatic DAG by Spiroplasma . Thus , our analyses show that the proliferation of Spiroplasma in flies is associated with a specific depletion of hemolymph DAG as well as a decrease in the amount of fat body lipid storage . These findings , together with the observation that DAG is depleted under nutrient deprivation , suggest that DAG availability limits Spiroplasma proliferation . 10 . 7554/eLife . 02964 . 011Figure 5 . Spiroplasma infection depletes lipids of Drosophila maintained under normal conditions . Quantification of metabolites in flies that have been maintained on rich media for 12 days . Glucose ( A ) , trehalose ( B ) , and L-amino acid ( C ) concentration within the hemolymph of uninfected flies ( Sp ( − ) ) and Spiroplasma-infected flies ( Sp ( + ) ) . L-amino acid concentration in the hemolymph is significantly higher in Spiroplasma-infected flies while glucose and trehalose concentrations remain unchanged . Mean ± SEM of three independent experiments is shown , NS ( p=0 . 798 and p=0 . 977 ) and p=0 . 0056 . ( D–E ) Quantifications of protein ( D ) and sterol ( E ) concentration in hemolymph from flies that harbor Spiroplasma ( Sp ( + ) ) and uninfected flies ( Sp ( − ) ) . Hemolymph samples denoted as ‘supernatant’ have been subjected to an additional centrifugation to remove Spiroplasma cells , whereas all other hemolymph samples contain both Spiroplasma cells and hemolymph . Flies-harboring Spiroplasma have significantly higher total levels of protein and sterol in the hemolymph . Mean ± SEM of three independent experiments is shown , p=0 . 04 and p=0 . 037 . After centrifugation to remove bacteria from the hemolymph , there was no longer any significant difference in protein and sterol concentrations between Spiroplasma-infected and uninfected hemolymph . Mean ± SEM of three independent experiments is shown , NS ( p=0 . 881 and p=0 . 491 , respectively ) . ( F ) Quantification of DAG content of hemolymph extracts from flies that harbor Spiroplasma ( Sp ( + ) ) and flies that do not ( Sp ( − ) ) . p=0 . 0266 . Mean ± SEM of three independent experiments is shown . ( G ) Quantification of whole-fly ( reflecting mainly fat body ) TAG levels in flies that harbor Spiroplasma ( Sp ( + ) ) and uninfected flies ( Sp ( − ) ) . p=0 . 0043 . Mean ± SEM of three independent experiments is shown . ( H ) Survival of flies subjected to an acute starvation after being maintained on rich media for 12 days . Flies-harboring Spiroplasma ( Sp ( + ) ) have significantly greater mortality rate than flies that do not harbor Spiroplasma ( Sp ( − ) ) . *p<0 . 0001 . N = 20 flies per condition , shown is one representative experiment out of three independent . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 01110 . 7554/eLife . 02964 . 012Figure 5—figure supplement 1 . The impact of Spiroplasma proliferation hemolymph DAG concentration in mated flies and old flies . Quantification of DAG content of hemolymph extracts from 12-day-old mated flies ( A ) and 28-day-old virgin flies ( B ) that harbor Spiroplasma ( Sp ( + ) ) and flies that do not ( Sp ( − ) ) . p=0 . 03 and *p=0 . 0008 , respectively . Mean ± SEM of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 01210 . 7554/eLife . 02964 . 013Figure 5—figure supplement 2 . The impact of Spiroplasma proliferation on fat body glycogen stores . Whole fly ( reflecting mainly fat body stores ) glycogen concentrations in 12-day-old flies uninfected ( Sp ( − ) ) and infected ( Sp ( + ) ) with Spiroplasma . Mean ± SEM of three independent experiments is shown , NS ( p=0 . 9226 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 013 Since the hemolymph derived from Spiroplasma-infected flies contained Spiroplasma and had reduced DAG , it is likely that Spiroplasma is not directly incorporating DAG into their membrane but rather metabolize DAG into another compound . The main classes of lipids that are present in Spiroplasma are phosphatidylglycerols , sterols , sphingolipids , and cardiolipins ( Freeman et al . , 1976 ) . Cardiolipins are a class of lipids found exclusively in eubacteria and in the mitochondria of eukaryotic cells ( Hawthorne and Ansell , 1982 ) . In contrast to other major lipidic components of Spiroplasma membranes , cardiolipins are not present at detectable levels in the Drosophila hemolymph ( Carvalho et al . , 2012 ) . Cardiolipins are comprised of four acyl chains , two phosphate groups and three glycerols , and therefore are made up of similar components to DAGs , which consist of two acyl chains and a glycerol . Analysis of the Spiroplasma genome reveals that Spiroplasma is nearly devoid of lipid metabolic capacities but possess genes involved in the synthesis of cardiolipin from the precursor DAG through a pathway involving DAG-3-phosphate and cytidine diphosphate-DAG ( unpublished data ) . To investigate the production of cardiolipin by Spiroplasma , we conducted a MALDI time-of-flight mass spectrometry ( MALDI-Tof-MS ) analysis of lipid species in fly hemolymph and found that there were only peaks at m/z values that correspond to cardiolipin in the hemolymph of flies-harboring Spiroplasma ( Figure 6A ) . A major ion at m/z 1425 . 05 was isolated and fragmented , which confirmed the presence of cardiolipin and revealed that the cardiolipins were comprised of C16:0 and C18:1 acyl chains ( Figure 6B ) . In addition , we used liquid chromatography–tandem mass spectrometry ( LC-MS/MS ) to quantify the effect of Spiroplasma infection on the concentration of individual DAG species in Drosophila hemolymph . We observed that overall DAG concentration declined by 16 . 3% , however certain DAG species ( e . g . , C32:1 and C34:1 DAG ) declined to a much greater extent ( Table 1 ) . It is notable that DAG species that decline to the greatest extent in the presence of Spiroplasma are those likely to contain ( based on Drosophila fatty acid composition ) one saturated ( e . g . , C14:0 or C16:0 ) and one mono-unsaturated ( e . g . , C16:1 or C18:1 ) medium-length acyl chain ( Shen et al . , 2010 ) . The observation that Spiroplasma-generated cardiolipin contains the same configuration of one saturated and one mono-unsaturated medium-length acyl chain ( predominately C16:0 , C18:1 ) indicates that Spiroplasma most likely produce cardiolipin using fly hemolymph DAG . The transformation of DAG into cardiolipin offers an explanation for the observed decrease in hemolymph DAG levels in the presence of Spiroplasma . 10 . 7554/eLife . 02964 . 014Figure 6 . Spiroplasma produces cardiolipin in Drosophila hemolymph . ( A ) Negative MALDI-TOF/MS lipid profile of hemolymph from Spiroplasma-uninfected ( top ) and Spiroplasma-infected flies ( bottom ) . The m/z signal peaks in the 1380–1460 range of Spiroplasma-uninfected hemolymph do not correspond to m/z values of cardiolipin , whereas the peaks in this region for Spiroplasma-infected hemolymph profile ( e . g . , 1403 . 11 , 1425 . 05 ) do correspond to cardiolipin . ( B ) The isolation and fragmentation of the m/z 1425 . 05 parent ion resulted in the generation of daughter ions with peaks at m/z 281 . 36 and 255 . 35 that have been characterized as oleic acid ( C18:1 ) and palmitic acid ( C16:0 ) using Lipid MS Predict software with an error tolerance set to 0 . 1 u . m . a . The peak at m/z 79 . 02 detected in the same experiment corresponds to a phosphate ion . Six additional peaks ( m/z 153 . 11 , 311 . 09 , 391 . 29 , 417 . 33 , 491 . 29 and 831 . 33 ) were also detected corresponding to phosphatidyl moieties and the cardiolipin ‘backbone’ ( Hsu and Turk , 2006 ) . Altogether , this indicates that the molecular ion corresponds to the cardiolipid species [M-2H+Na]− m/z 1425 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 01410 . 7554/eLife . 02964 . 015Table 1 . Characterization and quantification of hemolymph of DAG speciesDOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 015Component nameSp − ( µg/ml ) Sp + ( µg/ml ) % Of total mass declineC28:0 DAG13 . 412 . 37C28:1 DAG0 . 80 . 61 . 2C30:2 DAG0 . 70 . 51 . 2C30:0 DAG0 . 10 . 2−1 . 0C32:3 DAG1 . 21 . 10 . 7C32:2 DAG23 . 121 . 88 . 2C32:1 DAG19 . 815 . 130 . 1C34:2 DAG11 . 510 . 84 . 9C34:1 DAG9 . 76 . 919 . 3C34:0 DAG1 . 40 . 55 . 6C36:4 DAG0 . 80 . 51 . 8C36:3 DAG2 . 92 . 25 . 3C36:2 DAG4 . 22 . 99 . 3C36:1 DAG0 . 80 . 9−0 . 3C36:0 DAG1 . 70 . 76 . 992 . 177Quantification of the absolute concentration of individual DAG species in the hemolymph of Spiroplasma-uninfected ( Sp ( − ) ) and Spiroplasma-infected ( Sp ( + ) ) mated flies by LC-MS/MS . The % of total mass decline reflects the percentage of the total decline between Spiroplasma-uninfected and infected samples ( a total of 15 . 1 μg/ml or 16 . 3% ) that can be attributed to each DAG species . It is notable that C32:1 and C34:1 DAG species decline to a greater extent than other common DAG species such as C28:0 and C32:2 . This suggests that Spiroplasma is preferentially incorporating DAGs that have one saturated and one mono-unsaturated acyl chain . Notably , C34:1 DAGs are likely to be made up of oleic ( C18:1 ) and palmitic ( C16:0 ) acids , which have exactly the same acyl chains that were identified in Spiroplasma-generated cardiolipins ( Figure 6B ) . In Drosophila , dietary lipids are broken down in the gut lumen by lipases prior to absorption by intestinal cells ( Sieber and Thummel , 2012 ) . In the enterocytes , these compounds are used for the synthesis of DAG , which is packaged together with phosphoethanolamine , sterol , other minor lipids , and the apolipophorin protein ( Lpp ) , to form lipoprotein particles . Lpp is produced in the fat body but travels to the gut where it gets loaded with lipids prior to trafficking throughout the body ( Palm et al . , 2012 ) . Lpp is the main hemolymph lipid carrier , since more than 95% of the hemolymph lipids in Drosophila co-fractionate with Lpp ( Palm et al . , 2012 ) . Our results are consistent with Spiroplasma subverting and utilizing the lipids contained in hemolymph lipoprotein particles prior to their arrival at the fat body . This has the consequence of decreasing the observed levels of stored lipids . To rule out the possibility that Spiroplasma induces the mobilization of lipid stores , which could also decrease TAG levels , we quantified the effect of Spiroplasma on TAG levels in AKHR and Bmm double mutant flies . Adipokinetic hormone receptor ( AKHR ) and the Brummer lipase ( Bmm ) are components of two independent pathways that mobilize lipids from Drosophila fat bodies . Signaling through AKHR initiates mobilization of stored lipid in the insect fat body ( Gäde and Auerswald , 2003 ) , and Bmm is a TAG lipase involved in mobilization of stored lipid ( Grönke et al . , 2005 ) . AKHR1;bmm1 double mutant flies exhibit an ‘obese’ phenotype because the fat bodies of these flies store lipids but are not subsequently able to mobilize or release lipids ( Grönke et al . , 2007 ) . We found that fat body TAG levels in AKHR1;bmm1 double mutants that harbored Spiroplasma were still significantly lower after 12 days than flies with the same genotype that did not harbor Spiroplasma ( Figure 7 ) . While we cannot rule out other , as of yet uncharacterized , lipid mobilization pathways that could theoretically be activated by Spiroplasma , these results are still a strong indication that Spiroplasma lipid acquisition most likely occurs in the hemolymph before the point of lipid entry into the fat body stores . 10 . 7554/eLife . 02964 . 016Figure 7 . Spiroplasma-induced lipid depletion is not caused by the mobilization of stored lipids . TAG levels in AKHR1;Bmm1 double mutants that harbor Spiroplasma ( AKHR1;Bmm1 Sp ( + ) ) relative to the same genotype without Spiroplasma ( AKHR1;Bmm1 Sp ( − ) ) . p<0 . 0153 . Flies were maintained on a rich Drosophila diet for 12 days prior to TAG analysis . Mean ± SEM of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 016 We have observed that: ( i ) Spiroplasma's proliferation is affected by host nutrient limitation and ( ii ) Spiroplasma consumes the hemolymph DAGs . These findings lead us to hypothesize that the proliferation of Spiroplasma could be constrained by the availability of lipids in the hemolymph , which is known to be heavily dependent on the insect nutritional status ( Canavoso et al . , 2001 ) . To further investigate the lipid-centric metabolic interplay between Spiroplasma and Drosophila , we monitored Spiroplasma proliferation in flies with a reduced capacity for inter-organ lipid transport and hence decreased hemolymph lipid concentrations . A recent study has shown that RNAi-mediated knockdown of Lpp in the fat body resulted in a decrease in circulating lipoprotein particles and a blockage of lipid export from the gut of Drosophila larvae ( Palm et al . , 2012 ) . Based on these findings , we used a similar RNAi strategy to knockdown Lpp in the fat body of adult flies . In this experiment , we specifically expressed Lpp-RNAi in the fat bodies of adult flies using a flippase-mediated activation strategy ( Marois and Eaton , 2007 ) . We established that RNAi-mediated knockdown of Lpp does not significantly decrease the overall levels of protein in the adult hemolymph ( Figure 8—figure supplement 1 ) but does decrease adult hemolymph-lipids , as shown by quantification of hemolymph DAG and sterol ( Figure 8A , B ) . We then investigated the effect of this lipid reduction on the Drosophila–Spiroplasma interaction . We found that under conditions of Lpp knockdown , Spiroplasma proliferation was severely inhibited resulting in lower Spiroplasma titers in flies after 14 , 21 , and 28 days of aging ( Figure 8C ) . To rule out any possibility that the genetic background of Lpp-RNAi flies was causing the observed decrease in Spiroplasma titers , we quantified Spiroplasma titers in the absence of activation of Lpp-RNAi by heat-shock and found no significant difference between flies containing the Lpp-RNAi construct and those that did not ( Figure 8—figure supplement 2A ) . In addition , we used an independent Lpp-RNAi construct with another fat body specific driver and activation strategy to confirm that the decrease in Spiroplasma titers was specifically caused by RNAi-mediated Lpp knockdown ( Figure 8—figure supplement 2B ) . It is noteworthy that Spiroplasmas are not known to have the capacity to utilize proteins as nutrient sources ( Chang and Chen , 1983 ) , supporting the claim that lipids carried by Lpp ( and not the Lpp protein itself ) are the factors required for the proliferation of Spiroplasma . Another striking effect of the RNAi-mediated knockdown of Lpp was strongly diminished Spiroplasma-induced old age mortality ( Figure 8D ) . 10 . 7554/eLife . 02964 . 017Figure 8 . Lpp-lipids are required for Spiroplasma proliferation . ( A ) Quantification of DAG levels in hemolymph of 12-day-old flies 8 days after knockdown of Lpp by RNAi . Mean ± SEM of three independent experiments is shown , p=0 . 015 . ( B ) Quantification of sterol concentration in hemolymph of 12-day-old flies 8 days after knockdown of Lpp by RNAi . p=0 . 0125 . Mean ± SEM of three independent experiments is shown . ( C ) Spiroplasma titers quantified by qPCR in flies which have Lpp expression knocked down by RNAi relative to control flies . Spiroplasma titers were quantified at 7 , 14 , 21 , and 28 days after activation of RNAi . Mean ± SEM of at least three samples is shown ( five flies pooled per sample ) . **p=0 . 0005 ( 14 days ) , p<0 . 01 ( 21 days ) , and p<0 . 01 ( 28 days ) . Shown is one representative experiment out of three independent experiments . ( D ) The survival of Spiroplasma-infected ( Sp ( + ) ) and Spiroplasma-uninfected ( Sp ( − ) ) flies with or without RNAi-mediated Lpp knocked down . p<0 . 0001 , N = 18 flies per condition . Shown is one representative experiment out of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 01710 . 7554/eLife . 02964 . 018Figure 8—figure supplement 1 . Hemolymph protein is not decreased by Lpp depletion . Protein concentration in the hemolymph of flies that are 12 days old , 8 days after activation of Lpp-RNAi . NS ( p=0 . 908 ) . Mean ± SEM of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 01810 . 7554/eLife . 02964 . 019Figure 8—figure supplement 2 . Spiroplasma titers are decreased by Lpp depletion . ( A ) Spiroplasma titers quantified by qPCR in flies which have Lpp-RNAi construct which has not been activated by heat shock relative to control flies without the Lpp-RNAi construct , NS ( p=0 . 587 ) , and an additional control where Lpp-RNAi has been activated , p=0 . 0037 . Spiroplasma titers were quantified at 14 days of fly age . Mean ± SEM of at least three samples is shown ( five flies pooled per sample ) . Shown is one representative experiment out of the three independent experiments . ( B ) Spiroplasma titers quantified by qPCR in flies in which Lpp has been knocked down by the expression of the UAS-LppRNAi ( 46 ) using the C564>gal4TS driver . Flies were raised at the restrictive temperature of 18°C where C564>gal4 activity is repressed by gal80ts and then shifted to 29°C at 4 days post-eclosion ( Lpp-RNAi is then activated ) where they were maintained for 25 days . p=0 . 0215 . Mean ± SEM of at least three samples is shown ( five flies pooled per sample ) . Shown is one representative experiment out of the three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 02964 . 019 We have demonstrated that Spiroplasma subverts specific host lipids and that its proliferation is limited by the availability of host hemolymph-lipids . This finding is based on several observations: ( i ) Spiroplasma proliferation rate is decreased in the hemolymph of nutrient deprived flies and old flies , two conditions in which hemolymph–lipid concentration is decreased; ( ii ) the proliferation of Spiroplasma depletes hemolymph and fat body lipids; and ( iii ) a genetically induced reduction in Lpp-lipids inhibits Spiroplasma proliferation . Natural selection is expected to favor vertically transmitted endosymbionts with adaptations that minimize fitness costs to their hosts ( Werren and O'Neill , 1997 ) . We hypothesize that Spiroplasma's dependence on hemolymph-lipid availability for proliferation could be of adaptive significance , since it would enable Spiroplasma to limit its proliferation in the face of host nutrient deprivation , and therefore avoid the costly depletion of its host's energy and other vital metabolites . If the proliferation rate of Spiroplasma were to be determined by metabolites that do not decline under nutrient-limiting conditions , for example sugars and L-amino acids , then Spiroplasma proliferation would not be limited under host nutritional deprivation . Such a lack of restriction could result in greater negative effects on its host's fitness . For example , we might expect a striking decrease in reproductive output or survival , which was not observed for Spiroplasma-infected flies maintained on a nutrient poor diet . We only observed that mated flies-harboring Spiroplasma had increased mortality after about 7 days under nutrient-limiting conditions , and this is not likely to have a major impact on host fitness since increased mortality occurred after the production of eggs had stopped . Under normal conditions , Spiroplasma's significantly decreases the life span of its host Drosophila ( also shown in Ebbert , 1991 ) . However , aspects of this Spiroplasma-induced pathogenesis , including the late onset of symptoms , suggest that Spiroplasma may be employing strategies to minimize host fitness costs . Spiroplasma's capacity to utilize available host lipids for proliferation results in depleted hemolymph-lipids and reduced fat body stores . This depletion also has a fitness consequence for flies , since Spiroplasma-infected flies die more rapidly when subjected to a period of acute starvation , but it is possible that Spiroplasma's reliance on these lipids might be less detrimental than usage of other more critical metabolites that would impact host fitness more directly . Drosophila fitness and longevity are linked to egg production rates ( Partridge et al . , 1987 ) , which can be affected by the presence of facultative endosymbionts ( Fast et al . , 2011 ) . On nutrient rich media , we found that despite increasing early egg production Spiroplasma infection status did not affect the total number of eggs laid over a 14-day period . The Spiroplasma-induced early increase in egg production has been described previously for other Spiroplasma strains ( Ebbert , 1991; Martins et al . , 2010 ) . While the mechanistic basis of this increase is unknown , it offers a possible explanation for the greater number of eggs laid by Spiroplasma-infected flies on nutrient poor media , where the majority of eggs laid by all flies are in the first 2 days post-eclosion . Under rich nutritive conditions , we observed that Spiroplasma increased by twofold the number of eggs laid by virgin flies over a 14-day period . This finding is suggestive of a Spiroplasma-mediated disruption in the balance between egg retention and laying of virgin flies . One could speculate that Spiroplasma interferes with the signaling pathways ( e . g . , Juvenile Hormone and/or Ecdysone ) , which have previously been shown to regulate the production of eggs ( Soller et al . , 1999 ) . We observed a decrease in Spiroplasma proliferation in flies after 28 days . A late decline in Spiroplasma proliferation has been demonstrated for a number of other Spiroplasma strains ( Anbutsu and Fukatsu , 2003; Haselkorn et al . , 2013 ) . It is notable that flies aged for 28 days also experience a decline in DAG concentration , which is likely to explain the declining Spiroplasma proliferation rate . We suspect that Spiroplasma proliferation over the life of flies results in elevated Spiroplasma titers that ultimately deplete host hemolymph lipid and constrain further proliferation . In the MSRO-Spiroplasma system , this decline in proliferation does not appear to limit Spiroplasma-induced mortality , however , other Drosophila–Spiroplasma strains appear to reach maximal titers at earlier time points and therefore this is an aspect of the endosymbiont's biology could play a more important role in limiting fitness costs in other systems . We noted that Spiroplasma-infected flies had higher levels of cardiolipin , suggesting that Spiroplasma synthesize cardiolipin , most likely using DAG acquired from host lipoproteins . We observed that Spiroplasma specifically depleted host DAG species that contained one saturated and one mono-unsaturated acyl chain , which was also the most abundant acyl chain configuration in Spiroplasma-produced cardiolipin ( which contained C16:0 and C18:1 acyl chains ) . The basis of this acyl chain specificity is not known , but it could be linked to biophysical properties of cardiolipins in the highly curved Spiroplasma membrane . The utilization of DAG as a precursor for cardiolipin synthesis is supported by the annotation of the Spiroplasma genome ( unpublished data ) , which enabled the identification of a biosynthetic pathway similar to that described in Mycoplasma synoviae and Mycoplasma hyopneumoniae for the synthesis of cardiolipin from acylglycerols through DAG-3-phosphate and cytidine diphosphate-DAG ( Arraes et al . , 2007 ) . A number of studies have demonstrated links between cardiolipin and cell death ( Gonzalvez and Gottlieb , 2007 ) , which raise the possibility that cardiolipin produced by Spiroplasma could be involved in its pathological effects in old flies . The other lipids that comprise a smaller fraction of Lpp's cargo , including sterols and sphingolipids ( Palm et al . , 2012 ) might also be important for Spiroplasma proliferation but these lipids would most likely be incorporated unchanged into the membrane of Spiroplasma . Of the other lipid classes contained in lipoproteins , sterols are of particular interest as they have been shown to be required for proliferation of all species of Spiroplasma , and to be highly abundant in the Spiroplasma plasma membrane ( Freeman et al . , 1976 ) . Amongst bacteria , this requirement for sterol is unique to mollicutes ( Dahl , 1993 ) . Since insects are auxotrophic for sterols ( Canavoso et al . , 2001 ) , their concentrations are determined by dietary uptake . Although it is therefore possible that the sterol availability could play a key role in coordinating Spiroplasma proliferation with host nutritional state , we were unable to recover normal Spiroplasma proliferation rates when flies were maintained on nutrient poor media complemented with sterol ( unpublished data ) . This finding suggests that , while sterol might be important for Spiroplasma proliferation , it is not sufficient to induce proliferation under conditions of nutrient limitation . The strategies employed in different insect endosymbioses to limit over-proliferation of endosymbionts are not well characterized . We have proposed a model in which Spiroplasma's dependence on host hemolymph-lipid availability could limit its over-proliferation , primarily in the face of host nutritional scarcity . This mechanism could be important for the controlling proliferation of diverse extracellular endosymbionts . Different mechanisms are likely to be involved in intracellular endosymbiont control . For example , proliferation and localization of the obligate Sitophilus primary endosymbiont ( SPE ) is controlled by the expression of a specific antimicrobial peptide from its host , Sitophilus weevils ( Login et al . , 2011 ) . In aphids , studies suggest the activation of the host lysosomal system is involved in controlling titers of the obligate intracellular endosymbiont , Buchnera ( Nishikori et al . , 2009 ) . The metabolic exchanges between endosymbiotic bacteria and their arthropod hosts are generally not well understood , partly due to challenges associated with high levels of integration and interdependence between partners . Most of the available information is indirect , coming from the study of endosymbiont genomes , which reveals limited metabolic capacities and high levels of dependence on hosts ( Zientz et al . , 2004; Gosalbes et al . , 2010 ) . There are a number of studies that examined metabolic exchanges between hosts and obligate endosymbionts , which provide their hosts with one or more vital metabolite that is missing from the host's diet ( Douglas , 1998; Russell et al . , 2013 ) . These studies usually consider metabolic transfers from endosymbiont to host and not in the reverse direction ( Dale and Moran , 2006 ) . We have identified a transfer of metabolites from host to endosymbiont , showing that Drosophila Lpp-lipids are used by Spiroplasma for proliferation and more specifically provide evidence indicating that host-derived DAG is converted to cardiolipin by Spiroplasma . A number of recent studies have revealed that facultative endosymbionts protect their hosts from parasites and pathogens ( Oliver et al . , 2003; Hedges et al . , 2008; Teixeira et al . , 2008 ) . Endosymbiotic Spiroplasma has been implicated in a number of cases , including protecting diverse hosts from various eukaryotic parasites including parasitoid wasps , parasitic nematodes , and fungi ( Jaenike et al . , 2010; Xie et al . , 2010; Łukasik et al . , 2012 ) . We speculate that Spiroplasma-mediated protection could be linked to lipid utilization . Indeed , many parasitoid wasps are unable to synthesize fatty acids and their development requires the acquisition of host lipids ( Visser et al . , 2010 ) . Thus , sequestration of lipid by Spiroplasma might limit availability to any parasites or pathogens that occupy the same niche . Here , we have shown that the rate of Spiroplasma proliferation and onset of fly mortality are decreased upon depletion of hemolymph lipids . It is noteworthy that a variety of other microorganisms known to proliferate in Drosophila hemolymph and to cause pathogenesis such as Erwinia carotovora strain 15 , Listeria monocytogenes , Candida albicans , and Enterococcus fecalis do not appear to have attenuated virulence when hemolymph lipids are depleted ( unpublished data ) . Many of these pathogens are free-living and are likely to have retained well-developed metabolic capacities ( including the capacity to synthesize lipids ) . It is interesting to speculate that increased reliance on host provision of lipids is part of a suite of adaptations that facilitate the evolution of chronic , low-virulence infection strategies . It is notable that a number of pathogens that have the capacity to form chronic infections in humans including Mycobacterium tuberculosis and Chlamydia trachomatis have been shown to be heavily dependent on lipids acquired from their hosts ( Ehrt and Schnappinger , 2007; Robertson et al . , 2009 ) . Thus , the findings discussed here for endosymbionts could be of more general importance for host–microbe interactions . We used a wild-type Oregon-R ( ORR ) fly stock that harbors MSRO Spiroplasma ( Pool et al . , 2006; Herren and Lemaitre , 2011 ) but not Wolbachia . The w;AKHR1;bmm1 stocks used have been described ( Grönke et al . , 2007 ) . RNAi-mediated knockdown of Lpp was achieved using a pFRiPE-mediated inducible RNAi element ( Marois and Eaton , 2007 ) , placed in the presence of a heat-shock inducible flippase and an Adh-GAL-4 driver that is mostly active in the fat body . The strategy and stocks used are analogous to a previously published study , except that we induced RNAi in adults , as opposed to larvae ( Palm et al . , 2012 ) . We induced the flippase by heat-shock ( 1 . 5 hr at 37°C in a water bath ) in 4- to 5-day-old adult flies . This results in the excision of an upstream spacer region and activation of the UAS-Lpp-RNAi construct driven by Adh-GAL-4 and the silencing of Lpp in the fat body , specifically at the adult stage . An additional strategy used to knockdown the expression of Lpp involved the C564-GAL-4 driver ( also mostly active in the fat body ) in conjunction with tubulin-gal80ts , a temperature-sensitive repressor of GAL-4 expression , which blocks GAL-4 expression at 18°C but not 29°C ( McGuire et al . , 2003 ) . Flies that contained both the C564-GAL-4 , tubulin-gal80ts , and UAS-Lpp-RNAi ( 46 ) elements were maintained at 18°C and then shifted to 29°C as adults to induce UAS-Lpp-RNAi ( 46 ) and knockdown Lpp expression . The UAS-Lpp-RNAi ( 46 ) stock is TRiP #HM05157 originating from the transgenic RNAi project at the Harvard Medical School . Since Spiroplasma MSRO is vertically transmitted and kills male embryos , Spiroplasma-infected stocks were generated in several steps: ( 1 ) Crossing an infected ORR female with males carrying appropriate balancer chromosomes ( 2 ) Crossing the balanced female progeny with males of the genotype of interest ( either w;AKHR;Bmm , w , HS-FLP;adh-GAL-4/bcg or Dipt-GFP , C564-GAL-4;tub-GAL80ts ) ( 3 ) Several back-crosses were then carried out , resulting in a homozygous Spiroplasma-infected female with the appropriate genotype . These stocks were then maintained by crossing females with non-infected males of the same genotype . Spiroplasma-infected females carrying a GAL-4 construct were then crossed with males carrying UAS-RNAi constructs or controls ( w background ) . All flies were maintained at 25°C unless otherwise specified . The density of animals per vial was equilibrated between Spiroplasma-infected and Spiroplasma-uninfected stocks for development under similar levels of larval competition . Unless otherwise specified , all flies used were females and virgins . For survival experiments , counts were made every 24 hr and flies were transferred to new tubes every 3–4 days ( 2 days for mated flies ) . Climbing assays were carried out as described ( Barone and Bohmann , 2013 ) . Acute starvation assays have been described previously ( Grönke et al . , 2007 ) . To quantify the number of eggs laid , flies were collected immediately post-eclosion and maintained in individual Drosophila vials that each contained five flies . Eggs were counted every 48 hr over a 14-day period , and the number of eggs laid per 48 hr per fly was calculated ( correcting for any fly mortality ) . For experiments that required mated flies , males ( three for every five females ) were placed in the Drosophila vials and then removed after 7 days , males that died prior to this were replaced . Flies were raised and maintained on a standard cornmeal-agar diet , referred to as ‘rich media’ . Normal media contain 4 . 5 g agar , 58 . 8 g inactivated yeast ( Springaline BA95/0; Biospringer , Milwaukee , WI , USA ) , 35 g maize flour ( Farigel Maize; Westhove , Ennezat , France ) , 34 . 8 ml of 1:1 mix of grape and multi-fruit juice ( approximately 8 . 2 g of sugar ) , 3 . 6 ml of propionic acid , and 18 ml of a 10% solution of methyl paraben in 85% ethanol per 600 ml of water . For nutrient deprivation , adults were maintained on a restrictive diet referred to as ‘poor media’ , which contains 9 g agar , 1 . 9 g inactivated yeast , 7 . 5 g maize flour , 4 . 5 g sucrose , 9 g glucose , 0 . 3 g MgSO4 , 0 . 3 g CaCl2 , 3 . 6 ml propionic acid , and 18 ml of a 10% solution of methyl paraben in 85% ethanol per 600 ml of water ( Vijendravarma et al . , 2012 ) . Poor media complemented with inactivated yeast contain an additional 34 g of inactivated yeast . Poor media complemented with sucrose contain an additional 35 . 5 g of sucrose . In all cases , larvae were raised on rich media . To observe Spiroplasma in fly hemolymph , flies were dissected on microscope slides in 5 μl PBS containing 0 . 02 mM SYTO9 ( Invitrogen , Carlsbad , CA , USA ) . Slides were then mounted and observed on an Axioimager Z1 ( Zeiss , Oberkochen , Germany ) . Images were captured with an Axiocam MRn camera and Axiovision software . We extracted DNA from five flies per sample . The DNA extraction and quantitative PCR protocols have been previously described ( Anbutsu and Fukatsu , 2003; Herren and Lemaitre , 2011 ) . To determine the absolute number of bacteria per extraction , we extracted infected fly hemolymph and used fluorescence microscopy to calculate the concentration of Spiroplasma cells stained using SYTO9 ( as described above ) . A dilution series of known concentrations of Spiroplasma cell equivalents ( SE ) was then combined with five uninfected flies prior to DNA extraction and qPCR , which enabled us to generate a calibration curve . In subsequent analyses , to account for differences between qPCR runs , we always used a positive control of known Spiroplasma concentration . The results for all experiments involving Spiroplasma titers are given in SE per fly , which represents the absolute quantities of Spiroplasma per fly . A host gene , RPS17 , was also always quantified to verify the quality of the extraction but did not use this value in the analyses . To quantify bacteria in hemolymph samples , 0 . 5 μl of hemolymph ( collected as described for metabolite analyses ) was diluted in 300 μl cell lysis buffer prior to DNA extraction and qPCRs as previously described ( Herren and Lemaitre , 2011 ) . Drosophila hemolymph was collected from flies individually using a Drummond nanoject and pulled capillary needle . Metabolic quantifications are given as the mass ( µg ) of metabolite per µl or ml of Drosophila hemolymph . For each hemolymph sample , we collected 2 µl ( ∼25 flies ) , which was then diluted in 100 µl H2O prior to subsequent analyses . For all analyses hemolymph samples were centrifuged at 13 , 000×g for 2 min to remove Drosophila cells . Protein concentration was determined using a Bradford assay ( Bio-Rad , Hercules , CA ) . Glucose and trehalose concentration was quantified using glucose HK kit ( Sigma-Aldrich , St . Louis , MO , USA ) . Samples were treated with or without trehalase ( 11 mU , Sigma-Aldrich ) overnight at 37°C , and trehalose levels were obtained by subtracting the amount of free glucose in the untreated sample from the total glucose present in the sample treated with trehalase . L-amino acid concentration was quantified by a coupled enzyme reaction , using the L-amino acid quantitation kit ( Sigma-Aldrich ) . Hemolymph DAG and fat body TAG were analyzed using a coupled colorimetric assay ( Hildebrandt et al . , 2011 ) . Sterols were quantified using the Amplex Red Cholesterol Assay kit ( ThermoFisher Scientific , Waltham , MA , USA ) , which detects primarily cholesterol but also other sterols found in Drosophila ( e . g . , ergosterol ) . For protein and sterol assays , Spiroplasma were removed from hemolymph samples by centrifugation at 15 , 000×g for 15 min . For whole fly analyses , 10 flies were homogenized in 250 μl PBS and the quantifications are given per fly . Note that there was no significant difference in the mass of virgin flies with and without Spiroplasma at 7 and 14 days of age ( unpublished data ) . There was also no significant difference in the feeding rate of virgin flies with and without Spiroplasma ( unpublished data ) as measured by CAFE assay ( Ja et al . , 2007 ) over a period of 4 hr at 3 , 10 , and 12 days of age . Drosophila hemolymph ( collected as described for metabolite analyses ) was diluted in 1:10 in H2O . Lipids were extracted using a mixture of chloroform/methanol ( 50/50 vol:vol ) . After mixing and centrifugation , 1 μl of the lipid extract was diluted with 1 μl of 9-aminoacridine matrix ( 25 mg/ml , dissolved in isopropanol/acetonitrile [3/2 , vol/vol] ) . For MALDI-Tof-MS analysis , 1 μl of the mixture was rapidly spotted on an MTP 384 polished steel MALDI plate ( Bruker Daltonics , Billerica , MA , USA ) . MALDI-Tof mass spectra of the Drosophila hemolymph samples were acquired using FlexControl 3 . 0 ( Bruker Daltonics ) on a Bruker AutoFlex III Smartbeam ( Bruker Daltonics ) in a negative reflectron mode at a laser beam attenuation of 50 and focus of 40 at 100 Hz as laser repetition rate . A total of 1000 shots were acquired in the mass range of 400 to 2000 m/z . Data were processed with FlexAnalysis 3 . 0 ( Bruker Daltonics ) . Calibration of the instrument was performed using Peptide Standard Calibration II ( Bruker Daltonics ) . 1000 ion counts were accumulated in the mass spectrometer prior to fragmentation . Fragmentation was performed with the Bruker designed method and at a laser beam attenuation of 42% , a laser repetition rate of 100 Hz and a reflector detector voltage set to 1 . 861 kV in a negative mode . Hemolymph was extracted from 1250 7-day-old adult Spiroplasma-infected and Spiroplasma-uninfected mated female flies by pricking flies in the abdomen and thorax before centrifugation of flies at 16000×g at 4°C for 30 min in a 10-μm filter spin column ( Mobitec , Goettingen , Germany ) . Hemolymph was subsequently filtered in Ultrafree-MC Centrifugal Filter Units ( 0 . 22 µm pores ) at 16000×g at 4°C for 10 min . These samples were then sent to Avanti Polar Lipids analytical services division ( http://www . avantilipids . com ) where lipids were extracted ( Folch et al . , 1957 ) prior to extracts being spiked with internal standards for LC-MS/MS quantification of diacylglygerol species concentrations ( Hutchins et al . , 2008 ) . Statistical significance was calculated using a Gehan-Breslow-Wilcoxon test for survivals and an unpaired Student's t test for all other experiments ( with GraphPad Prism 5 . 0 ) and considered significant if p-values were lower than 0 . 05 . Asterisks indicate the level of significance: p<0 . 05 , p<0 . 01 , and *p<0 . 001 and NS ( non-significant ) .	All animals host a large number of harmless microbes . Often the two partners involved in these interactions will depend on each other to thrive: microbes support important host functions and in return the host provides a safe place to live and a continuous supply of food . Many microbes that are intimately associated with animals have lost the ability to gain nutrients from sources other than their host and are unable to survive on their own . However , in many cases , the source and the type of nutrients provided to the microbes are unknown . One of the most common microbial species found in insects is Spiroplasma . This microbe lives in very large numbers in the fluid that fills the body cavities of insects , called the hemolymph . The microbes are transmitted from mother to offspring , and in some circumstances can provide benefits to the insects; for instance , Spiroplasma-infested flies appear to be protected against infection by some parasites . Unfortunately , as it is difficult to study insect–microbe relationships , little else is known about the physiological interactions between these two species . Herren et al . studied the association between Spiroplasma and the fly Drosophila melanogaster . Under normal conditions , Spiroplasma only reduces the life span of the infested fly . This indicates that Spiroplasma has a low impact on the general fitness of its host , only negatively affecting the survival and egg laying ability of old flies . When flies had limited access to nutrients , the number of Spiroplasma they carried was reduced , without the flies losing fitness . This suggests that Spiroplasma growth is dependent on something in the flies' diet . To understand which nutrients are important for the growth of Spiroplasma in Drosophila , Herren et al . analyzed the hemolymph of flies and found that there are fewer fatty-molecules , called lipids , when nutrients are limited . Healthy flies carrying Spiroplasma also have fewer lipids in their hemolymph , suggesting that these are what Spiroplasma feed on . Indeed , inactivating a protein required by the fly to transport lipids to the hemolymph reduced the growth of Spiroplasma in these flies . Herren et al . concluded that the growth of Spiroplasma inside its host is limited by the availability of lipids in the hemolymph . Since this is dependent on diet , the dependence on lipids couples the growth of Spiroplasma to the nutritional state of its host . Herren et al . speculate that this mechanism reduces the fitness cost of harboring the microbes and prevents the damaging consequence of an uncontrolled proliferation of the microbes . Moreover , Spiroplasma's preference for lipids may explain why it helps to protect flies against parasitic infection , as many parasites also rely on lipids for their growth . Herren et al . suggest this strategy could also be used in other animal–microbe associations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2014	Insect endosymbiont proliferation is limited by lipid availability
Discrete populations of brain cells signal heading direction , rather like a compass . These ‘head direction’ cells are largely confined to a closely-connected network of sites . We describe , for the first time , a population of head direction cells in nucleus reuniens of the thalamus in the freely-moving rat . This novel subcortical head direction signal potentially modulates the hippocampal CA fields directly and , thus , informs spatial processing and memory . Nucleus reuniens ( NRe ) , one of the largest midline thalamic nuclei , receives extensive limbic inputs and provides a bridge linking the hippocampus ( especially area CA1 ) with medial prefrontal cortex ( McKenna and Vertes , 2004; Vertes , 2006; Prasad and Chudasama , 2013 ) . Its functions are not well-understood , but it has been suggested that , via these connections , NRe influences memory consolidation for spatial learning and generalisation of fear conditioning ( Eleore et al . , 2011; Loureiro et al . , 2012; Xu and Sudhof , 2013 ) . To elucidate its functions open-field single-unit recordings in NRe of freely-moving rats were made ( Mink et al . , 1983 ) . We have found an unexpected population of NRe cells signalling head direction ( HD ) in the horizontal plane , independent of location within the test arena . These cells resemble HD cells in the anterodorsal and anteroventral thalamic nuclei ( Shinder and Taube , 2011; Tsanov et al . , 2011 ) , the lateral mammillary nucleus ( Taube , 2007 ) , and certain parahippocampal regions ( Cassel et al . , 2013 ) . NRe cells maintain head directionality during light–dark transitions , and in environments of different shape . These cells establish directionality rapidly upon first entering an environment . NRe has not , to date , been part of the traditional HD circuit , which largely originates in the dorsal tegmental nucleus of Gudden and the lateral mammillary nucleus ( Taube , 2007; Cassel et al . , 2013 ) . Subsequent processing is via the anterodorsal thalamic nucleus , dorsal presubiculum , entorhinal cortex and to the hippocampus ( Su and Bentivoglio , 1990; Vertes et al . , 2007 ) . Our findings , therefore , reveal a novel head direction signal potentially modulating the hippocampal CA fields and , therefore , hippocampal spatial processing ( Brandon et al . , 2013 ) . Lighting conditions were systematically varied across foraging sessions for 10 cells . The animal foraged during light-dark-light sessions ( each 20 min ) . Light removal did not affect NRe head directional activity ( Figure 1 ) . 10 . 7554/eLife . 03075 . 003Figure 1 . Head direction cells recorded in the nucleus reuniens . ( A ) 18 representative head direction ( HD ) cells in nucleus reuniens ( NRe ) ; ( B ) NRe location on a coronal ( left ) and corresponding sagittal ( right ) rat brain section ( adapted from Paxinos and Watson , 2005 ) ; ( C ) representative histological specimen showing electrode track ( left ) ; recording positions corresponding to cell locations presented in panel a ( upper right inset ) showing location of NRe and detailed atlas ( lower right inset ) ; ( D ) representative recordings showing multi-day stability of HD cells: a representative cell recorded on each day of 16 days ( multiple transitions from light-dark-light , and environmental transformations from circle to square to circle ) . The solid line is the mean spike waveform and dashed lines are M ± SD of the spike waveform . The green outline shows predicted firing rates given the proportion of time the animal spent looking in each direction , calculated according to the distributive hypothesis . DOI: http://dx . doi . org/10 . 7554/eLife . 03075 . 003 We transformed arena shape ( circle-square-circle; Figure 1D; circle-square; Figure 2A , D ) . There was no effect of environmental shape changes on any HD cells ( n = 10 ) . 10 . 7554/eLife . 03075 . 004Figure 2 . Nucleus reuniens head direction cells do not appear to be spatially modulated and are present from first exposure to the environment . ( A and D ) Representative waveforms for two units showing , ( top to bottom ) , light-dark-light and circle–square transitions; from left to right , autocorrelation histogram , polar plots , tuning curves for clockwise ( CW ) vs counter-clockwise ( CCW ) head movements; ( B ) spatial analysis for cardinal orientations showing no effect of spatial position on unit activity; ( C ) temporal evolution of HD firing for cumulative samples ( time ranges: 0–1 , 0–2 , 0–4 , 0–8 , 0–16 and 0–20 min ) and independent time-binned samples ( 0–1 , 1–2 , 2–4 , 4–8 , 8–16 and 16–20 min ) demonstrating that HD activity is present in the first minute of exposure to the arena; ( D , E , F ) as ( A , B , C ) . Red lines are formed by continuous firing activity when the rat walks with its head directed in the preferred HD . The firing map represents −22 . 5° to +22 . 5°; as the ring was formed by eight plots each representing a 45° extent of head direction . DOI: http://dx . doi . org/10 . 7554/eLife . 03075 . 004 Transitions between light and dark , arena shape , mean head direction , mean head direction for clockwise and counter-clockwise movement , peak head direction and peak firing rate were compared using t-test for paired two samples for means with Bonferroni correction . No significant differences were observed between conditions . 22 cells were recorded across at least 2 consecutive days ( 15 were recorded for three or more days , and three were stably recorded for 14 or more days . Initial head direction remained stable across days , even for extended recordings ) , indicating no effect of time or sleep/wake cycle on their preferred directionality . Figure 2C , F depict the temporal evolution of head directional firing for cumulative samples and independent time-binned samples , demonstrating head directional activity is present from the first minute of exposure to the environment ( irrespective of environment shape ) . Separation angle ( the offset in mean peak firing rate for clockwise vs counter-clockwise head movements ) is present in about 50% of recorded HD cells . There were no significant differences between mean head direction measured in degrees for clockwise and counter-clockwise movement in the t test for paired two samples for means in the whole population of HD cells recorded in NRe . Brandon et al . ( Paxinos and Watson , 2005 ) recorded in medial entorhinal cortex ( mEC ) and found units firing in a fixed synchronous or anti-synchronous relationship with alternate theta cycles . We find a similar population of cells in NRe ( see online material ( OM ) for detailed calculations and Figure 3A , B ) . These cells , which do not carry a HD signal , are found only in NRe , providing a physiological marker for electrode depth . The theta skipping index is always positive ( [mean ± SD] 0 . 334 ± 0 . 127; jump factor: 0 . 367 ± 0 . 075; frequency ratio: of 2 . 00 ± 0 . 018 ) . Frequencies ( mean ± SD ) of the higher and lower oscillations were 8 . 670 ± 0 . 259 Hz and 4 . 334 ± 0 . 112 Hz , respectively . 10 . 7554/eLife . 03075 . 005Figure 3 . Theta skipping cells are present in nucleus reuniens . ( A and B ) 12 representative theta cycle skipping cells which do not carry HD signal recorded in NRe; ( A ) from the left waveform , autocorrelation per 10 ms , interspike interval histogram ( ISIH ) , autocorrelations per 1000 ms with and without fitting to the model , firing intensity map and polar plot are presented for two representative cells . Red line: envelope of the autocorrelation histogram obtained by fitting histogram data to the equation ( see OM ) ; ( B ) waveforms , ISIHs , and autocorrelation histograms with redline envelope of data fitting to the equation ( OM ) for 10 theta cycle skipping cells recorded in NRe . DOI: http://dx . doi . org/10 . 7554/eLife . 03075 . 005 All described HD cells were analysed to test for spatial/place cell modulation ( OM ) . In no case was any spatial modulation of the HD cell signal detected ( see , e . g . , Figure 2B , E ) . Nucleus Reuniens has been investigated anatomically , but remains largely unexplored electrophysiologically ( Cassel et al . , 2013 ) . The HD cells in NRe share many characteristics of other HD cells , such as those in the anterodorsal and anteroventral thalamic nuclei , the dorsal presubiculum and retrosplenial cortex ( Taube , 2007 ) . Current theoretical models of HD information suggest that the HD signal derives , in part , from the dorsal tegmental nucleus of Gudden and the lateral mammillary nuclei ( Taube , 2007; Cassel et al . , 2013 ) . In addition to projecting to the anterodorsal thalamic nucleus , the lateral mammillary nucleus innervates NRe ( McKenna and Vertes , 2004 ) . NRe also receives inputs from presubiculum ( postsubiculum ) and retrosplenial cortex , other sites containing HD cells . This pattern of connectivity suggests various potential sources for the heading information found in NRe . The presence of spatial cells in NRe is highly significant , as it is the major source of direct thalamic projections to the CA fields of the rat hippocampus ( Su and Bentivoglio , 1990 ) . NRe also receives widespread cortical and limbic inputs , involving both interoceptive and exteroceptive information . This nucleus has a notable role linking medial prefrontal cortex with the hippocampus ( Prasad and Chudasama , 2013 ) . These NRe HD cells are , therefore , pivotally positioned to influence hippocampal spatial processing directly because of its dense , direct connections with both the prefrontal cortex and hippocampus ( Vertes , 2006; Prasad and Chudasama , 2013 ) . Finally , the theta-skipping cells may provide a pace-maker like function for synchronising some early components of the HD system . The data from which this paper was generated are archived at doi: 10 . 5061/dryad . j68v0 and may be accessed via http://dx . doi . org/10 . 5061/dryad . j68v0 . Eleven ( 4–6 months ) male Lister-Hooded rats ( B&K , UK ) weighing 420–530 g were used . Upon arrival , animals were housed individually and handled by the experimenter daily for a week before being trained in the pellet-chasing task ( see below ) . Rats were food-restricted to 85% of their ad libitum body weight and kept in a temperature-controlled laminar airflow unit and maintained on a 12-hr light/dark cycle ( lights on from 08:00 to 20:00 hr ) . Experiments were carried out in strict accordance with regulations laid out by LAST Ireland and were compliant with the European Union directives on animal experimentation ( 86/609/EEC ) . Experiments were conducted in a circular arena ( diameter 96 cms ) and square arena ( 60 × 60 cm ) . The insides of the arenas were a uniform matt black , and low-level lighting was used during light testing; all lights were extinguished during dark testing . All experiments were conducted during the day between 0900 and 2000 hr . Session lengths were typically 20 min duration . Rats performed a pellet-chasing task during the course of the experiments . During testing , 20 mg food pellets ( TestDiet , 5TUL formula ) were thrown in the arena at random locations ca . every 20 s . During the weeks of recordings , animals were allowed 20 g of food daily . The environment is partially curtained with a visual cue card in a constant location . We leave the rat in the environment during the LDL transitions . Detailed descriptions of the surgical protocol and recording techniques can be found elsewhere ( Brotons-Mas et al . , 2010; Tsanov et al . , 2011 ) . Briefly , rats were implanted with tetrodes of either four or eight bundles of ø 25 µm platinum–iridium wires ( California Fine Wire Ltd . , USA ) mounted onto small driveable microdrives ( Axona Ltd . , UK ) at the following coordinates targeted at the nucleus reuniens ( see Figure 1 for histological verification ) : −1 . 60 mm posterior to bregma , −1 . 20 mm lateral to the midline and at angle of about 5 . 5° . Depth varied depending on the target structure and ranged from 4 . 8 to 5 . 6 mm below the brain surface ( Figure 1; Paxinos and Watson , 2005 ) . Rats were allowed at least 1 week of recovery post-surgery . Tetrodes were lowered slowly through the brain ( maximal rates 25–50 µm/day ) , typically over a period of weeks to prevent tissue damage and to ensure successful NRe electrode targeting and penetration . Based on the daily record of the electrode position and post-mortem histological verification , each recording could be located along the tetrode trace . The recording sessions took place in arenas located in the centre of the test room , which contained multiple , large visual cues made salient to allow the animals to orient themselves in the environment . An example of a NRe HD cell is provided in Video 1 . 10 . 7554/eLife . 03075 . 006Video 1 . An example of a well-discriminated head direction unit recorded in NRe while the rat is engaged in pellet chasing in a circular arena . The unit corresponds to unit 4 of Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 03075 . 006 Standard statistical testing used Matlab scripts and Axona software . Unit identification involved several criteria . First , neurons had to be active in all conditions and had to present the same waveform characteristics ( amplitude , height , and duration ) in those conditions . Furthermore , units had to demonstrate a clean refractory period ( >2 ms ) in the inter-spike interval ( ISI ) histogram . Units were sorted using conventional cluster-cutting techniques and classified by the environmental manipulation to which they were exposed . Once well-defined neuronal signals were isolated and the rats explored the arena sufficiently ( rats had to explore at least 90% of the open field in either session to be included in analysis to allow reliable calculation of spatial characteristics ) , recording commenced . In total , 758 well-isolated units were recorded in 3 rats in NRe and the adjacent anteromedial thalamic nucleus ( AM ) . After post-mortem histological verification , 483 cells were assigned to NRe . To select animals for analysis we set following criteria: histologically-verified electrodes should be placed in the lateral part of nucleus reuniens , therefore bypassing laterally the Rhomboid nucleus , avoiding the possibility that by mistake some cells would be wrongly assigned to NRe or the Rhomboid nucleus . Further , an observed electrophysiological criterion required co-localised HD cells and theta skipping cells in NRe . We did recordings in many animals in surrounding nuclei but as described in the present paper theta skipping cells are characteristic only for NRe , are electrophysiologically colocalised with HD cells therefore can serve as an electrophysiological marker of lateral part of NRe . Electrode tracks were localised predominantly in the lateral portion of NRe . The numbers and percentages of cells recorded in NRe were: 42 HD cells ( 8 . 7% ) ; 19 theta cycle skipping cells ( 3 . 9%; these cells characteristically appeared in NRe , electrophysiologically co-localised with HD cells ) ; 55 other theta modulated cells ( 11 . 3% ) ; 23 fast firing cells ( 4 . 7% ) ; 21 weakly-theta modulated cells ( 4 . 3% ) ; 13 other spatially-tuned cells ( 2 . 2% ) . 309 cells ( 63 . 9% ) were classified as unidentified low firing units–cells that did not exhibit any particular temporal or spatial properties or formed groups smaller than four cells with similar phenotype . Among unidentified low firing units 53 cells fired with maximum frequency lower than 1 Hz . Directional analyses were performed for all recorded cells in nucleus reuniens , 42 units in total ) . The rat's HD was calculated for each tracker sample from the projection of the relative position of the LEDs onto the horizontal plane . The directional tuning function for each cell was obtained by plotting the firing rate as a function of the rat's directional heading , divided into bins of 5° . The firing rate was computed based on the total number of spikes divided by the total time in that bin ( Taube et al . , 1990 ) . To restrict the influence of inhomogeneous sampling on directional tuning , we accepted data only if all directional bins were sampled by the rat . The directionality of the HD units in the horizontal plane ( measured in degrees ) was normalized for comparison of the HD firing rate properties . The peak firing rate of cells that respond to different direction of heading was aligned to a HD of 180° ( Bassett et al . , 2005 ) . The firing rate was normalized ( with values between 0 and 1 ) with respect to the peak firing rate for each unit ( Bassett et al . , 2005 ) . The firing rate is calculated by dividing the number of spikes by the number of visits at a particular head direction bin . The CW/CCW separation was calculated by considering a particular angular head velocity threshold ( 120°/s ) . When the rat moves at +120°/s , it was placed into the CW head direction category ( similarly , for CCW , −120°/s ) . If it coincidentally happens that spikes occur at a lower rate during slower head movements , the visit count will increase in the non-separating case , the spike count will be relatively decreased ( causing a decrease in the firing rate compared to the CW/CCW separation ) . Additional analyses examined for spatial modulation of these cells . The spatial specificity ( or spatial information content ) is expressed in bits per spike ( Skaggs et al . , 1996 ) . Place field size is computed as the region of the arena in which the firing rate of the place cell is above 20% of the maximum firing frequency ( Hollup et al . , 2001 ) . A place field was identified if nine neighbouring pixels ( sharing a side ) were above 20% of the peak firing rate . Place field size was represented in number of pixels . The spatial selectivity of a firing field ( ratio of maximal signal to noise ) was calculated by dividing the firing rate of the cell in the bin with the maximum average rate by its mean firing rate over the entire apparatus ( Skaggs et al . , 1996 ) . Average frequency is the total number of spikes divided by the total recording time and is expressed in Hz . Where recordings were conducted on successive days , units were matched based on their spike amplitude , height , and spike duration with the respective units from the control recording . Exploration was assessed by comparing the occupancy of bins and the number of visits per bin between the two recording conditions . On completion of the recording studies , the rats received an overdose of anaesthetic ( 1 . 5 g of urethane ( Sigma-Aldrich , Dublin , Ireland ) dissolved in 4 . 5 ml water ) and were then perfused intracardially with 250 ml of 0 . 1 M phosphate-buffered saline ( PBS ) at room temperature followed by 350 ml of 4% paraformaldehyde in 0 . 1 M PBS at ∼4°C , after which the brains were removed and placed in 4% paraformaldehyde ( for at least 72 hr ) . Brains were blocked , placed on a freezing platform , and 40 µm coronal sections were cut with a sledge microtome ( Leica 1400 ) . Two , alternate series that used all sections were taken through the rostral thalamus . One series was mounted directly onto gelatine-subbed slides , and then allowed to dry overnight before staining with cresyl violet , a Nissl stain . The second series was immunologically reacted with the neuronal marker α-NeuN ( MAB 377; Chemicon , Watford , UK ) , then with a secondary horse anti-mouse rat adsorbed antibody ( AI-2001; Vector Laboratories Ltd , Peterborough , UK ) and subsequently visualised with Vector Elite ABC ( PK-6100; Vector Laboratories Ltd ) and diaminobenzidine . A Leica DM5000B microscope with Leica DFC310FX digital camera and Leica Application Suite image acquisition software was used for brightfield microscopy . NeuN provides a stain assumed to be selective for neurons , that is it does not label glial cells . As a consequence it can sometimes make it easier to see underlying cytoarchtectonic features . Recording positions were determined by calculating distance above the deepest electrode position , and calculating distance below the first penetration into the tissue . The electrodes often caused tissue distortion and this was carefully allowed for in the position calculations . Positions of recorded cells were estimated as follows: theoretical positions of electrodes tips and NRe borderlines were estimated by reference to atlas ( Paxinos and Watson , 2005 ) and reconstructed histological specimens . The electrodes sometimes caused tissue distortion even with very slow penetrations , and this was allowed for in the position calculations . The position of electrodes below brain surface was known for each recording session and expressed in µm , thereby allowing estimates of each cell position to be derived .	Whether it is foraging for food or finding its way back to its nest , an animal often needs to know which direction it is heading in . Some neurons in a mammal’s brain have been shown to act like a compass , and send out nerve impulses whenever the animal points its head in a certain direction . For example , some of these neurons will fire when the animal faces northeast , but not when it faces northwest , and vice versa . Importantly these neurons , called ‘head direction’ cells , do not actually measure the Earth's magnetic field . Rather , they respond to information about landmarks in the environment and the animal's movements of its head or body to work out which way the animal is facing . Head direction cells are largely found in a closely-connected network of a few sites in the brain . However , Jankowski et al . have now discovered more of these cells in another region found deep within the centre of the brain . Measuring the nerve impulses from these neurons in rats that were moving freely around a test arena revealed that the neurons fired in exactly the same way as some other head direction cells in other regions of the brain . For example , they fired whenever the rat faced one direction , but stopped firing when it turned its head to face another . Jankowski et al . showed that the head direction cells in this region of the brain continued to work when the lights were turned off in the test arena , or when the shape of the arena was changed from a circle to a square . These neurons began sending information about head direction as soon as the rat entered the test arena , and many continued to fire when the rat faced the same direction even when they were retested on several different days . The head direction cells discovered by Jankowski et al . are connected to another region of the brain that is involved in remembering different locations in the environment and navigating between them . This suggests that these neurons might provide some of the information required to carry out these tasks . It also means that areas of the brain close to those that receive input from the outside world may perform more complex cognitive functions than previously thought .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2014	Nucleus reuniens of the thalamus contains head direction cells
The activation of the dodecameric Ca2+/calmodulin dependent kinase II ( CaMKII ) holoenzyme is critical for memory formation . We now report that CaMKII has a remarkable property , which is that activation of the holoenzyme triggers the exchange of subunits between holoenzymes , including unactivated ones , enabling the calcium-independent phosphorylation of new subunits . We show , using a single-molecule TIRF microscopy technique , that the exchange process is triggered by the activation of CaMKII , and that exchange is modulated by phosphorylation of two residues in the calmodulin-binding segment , Thr 305 and Thr 306 . Based on these results , and on the analysis of molecular dynamics simulations , we suggest that the phosphorylated regulatory segment of CaMKII interacts with the central hub of the holoenzyme and weakens its integrity , thereby promoting exchange . Our results have implications for an earlier idea that subunit exchange in CaMKII may have relevance for information storage resulting from brief coincident stimuli during neuronal signaling . Ca2+/calmodulin-dependent protein kinase II ( CaMKII ) plays a critical role in neurons , where it is a component of the molecular networks that lead to the strengthening of synaptic connections between co-active neurons , as exhibited in long-term potentiation ( LTP ) ( Lisman et al . , 2002; Coultrap and Bayer , 2012; Lisman et al . , 2012 ) . The early phases of LTP involve the activation of CaMKII and the consequent phosphorylation of downstream targets , such as the AMPA receptor and stargazin , which leads to increased ionic currents into the neuron as well the targeting of CaMKII to other proteins involved in LTP ( Lisman et al . , 2002; Wayman et al . , 2008 ) . The activation of CaMKII , for example , results in its recruitment to the NMDA receptor , and this NMDA:CaMKII complex may contribute to the maintenance of LTP ( Bayer et al . , 2001; Sanhueza et al . , 2011; Sanhueza and Lisman , 2013 ) . CaMKII also plays an essential role in modulating cardiac pacemaking and excitation-contraction coupling . Several heart diseases are associated with the hyper-activation of CaMKII ( Backs et al . , 2009; Chelu et al . , 2009; Neef et al . , 2010; Rokita and Anderson , 2012 ) . CaMKII is unusual among protein kinases because it is organized into a dodecameric holoenzyme ( Bennett et al . , 1983; Woodgett et al . , 1983; Kolodziej et al . , 2000; Morris and Torok , 2001; Rosenberg et al . , 2005; Chao et al . , 2011; see Stratton et al . , 2013 for a recent review of CaMKII structure , shown schematically in Figure 1A ) . The holoenzyme consists of a central hub formed by the association of the C-terminal domains of CaMKII , to which the catalytic modules , consisting of the kinase domain and a regulatory segment , are attached by a variable linker ( see Figure 1B for a schematic representation of the domains of CaMKII ) . There are four CaMKII genes in humans , termed α , β , γ , and δ and these generate ∼40 isoforms through alternative splicing that results in variations in the linker ( Tombes et al . , 2003 ) . The α and β isoforms are found predominantly in neurons , while the γ and δ isoforms are distributed more broadly . In this paper we focus on the human α isoform , CaMKIIα , the major isoform in neurons , which has a 30-residue linker connecting the catalytic module to the hub domain . 10 . 7554/eLife . 01610 . 003Figure 1 . CaMKII architecture . ( A ) The architecture of a dodecameric CaMKII holoenzyme . The inactive holoenzyme is shown as a more compact configuration . Upon activation by Ca2+/CaM , or phosphorylation of Thr 286 in the regulatory segment ( purple circles ) , the kinase domains are extended from the hub assembly . ( B ) The domains of a CaMKII subunit . ( C ) Phosphorylation control in CaMKII . The R1 element of the regulatory segment leads into a helical R2 element that blocks the substrate binding channel of the kinase domain in the inactive form . The R3 element contains the calmodulin-recognition motif , and upon CaM binding , CaMKII is autophosphorylated at Thr 286 in the R1 element . After CaM dissociates , Thr 305 and 306 are phosphorylated if Thr 286 is already phosphorylated . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 003 The regulatory segment of autoinhibited CaMKII docks within the substrate-recognition groove of the kinase domain and keeps it inactive ( Hook and Means , 2001 ) . An increase in calcium levels results in the binding of Ca2+/calmodulin ( Ca2+/CaM ) to the regulatory segment , thereby activating CaMKII ( Figure 1C; Colbran et al . , 1989; Ikura et al . , 1992 ) . Subsequent to activation by Ca2+/CaM , the phosphorylation of at least three critical threonine residues within the regulatory segment modulates kinase activity and the response to calmodulin ( Figure 1C ) . In contrast to most other kinases , CaMKII has no phosphorylation sites within the activation loop . A critical phosphorylation site , Thr 286 , is located in the R1 element of the regulatory segment ( see Figure 1B; the residue numbering we use corresponds to that of a human CaMKIIα construct in which the 30-residue linker between the catalytic module and the hub domain is deleted [Chao et al . , 2011] ) . The phosphorylation of Thr 286 requires one activated kinase domain to phosphorylate another kinase domain in the same holoenzyme , and trans-phosphorylation between different holoenzymes is not observed ( Hanson et al . , 1994; Rich and Schulman , 1998 ) . Phosphorylation of Thr 286 prevents the R1 element from rebinding to the kinase domain , thereby conferring partial calcium-independence to the catalytic module ( Lou et al . , 1986; Lisman et al . , 2002 ) . This property , referred to as autonomy , prolongs the active state of CaMKII and is likely to be critical for the generation of LTP ( Saitoh and Schwartz , 1985; Lai et al . , 1986; Miller and Kennedy , 1986; Schworer et al . , 1986; De Koninck and Schulman , 1998 ) . Mutant mice with Thr 286 in CaMKIIα replaced by alanine have limited LTP generation and display impairments in learning and memory ( Giese et al . , 1998 ) . Two other key sites of phosphorylation , Thr 305 and Thr 306 , are located within the calmodulin-binding portion of the regulatory segment , in the R3 element . The functional significance of these phosphorylation sites is less well understood but , as for Thr 286 , studies using knock-in mice have emphasized their importance in learning and memory ( Elgersma et al . , 2002 ) . Phosphorylation of either residue weakens the affinity of CaMKII for Ca2+/CaM ( Hanson and Schulman , 1992; Colbran , 1993 ) . Using in vitro experimentation , we now report that activation of the CaMKII holoenzyme enables CaMKII subunits to exchange between different holoenzymes and to spread their state of activation in the process . The possibility that memory storage in the brain might rely on the exchange of subunits between autonomous ( calcium-independent ) forms and unactivated forms of CaMKII had been proposed earlier by John Lisman , following an initial conjecture by Francis Crick about memory-storage molecules that have the ability to instruct newly synthesized unactivated molecules of earlier activation events ( Crick , 1984; Lisman , 1994 ) . Our results provide a mechanism for maintaining a pool of active CaMKII despite protein turnover , a process that could potentially be important for long-term information storage in the brain . Our general experimental strategy is to label two samples of CaMKII separately with two different fluorophores , followed by mixing and analysis of whether holoenzymes containing both of the fluorophores are obtained . This type of analysis is complicated in bulk solution by the difficulty in distinguishing between aggregation of holoenzymes and true subunit exchange . Gel filtration analysis of a constitutively active mutant ( CaMKIIT286D ) shows no evidence for aggregation , and incubation of the wild-type holoenzyme with Ca2+/CaM and ATP shows no changes in the UV-vis absorption spectrum ( data not shown ) . We also used dynamic light scattering ( DLS ) to determine the size distribution of activated CaMKII assemblies in solution ( for both wild-type , activated by Ca2+/CaM , and a constitutively active mutant in which Thr 286 is replaced by aspartate [CaMKIIT286D] in the absence of Ca2+/CaM ) . DLS measurements indicate that the addition of ATP to CaMKIIT286D , under the conditions used in our mixing experiments , results in a small population with sizes that are larger than that of a single holoenzyme ( Figure 2—figure supplement 1 ) . The DLS data for both wild-type CaMKII and CaMKIIT286D demonstrate that there is no time-dependent change in the particle size distribution over the time scale of our experiments . To rule out aggregation as a complicating factor definitively , we developed a single-molecule assay in which individual CaMKII holoenzymes are visualized directly . In this assay , two samples of CaMKII are labeled separately with the donor ( Alexa 488 , green fluorophore ) and acceptor ( Alexa 594 , red fluorophore ) using cysteine chemistry , and then mixed and incubated together in the presence or absence of Ca2+/CaM and ATP . CaMKII holoenzymes are then immobilized on a glass slide such that individual holoenzyme assemblies are separated spatially from each other , and visualized by TIRF microscopy ( see schematic diagram in Figure 2A ) . An exchange event is detected by the colocalization of red and green fluorophores , and potential aggregation is monitored by analysis of the distribution of fluorophore intensities . 10 . 7554/eLife . 01610 . 004Figure 2 . Single-molecule assay for subunit exchange reveals activation-dependent subunit exchange . ( A ) A representative single-molecule TIRF image , with red and green channels overlaid ( left ) . For analysis , CaMKII holoenzymes are immobilized on glass slides via biotin/streptavidin interactions ( right ) . ( B ) The rate of increase in colocalization is significantly faster at 37°C ( red ) compared to 25°C ( blue ) when Ca2+/CaM and ATP are added . At 37°C , the unactivated sample ( i . e . , with no addition of Ca2+/CaM and ATP ) shows only a low level of exchange even at long time points ( green ) . ( C ) Under activating conditions , decreasing the concentration of CaMKII from 8 µM ( red ) to 1 µM ( blue ) results in reduction of the rate of colocalization . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 00410 . 7554/eLife . 01610 . 005Figure 2—figure supplement 1 . Dynamic light scattering measurements on CaMKII . ( A ) CaMKIIT286D was mixed under the same conditions as in the single-molecule experiments ( ATP , 37°C ) , and samples were examined by dynamic light scattering ( DLS ) after 1 hr . All experiments were performed in duplicate . The autocorrelation function ( left ) calculated from the data before ATP addition ( blue ) is slightly different from the function calculated 1 hr after ATP addition ( red ) . The autocorrelation functions were used to calculate the molecular size distribution in the sample ( right ) . After 1 hr incubation with ATP ( red ) , there exists the same major species corresponding to single holoenzymes , as seen without incubation ( blue ) . In addition , there is a small percentage of larger particles that accumulates over time , but this is a very minor population . ( B ) Wild-type CaMKII was activated using Ca2+/CaM and ATP and incubated at 37°C for 1 hr . The autocorrelation curves do not change significantly over this time course ( left ) . The major species in this solution also remains the same over time ( right ) . The size of this species is consistent with fully extended CaMKII with CaM bound . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 00510 . 7554/eLife . 01610 . 006Figure 2—figure supplement 2 . Custom particle-tracking program . Analysis of colocalized particles using a single-molecule particle-tracking program . Two images from the same area are shown , one illuminated with the excitation laser for Alexa 488 and the other with the excitation laser for Alexa 594 . The left panel shows the raw images obtained using the two illuminations . The fitting procedure for each particle is based on a threshold subtraction and vector gradient analysis for image processing followed by two-dimensional Gaussian fitting ( see ‘Materials and methods’ ) . Particles that are too bright or not bright enough are omitted from analysis . The middle panel shows the results from the particle-tracking program , which locates each particle ( red circles ) and discounts particles that are not circular ( purple circles ) or misshaped due to edge effects ( green circles ) , and would not fit properly to a two-dimensional Gaussian curve ( double purple circles ) . The right panel shows the results of overlaying these two images to determine which particles are occupying the same position ( red circles ) . These are counted as the colocalized particles , and percent colocalization is determined by taking the ratio of colocalized particles/total particles in the less populated channel . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 00610 . 7554/eLife . 01610 . 007Figure 2—figure supplement 3 . Intensity distribution analysis of single-molecule images . The intensity distribution of each image over the course of an isolated experiment reveals a population of particles that have similar brightness , which is exemplified by a single population distribution on the histogram . Importantly , this distribution does not change over time . Data are shown for wild-type CaMKII activated by Ca2+/CaM and ATP ( A ) and CaMKIIT286D mixed with ATP ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 00710 . 7554/eLife . 01610 . 008Figure 2—figure supplement 4 . Comparison of activation methods and subunit exchange . The extent of colocalization is calculated for two separate samples of wild-type CaMKII at 25°C . In one sample , CaMKII separately labeled with red or green dye was mixed together and then activated by Ca2+/CaM and ATP during the mixing reaction ( purple ) . In the second sample , red CaMKII and green CaMKII were separately activated by Ca2+/CaM and ATP for 10 min at 25°C and then mixed together ( red ) . Samples were mixed at 25°C for 1 hr and then analyzed for colocalization . It is clear that pre-activating the samples does not affect the final level of colocalization . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 00810 . 7554/eLife . 01610 . 009Figure 2—figure supplement 5 . FRET mixing experiments corroborate single-molecule results . ( A ) The crystal structure of the dodecameric hub assembly of human CaMKIIγ ( PDB code: 2UX0 ) . The sites of fluorophore labeling are shown as purple spheres ( distances: a ∼33 Å [bright red subunit] , b ∼58 Å [light red subunit] ) . ( B ) Two separate CaMKII samples , one labeled with Alexa 488 ( green ) and the other labeled with Alexa 594 ( red ) , are mixed under various conditions and the FRET signal is measured , as defined by the ratio of the acceptor fluorescence to the donor fluorescence . ( C ) The extent of subunit exchange is monitored by FRET . At each time point , a fluorescence scan is taken with excitation at the donor wavelength . CaMKII is activated by Ca2+/CaM and ATP in one mixed sample ( blue squares ) while the other mixed sample is unactivated ( red circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 009 For the single-molecule experiments , wild-type CaMKII was expressed with a C-terminal Avitag ( i . e . , the tag is located after the hub domain ) ( Howarth and Ting , 2008 ) . The purified protein was biotinylated in vitro , and labeled separately with either the green or the red fluorophore . Mass spectrometric analysis showed that the labeling occurs at Cys 280 , in the C-terminal lobe of the kinase domain . Labeling efficiencies ranged between 20 and 40% for the wild-type protein ( see ‘Materials and methods’ ) , and this is sufficient for our analysis because of the direct visualization of holoenzymes . After mixing the two labeled species , a small amount of the CaMKII solution was removed at each time point , diluted , and then immobilized on the PEG-treated glass slides that were coated with streptavidin ( Figure 2A ) . Multi-color TIRF imaging was performed and the extent of colocalization was determined using a custom particle-tracking program ( see ‘Materials and methods’ ) . We score exchange events by treating all particles that have both red and green fluorophores equally . That is , our analysis does not distinguish between holoenzymes in which a single subunit has been exchanged and those in which more than one subunit has been exchanged . The positions of CaMKII holoenzymes were determined by fitting a two-dimensional Gaussian function to particle intensities ( see ‘Materials and methods’ ) . Importantly , the software filters out particles of insignificant brightness ( i . e . , noise ) and aggregates ( i . e . , unusually bright particles ) ( Figure 2—figure supplement 2 ) . On average , 6–20% of particles were eliminated from the analysis . Since samples from different time points are from the same reaction , the distribution of fluorescence intensities of each sample should not change over time in the absence of aggregation . Histograms of the fluorescence intensities were created for each image , after processing in the particle-tracking program , and monitored to check for any population of extra-bright particles ( most likely aggregates ) . The distribution of fluorescence intensities did not change significantly over the time course of the subunit exchange reaction , indicating that the colocalization we report is not due to aggregation , which would cause the distribution to shift towards values of greater intensity ( Figure 2—figure supplement 3 ) . This does not , however , rule out that aggregation or self-association of CaMKII holoenzymes may be necessary as an intermediate step for subunit exchange to occur , since such self-association might be reversed by dilution prior to the single-molecule analysis . We compared the results of mixing experiments in which CaMKII had either not been activated or had been activated by Ca2+/CaM and ATP . Protein samples were mixed and incubated together at 37°C for varying lengths of time in the absence of Ca2+/CaM and ATP and analyzed using the single-molecule assay . The level of colocalization detected under these conditions is very low , even at long times ( ∼5% at 240 min , Figure 2B ) . A dramatically different result is obtained when the samples are activated by incubation for 240 min at 25°C with Ca2+/CaM and ATP ( ∼40% colocalization in about 50 min , Figure 2B ) . The absolute levels of colocalization observed in our experiments are likely to be underestimates of the true degree of colocalization . Incomplete labeling , a population of fluorescent molecules in a dark state and errors in image processing would lower the total percentage of colocalization detected . The maximum extent of colocalization ( 62% ) was observed using a sample of CaMKII that has been labeled with both dyes simultaneously . We report the degree of colocalization in any particular experiment as a fraction of this maximal value . The values for colocalization are variable between experiments done on different days , due to labeling efficiencies and instrumental variability ( compare Figure 2B and C ) . All direct comparisons made in this paper are done using data from experiments performed on the same day . The extent of colocalization increases markedly with CaMKII concentration and with temperature , as shown in Figure 2B , C . For example , activated wild-type CaMKII reaches 40% of maximal colocalization within 6 min at 37°C , but at 25°C the time taken to achieve the same degree of colocalization is ∼250 min . If the concentration of CaMKII subunits is reduced from 8 μM to 1 μM , the extent of colocalization in 10 min at 37°C is reduced from ∼20% of maximal to ∼10% ( Figure 2C ) . The concentration of CaMKII subunits in dendritic spines is estimated to be ∼100 μM ( Otmakhov and Lisman , 2012 ) , suggesting that in this environment the timescale of subunit exchange may be substantially shorter than seen in our experiments . We have not studied subunit exchange in vitro at concentrations higher than 8 μM . We asked whether the activation-triggered colocalization involves a dead-end conversion in which holoenzymes that have undergone one cycle of exchange are inert to further exchange . To test this , we added saturating Ca2+/CaM and ATP to red-labeled and green-labeled forms of the holoenzyme and incubated these activated species separately for 15 min at 25°C . We then mixed the two samples and observed robust colocalization that is similar for samples that had not been incubated separately after activation ( Figure 2—figure supplement 4 ) . These data indicate that once CaMKII is activated , its subunits are capable of multiple cycles of exchange . The results of the single-molecule assay are corroborated by fluorescence resonance energy transfer ( FRET ) measurements in solution . The site of labeling in the wild-type protein , Cys 280 , is located at the periphery of the holoenzyme , and does not give a good FRET signal . We therefore labeled the hub domain at a site near the central hole in the hub assembly ( residue 335 ) , where any two such sites in the same hexameric ring would be close enough to generate a FRET signal . We replaced surface-exposed cysteines ( residues 280 and 289 ) in the kinase domain by serine and replaced Asp 335 in the hub domain by cysteine ( D335C ) . Labeling at this site should bring donor–acceptor pairs within ∼33 Å for adjacent subunits and ∼58 Å for non-adjacent subunits within a holoenzyme ( Figure 2—figure supplement 5A ) . The D335C mutant CaMKII was labeled with ∼80% efficiency . The FRET signal obtained for experiments using the D335C mutant showed a marked contrast for mixed samples that were either activated by Ca2+/CaM and ATP or not . There is a substantial increase in the FRET signal that occurs within minutes of mixing the two labeled species for the activated sample ( Figure 2—figure supplement 5B , C ) . In the unactivated sample ( i . e . , in the absence of Ca2+/CaM and ATP ) , there is only a modest , slow increase in the FRET signal with time , indicating that any exchange that might occur in the unactivated sample does so at a very slow rate . We interpret the results of the single-molecule experiments to mean that activation results in the weakening of inter-subunit contacts , facilitating exchange . To check whether this was happening , we sought to stabilize the holoenzyme assembly . To do this , we fused CaMKII to hemolysin-coregulated protein from Pseudomonas aeruginosa ( Hcp1 ) , a protein that forms hexameric rings with roughly the same diameter as the hub domain of CaMKII ( CaMKII-Hcp1 ) ( Mougous et al . , 2006 ) ( PDB code 1Y12 ) . The fusion protein was generated by linking the C-terminal end of the hub domain of CaMKII to the N-terminal end of Hcp1 by a 10-residue linker with a sequence that is designed to be flexible ( see ‘Materials and methods’ ) . The Avitag used to immobilize CaMKII to the glass slide was incorporated after the Hcp1 sequence . The kinase activity of the CaMKII-Hcp1 fusion was tested using a peptide substrate ( syntide ) and it displayed cooperative activation by Ca2+/CaM , with an activation profile similar to that of wild-type CaMKII ( Gaertner et al . , 2004; Rosenberg et al . , 2005 , data not shown ) . We carried out a mixing experiment using the CaMKII-Hcp1 fusion protein in which this construct was labeled separately with either red or green dye and the two samples were mixed and incubated at 37°C . Colocalization of the two fluorophores is only ∼10% even after 1 hr , compared to ∼70% for the wild-type holoenzyme ( Figure 3A ) . In an analogous experiment , we labeled wild-type CaMKII with the red fluorophore ( Alexa 594 ) and the CaMKII-Hcp1 fusion protein with the green fluorophore ( Alexa 488 ) and measured colocalization after activation ( Figure 3A ) . The level of colocalization is much below that observed with the wild-type protein in this case as well . 10 . 7554/eLife . 01610 . 010Figure 3 . Analysis of the exchange process . ( A ) Single-molecule experiments show that fusion of CaMKII to a hexameric protein ( Hcp1 ) slows the rate of colocalization . All samples are activated with Ca2+/CaM and ATP and mixing is done at 37°C . Mixing activated wild-type CaMKII yields about 70% of maximal colocalization ( blue ) . Mixing wild-type and CaMKII-Hcp1 shows a marked decrease in colocalization ( red ) . Mixing CaMKII-Hcp1 species results in nearly no colocalization ( green ) . ( B ) The isolated hub assembly does not result in colocalization when labeled subunits are mixed . ( C ) Deletion of the variable linker region does not affect colocalization significantly . Comparison of the short-linker construct ( red ) , short-linker construct mutated at the hub–kinase interface ( green ) , and wild-type CaMKII ( blue ) shows minimal differences in colocalization . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 010 The strong suppression of colocalization seen with the CaMKII-Hcp1 fusion protein lends further support to the idea that CaMKII holoenzymes exchange subunits upon activation . Since fusion of Hcp1 to the hub domain is unlikely to impede the separation of a holoenzyme into two hexameric rings , these data also suggest that exchange involves some other disassembly process . The fact that activation leads to subunit exchange in the intact holoenzyme made us wonder whether the hub domain assembly might be intrinsically unstable , and that the release of stabilizing contacts between the kinase domains and the hub upon activation might allow the subunits of the hub domain to separate and exchange . To test whether the hub domain is intrinsically capable of subunit exchange , we purified the hub domain and monitored colocalization . We found that the subunits of the isolated hub domain assembly do not exchange subunits ( Figure 3B ) . These data suggest that some combination of the kinase domain , the regulatory segment or the linker connecting the regulatory segment to the hub domain must be required for the exchange process . To examine the role of the linker we carried out single-molecule experiments using a construct of CaMKIIα in which the linker is eliminated entirely . This short-linker construct is similar to the construct used to obtain the crystal structure of CaMKIIα ( Chao et al . , 2011 ) . As shown in Figure 3C , the short-linker construct exhibits fluorophores colocalization with the same rate as full-length CaMKIIα when activated by Ca2+/CaM and ATP , indicating that the linker is not required for subunit exchange . We also wondered whether the release of interactions between the kinase domain and the hub might be the trigger for subunit exchange . In the crystal structure of the autoinhibited short-linker CaMKII holoenzyme , the kinase domains dock against the hub domains ( Chao et al . , 2011 ) . Mutation of lIe 321 in the hub domain to glutamate disturbs this docking and results in an opening of the holoenzyme assembly ( Chao et al . , 2011 ) . Introduction of the same mutation ( I321E ) in the context of the short-linker construct has no effect on the rate of colocalization ( Figure 3C ) . This suggests that the trigger for subunit exchange does not involve the disruption of the interface between the kinase domain and the hub that is seen in the structure of the autoinhibited holoenzyme . The experiments presented so far relied on activation by Ca2+/CaM and ATP to trigger exchange . Since the regulatory segment binds to Ca2+/CaM , it is difficult to introduce mutations into this segment without disturbing the interaction with calmodulin . To identify the role played by Ca2+/CaM in the exchange process , we used the constitutively activated form of the enzyme ( CaMKIIT286D ) in which the regulatory segment is expected to be displaced from the kinase domain even in the absence of Ca2+/CaM . CaMKIIT286D displays a vigorous degree of colocalization when ATP is added in the absence of Ca2+/CaM , comparable to that of the wild-type enzyme in the presence of Ca2+/CaM and ATP ( Figure 4A ) . These data demonstrate that although activation of the catalytic module is necessary for colocalization , the physical presence of Ca2+/CaM is not required . 10 . 7554/eLife . 01610 . 011Figure 4 . Phosphorylation of the calmodulin-recognition element is crucial for exchange . ( A ) Mixing CaMKIIT286D in the absence of Ca2+/CaM results in robust colocalization ( red ) . Mutating the CaM-recognition element ( R3 ) reduces colocalization significantly ( blue ) . ( B ) Both mixing experiments shown use wild-type CaMKII in the presence of Ca2+/CaM . The addition of a kinase inhibitor and 1 µM ATP significantly reduces colocalization ( green ) compared to a condition with full kinase activity ( 250 µM ATP , no kinase inhibitor ) ( red ) . ( C ) All species are mixed with CaMKIIT286D in the absence of Ca2+/CaM . Compared to the colocalization resulting from mixing CaMKIIT286D ( red ) , mixing CaMKIIT286D in the presence of a kinase inhibitor results in reduced colocalization ( blue ) . Replacement of either Thr 305 or Thr 306 by alanine also results in a reduction in colocalization ( pink and green , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 01110 . 7554/eLife . 01610 . 012Figure 4—figure supplement 1 . Kinase activity is crucial for exchange . Solution FRET experiments show that the addition of the kinase inhibitor , bosutinib , in the presence of Ca2+/CaM with no ATP significantly reduces the FRET signal ( green squares ) compared to activated CaMKII ( red circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 012 With the elimination of a direct role for Ca2+/CaM in the exchange process our attention was now focused on the kinase domain and the regulatory segment . We were particularly interested in the part of the regulatory segment that spans the calmodulin-recognition motif ( the R3 element ) , because this segment interacts with the hub domain in the crystal structure of the autoinhibited CaMKII holoenzyme ( Chao et al . , 2011 ) . Starting with the constitutively active variant ( CaMKIIT286D ) , we replaced residues 302 to 313 in the R3 element by an arbitrary linker sequence that is not expected to form regular secondary structure . Specifically , the sequence 302A I L T T M L A T R N F313 in the R3 element was replaced by 302S T G G G T G S T G S T313 . We monitored colocalization after mixing for this construct and found that colocalization was essentially eliminated ( Figure 4A ) . Given that the calmodulin-recognition element plays a crucial role in subunit exchange , how might activation control the ability of this segment to mediate exchange ? Activation by Ca2+/CaM releases the R3 element from the kinase domain , but it remains hidden within calmodulin . Thr 286 is , however , outside the footprint of calmodulin , and it is phosphorylated upon Ca2+/CaM binding to the holoenzyme . The release of Ca2+/CaM from a subunit after Thr 286 is phosphorylated facilitates phosphorylation at Thr 305 and Thr 306 , which prevents Ca2+/CaM rebinding ( Hashimoto et al . , 1987; Hanson and Schulman , 1992; Colbran , 1993 ) . Displacement of Ca2+/CaM and phosphorylation of the R3 element on Thr 305 and/or Thr 306 may therefore facilitate subunit exchange . Using this reasoning , the kinase activity of CaMKII is expected to be crucial for the exchange process . We carried out two mixing experiments using wild-type CaMKII and Ca2+/CaM in parallel ( Figure 4B ) . In one experiment , as described previously , the fluorescently labeled samples were incubated separately with ATP and Ca2+/CaM , then mixed together and monitored for CaMKII colocalization using the single-molecule assay . In the other experiment , the ATP concentration was lowered to 1 μM , which is below the value of the KM for ATP ( Colbran , 1993 ) . In addition , the kinase inhibitor bosutinib was added during the incubation phase . Although bosutinib was developed as an inhibitor of Abl tyrosine kinase , it has off-target activity against CaMKII ( Remsing Rix , 2009 ) and was used in the crystallization of autoinhibited CaMKII ( Chao et al . , 2011 ) . The fluorescently labeled samples were then mixed , and monitored for colocalization , as before . As shown in Figure 4B , the reduction in ATP concentration and the presence of bosutinib result in suppression of colocalization . For example , after 30 min of mixing , the wild-type enzyme exhibits 70% of the maximal degree of colocalization , whereas under conditions where phosphorylation is inhibited the level of colocalization is only ∼20% after 30 min . We confirmed the importance of kinase activity by using the FRET assay ( Figure 4—figure supplement 1 ) . In this case , we added Ca2+/CaM and the kinase inhibitor in the absence of ATP , and found that there is no FRET signal under these conditions . We compared the colocalization of the constitutively active variant CaMKIIT286D in the presence and absence of the kinase inhibitor and low ATP . As shown in Figure 4C , inhibition of kinase activity suppresses colocalization for CaMKIIT286D , although to a lesser extent than for the wild-type protein . These data indicate that phosphorylation at sites other than Thr 286 is also important for exchange , and drew attention to a possible role for Thr 305 and Thr 306 in the exchange process . We replaced these residues with alanine separately to generate two constructs ( CaMKIIT286D , T305A and CaMKIIT286D , T306A ) . Attempts to purify a construct in which both residues were mutated to alanine were not successful because this variant did not express well . Mixing experiments demonstrated substantial reduction in colocalization for both individual mutations , although somewhat less than the reduction obtained with treatment of CaMKIIT286D with bosutinib ( Figure 4C ) . We conclude that phosphorylation of both Thr 305 and Thr 306 , in addition to Thr 286 , is important for subunit exchange . An important question is whether subunits that are activated can exchange with holoenzymes that have not been activated , and further , whether such exchange could result in the phosphorylation of Thr 286 in the unactivated subunits . To study this , we used mixtures of wild-type CaMKII and the constitutively active CaMKIIT286D variant in mixing experiments . Because CaMKIIT286D lacks Thr 286 , this allows us to detect the phosphorylation of Thr 286 in the unactivated holoenzymes by using an antibody that is specific for the phosphorylated form of Thr 286 ( pT286 , see ‘Materials and methods’ ) . The antibody is labeled with Alexa 647 , and it does not detect CaMKIIT286D ( Figure 5—figure supplement 1 ) . The first important result is that subunits in constitutively active CaMKIIT286D can exchange with subunits from holoenzymes that have not been activated , when ATP is added ( Figure 5A ) . Fusion of the hexameric protein Hcp1 to the wild-type protein reduces the rate of colocalization , indicating that the unactivated CaMKII holoenzyme has to open up in some way for the exchange process to happen . 10 . 7554/eLife . 01610 . 013Figure 5 . Evidence for the spread of phosphorylation into unactivated holoenzymes . ( A ) Both mixing experiments shown use wild-type CaMKII in the presence of Ca2+/CaM and ATP . CaMKIIT286D mixed with unactivated wild-type CaMKII results in high colocalization ( red ) . This colocalization is suppressed by the addition of the Hcp1 module ( CaMKII-Hcp1; blue ) . ( B ) Levels of pThr286 labeling in each mixing experiment from ( A ) . The pThr286 antibody is modified with Alexa 647 , which is then added to the mixed samples . Subsequent analysis is for 3-color colocalization between Alexa dyes 488 , 594 , and 647 . The phosphorylation spreads significantly more in the CaMKIIT286D sample ( red ) compared to the sample mixed with CaMKIIT286D-Hcp1 ( blue ) . ( C ) There is colocalization of the pThr286 antibody with particles that have both Alexa 488 and 594 , indicating that the antibody is only binding to those CaMKII holoenzymes that have already exchanged subunits . The graph shows the fraction of antibody label that is colocalized to particles that contain both the red and green fluorophores . Note that this fraction is close to 100% . ( D ) Kinase activity against a peptide substrate ( syntide ) was monitored in solution using the ADP Quest assay , where the fluorescence of resorufin is an indicator of ATP consumption . The reaction rates are plotted for three separate samples . First , 3 µl of an unactivated CaMKII sample , then 1 µl of an activated CaMKII sample ( both are at the same final protein concentration ) , and finally a mixture of these components . The value of the reaction rate is indicated above each bar . It is clear that the reaction rate in the mixture is higher than just the addition of the rates of the individual components . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 01310 . 7554/eLife . 01610 . 014Figure 5—figure supplement 1 . Controls for the pThr286 antibody . Labeling activated wild-type CaMKII ( in which Thr 286 is expected to be phosphorylated ) with the pThr286 antibody results in significant colocalization of the antibody with CaMKII ( ∼78% ) . Labeling CaMKIIT286D with the antibody yields very little signal corresponding to the antibody label , within the noise of the experiment ( <4% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 014 We also examined the ability of the activated subunits to phosphorylate subunits from holoenzymes that had not been activated . To do this we devised a three-color single-molecule experiment , in which the Alexa 488 and Alexa 594 channels monitor the locations of the labeled CaMKII variants , and the Alexa 647 channel detects phosphorylated Thr 286 in wild-type CaMKII . To determine the extent of antibody labeling that can be detected in a fully activated sample of CaMKII , we first activated CaMKII by Ca2+/CaM and ATP for 10 min . This is expected to result in robust phosphorylation of each holoenzyme assembly because of the direct activation of each subunit by Ca2+/CaM . This resulted in ∼75% colocalization of fluorescent signals from CaMKII and the pT286 antibody ( Figure 5—figure supplement 1 ) . We then studied the extent of Thr 286 phosphorylation that is detected when the constitutively active CaMKIIT286D is mixed with unactivated wild-type CaMKII . As shown in Figure 5B , there is an increase in the degree of phosphorylation detected with time , demonstrating that CaMKIIT286D is able to phosphorylate subunits in unactivated wild-type CaMKII . It has been shown previously that trans-phosphorylation of Thr 286 occurs principally within a holoenzyme ( Hanson et al . , 1994; Rich and Schulman , 1998 ) . In order to check that the observed phosphorylation is due to the simultaneous presence of activated ( CaMKIIT286D ) and unactivated ( wild type ) subunits within the same holoenzyme , we carried out two analyses . First , we employed the Hcp1 fusion construct , for which subunit exchange was suppressed ( Figure 3A ) . Mixing CaMKIIT286D-Hcp1 with wild-type CaMKII resulted in decreased Thr 286 phosphorylation ( Figure 5B ) , indicating that stabilizing the activated CaMKII holoenzyme reduces the observed phosphorylation . In the second analysis , we showed that antibody binding occurs principally in those species in which the red and green fluorophores are colocalized ( Figure 5C ) , consistent with subunit exchange being required for the trans-phosphorylation reaction . We note that the signal from pT286 rises much more slowly than the rate of colocalization . Colocalization reaches a plateau value of ∼60% colocalization in ∼50 min , whereas the level of three-color colocalization at a comparable time point is only ∼5% . The replacement of Thr 286 by aspartate in CaMKIIT286D may not be a good mimic of the phosphorylated form of Thr 286 in terms of its ability to promote phosphorylation . We found that the activity of CaMKIIT286D towards a peptide substrate ( syntide , see ‘Materials and methods’ ) is only ∼20% that of activated wild-type CaMKII ( data not shown ) . Also , the ability of CaMKIIT286D to transphosphorylate an adjacent subunit within an unactivated holoenzyme may be inherently slow without calmodulin bound to the subunit being phosphorylated ( Rich and Schulman , 1998 ) . The activation of CaMKII is highly cooperative ( Gaertner et al . , 2004; Rosenberg et al . , 2005 ) , so low levels of Ca2+/CaM that are insufficient for autophosphorylation of an unactivated holoenzyme may be sufficient for binding the target subunit and enhancing its autophosphorylation in holoenzymes that contain some activated subunits through exchange . To get a better sense of the extent to which activated subunits can potentiate the activity of holoenzymes that have not been activated , we turned to a solution assay using two different samples of wild-type CaMKII , without fluorescent labeling . One sample , the activated sample , was prepared by incubation with ATP and Ca2+/CaM for 15 min , followed by removal of Ca2+/CaM ( see ‘Materials and methods’ ) . The second sample , the unactivated sample , was not exposed to Ca2+/CaM . We then measured the activity of these samples using a peptide substrate ( syntide ) and a continuous enzyme-coupled assay ( ADP Quest , see ‘Materials and methods’ ) . This assay relies on the conversion of ADP to a fluorescent product , resorufin , the accumulation of which is monitored over time ( Charter et al . , 2006 ) . A sample containing 1 μl of activated CaMKII at 6 μM concentration exhibits a peptide phosphorylation rate of 132 units , where the units are the change in the fluorescence of resorufin in unit time ( Figure 5D ) . A second sample containing 3 μl of unactivated CaMKII at the same concentration exhibits a rate of 49 units ( this activity represents the futile ATP hydrolysis rate of CaMKII , also observed without peptide substrate , data not shown ) . A third sample consisting of 1 μl of activated CaMKII at 6 μM mixed with 3 μl of unactivated CaMKII at 6 μM exhibits a rate of 592 units . In the absence of any crosstalk between the unactivated and activated holoenzymes , the reaction rate exhibited by the mixture is expected to be the sum of the reaction rates of the first two samples , that is , 181 units . The actual observed rate , 592 units , is about three times as high . This suggests that activated CaMKII holoenzymes have increased the activity of the holoenzymes that had not been activated previously . One difficulty in developing a conceptual model for the exchange process in CaMKII is that there appears to be little precedent in the biochemical literature , as far as we are aware , for oligomeric enzymes that spread their state of activation by exchanging subunits . We envisage two possible kinds of mechanisms for the exchange of subunits between CaMKII holoenzymes . The first mechanism is suggested by the unusual allosteric activation of porphobilinogen synthase ( Selwood and Jaffe , 2012 ) . This enzyme converts between an inactive hexamer and an active octamer , with the transition requiring dissociation of the assembly into dimers ( Selwood et al . , 2008 ) . We have developed a speculative structural model for how activation might promote the release of subunits , most likely dimers , from a CaMKII holoenzyme , discussed below . An alternative mechanism involves the transient formation of higher order aggregates of two or more holoenzymes , with exchange occurring within the aggregate followed by separation of holoenzymes with mixed subunits . The data available to us at present provide no clues as to how such a process might occur , so we do not discuss this further although it remains an important alternative mechanism for future investigation . Regardless of whether subunit exchange proceeds through the release of subunits or through transient aggregation , it seems likely that the regulatory segment is able to destabilize the hub assembly when this segment is released from the kinase domain . We developed a model for how the regulatory segment might weaken the hub assembly by analyzing molecular dynamics simulations of CaMKII ( see ‘Appendix’ ) , and by considering several characteristic features of the structure of the holoenzyme and the exchange process . These include ( i ) the ability of the hub assembly to interconvert between dodecameric and tetradecameric forms , ( ii ) the importance of phosphorylation of the regulatory segment , ( iii ) the presence within the hub domain of a deep cavity containing three conserved arginine residues and ( iv ) the known ability of the hub domain to serve as a docking site for peptide segments by virtue of its exposing an uncapped β sheet , with no steric hindrance to extending the sheet by an additional strand provided by the regulatory segment . For ease of discussion , the details of the molecular dynamics simulations and the reasoning that went into the development of this model are presented in an ‘Appendix’ that follows immediately after the main body of the text . The main features of the model are summarized schematically in Figure 6 . Note that the schematic in Figure 6 emphasizes the release of dimers , but we are uncertain as to whether the actual process involves a transient aggregation step that we do not yet understand . 10 . 7554/eLife . 01610 . 015Figure 6 . Schematic of a potential mechanism for subunit exchange in CaMKII . ( A ) In the unactivated state of CaMKII , the dodecameric hub domain undergoes fluctuations , which leads to transient lateral openings between vertical dimeric units . ( B ) Upon Ca2+/CaM binding , Thr 286 is trans-phosphorylated . For simplicity , just one kinase domain is depicted . When calcium levels drop , CaM falls off and Thr 305 and Thr 306 are subsequently phosphorylated . The now-released regulatory segment is free to bind the open β sheet of its own hub domain . This brings residues 305/306 in close proximity to the hub cavity ( blue triangle ) , which houses the three conserved Arg residues . Binding of the regulatory segment induces a crack to open in the hub domain , which exposes the Arg residues to the phosphate groups . We reason that phosphorylation of the regulatory segment leads to an interaction between the R3 element and the hub domain that weakens the lateral association between hub domain dimers , leading to their release from one holoenzyme . ( C ) Fluctuations in the autoinhibited holoenzyme create a hub assembly that resembles a ‘C’ shaped structure . This fluctuation may allow the capture of a vertical dimer that has been released from an active holoenzyme . The drawing depicts the adoption of a tetradecameric structure upon docking of an incoming vertical dimer , which may be an intermediate in the exchange process . As discussed in the main text , we have no direct experimental evidence at present concerning the exchange process . We cannot , therefore , rule out alternative mechanisms , such as those involving a transient aggregation of holoenzymes prior to exchange . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 015 The CaMKII holoenzyme can be thought of as being assembled from six ‘vertical dimers’ . In a view that has the mid-plane of the dodecamer horizontal , these vertical dimers each contribute one hub domain to the upper ring and one to the lower ring of the hub assembly . We propose that the hub assembly within a holoenzyme , whether activated or not , normally undergoes fluctuations that convert it from the closed double-ring form seen in crystal structures to an open C-shaped form with a gap between two adjacent vertical dimers . Molecular dynamics simulations suggest that this gap can easily be large enough to capture a vertical dimer provided by another holoenzyme . This could , for example , convert a dodecameric assembly to a tetradecameric one . In the absence of activation , however , the transient openings in the hub assembly of the holoenzyme simply reanneal , due to the stability of the lateral interactions between vertical dimers . The situation changes , however , when CaMKII is activated and Thr 305 and Thr 306 are phosphorylated . We propose that the phosphorylated regulatory segment , which is freed from interaction with the kinase domain and Ca2+/CaM , is now able to dock on the hub domain and form an additional strand of the β sheet that is the core scaffold of the hub domain . We speculate that when this happens , one or more of the phosphorylated sidechains in the regulatory segment reach into the interior of the hub and interact with the conserved arginine residues located in the hub cavity . This interaction might distort the structure of the hub domain , slightly weakening the lateral interactions in the hub assembly . A fluctuation in the hub assembly can now result in the release of an activated vertical dimer . A released vertical dimer can either be recaptured or , if it encounters an unactivated holoenzyme that is open transiently , it can be incorporated into that holoenzyme . While we believe that this model provides a reasonable framework for thinking about the exchange process , the schematic shown in Figure 6 is far from definitive . For example , close encounters between two holoenzymes may be necessary for an activated dimer to ‘hop’ directly from one holoenzyme to another , without being ever released completely . In such a situation , the interaction between the phosphorylated regulatory segment and the hub domain may occur in trans , between two holoenzymes , rather than within one holoenzyme . We anticipate that future studies will establish whether the model we are proposing is correct in essence , or whether the subunit exchange mechanism relies on features that we have failed to anticipate . The strengthening of synaptic connections between neurons that occurs following brief repetitive stimuli that induce LTP is likely to be important for the formation and maintenance of cognitive memory . Much attention has been focused on the role of CaMKII in this process , particularly its ability to maintain an activated state , through intra-holoenzyme phosphorylation of Thr 286 , after cessation of LTP-inducing stimuli . But can the active state be maintained beyond the lifetime of the kinase molecules present during the initial stimulus ? Lisman introduced the idea that information about previous activating events could be transmitted to newly synthesized CaMKII molecules if the subunits of the activated holoenzyme could exchange with unactivated ones ( Lisman , 1994 ) . We now demonstrate that the CaMKII holoenzyme has precisely this property when studied in vitro . Activated CaMKII holoenzymes can indeed instruct previously unactivated holoenzymes as to their phosphorylation state through subunit exchange , and this ability is switched on by activation and autophosphorylation . In the time since Lisman’s earlier speculation about subunit exchange , there has been no experimental precedent that has pointed to such a mechanism being operative for CaMKII . Translocation of CaMKII to the NMDA receptor requires active subunits and has been postulated to serve as a molecular memory ( Bayer et al . , 2001; Jiao et al . , 2011; Lemieux et al . , 2012; Neant-Fery et al . , 2012 ) . This population of CaMKII bound to the NMDA receptor may maintain its activity for longer times ( Otmakhov et al . , 2004 ) . Subunit exchange could facilitate the recruitment of additional holoenzymes to the receptor by spreading activation to unactivated holoenzymes . Recent experiments using a CaMKII construct engineered to be a FRET reporter of its activation state indicate that the bulk of CaMKII holoenzymes in dendrites maintain their activation state for not much longer than a minute after the stimulation is withdrawn ( Takao et al . , 2005 ) . Interestingly , another study demonstrated that localized stimulation of a neuron resulted in the translocation of CaMKII from the main body of the axon to synapses throughout the dendritic arbor for much longer times ( 15–40 min ) after the local stimulation is stopped; the authors of this report commented that subunit exchange may play a role in this process ( Rose et al . , 2009 ) . Given the high concentrations of CaMKII in synapses and the fact that calmodulin is limiting ( Liu and Storm , 1990; MacNicol and Schulman , 1992; Persechini and Stemmer , 2002 ) , subunit exchange provides a potential mechanism for increasing the spread of activation caused by an initial activating pulse . Now that we have demonstrated unambiguously that activation triggers the exchange of subunits in CaMKII in vitro , future experiments will address several important questions . We have focused on human CaMKIIα in this study . Is subunit exchange common to the other isoforms of the human enzyme ? Did the property of subunit exchange arise early in evolution , or did it evolve with the specialization of CaMKII to neuronal function ? Finally , and most importantly , how has nature exploited subunit exchange in activated CaMKII to determine the actual outcomes of neuronal signaling as well as in heart and other systems ? It will be challenging , but ultimately most informative , to devise experiments to address this last question . The CaMKII dodecamer can be described in terms of the lateral association of six vertical dimeric units , as shown in Figure 7 . We imagine that a critical step in the exchange process is a lateral opening of the ring formed by the hub domains ( Figure 7A , B ) . The exchange process might involve the release of vertical dimers . Such a vertical dimer could not be released easily from the Hcp1 fusion construct and indeed subunit exchange is suppressed between these variants ( Figure 3A ) . The release and capture of vertical dimers also provides a route for the interconversion of dodecameric and tetradecameric holoenzymes without requiring a more complete disassembly . As noted in the main text , we are uncertain about whether the transient aggregation of holoenzymes might be necessary for exchange to occur . In this Appendix , we focus on potential interactions between the regulatory segment and a single hub domain assembly . 10 . 7554/eLife . 01610 . 016Figure 7 . A vertical dimeric unit of the CaMKII assembly may be the unit of exchange . ( A ) The CaMKII hub assembly can be described as a set of six vertical dimers , and each dimer is labeled as A/B; C/D; etc . One of these dimers ( A/B ) is highlighted in blue/magenta ( black dashed line ) . The lateral interfaces and equatorial interface for this dimer are indicated by orange and black lines , respectively . ( B ) A schematic diagram that indicates how one vertical dimer may be released from the holoenzyme . ( C ) The structure of the vertical hub dimer from CaMKII is shown in comparison to a dimer of NTF2 ( PDB codes: 2UX0 and 1OUN , respectively ) . The notation for the secondary structural elements of the CaMKII hub domain are shown . ( D ) A molecular dynamics simulation was started from the dodecameric crystal structure of the hub domain ( PDB code: 2UX0 ) . The starting crystal structure is overlaid with an instantaneous structure from the molecular dynamics trajectory at 100 ns , aligning onto just one subunit ( indicated on the schematic ) . It is clear that the vertical dimeric unit is relatively stable ( blue/magenta ) , but there is a significant change in the relative positioning of the blue/magenta dimer with respect to the yellow/green dimer . This indicates that the lateral interfaces are more dynamic than the equatorial interfaces . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 01610 . 7554/eLife . 01610 . 017Figure 7—figure supplement 1 . Molecular dynamics simulations suggest that the contacts across equatorial interfaces are stronger than those across the lateral interfaces . Molecular dynamics simulations provide a clear indication that the lateral interfaces in the hub assembly are more likely to be disrupted than the equatorial ones . We generated a 100 nanosecond ( ns ) trajectory of the hub domain of human CaMKIIγ ( PDB code: 2UX0 ) . We aligned each instantaneous structure from the trajectory onto the initial structure by using each subunit , one at a time , and plotted the root mean square deviations in the positions of the Cα atoms in the two subunits across the lateral interfaces ( red , blue ) and the one subunit across the equatorial interface ( green ) . For comparisons across the equatorial interfaces , the rms displacements in the neighboring subunits are ∼2 Å across the ring . In contrast , for comparison across the lateral interfaces , the rms displacements of the neighboring subunits have shifted by as much as 6 to 10 Å for several of the interfaces , consistent with the idea that the lateral interfaces are more flexible than the equatorial ones . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 017 That the unit of exchange might be a vertical dimer is suggested by the fact that the hub domain is very closely related in structure , although not in sequence , to members of a large family of enzymes and binding proteins , such as nuclear transport factor 2 ( NTF2 ) and ketosteroid isomerase like proteins , which we refer to as the NTF2 family ( Figure 7C; Hoelz et al . , 2003 ) . A common quaternary structural unit within the NTF2 family corresponds to the vertical dimer in the CaMKII hub . Of the top 50 unique hits in a Dali search ( Holm and Rosenstrom , 2010 ) for structures similar to the mouse CaMKIIα hub domain ( PDB code: 1HKX ) , 29 form dimers in the crystal lattice that are similar to the vertical CaMKII hub dimer ( Figure 7C ) . Molecular dynamics simulations provide a clear indication that the lateral interfaces in the hub assembly are more likely to be disrupted than the equatorial ones ( Figure 7—figure supplement 1 ) . The hub domain of human CaMKIIγ has been crystallized as a dodecameric assembly ( Rellos et al . , 2010 ) ( PDB code: 2UX0 ) , and we generated a 100 nanosecond ( ns ) trajectory of this assembly . The internal motions of the hub can be described as rigid-body motions of individual vertical-dimer units , with alterations in the lateral dimer–dimer interfaces . This is demonstrated by aligning each instantaneous structure from the trajectory on each subunit in turn , and graphing the root mean square deviations in the positions of the Cα atoms in the two subunits across the lateral interfaces and the one subunit across the equatorial interface ( Figure 7D , Figure 7—figure supplement 1 ) . For comparisons across the equatorial interfaces , within a vertical dimer , the rms displacements in the neighboring subunits are ∼2 Å across the ring . In contrast , for comparison across the lateral interfaces , the rms displacements of the neighboring subunits are as much as 6 to 10 Å for several of the interfaces , consistent with the lateral interfaces being much more flexible than the equatorial ones ( Figure 7—figure supplement 1 ) . Further analysis of the simulation suggests that there is strain associated with the closed ring formed by the dodecameric hub assembly of CaMKII . Although the hub assembly is symmetric initially , the sixfold symmetry of the ring breaks down during the course of the simulation . This is due to the tightening of some of the lateral interfaces , for which the buried surface area increases . For the CaMKIIγ structure used in the simulation , the closer packing involves the sidechains of Phe 364 , Phe 367 , Tyr 368 , Asn 371 , Leu 372 on helix αD in one subunit and Gln 406 , Pro 414 , Thr 416 and Ile 418 ( Gln 418 in the α isoform ) from strands β4 and β5 , as well as Pro 379 in strand β3 in the other subunit ( see Figure 7C for the notation ) . With the exception of Ile 418 , these residues are all conserved in CaMKIIα . The tighter packing at some of the interfaces in the simulation of the dodecameric hub assembly is coupled to looser packing at some of the other interfaces . At one of the interfaces , the residues on helix αD in one subunit and the β4 and β5 strands on the other are splayed apart so that the sidechains of Phe 364 and Phe 367 , which are buried in the more stable interfaces , are now partially solvent exposed ( Figure 8 ) . These results suggest that the constraint of ring closure prevents the simultaneous optimization of all of the lateral interfaces . 10 . 7554/eLife . 01610 . 018Figure 8 . Strain associated with the closed ring formed by the dodecameric hub assembly of CaMKII . During the simulation of the CaMKIIγ dodecamer , the sixfold symmetry of the hub assembly breaks down due to the tightening of some of the lateral interfaces and loosening at others . ( A ) A view of one of the lateral interfaces , with a close-up view in ( B ) . At this interface , the residues on helix αD in one subunit and the β4 and β5 strands on the other are splayed apart after 100 ns of simulation , so that the sidechains of Phe 364 and Phe 367 , which are buried in the crystal structure ( left ) and more stable interfaces , are now partially solvent exposed ( right ) . These results suggest that the constraint of ring closure prevents the simultaneous optimization of all of the lateral interfaces . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 018 To further explore the extent to which the closed-ring form of the dodecamer is strained , we removed one vertical dimer from the crystal structure of the dodecamer , generating a C-shaped decameric model ( Figure 9A ) . This allows us to use molecular dynamics to study the effect of relaxing the ring-closure constraint , following a strategy that was used effectively to analyze strain in circular sliding DNA clamps ( Jeruzalmi et al . , 2001; Kazmirski et al . , 2005 ) . 10 . 7554/eLife . 01610 . 019Figure 9 . Molecular dynamics simulations of an open-ringed ( decameric ) hub assembly . ( A ) The dodecamer consists of six vertical dimers , denoted A:B , C:D…K:L . The decamer used in the simulations is created by removing the A:B dimer . ( B ) We initiated two independent molecular dynamics trajectories from this decameric structure and the results for one are shown in this diagram . Instantaneous structures from this simulation are shown overlaid with the crystal structure for either the dodecameric ( PDB code: 2UX0 ) or tetradecameric ( PDB code: 1HKX ) hub assembly . At 29 ns , it is clear that the decamer has relaxed to the tetradecameric conformation , with further opening evident at 91 ns . ( C ) There are two internal vertical interfaces in the decamer , between the E:F and G:H vertical dimers and between the G:H and I:J vertical dimers ( colored in the structural diagram shown at the top ) . To demonstrate the convergence of the two vertical interfaces to an arrangement that is distinct from the interfaces in the crystal structure , we calculated the displacement of atomic positions in interfacial subunits after the E:F/G:H and G:H/I:J interfaces were brought into spatial alignment using only one subunit , for a series of instantaneous structures extracted from the trajectories . To do this , we made a copy of the trajectory . The subunits in the original are labeled C through K , and in the copy they are labeled C′ through K′ . The two internal interfaces ( E:F/G:H and G:H/I:J ) are brought into alignment by superimposing subunit E from the original onto subunit G′ of the copy of the trajectory , for pairs of structures at the same point in the trajectory . The overlaid structures are shown at the bottom left . The close overlap shows that the two vertical interfaces in this instantaneous structure are similar . We then aligned subunit E of the instantaneous structure with subunit G of the crystal structure ( bottom right ) . The poor overlap between the crystal structure and the instantaneous structure is evident . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 01910 . 7554/eLife . 01610 . 020Figure 9—figure supplement 1 . Relieving strain in the dodecameric ring tends towards a tetradecameric conformation . ( A ) A schematic is shown for the calculation of the angle of opening within hub assemblies . This inter-subunit angle was calculated over the course of the trajectory for the decameric CaMKII as defined by the center of mass of one subunit at the apex and between the centers of mass of two subunits across the ring . ( B ) The value of the inter-subunit angle is graphed over the time course of the trajectory . In a short amount of time ( ∼10 ns ) , the value of the inter-subunit angle for the decameric CaMKII changes from that corresponding to a dodecamer to a value closer to that for a tetradecamer , and does not extend significantly beyond this point . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 02010 . 7554/eLife . 01610 . 021Figure 9—figure supplement 2 . Structural changes during molecular dynamics , showing the relaxation of the decamer to a specific interfacial arrangement . ( A ) Two vertical interfaces within one instantaneous structure were aligned as described in the legend for Figure 9C . After the alignment , the rms displacement in Cα positions between the E subunit in the original and the G′ subunit in the copy is calculated , and the process is repeated for each time point along the trajectory ( top ) . Relative convergence of orientations is evident from the low deviations in Cα positions between the E and G′ subunits ( 2–3 Å ) towards the end of the simulation . We also calculated the deviations in Cα positions across two replicate trajectories for pairs of structures at the same time point , revealing a similar trend ( bottom ) . ( B ) When the trajectory is compared in the same way to the crystal structure of the dodecameric hub assembly , the rms displacements in Cα positions are calculated between the E subunit from the trajectory and G subunit from the crystal structure . These values are significantly higher ( ∼6 Å ) towards the end of the trajectory . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 021 This decameric structure , derived from the crystal structure of the dodecameric CaMKIIγ hub assembly , was used to initiate two independent molecular dynamics trajectories ( 100 ns and 50 ns , respectively ) . There is a rapid relaxation of the curvature of the ring in both trajectories . The decameric structure opens up with respect to the starting structure in both trajectories within ∼10 ns ( Figure 9B ) . The principal outward displacement occurs in the plane of the ring so that the structures move closer to the tetradecameric form of the hub assembly ( PDB code: 1HKX; Figure 9B , Figure 9—figure supplement 1 ) . The constraint of ring closure undoubtedly imposes an entropic strain on the ring , and removal of the constraint is expected to yield a more floppy and open structure . What is notable , however , is that in both molecular dynamics trajectories we observe relaxation to a specific interfacial arrangement that is different from the interfacial arrangement seen in the crystal structure of the dodecamer . There are two internal vertical interfaces in the decamer , between the G:H and E:F vertical dimers and between the G:H and I:J vertical dimers . The relative orientations of subunits across these two interfaces converge to a similar arrangement , with low rms displacements , both within one trajectory and between two independent trajectories ( Figure 9C , bottom left , Figure 9—figure supplement 2A ) , and this arrangement is different from that seen in the crystal structure of the dodecamer ( Figure 9C , bottom right , Figure 9—figure supplement 2B ) . This relaxation towards a specific structure , as seen in two independent trajectories , suggests that there is an enthalpic contribution to the strain in the closed ring . It is difficult , however , to assign this strain to any particular sets of interaction . The structural relaxation preserves the set of residues that interact at each interface , and PISA analysis ( Krissinel and Henrick , 2007 ) of instantaneous structures extracted from the trajectories does not reveal any systematic trend in the estimated interfacial free energy . The experimental data suggest that the release of the regulatory segment from the body of the kinase domain upon activation allows the regulatory segment and , possibly the kinase domain , to interact with the hub domain in a way that weakens the inter-subunit interfaces in the hub assembly . Given that the calmodulin-binding R3 element is particularly important for exchange , we looked for plausible ways in which the phosphorylated R3 element could interact with the hub domain . There is one very obvious binding site in the hub domain for negatively charged groups , and that is the deep cavity in the hub domain , which corresponds to the active site or peptide-binding site in other structurally related proteins ( Hoelz et al . , 2003 ) . We shall refer to this as the ‘hub cavity’ ( Figure 10A , B ) . CaMKIIα contains three arginine residues that are located deep within the hub cavity ( residues 403 , 423 , and 439 ) . These residues are highly conserved either as arginine or lysine , with three positively charged residues present in CaMKII from C . elegans , mouse , and drosophila . No function has been ascribed to the hub cavity in CaMKII nor to the arginine residues within it . 10 . 7554/eLife . 01610 . 022Figure 10 . Side entrance to the hub cavity and docking of the regulatory segment onto the hub . ( A ) One vertical dimer unit has been removed from the side view of the decameric hub assembly ( shown as a schematic on the left ) . There is limited access to this cavity in the context of the holoenzyme . A close up of the G/H dimer from the crystal structure of CaMKIIγ ( PDB code: 2UX0 ) is highlighted in blue/pink ( right ) . There are three arginine residues that are located deep within the hub cavity ( residues 403 , 423 , and 439 ) , and these are highly conserved in CaMKII . ( B ) Shown is a vertical dimer from the mouse CaMKIIα crystal structure ( PDB code: 1HKX ) . The crystal structure is shown in cartoon representation ( left ) with the regulatory segment ( magenta ) docked onto β3 ( yellow ) . On the right , the crystal structure is shown as a surface representation where the hub cavity that contains the arginine residues ( blue ) is apparent through a small crack ( between αA and β3 ) , which we refer to as the lateral opening . ( C ) The regulatory segment ( magenta ) was docked onto β3 in the hub assembly ( yellow ) . This docked model was used to generate three independent molecular dynamics trajectories ( 100 ns each ) . In each of these trajectories the R3 element retains the hydrogen bonding pattern ( right ) that is consistent with its incorporation into the β sheet of the hub domain and the phosphate group maintains its proximity to the sidechain of Arg 403 , suggesting that this interaction is plausible . Each trajectory was sampled every 2 ns and each dot represents the formation of a hydrogen bond ( <3 . 5 Å ) . ( D ) Peptide binding disrupts lateral contacts within the hub domain . Docking of the regulatory segment peptide ( magenta ) prevents the crack between αA ( yellow ) and β3 ( yellow ) from closing . This is coupled to changes in the loops that make lateral contacts with the adjacent vertical dimers . This disruption may be sufficient to allow the release of a vertical dimer unit from the hub assembly and facilitate the exchange of subunits between holoenzymes . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 022 A feature of the hub assembly that might allow interaction between Thr 305 and Thr 306 in the regulatory segment and the arginine residues in the interior of the hub cavity is suggested by an interesting variability in the structures of individual subunits of the hub assembly in crystal structures . Although in many cases the only access to the interior of the hub cavity is through the main opening to the cavity , which is far from where the regulatory segment connects to the hub , some individual subunits in hub domain assemblies exhibit cracks in the surface at the interface between helix αA and the β sheet against which it is packed ( Figure 10B ) . These cracks make Arg 403 , located within the cavity , partially accessible from the side of the hub domain . If such cracks were to open up further then phosphorylated residues in the R3 element might gain access to the interior of the cavity from the side of the hub domain , in the vicinity of the lateral interfaces between vertical hub domain dimers in the holoenzyme . We carried out a 1 μs molecular dynamics simulation of a vertical hub domain dimer that was extracted from the crystal structure of mouse CaMKIIα ( PDB code: 1HKX , Video 1 ) . This simulation revealed transient fluctuations in the hub domain structure that opened the interface between helix αA and the underlying β sheet substantially . The simulation suggests that if the R3 element was to dock alongside the lateral interface then it might be positioned to insert phosphorylated sidechains into the interior of the hub domain when such a transient fluctuation occurs . 10 . 7554/eLife . 01610 . 023Video 1 . Lateral opening in the hub domain . Video of a 1 μs molecular dynamics simulation of a vertical hub domain dimer that was extracted from the crystal structure of mouse CaMKIIα ( PDB code: 1HKX ) . This simulation revealed transient fluctuations in the hub domain structure that opened the interface between helix αA and the underlying β sheet substantially . These fluctuations provide access to the hub cavity . DOI: http://dx . doi . org/10 . 7554/eLife . 01610 . 023 The molecular details for such a docking mechanism are suggested by the presence of an open edge on the β sheet of the hub domain right alongside the site where the transient openings into the hub cavity occur . This open edge of the β sheet is in fact utilized for a docking interaction by an 8 residue portion of the R3 element in the structure of autoinhibited CaMKII holoenzyme , in which this portion of the R3 element extends the β sheet by one strand ( Chao et al . , 2011 ) . In that structure , however , further interaction between the R3 element and the hub domain is prevented by the fact that the kinase domain sequesters the rest of the regulatory segment . We modeled the regulatory segment , with Thr 305 phosphorylated , as an additional strand that extends the β sheet of the hub domain , as shown in Figure 10C ( see ‘Materials and methods’ for details of the molecular docking ) . This model was used to generate three independent molecular dynamics trajectories , each extending for 50 ns . In each of these trajectories the R3 element retains the hydrogen bonding pattern that is consistent with its incorporation into the β sheet of the hub domain and the phosphate group maintains its proximity to the sidechain of Arg 403 , suggesting that this interaction is plausible ( Figure 10C ) . We analyzed these ‘docked’ trajectories for fluctuations that would disturb the contacts made within the hub assembly . The peptide stays docked onto the β3 strand of the hub domain throughout the trajectories , which , in turn , prevents the crack between helix αA and the β sheet from closing ( Figure 10D ) . Notably , the widening of the crack is coupled to changes in the disposition of the loops that make lateral contacts with the adjacent vertical dimers ( Figure 10D ) . This disruption may be sufficient to allow the release of a vertical dimer from the hub assembly ( see Figure 6 for a schematic representation of such a process ) . Experimental validation of this model will require an extensive set of studies that are beyond the scope of this paper . For example , we mutated the arginine residues in the hub cavity that we have implicated in the exchange process , but this resulted in a substantial loss of fluorophore labeling at residue 335 , indicating that the structure of the hub domain might be compromised . A systematic test of the predictions of this model will be carried out in future studies . Full-length constructs of human CaMKII were cloned in a pSMT-3 vector containing an N-terminal sumo expression tag ( LifeSensors , Malvern , PA ) . Mutants were generated using a Quikchange protocol ( Agilent Technologies , Santa Clara , CA ) . The linker connecting Hcp1 to CaMKII ( GGC GCG TCT GGC GCG TCT GGC GCG TCT ) and the hub domain construct were made using standard PCR techniques . All CaMKII variants were expressed using E . coli and prepared as described previously ( Chao et al . , 2010 ) . Briefly , protein expression was done in Tuner ( DE-3 ) pLysS cells that contained an additional plasmid for λ phosphatase production . Cells were induced by addition of 1 mM isopropyl β-D-1-thiogalactopyranoside and grown overnight at 18°C . Cell pellets were resuspended in Buffer A ( 25 mM Tris , pH 8 . 5 , 150 mM potassium chloride ( KCl ) , 1 mM DTT , 50 mM imidazole , and 10% glycerol ) and lysed using a cell disrupter . All purification steps were carried out at 4°C and all columns were purchased from GE Healthcare ( Piscataway , NJ ) . Cleared lysate was loaded on 5 mL Ni-NTA column , eluted with 0 . 5 M imidazole , desalted using a HiPrep 26/10 desalting column into Buffer A with 0 mM imidazole , and cleaved with Ulp1 protease ( overnight at 4°C ) . The cleaved samples were loaded onto the Ni-NTA column and the flow through was loaded onto a Q-FF 5 ml column , and then eluted with a KCl gradient . Eluted proteins were concentrated and then buffer-exchanged using a Superpose 6 gel-filtration column equilibrated in 25 mM Tris , pH 8 . 0 , 150 mM KCl , 2 mM tris ( 2-carboxyethyl ) phosphine [TCEP] and 10% glycerol . Fractions with pure protein were frozen at −80°C . Calmodulin ( from Gallus gallus ) was expressed using a pET-15b vector ( generous gift of Angus Nairn ) , and purified as described previously ( Putkey and Waxham , 1996 ) . Samples for mass spectrometry were prepared by the addition of 80% acetonitrile and 250 ng trypsin to ∼10 μM CaMKII . Reactions were carried out at 25°C for 2–3 hr then put on a Speed-Vac for an additional 2–3 hr to remove residual acetonitrile . Samples were analyzed using LC/MS ( QB3/Chemistry Mass Spec Facility at UC Berkeley ) . CaMKII is expressed with a C-terminal biotinylation sequence ( Avitag ) ( Howarth and Ting , 2008 ) , and after purification , labeled with biotin using the biotin ligase BirA . This reaction mixture contained 1 mM ATP , 10 μM biotin ( dissolved in DMSO ) , 0 . 6 μM BirA ligase in a final volume of 0 . 5–1 ml . The reaction was carried out on ice for 1 hr and desalted into labeling buffer using a PD-25 column ( GE Healthcare ) . Mass spectrometry indicates complete biotinylation . Following biotinylation , CaMKII was labeled with Alexa fluor dyes . Purified CaMKII was desalted into a buffer containing 25 mM Tris , pH 8 . 0 , 150 mM potassium chloride , 1 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) and 10% glycerol just prior to labeling with Alexa Fluor C5-maleimide dyes ( 488 and 594 , Life Technologies ) . Alexa dye was resuspended in 25 mM Tris ( pH 8 . 0 ) at a final concentration of 10–15 mM . Labeling was completed by addition of threefold to fivefold molar excess dye over CaMKII subunit concentration , and incubated for 2–3 hr at 25°C . Samples were desalted ( 1 or 2X depending on efficiency and amount of protein ) using PD-25 columns ( GE healthcare ) into an imaging buffer ( 25 mM Tris , pH 8 . 0 , 150 mM potassium chloride , 1 mM TCEP ) . Samples were concentrated using Amicon filters to a final subunit concentration of 6 μM . Dye incorporation was estimated using spectrophotometric analysis ( Nanodrop , Thermo Scientific , DE ) . The absorbance of each dye at 280 nm ( A280 ) was estimated using free dye resuspended in buffer . The dye-labeled protein samples were then scanned and the A280 was corrected for dye contribution . The concentration of protein and dye was calculated using the corresponding extinction coefficients and the ratio of protein labeled is estimated as: [dye]/[protein] ( Alexa 488: ε = 71 , 000 cm−1M−1 , Alexa 594: ε = 71 , 000 cm−1M−1 , and extinction coefficients for CaMKII variants were estimated using an online protein calculator tool: http://web . expasy . org/protparam/ ) . CaMKII labeled with Alexa 488 ( 6 μM ) was mixed with CaMKII labeled with Alexa 594 ( 6 μM ) with or without ATP ( 250 μM ) , MgCl2 ( 8 mM ) , calcium ( 500 μM ) and calmodulin ( 6 μM ) . The final concentration of CaMKII in these mixtures is 4 . 8 μM . In the main text we refer to the stock concentration of CaMKII used to make the mixtures ( 6 μM ) . At each time point , 2 μl of the CaMKII mixture was diluted into 1000 μl of desalting buffer and spread onto the chamber housing the functionalized cover slip . CaMKII was incubated on the cover slip for 1 min and then the surface was washed 3X with 1 ml desalting buffer . These slides were then imaged using TIRF microscopy . In experiments where the inhibitor , bosutinib , was added , 6 μM CaMKII ( after labeling with Alexa fluor ) was incubated with 50 μM bosutinib for 15 min at 25°C . Proteins were then mixed as above , except with less ATP ( 0 or 1 μM ) . In the 3-color colocalization experiments , the pThr286 antibody ( Pierce antibodies #A-20186 , Thermo Scientific ) was labeled with Alexa 647 prior to the experiment , following the protocol for the monoclonal antibody labeling kit ( Life Technologies ) . At each time point in these experiments , antibody was added to the CaMKII mixture at 35-fold molar excess over CaMKII concentration and binding was carried out at 25°C for 3 min . The sample was then diluted 500-fold and spread onto the functionalized cover slip , as above . For the preparation of phosphorylated CaMKIIα , the same procedure as above was followed , except that the His tag was not removed by Ulp1 cleavage , and a subtractive Ni-NTA purification step was not done . This protein was then labeled with Alexa 594 as previously described . His-tagged CaMKII ( 6 μM ) was activated at 25°C by addition of Ca2+/CaM ( 40 μM/6 μM ) , ATP ( 250 μM ) and MgCl2 ( 8 mM ) for 10 min . The reaction was quenched by addition of EDTA ( 50 mM ) and EGTA ( 20 mM ) . This reaction volume was mixed with free Ni-NTA resin ( 1 ml ) , equilibrated with Buffer A ( 5% glycerol ) and allowed to bind for 1–2 hr ( 4°C ) . After washing with Buffer A , bound CaMKII-His was eluted with 0 . 5 M imidazole ( 5% glycerol ) , desalted into imaging buffer using a PD-25 column and concentrated to at least 6 μM . A mutant of CaMKIIα was used for these experiments , in which all surface exposed cysteine residues ( 280 , 289 ) were mutated to serine , and aspartate 335 was mutated to cysteine ( D335C ) . The same labeling and mixing experiments for the single-molecule assay were done . Samples were incubated at 25°C or 37°C . At each time point , 25 μl from the mixed sample was removed and diluted to a final volume of 150 μl . An emission spectrum ( 500–700 nm ) was acquired for each diluted sample excited at 490 nm using a Fluoromax-3 fluorometer ( Horiba Scientific , Edison , NJ ) . Data were analyzed by calculating the FRET ratio ( acceptor emission at 610 nm divided by donor emission at 515 nm ) . Error bars are calculated from the standard deviation between separate experiments on different days . Kinase activity was monitored using an ADP quest assay ( Charter et al . , 2006 ) . Phosphorylated CaMKII ( activated sample ) was prepared as above . Activated and unactivated samples were diluted to 1 μM . For the mixing experiments , these samples were mixed in a 3:1 ratio of unactivated:activated protein ( 15 μl unactivated + 5 μl activated ) . For controls , each sample was measured individually at these same concentrations , and brought to the same final volume by addition of buffer . The kinase assay was carried out in a 384-well plate format at 30°C in a 27 . 5 μl reaction volume . The reaction mixture contained the following components ( listed at final concentration ) : 10 mM MgCl2 , 20 μM DTT , 0 . 3–0 . 5 mM peptide substrate syntide ( PLARTLSVAGLPGKK ) , and Tris . For each reaction , 5 μl of the protein solution was mixed with 4 μl concentrated reaction mixture , and 1 μl of either water or Ca2+/CaM ( 200 μM/4 μM ) was added . The protocol for the ADP quest assay was followed for the remaining steps and reactions were initiated by the addition of 250 μM ATP to the mix . The increase in fluorescence was monitored at 590 nm ( excitation at 530 nm ) in a fluorescent microplate spectrophotometer ( Synergy H4 , Biotek ) with sampling every 5 min . Slopes ( Δfluorescence590 nm/time ) were calculated from the early data points where an initial linear rate is observed . The molecular dynamics trajectories were generated using the Gromacs 4 . 6 . 2 package ( Berendsen et al . , 1995; Pronk et al . , 2013 ) . For all the hub dimer molecular dynamics simulations , with the docked pThr containing peptide , Amber12 ( Case et al . , 2012 ) was used for ease of handling of modified backbone residues . The ff99SB-ILDN force field was used for all the calculations ( Lindorff-Larsen et al . , 2010 ) . All simulations were carried out in aqueous medium using the TIP3P water model and appropriate counterions ( Na+ and Cl− ) were added to neutralize the net charges . After initial energy minimization , the systems were subjected to 30–100 ps of constant number , volume and temperature ( NVT ) equilibration , during which the system was heated to 300K . This was followed by a short equilibration at constant number , pressure and temperature ( NPT , 20–100 ps ) . The equilibration steps were performed with harmonic positional restraints on the protein atoms . Finally , the production simulations were performed under NPT conditions , with the Berendsen and v-rescale thermostats in Amber12 and Gromacs 4 . 6 . 2 respectively , in the absence of positional restraints . Periodic boundary conditions were imposed , and particle-mesh Ewald summations were used for long-range electrostatics and the van der Waals cut-off is set at 1 nm . A time step of 2 fs was employed and the structures were stored every 2 ps . SHAKE and LINCS constraint algorithms were used with Amber12 and Gromacs 4 . 6 . 2 , respectively , to fix covalent bonds . We modeled a 16-residue stretch of the R3 element as an additional strand of the β sheet in the hub domain . We created the model by first docking a β strand taken from an arbitrary protein so that the last 9 residues of the strand were aligned roughly with the 8 residue strand formed by R3 element in the autoinhibited CaMKII holoenzyme ( PDB code: 3SOA ) . This model does not have the strand aligned optimally with the rest of the β sheet , and to improve the model we searched the protein databank , using the PDBeFOLD server ( Krissinel and Henrick , 2004 ) , for structures that had β strands arranged in the same topology and with close spatial overlap to the three strands from the hub domain and the roughly modeled fourth β strand , yielding a match in the human core Snrnp domain ( PDB code: 1D3B ) . We then created a new model in which the β sheet from the crystal structure of the hub domain was retained and the fourth strand was taken from Snrnp , with appropriate changes to the sequence . Using this fourth strand as a template we modeled in the sequence of the CaMKII R3 element in more than one sequence register , and carried out a series of short molecular dynamics simulations ( ∼100 ns each ) . In one of these simulations a conformational change that is similar to the transient fluctuations noted in the Appendix happened to occur , opening up access to the hub cavity . The sidechain of Met 307 from the R3 element was seen to approach Arg 403 closely in this simulation . We then changed the sequence register of the peptide so that Thr 305 occupied the position of the methionine sidechain , and we added a phosphate group to the threonine sidechain .	How do fleeting signals passing through the neurons of our brains become memories that can last for years or even decades ? An enzyme called CaMKII is known to have an important role in the formation of memories . CaMKII adds phosphate groups to proteins—a process that is called phosphorylation—and is itself activated when calcium levels increase inside the neurons where the enzyme is found . Individual CaMKII proteins bind together in groups of 12 to form a ‘holoenzyme’ . When one of the 12 subunits is activated by calcium , it can phosphorylate the other subunits in the same holoenzyme . Once this happens , the activation of CaMKII can continue after the initial rise in calcium has ceased , and this effect is thought to be involved in the formation of long-term memories . 30 years ago , Francis Crick—famous for his role in the discovery of the double helix—proposed that memory formation might involve ‘memory-storage molecules’ passing an activated state to unactivated molecules , and John Lisman later suggested that CaMKII could fulfil this role by swapping subunits of holoenzymes between activated and unactivated ones . Now , Stratton , Lee et al . have tested whether CaMKII can exchange subunits by using advanced microscopy to track single molecules of CaMKII labelled with fluorescent markers . This revealed that activation can cause CaMKII subunits repeatedly to mix between holoenzymes—and this only happens once a first holoenzyme has been activated . Subunits of CaMKII join together via a central ‘hub’ region , but when a subunit is activated , the phosphorylated segment may interact with the hub . This weakens the connections between the subunits , thereby making it easier for subunits to exchange between holoenzymes . This process provides a mechanism by which a level of activated CaMKII can be maintained , even if some subunits become degraded and long after the disappearance of the initial activation signal .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	Activation-triggered subunit exchange between CaMKII holoenzymes facilitates the spread of kinase activity
At vertebrate neuromuscular junctions ( NMJs ) , the synaptic basal lamina contains different extracellular matrix ( ECM ) proteins and synaptogenic factors that induce and maintain synaptic specializations . Here , we report that podosome-like structures ( PLSs ) induced by ubiquitous ECM proteins regulate the formation and remodeling of acetylcholine receptor ( AChR ) clusters via focal ECM degradation . Mechanistically , ECM degradation is mediated by PLS-directed trafficking and surface insertion of membrane-type 1 matrix metalloproteinase ( MT1-MMP ) to AChR clusters through microtubule-capturing mechanisms . Upon synaptic induction , MT1-MMP plays a crucial role in the recruitment of aneural AChR clusters for the assembly of postsynaptic specializations . Lastly , the structural defects of NMJs in embryonic MT1-MMP-/- mice further demonstrate the physiological role of MT1-MMP in normal NMJ development . Collectively , this study suggests that postsynaptic MT1-MMP serves as a molecular switch to synaptogenesis by modulating local ECM environment for the deposition of synaptogenic signals that regulate postsynaptic differentiation at developing NMJs . Cell-cell communication in the nervous system occurs at specialized structures called synapses , where nerve signals propagate from a presynaptic neuron to a postsynaptic target cell . The postsynaptic apparatus containing a high density of neurotransmitter receptors represents an important feature of chemical synapses . In the past decades , tremendous effort has been made in understanding how the synthesis , trafficking , localization , and clustering of neurotransmitter receptors at the postsynaptic sites regulate synaptic formation , function , and plasticity in both central and peripheral synapses ( Li et al . , 2018; Sanes and Lichtman , 2001; Song and Huganir , 2002; Wu et al . , 2010 ) . Many of those studies were performed using the neuromuscular junction ( NMJ ) , a simple peripheral chemical synapse . At the postsynaptic apparatus of NMJs , acetylcholine receptor ( AChR ) molecules are clustered at a nearly crystalline density of ~10 , 000 receptors/μm2 ( Fertuck and Salpeter , 1976 ) ; in contrast , its density drops drastically to <10 receptors/μm2 at the extra-synaptic regions . However , the detailed molecular mechanisms underlying the formation and maintenance of densely clustered AChR molecules at the postsynaptic apparatus are not fully understood . Previous knockout studies showed that even in the absence of motor neurons , AChR clusters can still be found in muscle fibers in vivo ( Lin et al . , 2001; Yang et al . , 2001 ) , suggesting that such AChR pre-patterns are spontaneously formed via muscle-intrinsic mechanisms . Interestingly , a similar structure of aneural AChR clusters can be induced in Xenopus primary muscles or immortalized C2C12 myotubes by culturing them on substratum coated with laminin , a major structural glycoprotein in the extracellular matrix ( ECM ) ( Kummer et al . , 2004; Lee et al . , 2009 ) . These aneural AChR clusters , located at the bottom surface of cultured muscles in direct contact with ECM proteins , may undergo topological transformation , which is mirrored by the progressive structural changes in synaptic AChR clusters during NMJ maturation in vivo ( Kummer et al . , 2004 ) . Our previous work showed that actin-rich structures are highly concentrated at the perforated regions of aneural AChR clusters ( Lee et al . , 2009 ) . These structures were later shown to share the typical characteristics of podosome-like structures ( PLSs ) ( Proszynski et al . , 2009 ) . PLSs , initially identified as dynamic foot-like structures in motile or invasive cells , have been linked to pathophysiological processes such as cancer cell invasion and metastasis via local proteolysis of ECM proteins ( Linder , 2007 ) . It is worth to note that a small proportion of aneural AChR clusters can also be identified at the top surface of cultured muscles , and these spontaneously formed clusters are likely mediated through ECM- and PLS-independent mechanisms . At the NMJ , the exact functions of synaptic PLSs in regulating AChR cluster formation and remodeling remain largely unclear . In this study , we show that the assembly of PLSs , which can be induced by different ECM proteins , focally regulates matrix degradation for AChR clustering and topological remodeling . Next , we further demonstrate that intracellular trafficking and surface insertion of membrane-type 1 ( MT1- ) matrix metalloproteinase ( MMP ) are mediated via microtubule-capturing mechanisms at PLSs , which in turn spatiotemporally regulate the topological remodeling of aneural AChR clusters and their dispersal upon synaptic induction . Inhibition of MMP activity or reduced expression of muscle MT1-MMP greatly suppresses nerve-induced AChR cluster formation and stabilizes aneural AChR clusters against dispersal . Lastly , we show that MT1-MMP is required for the recruitment of AChR molecules from aneural to synaptic AChR clusters at developing NMJs in vitro and in vivo . Taken together , this study revealed the significance of PLS-directed MT1-MMP trafficking and surface insertion in modulating the assembly and topological remodeling of AChR clusters via focal matrix degradation at developing neuromuscular synapses . When dissociated myotomal tissues from early Xenopus embryos were cultured on glass coverslips coated with a mixture of ECM proteins , containing entactin/nidogen , collagen and laminin ( ECL ) , topologically complex aneural AChR clusters were observed mostly on the bottom surface of muscle cells in contact with ECL-coated substratum ( Figure 1A ) . Some aneural AChR clusters could also be found within the top surface of muscle cells but exhibited a sparsely scattered morphology . To further identify if specific ECM proteins are required for the formation of these topologically complex AChR clusters , we tested several key ECM proteins individually , including laminin , collagen , and gelatin , for their ability to induce AChR cluster formation in cultured Xenopus muscle cells . We found that the formation of bottom AChR clusters could be effectively induced by all ECM proteins tested , in contrast to the negative control using poly-D-lysine ( PDL ) , a polypeptide commonly used to promote cell attachment ( Figure 1B ) . In immortalized C2C12 myotubes , aneural AChR clusters can undergo a topological transformation from a plaque to a perforated pretzel-shaped , and eventually to C-shaped arrays ( Kummer et al . , 2004 ) . Hence , we further classified the aneural AChR clusters in Xenopus muscle cultures into scattered , plaque , perforated , and C-shaped based on their morphological features . A majority of ECM-induced aneural AChR clusters exhibited perforated structures . While the percentage of muscle cells with AChR clusters was largely reduced on PDL-coated substrate , the largest share of AChR clusters exhibited scattered structures ( 9 . 59% out of 13 . 33% total ) ( Figure 1C ) , similar to those structures found on the top muscle surface . As opposed to a previous study showing the topological transformation of AChR clusters in C2C12 myotubes ( Kummer et al . , 2004 ) , most AChR clusters , regardless of their topological features , were gradually dispersed over 6 days in cultured Xenopus muscle cells ( Figure 1—figure supplement 1 ) . Our previous study indicated that actin depolymerizing factor ( ADF ) /cofilin , one of the molecular components in PLSs , regulates the formation and maintenance of AChR clusters via vesicular trafficking ( Lee et al . , 2009 ) . Here , we further performed immunostaining of several typical core markers ( composed of F-actin and actin-associated proteins , such as ADF/cofilin , Arp2/3 complex , and cortactin ) and cortex markers ( composed of adhesion molecules , such as vinculin , talin , and paxillin ) of PLSs . All these markers were found to be enriched in the perforations ( arrows ) of AChR clusters in Xenopus muscle cultures ( Figure 1—figure supplement 2 ) . As cortactin is highly enriched and involved in both early and late stages of PLS formation ( Webb et al . , 2006 ) , we then used cortactin as an indicator of PLS localization in AChR clusters . PLSs were spatially concentrated in a majority of perforated AChR clusters induced by different ECM proteins ( Figure 1D and E ) . In contrast , PLSs were not localized in AChR clusters , even in a small fraction of AChR clusters ( 2 . 16% out of 13 . 33% total ) exhibiting perforated structures , on PDL substrates . These data suggest that different ECM proteins can induce the assembly of PLSs , which may in turn promote the formation of topologically complex structures of AChR clusters . The data above showed that perforated AChR clusters can be effectively induced by gelatin ( Figure 1B and C ) , an ECM protein that was previously shown to induce simple AChR plaques only ( Kummer et al . , 2004 ) . This novel finding prompted us to adopt the well-established fluorescent gelatin degradation assay ( Starnes et al . , 2011 ) to investigate the possible association between matrix degradation and PLS localization in AChR clusters . After plating the muscle cells on FITC-gelatin-coated substratum , confocal microscopy was performed to visualize the extent of gelatin degradation in 3 day old live Xenopus primary muscle cells . We observed FITC-gelatin that was extensively degraded in close association with the perforated regions of aneural AChR clusters ( Figure 2A ) . Next , we quantitatively measured the fluorescence intensity of FITC-gelatin in different AChR clusters and identified a linear relationship between the extent of gelatin degradation and the perforation area of aneural AChR clusters ( Figure 2B ) . A similar spatially restricted gelatin degradation pattern was observed in cultured C2C12 myotubes ( Figure 2C ) . Besides gelatin , immunostaining experiments also showed the spatial degradation of laminin at the perforated regions of AChR clusters in Xenopus primary muscle cells ( Figure 2D ) . To examine if the focal degradation pattern of ECM proteins is mediated by MMP activity , we first used two broad-spectrum MMP inhibitors , BB-94 and Marimastat ( BB-2516 ) , which are known to be highly potent against MMP-1 , -2 , -3 , -7 , -9 and MT1-MMP ( Cathcart et al . , 2015 ) . Although perforated AChR clusters could be identified in muscle cells treated with 5 µM BB-94 or 10 µM BB-2516 , the degradation of fluorescent gelatin associated with these perforated AChR clusters was completely abolished ( Figure 2E and F ) . Quantification further demonstrated that the initial formation of aneural AChR clusters in muscle cells was largely unaffected in the presence of MMP inhibitors ( Figure 2G ) . To investigate if PLS-associated MMP activity regulates the topological remodeling of AChR clusters , we monitored the same group of muscle cells over 2 days after aneural AChR clusters had been formed . We observed a significant reduction in the number of aneural AChR clusters in control muscle cells , in contrast to BB-94 or BB-2516 treatment that caused a slight increase in the number of aneural AChR clusters ( Figure 2H ) . To further explore if MMP activity regulates the topological remodeling of AChR clusters , we performed time-lapse imaging to monitor the dynamic changes in AChR cluster morphology and intensity in response to BB-94 treatment . AChR clusters were first labeled with fluorescent α-bungarotoxin ( 0 hr ) , then they were monitored for 96 hr after treatment . In control untreated cells , we observed a gradual dispersal of aneural AChR clusters over the 96 hr imaging period; however , such spontaneous dispersal and topological remodeling of AChR clusters were largely inhibited by BB-94 treatment within the same imaging period ( Figure 2I and J ) . These data demonstrated that MMP activity is involved in the topological remodeling and dispersal of aneural AChR clusters . The spatially restricted patterns of gelatin degradation suggested the possible involvement of membrane-type MMPs ( MT-MMPs ) in modulating ECM environment focally to regulate AChR cluster remodeling and dispersal . To test this , we first performed immunostaining of Xenopus muscle cells and detected endogenous MT1-MMP that was enriched in the perforations of aneural AChR clusters ( Figure 3A ) . Next , we over-expressed MT1-MMP-mCherry construct in muscle cells via Xenopus embryo microinjection . In MT1-MMP-mCherry-overexpressing muscle cells , we observed extensive degradation of fluorescent gelatin in the entire cell area ( Figure 3B ) , in contrast to the spatially restricted degradation pattern at the perforated regions of AChR clusters in control cells . In addition , MT1-MMP-mCherry overexpressing cells showed a lower percentage of ECM-induced bottom AChR clusters , but a higher percentage of top AChR clusters , in comparison to the wild-type control cells ( Figure 3C ) . Quantitative analyses on those bottom AChR clusters in MT1-MMP-mCherry-overexpressing cells showed that the intensity of aneural AChR clusters was significantly reduced , which was accompanied by the extensive gelatin degradation ( Figure 3D ) . Interestingly , BB-94 or BB-2516 treatment was able to partially revert the distribution of aneural AChR clusters from the top surface back to the bottom surface of MT1-MMP-mCherry-overexpressing muscle cells ( Figure 3C ) . Consistent with that , the extensive degradation of fluorescent gelatin and the reduction of AChR intensity caused by MT1-MMP-mCherry overexpression were largely rescued by either BB-94 or BB-2516 treatment ( Figure 3D ) . Our results indicated that the precise control of focal ECM degradation by MT1-MMP at the perforated sites plays an essential role in the formation of bottom aneural AChR clusters . To investigate the cytoskeletal involvement for intracellular trafficking of MT1-MMP , we next performed total internal reflection fluorescence ( TIRF ) microscopy to visualize the perimembrane fraction of GFP-tagged microtubule plus end-binding protein 1 ( EB1-GFP ) in live cultured muscle cells . Interestingly , we observed EB1-GFP comets to be highly enriched at the perforated aneural AChR clusters ( Figure 4A ) . This pattern of EB1-GFP comets was not an artifact of overexpression , as a similar localization pattern of endogenous EB1 signals was also detected in fixed Xenopus muscle cells ( Figure 4—figure supplement 1 ) . By analyzing the density and trajectory of EB1-GFP comets in time-lapse images , we found that EB1-GFP comets were densely localized , but they showed lower mobility , at sites of aneural AChR clusters than other regions of the cell ( Figure 4B and C ) . These data suggested that AChR clusters are the focal point of directed microtubule-based intracellular transport . Previous studies showed that cytoplasmic linker-associated protein ( CLASP ) and LL5β , a PLS cortex protein involved in NMJ development ( Kishi et al . , 2005 ) , interact with microtubules to direct the vesicular trafficking of different proteins to the NMJ ( Basu et al . , 2015 ) . Thus , we examined if CLASP is required for directing the movement of EB1-GFP comets to PLS-enriched AChR clusters using antisense morpholino oligonucleotide ( MO ) -mediated knockdown approach . As validated by western blot analysis , endogenous CLASP protein level was largely reduced in Xenopus embryos microinjected with CLASP-MO ( Figure 4—figure supplement 2 ) . In CLASP-MO muscle cells , TIRF microscopy showed the perimembrane signals of EB1-GFP that were primarily found at the center of perforations within aneural AChR clusters ( Figure 4A ) . Although the density of EB1-GFP comets was only slightly reduced at AChR cluster region ( Figure 4B ) , the average speed of EB1-GFP comets in the region of AChR clusters was significantly increased by 26 . 7% to 0 . 12 ± 0 . 004 μm/s in CLASP-MO muscle cells ( Figure 4C ) , indicating that CLASP is involved in microtubule capturing at aneural AChR clusters . To further study the directed trafficking and local capturing of EB1-GFP at aneural AChR clusters , we performed TIRF in combination with fluorescence recovery after photobleaching ( FRAP ) experiments . Before photobleaching , EB1-GFP signals were primarily enriched at the edge of perforations of aneural AChR clusters in control cells but at the center of perforations in CLASP-MO muscle cells ( Figure 4D ) . To better show the spatial patterns of EB1-GFP signals at the perforations of aneural AChR clusters in control versus CLASP-MO muscle cells , we have plotted the fluorescence intensity profiles of EB1-GFP and AChR across a perforated region of AChR clusters , as indicated in Figure 4D . In the control cell , the peaks of EB1-GFP intensity showed a slight lateral shift towards the center of a perforated region ( Top chart , Figure 4E ) , indicating the preferential enrichment of EB1-GFP at the edge of perforations . In contrast , CLASP knockdown caused the peak of EB1-GFP intensity to cover a larger area of a perforated region , indicating that EB1-GFP is enriched relatively closer to the center of perforations ( Bottom chart , Figure 4E ) . After photobleaching the region of perforated AChR clusters ( dotted rectangles in Figure 4F ) , we performed time-lapse imaging that reflects the rapid recovery of EB1-GFP fluorescence signals ( in seconds after photobleaching ) . Spatially , the recovery of EB1-GFP signals was detected near the edge of those perforations within AChR clusters in control cells ( arrows in Figure 4F , Figure 4—video 1 ) , but it was largely reduced in CLASP-MO muscle cells ( arrowheads in Figure 4F , Figure 4—video 2 ) , leading to a significant increase in the half-time of EB1-GFP signal recovery by 58% from 5 . 71 ± 0 . 83 s in control cells to 9 . 01 ± 0 . 71 s in CLASP-MO muscle cells ( Figure 4G and H ) . Collectively , our data suggested that PLSs may direct the vesicular trafficking to AChR clusters through EB1-/CLASP-mediated microtubule-capturing mechanisms . As microtubules serve as major tracks for vesicular trafficking in mammalian cells , the cortical microtubule organization plays a crucial role in the targeted delivery of secretory and membrane proteins via exocytosis ( Noordstra and Akhmanova , 2017 ) . To further investigate if MT1-MMP intracellular trafficking is mediated by microtubule-based transport , we performed dual-channel live-cell imaging on Xenopus muscle cells over-expressing both MT1-MMP-mCherry and EB1-GFP ( Figure 5A ) . In region ‘i’ , some MT1-MMP-mCherry vesicles ( arrows ) were found to be initially immobile at the beginning of this time-lapse series , until they were captured by a moving EB1-GFP comet ( arrowhead ) , leading to a coordinated movement of both EB1-GFP and MT1-MMP-mCherry signals from 27 s to 29 s time-points in this imaging period ( middle row , Figure 5A ) . Another example in region ‘ii’ indicated that MT1-MMP-mCherry vesicles ( arrows ) were able to move bidirectionally on microtubule structures , as visualized by the fading tracks ( arrowheads ) generated by moving EB1-GFP comets ( bottom row , Figure 5A ) . The spatiotemporal correlation between MT1-MMP-mCherry and EB1-GFP signals was better appreciated by constructing kymographs using multiple timeframes along ‘K1’ and ‘K2’ lines ( Figure 5B ) . In these two kymographs , most MT1-MMP-mCherry vesicles were found to be relatively immobile . Interestingly , we observed some displacement of MT1-MMP-mCherry vesicles immediately after EB1-GFP comets had come into contact with them ( arrows in Figure 5B ) . Consistent with the subcellular localization of MT1-MMP-mCherry , immunostaining data revealed that endogenous MT1-MMP exhibited punctuated structures inside the muscle cells , and some of them were co-localized with vesicle-associated membrane protein 1 ( VAMP1 ) ( Figure 5—figure supplement 1 ) . Taken together , these data illustrated that EB1 coordinates the vesicular trafficking of MT1-MMP in cultured muscle cells . To further visualize the directed trafficking and capturing of MT1-MMP vesicles at PLS-associated AChR clusters , we performed TIRF-FRAP experiments on MT1-MMP-mCherry-expressing muscle cells after AChR labeling . The cells with low expression level of MT1-MMP-mCherry were chosen , in which the formation of aneural AChR clusters was largely unaffected . After photobleaching , we detected local capturing of MT1-MMP-mCherry vesicles at the edge of perforations within aneural AChR clusters ( white arrows , region ‘i’ in Figure 5C ) , as well as at the edge of AChR cluster periphery ( red and green arrows , region ‘ii’ in Figure 5C ) . In kymographs constructed along ‘K3’ and ‘K4’ lines , we observed MT1-MMP-mCherry vesicles that were frequently captured at both perforations ( arrowheads ) and periphery ( arrows ) of AChR clusters over the 120 s imaging period after photobleaching ( Figure 5D ) . As both sites are enriched with PLS cortex markers , vinculin and talin ( arrowheads in Figure 1—figure supplement 2 ) , we speculated that PLSs , through the cortex domains , can mediate site-directed capturing of MT1-MMP vesicles for their subsequent surface insertion . As MT1-MMP-mCherry over-expression significantly inhibited the formation of AChR bottom clusters ( Figure 3C ) , we further used MT1-MMP tagged with a pH-sensitive green fluorescent protein pHluorin ( MT1-MMP-pHluorin ) to verify if the surface expression of MT1-MMP produces a similar inhibitory effect on AChR clustering . In cultured muscle with low MT1-MMP-pHluorin expression level , MT1-MMP-pHluorin signals were spatially enriched at the perforated regions of aneural AChR clusters ( arrow ) and caused only a slight inhibition on aneural AChR clustering ( Figure 6A and B ) . At high MT1-MMP-pHluorin expression , the formation of AChR bottom clusters was significantly inhibited , leading to sparsely scattered appearance if observed . These data further confirmed that MT1-MMP activity is crucial for the formation of topologically complex AChR clusters by precisely controlling focal ECM degradation at the perforated sites . To validate the surface localization of MT1-MMP in AChR clusters , we treated MT1-MMP-pHluorin-expressing muscle cells with a non-permeable buffer at pH 5 . 0 and observed a large reduction in MT1-MMP-pHluorin fluorescence intensity ( arrows in Figure 6C ) , mimicking the quenching of MT1-MMP-pHluorin signals in acidic vesicular compartments . The signals were then restored after changing the culture medium back to pH 7 . 8 , validating the effectiveness of this probe for visualizing surface localization of MT1-MMP , as it emits bright fluorescence upon surface insertion . Using this probe , we further determined the correlation between surface localization of MT1-MMP and topological remodeling of AChR clusters by monitoring the changes in the same AChR clusters over 24 hr in cultured muscle cells with different expression levels of MT1-MMP-pHluorin ( Figure 6D ) . It should be noted that muscle cells with low MT1-MMP-pHluorin expression level were chosen to minimize the non-specific effects of exogenous overexpression . In this example , AChR cluster intensity was moderately reduced by 26% after 24 hr in the muscle cell with a very low expression level of MT1-MMP-pHluorin . In contrast , AChR cluster intensity in a muscle cell with a relatively higher ( low ) MT1-MMP-pHluorin expression showed about 78% reduction during the same period . Quantitative analyses on a pool data of 13 AChR clusters further indicated that MT1-MMP-pHluorin intensity in AChR cluster region was linearly correlated with the change in AChR cluster intensity over the next 24 hr ( Figure 6E ) . These data indicated the functional role of surface MT1-MMP in promoting the topological remodeling of AChR clusters . Next , we performed TIRF-FRAP experiments that aimed to identify the exact locations of MT1-MMP surface insertion at AChR clusters . In this example , strong and discrete MT1-MMP-pHluorin signals re-appeared at the perforation ( arrows ) and periphery ( arrowheads ) of AChR clusters at time-points of 118 s and 48 s , respectively , in control muscle cells after photo-bleaching ( Figure 6F ) . Kymographs constructed along ‘K1’ and ‘K2’ lines indicated that some newly inserted MT1-MMP-pHluorin signals were relatively stable ( arrows ) , while some were short-lived but with multiple events of MT1-MMP surface insertion at the same location in AChR clusters ( arrowheads ) within a short imaging period ( Figure 6G ) . By quantifying the number of events of MT1-MMP surface insertion , we detected a significant reduction of MT1-MMP-pHluorin surface insertion at AChR clusters in CLASP-MO muscle cells when compared to that in control cells ( Figure 6H ) . Together with the above data showing the local capture of EB1-GFP comets , this data further confirmed that vesicular trafficking and surface targeting of MT1-MMP are spatially regulated by PLS cortex proteins at the perforation and periphery of AChR clusters through EB1-/CLASP-mediated microtubule capturing mechanisms . To understand the functional role of PLS-associated MMP activity at developing NMJs , we first examined the effects of MMP inhibitors on ECM degradation and synaptic AChR clustering in Xenopus nerve-muscle co-cultures . Along the trail of neurites in contact with the basal membrane of muscle cells ( arrows ) , we observed spatial degradation of fluorescent gelatin-coated substratum and synaptic AChR clustering at the nerve-muscle contacts ( Figure 7A ) . Like the focal gelatin degradation patterns observed in AChR-poor regions of aneural clusters ( Figure 2A ) , synaptic AChR clusters at the nerve-muscle contacts were found to be closely associated to , but not perfectly co-localized with , the sites of gelatin degradation , with only 17 . 15 ± 15% of area in AChR clusters to be associated with gelatin degradation in control co-cultures ( Figure 7C ) . However , both gelatin degradation and nerve-induced AChR clustering were effectively inhibited by BB-94 or BB-2516 ( Figure 7A and B ) . Nerve innervation of skeletal muscles involves local signals to initiate the formation of synaptic AChR clusters and global signals to induce the dispersal of aneural AChR clusters ( Dai and Peng , 1998 ) . Thus , we further examined the involvement of MMP activity in the dispersal of aneural AChR clusters induced by latex beads coated with agrin , a heparan sulfate proteoglycan that induces postsynaptic differentiation . Local application of agrin-coated beads to cultured muscle cells is capable of inducing postsynaptic differentiation in a spatiotemporally controllable manner ( Lee et al . , 2009 ) . In control cells , we detected a gradual dispersal of aneural AChR clusters upon agrin-bead stimulation for 24 hr; however , those clusters in agrin bead-contacted muscle cells were greatly stabilized by BB-94 or BB-2516 ( Figure 7D and Figure 7—figure supplement 1A ) . To determine if agrin stimulation enhances the localization of MT1-MMP in aneural AChR clusters prior to their dispersal , we performed immunostaining experiments , in which we demonstrated an increase of endogenous MT1-MMP signals at aneural AChR clusters upon agrin stimulation ( Figure 7—figure supplement 2 ) . These data implicated the important role of MT1-MMP activity in regulating not only the assembly of synaptic AChR clusters at nerve-muscle contacts , but also the disassembly of aneural AChR clusters , at developing NMJs . In contrast to nerve-induced AChR clustering , we noted that MMP inhibitors did not significantly affect the formation of AChR clusters induced by agrin beads ( Figure 7—figure supplement 1B and C ) . As the latex beads were coated with a high amount of exogenous recombinant agrin , and they were placed on the top surface of cultured muscle cells , it remained unclear if MMP activity regulates synaptic AChR clustering induced by the physiological amount of agrin and in the presence of different ECM proteins . Therefore , we further developed an ‘endogenous agrin track’ assay ( Figure 7E ) , by which we examined the effects of MMP inhibitors on synaptic AChR clustering in muscle cells cultured on ECL-coated substratum with endogenous agrin tracks . Using this assay , both BB-94 and BB-2516 were found to significantly inhibit the formation of synaptic AChR clusters induced by the endogenous agrin tracks ( Figure 7F and G ) , consistent with our data on nerve-induced AChR clustering . To identify the postsynaptic functions of MT1-MMP in NMJ development , antisense MO was used to knock down the endogenous expression of MT1-MMP specifically in muscle cells and then co-cultured with wild-type ( WT ) spinal neurons . The efficiency of MO-mediated knockdown of endogenous MT1-MMP protein level in Xenopus embryos was validated by western blot analysis ( Figure 8—figure supplement 1 ) . Pre-existing and newly inserted AChR molecules were differentially labeled by α-bungarotoxin conjugated with different fluorophores according to our previously established protocol ( Lee et al . , 2009 ) . In short , we first labeled all pre-existing AChR molecules with tetramethylrhodamine-conjugated α-bungarotoxin in muscle cells . After plating the spinal neurons for 1 day , the newly synthesized AChRs can then be labeled with Alexa Fluor 488-conjugated α-bungarotoxin . In control co-cultures , both pre-existing and newly inserted AChR signals were detected at the nerve-muscle contact sites ( Figure 8A ) , suggesting that both receptor pools contribute to the assembly of nerve-induced synaptic AChR clusters . Importantly , MT1-MMP knockdown caused a significant reduction , but not a complete inhibition , of nerve-induced AChR clustering in the chimeric co-cultures of MT1-MMP MO muscles with WT neurons ( Figure 8A and B ) . This could be explained by the partial knockdown of endogenous MT1-MMP protein level ( Figure 8—figure supplement 1 ) and/or the recruitment of diffuse AChRs to the nerve-muscle contacts via MT1-MMP-independent processes . Intriguingly , both pre-existing and newly inserted AChRs at the nerve-muscle contacts were equally affected in the chimeric co-cultures of MT1-MMP MO muscles with WT neurons ( Figure 8C and D ) . To rule out the possible off-target effects of MT1-MMP MO , we next performed rescue experiments by overexpressing MT1-MMP-mCherry ( MT1-mCherry ) in MT1-MMP knockdown ( MT1-MO ) muscle cells . In the chimeric co-cultures of MT1-MO + MT1-mCherry muscles with WT neurons , we found that the percentage of nerve-muscle contacts with AChR clusters was largely returned to the level comparable with the control groups , and both pre-existing and newly inserted AChRs were detected at the nerve-muscle contacts ( Figure 8 ) . These data suggested that postsynaptic MT1-MMP regulates both the surface delivery of newly synthesized AChR molecules and the redistribution/recruitment of pre-existing AChR molecules ( from either aneural AChR clusters or diffuse AChRs on muscle surface ) during the assembly of nerve-induced postsynaptic specializations at developing NMJs . To determine if aneural AChR clusters contribute to the formation of NMJs , we next performed laser-based photobleaching experiments to differentiate the recruitment of AChR molecules from aneural clusters versus diffuse AChRs for the assembly of nerve-induced synaptic AChR clusters ( Figure 9A ) . Without photobleaching , AChR signals were prominently detected at the nerve-muscle contact sites . The formation of nerve-induced synaptic AChR clusters was accompanied by the dispersal of aneural AChR clusters after 1 day in co-culture . In the experimental groups with photobleaching of aneural AChR clusters before co-culturing with the spinal neurons , we detected a significant reduction in AChR signals , which were contributed from diffuse AChRs only , at the nerve-muscle contacts in comparison to the control groups without photobleaching ( Figure 9B and C ) . These results provided compelling evidence to support the hypothesis that AChR molecules from aneural clusters serve as a major source for the assembly of synaptic AChR clusters . To further investigate the involvement of MT1-MMP in the recruitment of AChRs from aneural to synaptic clusters , we performed the same photobleaching experiments using chimeric co-cultures composed of either control MO or MT1-MMP MO muscle cells with WT spinal neurons ( Figure 9B ) . In the chimeric co-cultures with control MO muscle cells , both aneural clustered and diffuse AChRs were recruited to the nerve-muscle contacts at a level comparable to the wild-type co-cultures . However , knockdown of MT1-MMP in muscles significantly suppressed the recruitment of AChR molecules from aneural to synaptic clusters , while the recruitment from diffuse AChRs was unaffected ( Figure 9C ) . Consistent with the data above ( Figure 7D ) , we found that knockdown of muscle MT1-MMP caused a higher stability of aneural AChR clusters against nerve-induced dispersal ( arrowheads , lower left panels in Figure 9B ) . Interestingly , after photobleaching of aneural AChR clusters in the chimeric co-cultures with MT1-MMP MO muscles cells , the fluorescence intensity of photobleached aneural AChR clusters was partially recovered after 1 day in co-culture ( arrowheads , lower right panels in Figure 9B ) . These results suggested that diffuse AChRs were constantly recruited to aneural AChR clusters that were stabilized by muscle MT1-MMP knockdown , presumably through the diffusion-trap mechanisms in which diffuse AChRs can be immobilized by molecular scaffolds associated with the stable aneural AChR clusters ( Geng et al . , 2008 ) . Taken together , our data further supported that muscle MT1-MMP plays a dual-functional role in regulating the formation and dispersal of synaptic and aneural AChR clusters , respectively . To further investigate if MT1-MMP is required for NMJ development in vivo , we first performed immunostaining experiments to examine the localization of endogenous MT1-MMP at mature NMJs in rat soleus muscles . MT1-MMP was found to be highly enriched at the NMJs ( Figure 10A ) . On the other hand , some immunostaining signals , which were partly contributed by intracellular MT1-MMP proteins , were also observed outside of the AChR cluster regions , further indicating that the precise control of MT1-MMP trafficking and surface insertion is required for mediating local ECM degradation at NMJs . The specificity of MT1-MMP antibody was validated by pre-incubating the antibody with recombinant MT1-MMP proteins , which caused a large reduction of immunostaining signals at the NMJs . To further determine whether MT1-MMP is postsynaptically localized , a denervation experiment on adult rat soleus muscle was performed ( Figure 10B ) . In 4 days after sciatic nerve cut , we observed only a slight reduction in MT1-MMP signals at the NMJs of denervated muscles , suggesting that MT1-MMP is expressed primarily in postsynaptic muscles . Next , we examined the patterns of AChR clusters in embryonic diaphragm muscles between WT and MT1-MMP-deficient mice , which was previously established ( Sakamoto and Seiki , 2009; Zhou et al . , 2000 ) . Specifically , diaphragms from WT and MT1-MMP-/- mice at E13 . 5 and E18 . 5 were dissected and whole-mount stained for AChR and phrenic nerve . Aneural and synaptic AChR clusters were quantified based on whether the clusters were in direct contacts with the phrenic nerve branches . The endplate band width , which represents the zone of AChR cluster formation in the center of muscle fibers innervated by the phrenic nerve , was also measured . At E13 . 5 , MT1-MMP-/- mice showed a significantly higher density of aneural AChR clusters than the age-matched WT mice , but the density of synaptic AChR clusters was largely unchanged at this stage ( Figure 10C-E ) . At E18 . 5 , we observed a significant reduction in the density of synaptic AChR clusters and endplate band width in MT1-MMP-/- mice compared to those in the WT mice ( Figure 10E-F ) , indicating that synaptic AChR clusters formed in the knockout mice were immature with the altered number and density . Interestingly , the phrenic nerve in MT1-MMP-/- mice also exhibited abnormal phenotype with less axonal defasciculation and arborization in the diaphragm muscles ( Figure 10C and G ) , suggesting either the retrograde signaling of postsynaptic MT1-MMP or a possible role of neuronal MT1-MMP in regulating axonal growth and differentiation . Together , these findings further validated our in vitro data showing an important function of MT1-MMP in regulating the recruitment of aneural AChR clusters for the assembly of postsynaptic specializations at developing NMJs in vivo . At developing NMJs , AChR pre-patterns can be formed in muscle fibers without motor innervation in vivo ( Lin et al . , 2001; Yang et al . , 2001 ) , suggesting that muscle-intrinsic mechanisms are sufficient to regulate the initial formation of aneural AChR clusters . In this study , we showed that the formation of topologically complex aneural AChR clusters in Xenopus muscle cells requires several key ECM proteins ( Figure 1 ) . As ECM proteins can be secreted and deposited by skeletal myofibers and other different cell types that constitute the synaptic basal lamina at developing NMJs , the initial formation of AChR clusters may not be solely mediated by muscle-intrinsic mechanisms . The synaptic basal lamina is unique in composition , containing factors capable of inducing the assembly of synaptic specializations in both presynaptic and postsynaptic membranes , such as laminin-421 , collagen IV/VI/XIII , heparan sulfate proteoglycans ( e . g . agrin and perlecan ) , and other ubiquitously expressed ECM proteins ( e . g . laminin-111 , fibronectin and entactin/nidogen ) ( Cescon et al . , 2018; Chan et al . , 2020; Latvanlehto et al . , 2010; Singhal and Martin , 2011 ) . Among the different factors in the synaptic basal lamina , agrin plays the central role in the formation of postsynaptic specializations by activating the Lrp4/MuSK complex ( Kim et al . , 2008; Zhang et al . , 2008 ) . In this study , we observed extensive degradation of ubiquitously expressed ECM proteins , including gelatin and laminin-111 , at AChR clusters ( Figure 2A-D ) . A previous study suggested that MMP-3 is implicated for the removal of agrin from synaptic basal lamina at adult NMJs ( VanSaun and Werle , 2000 ) . In this study , however , the pattern and intensity of endogenous agrin tracks were largely unaffected by MMP inhibitors ( Figure 7F ) , ruling out the possible proteolytic degradation of secreted agrin by MT1-MMP activity at developing NMJs . Therefore , we hypothesize that PLS-associated MMP activity focally degrades different ubiquitously expressed ECM proteins , which clear the zone at the nascent synaptic basal lamina for the deposition of other secreted synaptogenic factors ( e . g . agrin or neuregulin ) and/or dismantle the extracellular anchorage of AChR molecules on ECM proteins that allow aneural clusters to undergo dispersal and re-distribution upon nerve induction ( Figure 11 ) . Interestingly , agrin can serve as a mechanotransduction signal , which may transduce ECM and cellular rigidity signals to the Hippo pathway effector YAP , that requires both the Lrp4-MuSK signaling and integrin-focal adhesion signaling in liver cancer cells ( Chakraborty et al . , 2017 ) . Importantly , muscle YAP mutation was found to impair the formation of NMJs and to inhibit NMJ regeneration after nerve injury ( Zhao et al . , 2017 ) . Therefore , the possible crosstalk between ECM-integrin and agrin-Lrp4-MuSK signaling pathways in regulating the postsynaptic differentiation at NMJs remains to be investigated . PLSs are highly dynamic actin-rich organelles found in transformed fibroblasts and many other motile or invasive cells . We detected several typical core and cortex markers of PLSs that were highly concentrated at perforated AChR clusters ( Figure 1—figure supplement 2 ) , further confirming the presence of PLSs in primary myotomal muscle cultures . Intriguingly , PLS cortex proteins , like vinculin and talin , were enriched not only at the perforated regions , but also at the periphery , of aneural AChR clusters ( Figure 1—figure supplement 2B ) . At those sites , some MT1-MMP-mCherry signals were captured ( Figure 5D ) , and some MT1-MMP-pHluorin signals were spatially recovered after photobleaching ( Figure 6F ) . These data suggested that PLS cortex proteins may capture MT1-MMP-containing vesicles and facilitate their surface insertion at the perforation and periphery of AChR clusters . It is important to note that both signals of MT1-MMP immunostaining ( Figure 3A ) and MT1-MMP-mCherry overexpression ( Figure 5 ) were not selectively localized to aneural AChR clusters , as both surface and intracellular MT1-MMP at the vesicular compartments were highlighted . In contrast , the localization patterns of MT1-MMP-pHluroin at a low expression level ( Figure 6A ) and fluorescent gelatin degradation ( Figure 2A ) were primarily restricted to the perforated AChR clusters . These results suggested that the vesicular trafficking and surface insertion of MT1-MMP is precisely controlled to mediate ECM degradation during NMJ formation . Microtubules are important cytoskeletal structures for intracellular transport of proteins . EB1 , a key regulator of the plus-end-tracking protein complex , functions to modulate microtubule dynamics and interaction with other intracellular organelles . Although endogenous EB1 comets can be found in the entire cell , they are highly enriched at the sites specialized for efficient coordination of cargo transport and release , such as focal adhesions and NMJs ( Noordstra and Akhmanova , 2017; Schmidt et al . , 2012 ) . Direct interactions between EB1 and other microtubule plus-end-tracking proteins ( CLASPs and CLIP-170 ) are implicated in the ‘recruitment mechanism’ , by which EB1 functions to search-and-capture post-Golgi vesicles and to spatially capture microtubules at the cell cortex ( Vaughan , 2005 ) . LL5β , one of the PLS cortex proteins , is localized at the postsynaptic membrane to serve as a corral for AChRs within the clusters ( Kishi et al . , 2005 ) . Interestingly , LL5β and CLASP1 interact with microtubules for directing the vesicular trafficking of different proteins to the postsynaptic membrane of NMJs ( Basu et al . , 2015 ) and to the focal adhesions ( Stehbens et al . , 2014 ) . In this study , antisense MO-mediated knockdown of endogenous Xenopus CLASP , the homolog of mammalian CLASP1 , in cultured muscle cells significantly reduced the local capture of microtubules at AChR clusters , as reflected by the increased average speed of EB1-GFP comets within AChR regions ( Figure 4C ) . It was also noted that microtubule plus-end-tracking proteins are capable to hop on and off from microtubule ends ( Dragestein et al . , 2008 ) , thus we cannot exclude the possibility of EB1-GFP signal recovery at the photobleached region that may be partly contributed by the exchange of EB1-GFP molecules between the free and microtubule-bound pools . Given that MT1-MMP surface insertion at aneural AChR clusters was significantly inhibited in CLASP-MO muscle cells ( Figure 6H ) , we therefore hypothesize that EB1-mediated microtubule capturing mechanisms can be regulated by PLSs at both cortex ( adhesion molecule-based ) and core ( actin-based ) domains via CLASP-dependent and -independent manner respectively , which is in agreement with other previous studies demonstrating the involvement of microtubule-dependent vesicular transport in AChR clustering ( Basu et al . , 2015; Marchand et al . , 2002; Schmidt et al . , 2012 ) . It has long been suggested that innervation of the skeletal muscle involves local signaling to initiate the formation of synaptic AChR clusters at the nerve-muscle contacts , and global signaling to induce the dispersal of aneural AChR clusters ( Dai and Peng , 1998 ) . However , a compelling evidence supporting that aneural AChR clusters serve as source for synaptic AChR clusters at developing NMJs is still lacking . In this study , we used laser-based photobleaching approach to demonstrate the differential contribution of AChR molecules from aneural clusters and diffuse AChRs for the assembly of nerve-induced synaptic AChR clusters ( Figure 9 ) . In agreement with previous studies ( Anderson and Cohen , 1977a; Cohen et al . , 1987 ) , most nerve contacts in Xenopus primary cultures are located at either the basal surface or the lateral side of muscle cells , which are influenced by MMP-mediated proteolytic degradation of ECM proteins . Importantly , we demonstrated that the recruitment of AChRs from aneural clusters , but not diffuse AChRs , is mediated by MT1-MMP-dependent processes . While MT1-MMP knockdown caused a significant inhibition in synaptic AChR clustering at the nerve-muscle contacts , it is intriguing to note that diffuse AChRs were constantly recruited to aneural AChR clusters that were stabilized by MT1-MMP knockdown ( Figure 9B ) . This recruitment may be mediated by the diffusion-trap mechanism , in which diffuse AChRs can be tethered to molecular scaffolds associated with the stable aneural AChR clusters ( Geng et al . , 2008 ) . Therefore , we hypothesize that MT1-MMP facilitates the disassembly of extracellular ( e . g . ECM proteins ) and/or intracellular ( e . g . cytoskeletal proteins ) molecular scaffolds that are tightly associated with aneural AChR clusters , such that AChR molecules from dispersing aneural clusters can be recruited to synaptic clusters in response to nerve induction . In a minority of cases , some aneural and nerve-induced AChR clusters can be found at the top muscle surface , where no ECM proteins are directly involved . Particularly , those top aneural AChR clusters are spontaneously formed with possibly no physiological relevance , whether MT1-MMP regulates the formation and remodeling of ECM-independent top AChR clusters remains unknown . Collectively , our present study further supported that postsynaptic MT1-MMP plays a dual-functional role in regulating the formation and dispersal of synaptic and aneural AChR clusters , respectively , via focal ECM degradation . MT1-MMP is the best characterized and the most prevalent isoform of the large MMP family , including membrane-type or soluble MMPs . MT1-MMP deficiency causes myogenic impediments and central nucleation of myofibers , which are typically found in muscular dystrophy ( Ohtake et al . , 2006 ) . Besides , MT1-MMP knockout mice develop multiple abnormalities and die between 50–90 postnatal days ( Holmbeck et al . , 1999 ) , suggesting that the possibility of functional redundancy by other MMP isoforms is low . In this study , we used mutant mice lacking MT1-MMP in all cells and observed that the density of aneural AChR clusters was significantly higher than that in wild-type mice at E13 . 5 ( Figure 10D ) , consistent with the in vitro findings on the involvement of MT1-MMP in the recruitment of AChR molecules from aneural to synaptic AChR clusters ( Figure 9 ) . However , at E18 . 5 , the density of aneural AChR clusters between MT1-MMP-/- and wild-type mice was comparable . While aneural AChR clusters in MT1-MMP-/- mice were not recruited to synaptic AChR clusters , they were subjected to the regular metabolic turnover/degradation , leading to the disappearance of aneural AChR clusters at E18 . 5 . This is supported by previous studies showing that the half-life of extra-synaptic AChRs is only about 19 hr , compared with 8–14 days for synaptic AChRs at mature NMJs ( Chang and Huang , 1975; Steinbach et al . , 1979 ) . On the other hand , reduced synaptic AChR density in MT1-MMP-/- mice ( Figure 10E ) is attributed by ( 1 ) significantly declined recruitment of AChR molecules from aneural clusters , and ( 2 ) reduced amount of newly inserted AChRs into the postsynaptic regions . These data provided a solid validation of our in vitro findings that demonstrate the important role of MT1-MMP in regulating the recruitment of pre-existing and newly inserted AChRs to the nerve-muscle contact sites ( Figure 8 ) . It is also important to note that the phrenic nerve in MT1-MMP-/- mice showed a significant reduction of axonal growth and arborization in the diaphragm muscles at both E13 . 5 and E18 . 5 ( Figure 10G ) . This phenotype on defects in axonal growth could be explained by the retrograde signaling initiated by postsynaptic MT1-MMP , which could mediate the extracellular cleavage of Lrp4 to generate ecto-Lrp4 fragments that activate agrin signaling in trans ( Wu et al . , 2012 ) . In fact , Lrp4 is also considered as a retrograde signal to induce presynaptic differentiation in vivo ( Yumoto et al . , 2012 ) . Alternatively , postsynaptic MT1-MMP could also mediate the proteolytic conversion of proBDNF to mature BDNF ( Je et al . , 2012 ) that promotes neuronal survival and outgrowth in cultured neurons ( Peng et al . , 2003 ) . Nevertheless , by using the whole-body MT1-MMP knockout mice in the present study , we cannot rule out the possible secondary effects on postsynaptic AChR clustering and remodeling that are contributed by the axonal growth and presynaptic differentiation defects in the absence of neuronal MT1-MMP . Therefore , our future studies plan to develop muscle-specific inducible MT1-MMP knockout animal models to further understand how muscle MT1-MMP controls the formation and maturation of NMJs in vivo . Particularly , it would be of great interest to address whether synaptic maturation , as reflected by the plaque-to-pretzel topological transition of AChR clusters during the early postnatal stages in mice , is critically driven by neuronal or muscle MT1-MMP . From a clinical perspective , increased serum levels of several MMPs have been identified in patients with myasthenia gravis , an autoimmune neuromuscular disease ( Helgeland et al . , 2011 ) . As inhibiting MMP activity could greatly stabilize aneural AChR clusters against spontaneous and nerve-induced dispersal ( Figure 2I and Figure 7D ) , it would be interesting to test if manipulating muscle MT1-MMP expression level or its activity can suppress the pathogenic action in causing NMJ disassembly using our recently established Xenopus cell-based assay for investigating the pathogenesis of myasthenia gravis ( Chan et al . , 2017; Yeo et al . , 2015 ) . Given the physiological and pathophysiological roles of MMPs at the NMJs , understanding the cell biology of MT1-MMP trafficking and surface delivery in this study may provide additional insight into the regulation of synaptic function and dysfunction by modulating MMP trafficking and activity in neuromuscular development and diseases . Adult Xenopus laevis animals were purchased from Xenopus 1 . Xenopus oocytes were fertilized in vitro , and the embryos were raised in Holtfreter’s solution ( vol/vol; 60 mM NaCl , 0 . 6 mM KCl , 0 . 9 mM CaCl2 , 0 . 2 mM NaHCO3 , pH 7 . 4 ) at 22°C . 20–100 pg of DNA constructs encoding MT1-MMP-pHluorin , MT1-MMP-mCherry ( a gift from Dr Cheng-Han Yu , The University of Hong Kong ) or EB1-GFP ( Addgene , 39299 ) were microinjected into 1 cell stage Xenopus embryos using an oocyte injector , Nanoject ( Drummond Scientific ) . GFP- or mCherry-expressing embryos were screened for primary culture preparation , as previously described ( Lee et al . , 2009; Peng et al . , 1991 ) . In short , dorsal parts of wild-type or microinjected embryos at Nieuwkoop and Faber stage 19–22 were dissected . After the enzymatic digestion by collagenase ( Sigma , C98191G ) , myotomal tissues and neural tubes were isolated , followed by dissociation with calcium-magnesium-free solution . The dissociated cells were plated on either coverslips or glass-bottom dishes coated with a mixture of cell attachment substrate , entactin-collagen IV-laminin ( ECL ) ( Millipore , 08–100 ) , or specific ECM proteins including laminin ( Sigma , L2020 ) , collagen I ( Fisher Scientific , C354249 ) , and gelatin ( Sigma , G1393 ) at 10 µg/ml , or poly-D-lysine ( Sigma , P1024 ) at 100 µg/ml . Cells were maintained in culture medium containing 87% Steinberg’s solution ( vol/vol; 60 mM NaCl ( Sigma , S3014 ) , 0 . 67 mM KCl ( Sigma , P5405 ) , 0 . 35 mM Ca ( NO3 ) 2 ( Sigma , 202976 ) , 0 . 83 mM MgSO4 ( Sigma , M2773 ) , 10 mM HEPES ( Sigma , H3375 ) , pH 7 . 4 ) , 10% Leibovitz's L-15 medium ( Sigma , L4386 ) ( vol/vol ) , 1% fetal bovine serum ( Gibco , 10270 ) ( vol/vol ) , 1% penicillin/streptomycin ( Thermo Fisher Scientific , 15140122 ) ( vol/vol ) and 1% gentamicin sulfate ( Thermo Fisher Scientific , 15750060 ) ( vol/vol ) . Muscle cells were kept at 22°C for at least 24 hr to allow cell attachment and development of aneural AChR cluster before treatments , if any . To make nerve-muscle co-cultures , dissociated spinal neurons were plated into 2 d-old muscle cultures and grew for 1 day before imaging . Most of the nerve-contacted sites can be found in the basal membrane of muscle cells on ECM-coated substrates , where the growth cones of spinal neurons are able to crawl and migrate under the muscle cells ( Anderson and Cohen , 1977a; Anderson et al . , 1977b ) . All experiments involving Xenopus frogs and embryos were carried out in accordance with the approved protocol ( #4627–18 ) by the Committee on the Use of Live Animals in Teaching and Research ( CULATR ) of The University of Hong Kong . Authenticated C2C12 cell line was purchased from ATCC . Cell identity has been confirmed by STR profiling and cell line was found to be free of Mycoplasma . Cells were cultured in Dulbecco’s Modified Eagle’s Medium ( DMEM ) containing 20% fetal calf serum supplemented with penicillin/streptomycin . Only cells with low passage ( <5 ) were used in this study . After trypsinization , cells were cultured on glass coverslips coated with 1 mg/ml fluorescent gelatin . To induce myoblast fusion , the culture medium was replaced with the fusion medium containing 2% horse serum ( Thermo Fisher Scientific , 16050130 ) in DMEM supplemented with penicillin/streptomycin . MT1-MMP+/- mice in C57BL/6 background were obtained from Motoharu Seiki ( Kanazawa University , Japan ) . In short , exons 1–5 encoding the catalytic domain of MT1-MMP were substituted by LacZ coding sequence ( Sakamoto and Seiki , 2009 ) . MT1-MMP+/- mice were backcrossed against C57BL/6 background mice at least 12 times . MT1-MMP+/- mice were mated to produce homozygous MT1-MMP-/- mice , which were confirmed by genotyping . All animal experiments were carried out in accordance with the approved protocol ( #5149–19 ) by the Committee on the Use of Live Animals in Teaching and Research ( CULATR ) of The University of Hong Kong . 6-week-old rats received anesthesia by intraperitoneal injection of 130 mg/kg ketamine with 13 mg/kg xylazine . After anesthesia , the left sciatic nerve was exposed with a lateral longitudinal straight incision from the greater trochanter to mid-thigh , and then by blunt dissection between the quadriceps femoris and biceps femoris . After clearing the surrounding connective tissues , the left sciatic nerve was cut with scissors . A sham operation was performed , where the sciatic nerve was exposed , and then the skin was closed immediately afterwards . After the surgery , the rats were returned to the cages in a pathogen-free environment , in a 12 hr light/dark cycle and with water and food ad libitum . Rats were euthanized 4 days after surgery with an overdose of pentobarbital ( 300 mg/kg ) . Then , soleus muscles were harvested from the control and denervated rats . After the fixation of the harvested muscles with 2% paraformaldehyde for 30 min , the tissues were then covered with tissue-tek O . C . T . compound ( Sakura ) , and frozen with methanol in liquid nitrogen . Serial longitudinal and cross sections ( 20 μm thick ) were cut in a cryostat microtome at −15°C ( CryoStar NX50 Cryostat , Thermo Scientific ) . All animal experiments were carried out in accordance with the approved protocol by the Committee on the Use of Live Animals in Teaching and Research ( CULATR ) of The University of Hong Kong . Knockdown of endogenous proteins in Xenopus was achieved by embryonic injection of antisense morpholino oligonucleotides ( MO , Gene Tools ) , which bind to the target mRNA sequence to block its protein translation . The following MO sequences were used in this study: Xenopus MT1-MMP MO: 5′-CCAGG CTGCT CTCAG AGGCT CCATC-3′ , Xenopus CLASP MO: 5’-GCCAG TAGTC CATTC CCTGT TCCAT-3’ , and standard control MO: 5’-CCTCT TACCT CAGTT ACAAT TTATA-3’ . To visualize the presence of MO in the microinjected embryos , MOs were co-injected with Alexa Fluor 488- or Alexa Fluor 546-conjugated dextran ( Thermo Fisher Scientific , D22910 or D22911 ) as a cell lineage tracer . The effectiveness of MO-mediated knockdown of endogenous proteins was validated by Western blot analyses . For the experiments studying the effect of MMP inhibitors on the formation of aneural AChR clusters , 5 μM BB-94 ( ApexBio , A2577 ) or 10 μM Marimastat/BB-2516 ( ApexBio , A4049 ) was added in the culture medium before muscle cell plating . For the experiments studying the nerve-induced or agrin bead-induced AChR clusters , BB-94 or BB-2516 was added to the muscle cultures about 1 hr before adding the spinal neurons or agrin beads . Surface AChRs were labeled with 0 . 1 μM tetramethylrhodamine- , Alexa Fluor 488- , or Alexa Fluor 647-conjugated α-bungarotoxin ( Thermo Fisher Scientific , T1175 , B13422 , or B35450 ) for 45 min in culture medium , followed by extensive washing with culture medium . To differentially label the pre-existing and newly-inserted pools of AChRs ( Lee et al . , 2009 ) , the pre-existing AChRs were first labeled with tetramethylrhodamine-conjugated α-bungarotoxin for 45 min , followed by saturating all unlabeled surface AChRs with a high dose ( 6 μM ) of unconjugated α-bungarotoxin ( Thermo Fisher Scientific , B1601 ) for 30 min and then washed extensively with culture medium for at least 3 times . After 1 day , newly inserted AChRs were labeled with 1 μM Alexa Fluor 488-conjugated α-bungarotoxin . Glass coverslips with live cultured cells were then mounted on custom-made sealed chambers containing culture medium for live-cell imaging . Glass coverslips ( Fisher Scientific , 12-545-82 ) or glass-bottom dishes ( MatTek , P35G-1 . 5–14 . C-GRID ) were coated with 1 mg/ml Oregon Green 488-gelatin or FITC-gelatin ( Thermo Fisher Scientific , G13186 or G13187 ) for 10 min and followed by cross-linking with 0 . 5% glutaraldehyde ( Sigma , G5882 ) in PBS for 15 min . After that , gelatin-coated coverslips or glass-bottom dishes were treated with 5 mg/ml NaBH4 ( Sigma , 452882 ) for 3 min . Dissociated cells were plated on fluorescent gelatin-coated coverslips or glass-bottom dishes and were maintained for different time periods with or without treatment , as specified . All images were taken using the same acquisition settings across different experimental groups . The extent of gelatin degradation was quantified using ImageJ ( NIH ) , by which the mean intensity of fluorescent gelatin signals within the regions of aneural AChR clusters or nerve-muscle contact sites was measured , and then normalized with the mean intensity at the adjacent region in the same cell ( except in MT1-MMP-mCherry overexpression experiments , the mean intensity at the adjacent region without cell attachment was used for normalization ) . Dissociated Xenopus spinal neurons were first plated on ECL-coated coverslips and cultured for 1 day . Spinal neurons were then digested and removed with 1% Triton X-100 in PBS for 5 min , followed by extensive washing with PBS for at least 5 times for 5 hr . For the experimental groups , pharmacological agents were added into the culture medium before plating the dissociated muscle cells . After 1 day in culture , the location of agrin tracks was visualized by agrin immunostaining ( 1:100; DSHB , 6D2 ) and then confirmed if neurites were not found in the phase contrast images . Cultured cells were either fixed with 4% paraformaldehyde ( Thermo Fisher Scientific , 28908 ) in PBS for 15 min , followed by permeabilization with 0 . 5% Triton X-100 in PBS for 10 min or fixed with −20°C 95% ethanol for 5 min . For EB1 immunostaining , muscle cells were fixed with 90% methanol containing 50 mM EGTA at −20°C for 15 min , followed by further fixation with 4% PFA at room temperature for 10 min . The fixed cells were washed extensively with PBS for at least 3 times , followed by blocking overnight with 2% Bovine Serum Albumin ( Sigma , A9418 ) at 4°C . Cells were incubated with primary antibodies , including β-tubulin ( 1:1000; DSHB , 6G7-s ) , cortactin ( 1:500; Santa Cruz Biotechnology , sc11408 ) , paxillin ( 1:100; Biolegend , 624001 ) , p34-Arc ( 1:100; Millipore , 07–227 ) , talin ( 1;100; Sigma , T3287 ) , vinculin ( 1:100; Sigma , V4504 ) , laminin ( 1:100; Thermo Fisher Scientific , PA516287 ) , EB1 ( 1:100; BD Biosciences , 610534 ) , Synaptobrevin/VAMP1 ( 1:300; Synaptic Systems , 104002 ) , MT1-MMP ( 1:50; Millipore , MAB3328 ) , or ADF/cofilin ( 1:500; a gift from Dr James Bamburg , Colorado State University ) at room temperature for 2 hr , followed by fluorophore-conjugated secondary antibodies ( Thermo Fisher Scientific ) for 45 min . For MT1-MMP immunostaining , muscle cells were fixed with −20°C 100% methanol for 5 min . Alexa Fluor 488 Tyramide SuperBoost kit ( Thermo Fisher Scientific , B40912 ) was also used . In short , endogenous peroxidase activity was first quenched by incubating the fixed muscle cells with 3% hydrogen peroxide for 1 hr , followed by blocking with 2% Bovine Serum Albumin for 1 hr at room temperature . The fixed muscle cells were then stained with MT1-MMP primary antibody at 4°C overnight , followed by incubating with poly-HRP-conjugated secondary antibody for 1 hr at room temperature . Endogenous MT1-MMP signals were developed by incubating the cells with Alexa Fluor 488 tyramide solution for 5 min . Coverslips were then mounted on glass slides with the anti-bleaching reagent fluoromount-G ( Thermo Fisher Scientific , 00-4958-02 ) for later observation . For mouse diaphragm muscle tissues , the entire diaphragms were dissected and fixed in 2% paraformaldehyde ( PFA ) in PBS at room temperature for 1 hr , and then incubated with 0 . 1 M glycine in PBS for 30 min . The fixed tissues were labeled with 0 . 1 μM tetramethylrhodamine α-bungarotoxin at room temperature for 30 min . After extensive washing with PBS , tissues were labeled with the primary antibody against neurofilament NF200 ( 1:1000; Sigma , N4142 ) in the blocking buffer ( 1% BSA and 0 . 3% Triton X-100 ) at 4°C for 2 days . After extensive washing with 0 . 5% Triton X-100 in PBS , tissues were incubated overnight with Alexa Fluor 488-conjugated secondary antibody at 4°C . Diaphragm muscles were flat-mounted with fluoromount-G onto glass slides . For rat soleus muscle tissues , the sections were incubated with 0 . 1 M glycine for 30 min to remove autofluorescence background and permeabilized by methanol . After blocking with 2% BSA and 0 . 2% Triton X-100 at room temperature for 2 hr , the sections were labeled with primary antibody against MT1-MMP ( 1:100; Santa Cruz Biotechnology , sc-373908 and 1:100; Millipore , MAB3328 ) , neurofilament NF200 ( 1:500; Sigma , N4142 ) , and synaptophysin ( 1:1000; Abcam , ab32127 ) for 2 days at 4°C . For MT1-MMP pre-incubation , 1 μg recombinant human MT1-MMP ( R and D Systems , 918-MP-010 ) was pre-incubated with 1 μg MT1-MMP antibody for 2 hr at room temperature . After extensive washing with PBS , the muscle sections were labeled with fluorophore-conjugated secondary antibodies ( Thermo Fisher Scientific ) for 2 hr at room temperature . Muscle sections were mounted with DAPI-Fluoromount-G ( Electron Microscopy Sciences , 17984–24 ) . Xenopus embryos at Nieuwkoop and Faber stage 19–22 were homogenized in RIPA buffer in the presence of protease inhibitor cocktail and EDTA , followed by incubation on ice for 5 min . After high-speed centrifugation ( 15000 x g ) , the supernatant was collected for protein concentration measurement using a BCA protein assay kit ( Thermo Fisher Scientific , 23227 ) . 30 μg protein lysates were used for protein separation by SDS-PAGE , and then blotted onto Immobilon-P membrane ( Millipore , IPVH00010 ) . After blocking with 5% non-fat milk in TBST , the blots were probed for the following primary antibodies: MT1-MMP catalytic domain ( 1:1000; Millipore , AB6005; kindly provided by Professor Timothy Gomez , University of Wisconsin-Madison ) ; CLASP1 ( 1:250; Abcam , Ab85919 ) ; anti-β actin ( 1:1000; Sigma , A2228 ) ; anti-α tubulin ( 1:2000; Sigma , T6074 ) overnight at 4°C . After extensive washing with TBST , the blots were reacted with secondary antibodies conjugated with HRP . Signals were detected by ECL western blotting substrate ( Thermo Fisher Scientific , 32106 ) , and images were taken by ChemiDoc XRS+ System ( Bio-Rad ) . Wide-field fluorescence imaging was performed on inverted microscopes ( IX81 or IX83 , Olympus ) using oil immersion PlanApo 60X NA 1 . 42 objective lens . On IX81 , digital still images were captured by iXon EMCCD camera ( Andor ) through the software ) through the software cell^R ( Olympus ) . On IX83 , digital still and time-lapse images were captured by ORCA-Flash4 . 0 LT+ digital CMOS camera ( Hamamatsu ) through the software MicroManager ( Open Imaging ) ( Edelstein et al . , 2014 ) . In FRAP experiments , total internal reflection fluorescence ( TIRF ) mode using Axio TIRF unit fitted in an inverted microscope equipped with oil immersion 100X NA 1 . 46 DIC objective lens was used . Images were captured through Metamorph ( Molecular Devices ) by Evolve 512 EMCCD camera ( Photometrics ) . To obtain the baseline of fluorescence intensity before photobleaching , 5 to 6 images were taken with 3 s intervals on cells expressing EB1-GFP and 1 s intervals on cells expressing MT1-MMP-pHluorin or MT1-MMP-mCherry . Photobleaching was performed by using Sapphire laser line ( 488 nm ) with 50% laser intensity , which was kept the same for all FRAP experiments . The photobleaching duration was adjusted based on the signal intensity , ranging from 0 . 13 to 4 . 5 s for photobleaching areas between ~ 27 and 34 μm2 . After photobleaching , images were taken every 3 s until the fluorescence intensity has reached the plateau level . In the experimental groups involving photobleaching of aneural AChR clusters , interactive bleaching function was used in a confocal microscope ( LSM 800 , Carl Zeiss ) using 20X NA 1 . 4 DIC objective lens . Photobleaching was performed by using an argon laser ( 488 nm ) with a 100% laser power . Identical settings were applied in all photobleaching experiments . A fluorescence image was taken immediately after photobleaching to ensure the signals of aneural AChR clusters had been completely bleached . For imaging of mouse diaphragm muscles and rat soleus muscles , multiple tile-scanned , z-stack images were captured on a confocal microscope ( LSM 800 , Carl Zeiss ) using 20X NA 1 . 4 DIC objective lens or oil immersion 63X NA 1 . 4 DIC objective lens . Images were captured with 32-channel GaAsp photomultiplier modules through the software ZEN 2 . 3 ( Carl Zeiss ) . All acquisition settings ( i . e . laser intensity and gain ) were kept the same for different experimental groups in the same experiment . All acquired images were processed and analyzed by ImageJ ( NIH ) or Imaris ( Bitplane ) . To quantify the percentage of AChR top and bottom clusters , muscle cells were first identified in the phase-contrast channel , and then the presence of top or bottom aneural AChR clusters was scored in the fluorescence channel . To quantify the extent of fluorescent gelatin degradation , the intensity of fluorescent gelatin in the region of interest ( ROI , created based on the shape and location of AChR clusters ) was measured and then normalized with the intensity of fluorescent gelatin at the adjacent area in the same muscle cell . For experiments involving MT1-MMP-mCherry overexpression , the fluorescent gelatin intensity in the region of AChR clusters was normalized with the intensity in the region without cell attachment . To quantify the change in the area of aneural AChR clusters , ROI was created at the region of AChR clusters in the image taken at the first time point , and then applied to the images taken at subsequent time points for measuring and calculating the percentage change in cluster size . To quantify the normalized intensity of aneural AChR clusters , the fluorescence intensity of AChRs in the ROI was first measured in all experimental groups and then normalized to the average AChR intensity in control group of the same experiment . To quantify the density and speed of EB1 comets , ROI for AChR region was created based on the shape and location of AChR clusters , while ROI for non-AChR region was also created at the adjacent area in the same muscle cell . The number and speed of EB1 comets were measured from multiple frames in 7 s time-lapse images using TrackMate plugin in ImageJ ( Tinevez et al . , 2017 ) . The density of EB1 comets was calculated by dividing the number of EB1 comets by the ROI area . To quantify the FRAP experiments , the relative fluorescence intensity in the photobleached area was first measured and then normalized by the mean fluorescence intensity of the same region in the images acquired before photobleaching . The recovery half-time was determined by the time required for 50% recovery of fluorescence signals in the FRAP regions . To quantify the number of MT1-MMP-pHluorin surface insertion events , ROI was created at the region of aneural AChR clusters . After photobleaching , the number of MT1-MMP-pHluorin signals in the ROI was measured from multiple frames in 60 s time-lapse images using TrackMate plugin in ImageJ . It was then normalized to the area of aneural AChR clusters . To quantify the percentage of nerve-muscle contacts with AChR clustering , muscle cells with nerve contact at the bottom surface were first identified in the phase-contrast images , and then the presence of AChR clusters was scored in the fluorescence images . To quantify the fluorescence intensity of nerve-induced AChR clusters per unit length of nerve-muscle contact , the integrated fluorescence value of AChR signals was first measured , and then it was normalized to the area of nerve-muscle contacts as determined and measured in the phase-contrast images . To quantify the density of aneural and nerve-induced synaptic AChR clusters in mouse diaphragm muscles , the total number of AChR clusters was first measured from stack images using Imaris software . The number of nerve-induced synaptic AChR clusters was calculated based on the overlapping of signals between AChR clusters and neurofilaments . The number of aneural AChR cluster was then calculated by subtracting the number of nerve-induced synaptic AChR clusters from the total number of AChR clusters . The calculated numbers were then normalized to the length of the main nerve trunk . Endplate band width was defined by the average distance between two farthest AChR clusters in the muscle fibers along the main nerve trunk . The length of nerve branches was quantified by measuring the average axonal length from the main nerve trunk . Approximately 100 measurements were taken in each diaphragm , and the quantification of the endplate band width and length of branches was performed using the whole diaphragm from at least 3 animals in each genotype . In all figures , mean and SEM values were shown in the graphs , unless otherwise specified . The numbers of biological replications and the statistical tests applied were specified in the figure legends .	Voluntary movement relies on skeletal muscle cells and nerve cells being able to communicate with one another . This communication occurs at a specialized region called the neuromuscular junction , or NMJ for short . These junctions are surrounded by a meshwork of proteins , known as the matrix , which structurally supports the nerve and muscle cells . Muscle cells contain proteins called acetylcholine receptors on their cell surface . When these receptors cluster together at the NMJ , this allows nerve cells to communicate with the muscle cell and tell the muscle to contract . However , these clusters can also form spontaneously without the help of nerve cells at regions away from the communication site . Alongside these spontaneous clusters of acetylcholine receptors are dynamic actin-enriched structures . These structures are responsible for releasing enzymes that digest the surrounding matrix and are commonly found in migrating cells . But as skeletal muscle cells do not migrate , it remained unclear what purpose these structures serve at the NMJ . Now , Chan et al . have used advanced microscopy techniques to show how these actin-enriched structures can help acetylcholine receptors cluster together at the site of communication between the nerve and muscle cells . The experiments showed that these structures direct a molecule called MT1-MMP to the muscle surface . This molecule then clears the surrounding matrix so that signals sent from the nerve can be effectively deposited at the narrow space between these two cells . When the muscle cells receive this initiating signal , acetylcholine receptors are recruited from the spontaneously formed clusters to the communication site , allowing the muscle to contract . When MT1-MMP was experimentally eliminated in mice , this disrupted the recruitment of acetylcholine receptors to the NMJ . Overall , these experiments help researchers understand how clearing the matrix between nerve and muscle cells contributes to the deposition of factors that build the communication site at developing NMJs . In the future this might help develop treatments for movement disorders caused by abnormalities that affect the clearing of matrix proteins in these junctions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2020	Site-directed MT1-MMP trafficking and surface insertion regulate AChR clustering and remodeling at developing NMJs
Mesenchymal stem cells ( MSCs ) and osteolineage cells contribute to the hematopoietic stem cell ( HSC ) niche in the bone marrow of long bones . However , their developmental relationships remain unclear . In this study , we demonstrate that different MSC populations in the developing marrow of long bones have distinct functions . Proliferative mesoderm-derived nestin− MSCs participate in fetal skeletogenesis and lose MSC activity soon after birth . In contrast , quiescent neural crest-derived nestin+ cells preserve MSC activity , but do not generate fetal chondrocytes . Instead , they differentiate into HSC niche-forming MSCs , helping to establish the HSC niche by secreting Cxcl12 . Perineural migration of these cells to the bone marrow requires the ErbB3 receptor . The neonatal Nestin-GFP+ Pdgfrα− cell population also contains Schwann cell precursors , but does not comprise mature Schwann cells . Thus , in the developing bone marrow HSC niche-forming MSCs share a common origin with sympathetic peripheral neurons and glial cells , and ontogenically distinct MSCs have non-overlapping functions in endochondrogenesis and HSC niche formation . Bone marrow stromal cells ( BMSCs ) are a heterogeneous population . The different mesenchymal cell types might either arise from a variety of resident progenitors or might ultimately be derived from a single population of rare MSCs ( Caplan , 1991 ) . In adult mammals , a multiple origin of skeletal MSCs is suggested by the distinct germ layer derivation of different bone structures , with craniofacial bones generated by the neuroectoderm , whereas the axial and appendicular bones are respectively derived from paraxial and lateral mesoderm . Mesoderm generates chondrocytes , which are progressively replaced by osteoblasts through the process of endochondral ossification ( Olsen et al . , 2000 ) . MSCs share cell-surface markers and localization with pericytes , suggesting that some pericytes might be MSCs ( Crisan et al . , 2008 ) . However , it remains unclear whether the bone marrow hosts ontogenically distinct MSCs in the same bones and whether they are endowed with specific functions . In adult bone marrow , a variety of mesenchymal cells regulate HSCs ( Calvi et al . , 2003; Zhang et al . , 2003; Arai et al . , 2004; Sacchetti et al . , 2007; Chan et al . , 2008; Naveiras et al . , 2009; Mendez-Ferrer et al . , 2010; Omatsu et al . , 2010; Raaijmakers et al . , 2010 ) . Nevertheless , the specialized functions and developmental origin of these cells are largely unknown . Adult HSCs are also regulated by certain other non-hematopoietic lineages , including endothelial cells ( Avecilla et al . , 2004; Kiel et al . , 2005; Ding et al . , 2012 ) , sympathetic neurons and associated non-myelinating Schwann cells ( Katayama et al . , 2006; Spiegel et al . , 2007; Mendez-Ferrer et al . , 2008; Yamazaki et al . , 2011 ) , perivascular cells expressing the leptin receptor ( Ding et al . , 2012 ) and mesodermal derivatives ( Greenbaum et al . , 2013 ) . However , the relationships and potential overlap among these populations remain unclear . It is also not known whether MSCs that form the HSC niche also generate other stromal cells or are a specialized population that arises earlier in embryogenesis and persists into adulthood . In this study , we investigated the developmental origin and functions of MSCs in the primordial marrow of long bones . We show that , like peripheral neural and glial cells , HSC niche-forming MSCs in perinatal bone marrow arise from the trunk neural crest and make only a modest contribution to endochondrogenesis . Thus , whereas mesoderm-derived MSCs are mostly involved in endochondral ossification , neural crest-derived cells have a specialized function in establishing the HSC niche in the developing marrow of the same bones . These results provide compelling evidence for functional segregation of MSCs derived from different germ layers . The data also show that three HSC niche components—peripheral sympathetic neurons , Schwann cells , and MSCs—share a common origin . In adult mouse bone marrow , stromal cells expressing the green fluorescent protein ( GFP ) driven by the regulatory elements of nestin promoter ( Nes-GFP+ ) display features of both MSCs and HSC niche cells ( Mendez-Ferrer et al . , 2010 ) . This finding prompted us to characterize Nes-GFP+ cells during marrow development in limb bones . GFP+ cells were already present in E16 . 5 bone marrow , associated preferentially with blood vessels infiltrating the cartilage scaffold ( Figure 1—figure supplement 1A–C ) . At E18 . 5 Nes-GFP+ cells were frequently associated with arterioles and sprouting endothelial cells within the osteochondral junction ( Figure 1A–C ) . Fetal bone marrow Nes-GFP+ cells were heterogeneous , composed of a majority of BMSCs but also including a small subset of CD31+ putative endothelial cells that increased during the postnatal period ( Figure 1D , E and Figure 1—figure supplement 1D ) . Compared with Nes-GFP- BMSCs , the Nes-GFP+ cell population was enriched in endogenous Nestin mRNA expression ( Figure 1F ) . Arterioles were associated with an intense fluorescence microscopy signal , due to the presence of several concentric GFP+ cells , including an outer layer that expressed smooth muscle actin and an inner layer of endothelial cells ( Figure 1G and Figure 1—figure supplement 1E , F ) . Fetal bone marrow Nes-GFP+ cells were distinct from S100-expressing chondrocytes and osteoblastic cells genetically labeled with the 2 . 3-kilobase proximal fragment of the α1 ( I ) -collagen promoter ( Dacquin et al . , 2002 ) ( Figure 1H–J ) . Contrasting the marked proliferation of Nes-GFP- BMSCs in perinatal life , Nes-GFP+ cells remained mostly quiescent ( Figure 1K and Figure 1—figure supplement 1G ) . As a result , whereas Nes-GFP- BMSCs steadily expanded , Nes-GFP+ BMSC number did not change significantly ( Figure 1L ) . Fetal bone marrow Nes-GFP+ cells thus include a small subset ( <10% ) of endothelial cells and a large population of non-endothelial stromal cells ( >90% ) . Unlike Nes-GFP- stromal cells , Nes-GFP+ cells proliferate slowly and do not express osteochondral protein cell markers . 10 . 7554/eLife . 03696 . 003Figure 1 . Fetal bone marrow nestin+ cells proliferate slowly and are distinct from osteochondral cells . ( A–C ) Nes-GFP+ cells in fetal bones undergoing endochondral ossification . Whole-mount confocal projection of E18 . 5 Nes-Gfp femoral bone marrow stained with CD31 ( magenta ) to mark endothelium . Note the perivascular distribution of GFP+ cells ( green ) in arterioles ( B–B′′ ) and small vessels invading the primary spongiosa ( C–C′′ ) . ( D–E ) Nes-Gfp transgene is expressed by a subset of bone marrow endothelial cells . Flow cytometry histograms show the frequency of CD45− Nes-GFP+ cells expressing CD31 . ( F ) Endogenous Nestin mRNA expression measured by qPCR in stromal populations isolated from Nes-Gfp mice at the indicated stages ( mean ± SD , n = 3–5 ) . ( G ) Nes-Gfp bone marrow section stained with smooth muscle actin antibodies ( αSma , red; asterisks ) to reveal arterioles . ( H ) Limb section from an E17 . 5 Nes-Gfp;Col2 . 3-Cre;KFP embryo showing Nes-GFP+ ( green ) and osteoblasts identified with antibodies to Katushka ( KFP ) protein ( red ) , driven by the 2 . 3-kb proximal fragment of the α1 ( I ) -collagen promoter . Arrowheads , endosteal surface . ( I ) Metaphysis of E17 . 5 Nes-Gfp embryo showing S100+ chondrocytes ( red ) . ( J ) Magnified view of boxed area in ( I ) . ( K ) Representative cell cycle profiles of bone marrow stromal Nes-GFP+/- cells at early postnatal stages . Frequencies of cells in G2/S-M ( % ) are indicated . ( L ) Number of stromal Nes-GFP+/− cells in postnatal bone marrow ( mean ± SEM , n = 3–4 ) . Scale bars: 200 μm ( A , A′ , B′′ , C , H ) , 100 μm ( G , I and J ) ; ( A′ , G–J ) dashed line indicates bone contour . BM , bone marrow; C , cartilage; PS , primary spongiosa . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 00310 . 7554/eLife . 03696 . 004Figure 1—figure supplement 1 . Perivascular and endothelial Nes-GFP+ cells invade the incipient bone marrow associated with blood vessels . ( A ) Longitudinal section of E16 . 5 Nes-Gfp forelimb . A profuse perivascular network of GFP+ cells can be observed , mostly associated with arterial blood vessels . ( B ) High magnification detail ( inset B’ ) of arterioles containing Nes-GFP+ cells during invasion of the distal shaft of fetal bone . ( C ) Section through an E16 . 5 Nes-Gfp metaphyseal region undergoing vascularization , revealed by CD31 immunostaining of endothelial cells ( red ) . ( D ) Representative FACS histograms of anti-CD31-stained CD45- bone marrow cells from 2-week old Nes-Gfp mice . The frequency of GFP+ putative endothelial cells is indicated . ( E and F ) Confocal high magnification detail is showing several layers of perivascular GFP+ cells ( arrows ) encircling an arteriole . Innermost endothelial cells immunostained with CD31 ( red ) also expressed GFP ( yellow overlay , asterisks ) . ( G ) Femoral bone marrow section from newborn Nes-Gfp mouse stained with Ki67 ( red ) to label proliferative cells . Arrows indicate GFP+ Ki67+ cells; arrowheads depict GFP+ Ki67− cells; dashed line marks bone contour . ( A-C , E , G ) Nuclei were counterstained with DAPI ( gray ) . Scale bars: 500 μm ( A ) ; 200 μm ( B ) ; 100 μm ( C , G ) ; 50 μm ( E and F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 004 We next studied whether Nes-GFP+ cells displayed osteoprogenitor activity in fetal bone marrow . The axial and appendicular skeleton is thought to originate solely from mesoderm . During endochondral ossification , cartilage is progressively replaced by osteoblast precursors that express the transcription factor osterix and infiltrate the perichondrium along the invading blood vessels ( Maes et al . , 2010 ) . To identify mesodermal derivatives , we performed lineage-tracing studies by crossing mice expressing the RCE reporter—a sensitive reporter that drives stronger GFP expression than other reporter lines ( Sousa et al . , 2009 ) —with mice expressing inducible Cre recombinase under the regulatory elements of the Hoxb6 gene , which is expressed in the lateral plate mesoderm ( Nguyen et al . , 2009 ) . The resulting double-transgenic mice were administered tamoxifen at E10 . 5 , a stage when the Hoxb6 gene is still expressed . These mice and newborn Nes-gfp embryos were analyzed for osterix protein expression , which marks cells committed to the osteoblast lineage . Unlike osteoblast precursors derived from lateral plate mesoderm , Nes-GFP+ cells in fetal-limb bone marrow did not express highly osterix protein ( Figure 2A , B ) . 10 . 7554/eLife . 03696 . 005Figure 2 . Bone marrow nestin+ cells are different from mesodermal osteo-chondroprogenitors . ( A and B ) Bone marrow sections from Nes-Gfp ( A–A′′ ) and Hoxb6-CreERT2;RCE ( B–B′′ ) E18 . 5 embryos ( tamoxifen-induced at E10 . 5 ) immunostained with Osterix antibodies ( Osx , red ) to label osteoprogenitor cells . GFP+ Osx+ mesodermal-derived osteoprogenitors are marked with asterisks ( insets 2–3 ) . ( C–C′′′ ) Perinatal recombination in Nes-CreERT2 mice efficiently targets bone marrow stromal Nes-GFP+ cells . Bone marrow section of a P7 Nes-Gfp;Nes-CreERT2;R26-Tomato mouse that received tamoxifen at birth , showing Nes-GFP+ cells ( green ) , Nes-derived progeny ( red ) , and double-positive cells ( arrowheads ) . ( D–F ) Fate mapping of the progeny of nestin+ cells and limb mesoderm in E18 . 5/19 . 5 femoral bone marrow from Nes-CreERT2;RCE ( D–D′′ , F–F′ ) and Hoxb6-CreER;RCE fetuses ( E ) . ( D ) GFP ( green ) and nuclei counterstained with DAPI ( gray ) in bone of E18 . 5 fetus induced with tamoxifen at E13 . 5 . Neither proliferating ( * ) nor hypertrophic ( ** ) chondrocytes showed GFP fluorescence ( inset 1 ) . ( D′–D′′ ) Nes-derived cells with a similar morphology and distribution to Nes-GFP+ cells were detected near the cartilage–perichondrium interface ( arrows ) and within the chondro–osseous junction ( arrowheads ) . ( E and F ) Bone marrow sections of ( E ) Hoxb6-CreER;RCE and ( F ) Nes-CreERT2;RCE E18 . 5 embryos induced with tamoxifen at E10 . 5 and E8 . 5 , respectively , stained with S100 antibodies to label chondrocytes ( red ) . High magnification views of cartilage ( inset 2 ) showing abundant double-positive chondrocytes ( arrowheads ) . ( F ) Nes-traced cells ( green ) were not chondrocytes ( red , * ) but infiltrated the chondro–osseous junction and trabecular bone ( arrowheads ) . Scale bars: 200 μm ( A–A′ , B–B′ ) , 100 μm ( A′′ , B′′ ) , 50 μm ( B′′3 , C ) . BM , bone marrow; C , cartilage; GIFM , genetic inducible fate mapping . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 00510 . 7554/eLife . 03696 . 006Figure 2—figure supplement 1 . Sub-fractionation of fetal bone marrow mesenchymal progenitors . ( A ) Neonatal Nes-Gfp bone marrow section immunostained with anti-osterix antibodies ( red ) . Right panels , perichondrium detail showing inner rim of GFP+ cells ( green ) adjacent to distinctive perichondrial osterix+ cells . Nuclei were counterstained with DAPI ( gray ) . Scale bar , 100 μm . PC , perichondrium . ( B ) Scheme showing FACS isolation of bone marrow stromal populations from Nes-Gfp mice and MSC assays . ( C ) Representative FACS plots and gating strategy to isolate stromal GFP+/- cells from Nes-Gfp bone marrow . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 006 We next performed genetically inducible fate mapping using Nes-CreERT2 mice ( Balordi and Fishell , 2007 ) . In these mice , tamoxifen administration triggers labeling of Nes-GFP+ cells and their progeny ( Figure 2C , D ) . Tamoxifen was administered at E13 . 5 ( when primary ossification centers start forming ) ( Maes et al . , 2010 ) , and at E8 . 5 , to mark earlier nestin+ embryonic precursors . Unlike Hoxb6-traced mesodermal derivatives , nestin+ cells did not contribute to cartilage formation during this period . In contrast , Nes-traced cells with a similar morphology and distribution to Nes-GFP+ cells were observed in the chondro–osseous junction ( Figure 2E , F ) . Similarly , Nes-GFP+ cells were not present inside the cartilage but were found in the innermost part of the perichondrium ( Figure 2—figure supplement 1A ) , a region enriched in MSCs ( Maes et al . , 2010; Yang et al . , 2013; Zaidi and Mendez-Ferrer , 2013 ) . The results thus show that , unlike mesodermal derivatives , Nes-GFP+ cells do not exhibit osteochondral progenitor activity in the fetal bone marrow . The lack of a contribution by nestin+ cells to fetal endochondrogenesis raised questions regarding their putative MSC properties in fetal bone marrow . We therefore measured mesenchymal progenitor activity in purified bone-marrow stromal subsets using the fibroblastic colony-forming unit ( CFU-F ) assay ( Friedenstein et al . , 1970 ) and the multipotent self-renewing sphere-forming assay ( Mendez-Ferrer et al . , 2010 ) ( Figure 2—figure supplement 1B ) . BMSCs were isolated according to Nes-GFP expression ( Figure 2—figure supplement 1C ) . CFU-F efficiency was nearly three times higher in Nes-GFP- cells than in Nes-GFP+ cells at E17 . 5 ( Figure 3A ) . Conversely , non-adherent sphere formation was markedly enriched in GFP+ cells , whereas most spheres derived from GFP− cells rapidly attached to plastic and spontaneously differentiated into adipocytes ( Figure 3B–G ) , suggesting that Nes-GFP- BMSCs are in a more committed state . Spheres formed by Nes-GFP+ bone marrow cells contained mesenchyme-like spindle-shaped GFP+ cells ( Figure 3E ) . At E18 . 5 and during the first postnatal week , CFU-F frequency was 6-fold higher in the GFP− stromal population than in GFP+ cells ( Figure 3H ) . However , at later postnatal stages , CFU-F activity was progressively restricted to Nes-GFP+ cells due to a sharp drop in activity in GFP− BMSCs ( >100-fold reduction between E18 . 5 and P14 , compared with a 0 . 5-fold reduction in Nes-GFP+ cells ) . At P7 , CFU-Fs derived from Nes-GFP- cells contained mostly preosteoblasts ( Figure 3I–K ) . The expression of genes associated with chondrocyte development was higher in Nes-GFP- than in Nes-GFP+ BMSCs at E18 . 5; in contrast , the expression of master regulators of chondrogenesis , osteogenesis , and adipogenesis was progressively enriched in postnatal Nes-GFP+ BMSCs ( Figure 3L ) , consistent with the increasing MSC enrichment in this population . Together , these results suggest that most fetal BMSCs do not express nestin and quickly differentiate towards committed skeletal precursors , losing most MSC activity by the second week after birth . In contrast , nestin+ cells conserve MSC activity throughout life . 10 . 7554/eLife . 03696 . 007Figure 3 . Perinatal enrichment of MSC activity in bone marrow nestin+ cells . ( A and B ) Fibroblast colony-forming units ( CFU-F ) and mesensphere-forming activities segregate in Nes-GFP- and Nes-GFP+ fetal bone marrow cells , respectively . Frequencies of CFU-F and mesensphere-forming efficiency in E17 . 5 Nes-Gfp embryos . ( C–E ) Representative sphere cultures from both mesenchymal subpopulations sorted from bone marrow of Nes-Gfp fetuses . Note the presence of GFP+ fibroblast-like cells ( E′ and insets1–3 ) . ( F–G ) Adherent colonies derived from GFP- population stained with Oil Red O ( red ) , to reveal mature adipocytes . ( H ) MSC activity is progressively restricted to bone marrow Nes-GFP+ cells . Frequency of CFU-Fs in cultures of stromal ( CD45− CD31− Ter119− ) GFP+/− cells , isolated from the bone marrow of Nes-Gfp mice of the indicated age . Right panels show representative CFU-Fs in cell populations from adult mice . ( I ) Frequency of osteoblastic colony-forming units ( CFU-OB ) in bone marrow stromal GFP+/− cells of the indicated age . ( J ) Representative Giemsa-stained CFU-F from 3- and 10-day old bone marrow subpopulations . ( K ) Stained CFU-OB from E18 . 5 ( alkaline phosphatase staining , left panels ) and 1-week old ( alizarin red staining , right panels ) bone marrow subpopulations . ( L ) qPCR analysis of mesenchymal genes in bone marrow stromal populations isolated from fetal ( E18 . 5 ) or 1-week old ( P7 ) Nes-Gfp mice , as depicted ( Figure 3—figure supplement 1 ) . ( A–B , H–L ) Mean ± SD , n = 3–6; p < 0 . 05 , unpaired two-tailed t test . Scale bars: 200 μm ( D , E′ , G ) , 100 μm ( G ) , 50 μm ( E′1–3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 007 Neural crest cells are characterized by nestin expression and sphere-forming ability . Although cells traced to neural crest origin have been reported in adult murine bone marrow ( Nagoshi et al . , 2008; Morikawa et al . , 2009b; Glejzer et al . , 2011; Komada et al . , 2012 ) , their precise identity , developmental dynamics , and function have remained elusive . Moreover , the neural crest-specific Wnt1-Cre line used in these studies displays ectopic Wnt1 activation . To trace neural crest derivatives , we performed genetic fate-mapping studies with a recent Wnt1-Cre2 line that does not induce ectopic Wnt1 activity ( Lewis et al . , 2013 ) . Unexpectedly , limb bones from Wnt1-Cre2;R26-Tomato double-transgenic neonates showed some neural crest-derived osteoblasts and osteocytes aligning the most recent layers of bone deposition , as well as similarly distributed chondrocytes in the outermost layers of the femur head ( Figure 4A , B ) . As expected , neural crest-traced Schwann cells expressing glial fibrillary acidic protein ( GFAP ) were also detected in the bone marrow of one-week old mice ( Figure 4—figure supplement 1A , B ) . Intriguingly , GFAP− perivascular cells with a similar morphology and distribution to Nes-GFP+ cells were also derived from Wnt1+ cells ( Figure 4—figure supplement 1C , D ) . The number of neural crest-traced osteochondral cells increased in the first postnatal week ( Figure 4C ) . By P28 , CFU-F activity was much higher in Wnt1-Cre2-traced cells than in non-neural crest-traced bone marrow stromal cells ( Figure 4D ) . These results show that the neural crest contributes to limb bones late in development . 10 . 7554/eLife . 03696 . 008Figure 4 . Contribution of trunk neural crest cells to mesenchymal lineages in long bones . ( A and B ) Fate mapping of neural crest derivatives in femoral bone marrow of neonatal Wnt1-Cre2;R26-Tomato mice . ( A ) Section through femoral distal epiphysis showing cortical neural-crest-derived chondrocytes ( arrowheads , red ) ; green signal corresponds to phalloidin staining . Nuclei were counterstained with DAPI ( gray ) . ( B ) Bone marrow section showing Wnt1-Cre2-derived Tomato+ ( red ) osteocytes . ( Inset 1 ) Neural-crest-derived osteocytes ( asterisks ) in endosteal region , showing their typical morphology revealed by phalloidin staining ( green ) . ( C–E ) The neural crest contributes to Pdgfrα+ BMSCs in long bones . ( C–C′′ ) Fluorescent signals of GFP , Tomato , and DAPI in bone marrow sections from 1-week old Wnt1-Cre2;R26-Tomato;Nes-Gfp mice . ( D ) Frequency of fibroblastic colony-forming units ( CFU-F ) in CD31- CD45- Ter119- Tomato+/- bone marrow cells sorted from 1-week old Wnt1-Cre2;R26-Tomato mice ( n = 3 ) ; N . D . , not detectable . ( D′ ) Examples of Giemsa staining ( top panel ) and Tomato fluorescence in neural crest-derived CFU-Fs . ( E ) Representative flow cytometry analysis of bone marrow stromal cells from 4-week old Wnt1-Cre2;R26-Tomato;Nes-Gfp mice . ( F ) Flow cytometry analysis of bone marrow stromal cells from Nes-Gfp;Sox10-CreERT2;R26-Tomato triple-transgenic mice stained with Pdgfrα antibody . ( E , F ) Frequencies of neural crest-traced BMSCs are indicated . Scale bars: 200 μm ( A–A′′ , C , D–D′ ) , 100 μm ( B ) , 20 μm ( B1 , C′–C′′′ ) . Dashed line depicts the bone and cartilage contour ( A–C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 00810 . 7554/eLife . 03696 . 009Figure 4—figure supplement 1 . Bone marrow Nes-GFP+ cells are distinct from mature Schwann cells . ( A–C ) Diaphysis section stained with antibodies against glial fibrillary acidic protein ( Gfap , cyan ) . Neural crest-traced cells included Gfap+ Schwann cells ( inset 1 , arrowheads ) and Gfap− putative BMSCs ( inset 2 , asterisks ) ( B-B′′ , C-C′′ ) . ( D-D′′′ ) Staining of basal lamina with collagen IV antibodies ( blue ) showing close association of neural crest-derived cells ( red ) with blood vessels . ( E ) Surface expression of Pdgfrα enriches for fetal bone marrow mesenchymal progenitors . Fibroblast colony-forming units ( CFU-F ) in bone marrow stromal cells ( CD45- CD31- Ter119- ) isolated from E17 . 5 Nes-Gfp mice according to GFP and Pdgfrα expression . ( F ) Frequency of CFU-F colonies obtained from sorted populations of Wnt1-Cre2;R26-Tomato;Nes-Gfp 1-week old pups . ( G ) Schwann cells are closely associated with distinctive Nes-GFP+ perivascular cells ( arrowheads ) . Immunofluorescence of P9 Nes-Gfp tibial bone marrow showing GFP+ cells ( green ) , Gfap+ Schwann cells ( red ) , and CD31+ endothelial cells ( pink ) ; nuclei were counterstained with DAPI ( gray ) . ( G′–G′′ ) Details of central diaphyseal region ( insets ) at high magnification . ( G′ ) Note the long Gfap+ Schwann cells extending through the arteriole , surrounded by distinctive Nes-GFP+ cells ( arrowhead ) . ( G′′ ) Magnified view of a long arteriole . Asterisks indicate Gfap- perivascular Nes-GFP+ cells . Scale bars: 200 μm ( A , D , G ) , 50 μm ( B , C , G′–G′′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 00910 . 7554/eLife . 03696 . 010Figure 4—figure supplement 2 . Contribution of trunk neural crest to bone marrow stromal lineages . ( A-A′′′ ) Bone marrow section from newborn Wnt1-Cre2;R26-Tomato pup stained with calcein to mark calcium deposition ( green ) , showing Wnt1-Cre2-traced Tomato+ osteoblasts ( red ) in calcifying areas ( asterisks ) . Scale bar = 100 μm . ( B ) Representative bone section of an E18 . 5 Sox10-CreERT2;R26-Tomato mouse showing Tomato+ osteocytes ( asterisks ) and bone-lining osteoblasts ( red ) . Dashed line depicts the bone contour . Nuclei were counterstained with DAPI ( gray ) . ( C and D ) Representative flow cytometry plots of bone marrow stained with Pdgfrα from neonatal ( D ) and 4-week old ( C ) Wnt1-Cre2;R26-Tomato;Nes-Gfp mice , after gating on the stromal ( CD45- CD31- Ter119- ) population . ( E ) Representative flow cytometry histogram of bone marrow cells from E18 . 5 Sox10-CreER;R26-Tomato;Nes-Gfp embryos , induced with tamoxifen at E9 . 5 , after gating on the tomato-bright population . Scale bars: 200 μm ( B ) , 100 μm ( A , B′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 010 We further characterized cell surface marker expression by neural crest-derived MSCs . Platelet-derived growth factor receptor alpha ( Pdgfrα ) is required in mesodermal and neural crest-derived mesenchyme during development ( Schatteman et al . , 1992; Soriano , 1997 ) . Mouse Pdgfrα+ BMSCs are highly enriched in CFU-F activity ( Takashima et al . , 2007; Morikawa et al . , 2009a ) and most adult mouse bone marrow nestin+ cells are also Pdgfrα+ ( Yamazaki et al . , 2011; Pinho et al . , 2013 ) . We found that Pdgfrα+ BMSCs were also enriched in CFU-F activity at fetal stages ( Figure 4—figure supplement 1E ) . In the bone marrow of 4-week old Nes-Gfp;Wnt1-Cre2;R26-Tomato mice , most neural crest-traced cells were also Pdgfrα+ and Nes-GFP+ ( Figure 4E ) . For confirmation , we intercrossed Nes-Gfp;R26-Tomato mice with a line expressing tamoxifen inducible Cre recombinase under the regulatory elements of the gene encoding the neural crest transcription factor Sox10 . Nes-Gfp;Sox10-CreERT2;R26-Tomato mice were administered tamoxifen at E9 . 5 to label migratory neural crest-derived cells . Similar to the situation in stage-matched Nes-Gfp mice—and also consistent with Wnt1-Cre2-traced cells—most Sox10-CreERT2-traced bone marrow stromal cells were Pdgfrα+ and Nes-GFP+ cells ( Figure 4F and Figure 4—figure supplement 2 ) . These results thus demonstrate definitively that the neural crest contributes to nestin+ BMSCs . The finding that some fetal bone marrow Nes-GFP+ cells expressed Pdgfrα while others did not prompted us to study the possible functional heterogeneity of this population . Recent work showed that most adult bone marrow nestin+ cells are Pdgfrα+ and also that nestin+ Pdgfrα+ Schwann cells contribute to HSC maintenance ( Yamazaki et al . , 2011 ) . We found that bone marrow Nes-GFP+ cells were closely associated with distinctive Gfap+ Schwann cells ( Figure 4—figure supplement 1G ) . After sorting of neonatal GFP+/− Pdgfrα+/− BMSCs ( Figure 5A ) , the two populations were analyzed by next-generation sequencing . Detection of endogenous Pdgfrα and Nes transcripts verified the isolation strategy . Interestingly , whereas Ly6a/Sca1 expression was higher in GFP−Pdgfrα+ cells ( Figure 5—figure supplement 1A , B ) , the expression levels of HSC maintenance genes ( Cxcl12 , Kitl and Angpt1 ) and the Leptin receptor , which marks HSC niche-forming mesenchymal cells ( Ding et al . , 2012 ) , was highly enriched in GFP+ Pdgfrα+ cells ( Figure 5B ) . This population also abundantly expressed other genes enriched in MSCs ( Figure 5—figure supplement 1C ) . In contrast , Nes-GFP+ Pdgfrα− cells expressed genes characteristic of Schwann cell precursors ( Sox10 , Plp1 , Erbb3 , Dhh ) but did not express mature Schwann cell genes , such as Gfap ( Figure 5C ) . Gene ontology analysis of differentially expressed genes between the two Pdgfrα+ subpopulations revealed enrichment of categories related to ossification , bone and blood vessel development , axon guidance , and Schwann cell differentiation ( Figure 5—figure supplement 1D and supplement 2 ) . 10 . 7554/eLife . 03696 . 011Figure 5 . The neonatal bone marrow Nestin-GFP+ population contains Pdgfrα+ MSCs and Pdgfrα− Schwann cell precursors . ( A ) Representative flow cytometry profiles showing Nes-GFP and Pdgfrα expression in postnatal BMSCs . ( B and C ) Relative mRNA expression levels of ( B ) HSC niche-related genes and ( C ) Schwann cell progenitor genes by GFP+ Pdgfrα+ ( black ) or GFP+ Pdgfrα− BMSCs . RNAseq data are expressed as fragments per kilobase of exon per million fragments mapped ( FPKM; n=2 independent samples from pooled newborns ) . Note that the neonatal GFP+ Pdgfrα− subpopulation has a Schwann cell progenitor signature ( C ) , whereas GFP+ Pdgfrα+ cells are enriched in HSC maintenance genes . ( D ) Principal component analysis comparing the transcriptome of neonatal Nes-Gfp bone marrow stromal subsets with available microarray expression data sets from neural crest-derived populations and primary adult mouse BMSCs ( Table 1 ) . ( E and F ) In vitro differentiation of neonatal subpopulations isolated as in ( A ) and cultured in mesenchymal ( mesenchymal ) and Schwann cell ( glial ) differentiation medium . Adipocytes were stained with Oil Red O ( red ) and counterstained with hematoxylin ( left panels ) ; Schwann cells were stained with antibodies against glial fibrillary acidic protein ( Gfap , red ) and overlaid with endogenous GFP fluorescence ( right panels ) . Scale bars: 200 μm ( top right insets: 50 μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 01110 . 7554/eLife . 03696 . 012Figure 5—figure supplement 1 . The neonatal Nes-GFP+ bone marrow population is enriched in primitive mesenchymal progenitors . ( A ) Representative FACS profile of GFP and Pdgfrα expression in CD45− CD31− Ter119− BMSCs from neonatal Nes-Gfp long bones . Four different stromal subpopulations identified based on expression levels of Pdgfrα and GFP were isolated and analyzed by RNAseq . ( B ) Endogenous transcript expression levels of Nestin , Pdgfra , Cspg4 ( NG2 ) , and Sca1 ( Ly6a ) in isolated stromal populations depicted in ( A ) , expressed in fragments per kilobase of exon per million fragments mapped ( FPKM ) . ( C ) Expression profiles of characteristic mesenchymal genes in the stromal subpopulations . ( D ) Functional gene ontology enrichment analysis ( DAVID software ) of genes differentially expressed ( p ≤ 0 . 10 ) between the GFP+ and GFP− subsets of P1 Pdgfrα+ bone marrow stromal cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 01210 . 7554/eLife . 03696 . 013Figure 5—figure supplement 2 . RNA-seq data analysis . Top 50 downregulated ( A ) and upregulated ( B ) genes in Nes-GFP+ Pdgfrα+ and Nes-GFP+ Pdgfrα− cells from P1 bone marrow stroma . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 013 To further characterize nestin+ subpopulations , we compared the transcriptome-wide profile of neonatal Nes-GFP+/− Pdgfrα+/− BMSCs with publicly available microarray expression data sets from primary adult BMSCs or neural crest derivatives ( Table 1 ) . Unbiased hierarchical clustering and principal component analysis ( Figure 5D ) revealed that Nes-GFP+ Pdgfrα+ cells were more similar to adult primitive BMSCs and distinct from more differentiated osteoblastic cells ( Nakamura et al . , 2010 ) . Pdgfrα+ Nes-GFP+/− cells clustered nearby , consistent with the increasing restriction of Pdgfrα to Nes-GFP+ cells in postnatal bone marrow . In addition , Nes-GFP+ Pdgfrα + cells clustered far from Nes-GFP+ Pdgfrα− cells , whose genomic profile was closest to that of E12 . 5 Schwann cell precursors . MSC-like and neural crest stem-cell-like derived clones ( Wislet-Gendebien et al . , 2012 ) were markedly different , probably because these were cultured cells . Intriguingly , we noted a maturation hierarchy of Schwann and osteolineage cells , from undifferentiated cells ( Figure 5D , lower corners ) to more mature lineages ( Figure 5D , contralateral upper corners ) . At the intersection of these differentiation waves , adult bone marrow CD45− Nes-GFP+ cells ( Mendez-Ferrer et al . , 2010 ) converged with bone marrow HSC niche cells identified by the expression of stem cell factor ( Ding et al . , 2012 ) . These results suggest the existence of two nestin+ populations with non-overlapping MSC and Schwann cell precursor features . To test this hypothesis functionally , we cultured neonatal Nes-GFP+/− Pdgfrα+/− BMSCs in differentiation medium , finding that mesenchymal and glial differentiation was segregated in Pdgfrα+ and Pdgfrα− cells , respectively ( Figure 5E , F ) . Thus two Nes-GFP+ neural crest derivatives occur in postnatal bone marrow: Pdgfrα+ MSCs enriched in HSC-supporting genes and Pdgfrα− Schwann cell precursors . 10 . 7554/eLife . 03696 . 014Table 1 . Description of publicly available data sets used for principal component analysesDOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 014Plot IDCell PopulationGEO samplesSample descriptionRef1BM osteoblastic cells ( Alcam- Sca1- ) GSM437794BM ( adult ) primary stromal† Alcam- Sca1-‡2BM osteoblastic cells ( Alcam+ Sca1- ) GSM437795BM ( adult ) primary stromal† Alcam+ Sca1-3BM osteoblastic cells ( Pdgfrα- Sca1- ) GSM437797BM ( adult ) primary stromal† Pdgfrα- Sca1-4BM osteoblastic cells ( Pdgfrα+ Sca1- ) GSM437798BM ( adult ) primary stromal† Pdgfrα+ Sca1-5BM MPC’s ( Pdgfrα+ Sca1+ ) GSM437799BM ( adult ) primary stromal† Mesenchymal progenitor cells ( MPC ) ( Pdgfrα+ Sca1+ ) 6BM MPC’s ( Alcam+ Sca1+ ) GSM437796BM ( adult ) primary stromal† Mesenchymal progenitor cells ( MPC ) ( Alcam+ Sca1+ ) 7BM MSC-like derived clonesGSM795638-40BM MSC-derived cell line , Wnt1Cre/R26R β-gal- selected clone passage>10§8BM NCSC-derived clonesGSM795641-43BM NCSC-derived cell line , Wnt1Cre/R26R β-gal- selected clone passage>109E13 . 5 vascular segment ( mesoderm-derived ) GSM261911E13 . 5 internal carotid artery vascular segment ( smooth muscle mesoderm-derived ) ¶10E13 . 5 vascular segment ( NC derived ) GSM261912E13 . 5 external carotid artery vascular segment ( smooth muscle NC-derived ) 11BM ( adult ) Nes-GFP+GSM545815-17BM ( adult ) primary ( CD45- ) Nes-GFP+ cells*12BM ( adult ) Scf-GFP+GSM821066-68BM ( adult ) primary Scf-GFP+ cells††13P0 Schwann cellsGSM15386-88P0 primary Schwann cells ( Plp-GFP+ ) from sciatic nerve‡‡14E12 . 5 Schwann cell precursors ( SCPs ) GSM15373-75E12 . 5 primary Schwann cell precursors ( Plp-GFP+ ) from sciatic nerve15E14 . 5 Enteric Neural crest cells ( ENCC ) GSM844492-94E14 . 5 primary ENCCs ( Wnt1Cre/R26-YFP+ ) from gut16E9 . 5 Neural crest stem cells ( NCSC ) GSM15370-72E9 . 5 trunk primary Plp-GFP+ cells ( migrating NCSCs ) 17E18 . 5 Schwann cellsGSM15383-85E18 . 5 primary Schwann cells ( Plp-GFP+ ) cells from sciatic nerveGene Expression Omnibus database ( http://www . ncbi . nlm . nih . gov/geo/ ) †CD45-CD31-Ter119-References:‡Nakamura et al . Isolation and characterization of endosteal niche cell populations that regulate hematopoietic stem cells . Blood ( 2010 ) vol . 116 ( 9 ) pp . 1422-32 . §Wislet-Gendebien et al . Mesenchymal stem cells and neural crest stem cells from adult bone marrow: characterization of their surprising similarities and differences . Cell Mol Life Sci ( 2012 ) vol . 69 ( 15 ) pp . 2593-608 . ¶Zhang et al . Origin-specific epigenetic program correlates with vascular bed-specific differences in Rgs5 expression . FASEB J ( 2012 ) vol . 26 ( 1 ) pp . 181-91 . *Méndez-Ferrer et al . Mesenchymal and haematopoietic stem cells form a unique bone marrow niche . Nature ( 2010 ) vol . 466 ( 7308 ) pp . 829-34 . ††Ding and Morrison . Haematopoietic stem cells and early lymphoid progenitors occupy distinct bone marrow niches . Nature ( 2013 ) pp . 1-6 . ‡‡Buchstaller et al . Efficient isolation and gene expression profiling of small numbers of neural crest stem cells and developing Schwann cells . J Neurosci ( 2004 ) vol . 24 ( 10 ) pp . 2357-65 . Under similar culture conditions used to grow neural crest cells , adult mouse bone marrow Nes-GFP+ cells can form self-renewing and multipotent mesenchymal spheres with the capacity to transfer hematopoietic activity to ectopic sites during serial transplantations ( Mendez-Ferrer et al . , 2010 ) . In addition , human bone marrow-derived mesenspheres secrete factors that can expand human cord blood HSCs through secreted factors ( Isern et al . , 2013 ) . These findings and the results presented so far together suggest that neural crest-derived MSCs might have a specialized function in establishing the HSC niche in the developing bone marrow . We further studied the role of neural crest-derived cells in this process using a loss-of-function model . Perineural migration of neural crest-derived cells requires the interaction of the receptor tyrosine-protein kinase ErbB3 with the ligand neuregulin-1 , produced by developing nerves ( Jessen and Mirsky , 2005 ) . Erbb3-deficient mice initially show normal development of peripheral nerves but later display impaired perineural migration of neural crest-derived cells and die at perinatal stage ( Riethmacher et al . , 1997 ) . Hematopoietic progenitors were increased in fetal liver of KO mice ( Figure 6A ) . In contrast , expression of the MSC marker CD90 , enriched in Nes-GFP+ cells ( Figure 6B , C ) , was reduced two-fold in Erbb3−/− limb bone marrow , associated with 5-fold drop in the number of bone marrow hematopoietic progenitors ( Figure 6D , G ) . To further dissect the contribution of neural crest to fetal hematopoiesis , we performed a similar analysis in mice conditionally lacking ErbB3 in Schwann-committed cells . To label Schwann cells , we intercrossed R26-Tomato reporter mice intercrossed with a line expressing Cre recombinase under the regulatory elements of desert hedgehog ( Dhh ) promoter ( Jaegle et al . , 2003 ) ( Figure 6—figure supplement 1A–C ) . These Dhh-Cre mice were then intercrossed with the ErbB3 conditional KO . Similar to the constitutive KO , Dhh-Cre;Erbb3fl/fl mice are virtually devoid of Schwann cells ( Sheean et al . , 2014 ) ; however , unlike the constitutive KO , Dhh-Cre;Erbb3fl/fl mice had a normal frequency of bone marrow hematopoietic progenitors ( Figure 6—figure supplement 1D ) . Together , these results suggest that neural crest cells not yet committed to the Schwann cell lineage migrate along developing nerves to the bone marrow , giving rise to HSC niche-forming MSCs . 10 . 7554/eLife . 03696 . 015Figure 6 . Perineural migration of neural crest-derived cells to long bones generates nestin+ MSCs with specialized HSC niche function . ( A ) E17 . 5-E18 . 5 fetal liver ( FL ) sections from wild-type ( wt ) and Erbb3-null embryos stained with antibodies for mature hematopoietic lineage ( blue ) and c-Kit ( red ) . Quantification of fetal liver Lin− c-Kit+ hematopoietic progenitors ( per 0 . 41 mm2 ) . ( B ) Representative FACS profile of CD45- CD31-Ter119- bone marrow cells from 2-week old Nes-Gfp mice stained with the mesenchymal marker CD90 , showing the expression enrichment in Nes-GFP+ cells . ( C ) Neonatal bone marrow section stained with anti-CD90 ( red ) , which labeled Nes-GFP+ ( green ) cells . Scale bar: 50 μm . ( D ) Representative bone marrow sections from wt and Erbb3-null E17 . 5/18 . 5 mice immunostained with anti-CD90 ( red ) . ( E ) Quantification of CD90 immunostaining of samples in ( D ) ; n = 3 . ( F ) Staining of bone marrow sections from wt and Erbb3-null embryos with antibodies for mature hematopoietic lineage ( blue ) and c-Kit ( red ) . ( G ) Quantification of bone marrow Lineage− c-Kit+ hematopoietic progenitors in E17 . 5/18 . 5 wt and Erbb3-null mice ( n = 3 ) . ( E , G ) Mean ± SEM; p < 0 . 05 , unpaired two-tailed t test . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 01510 . 7554/eLife . 03696 . 016Figure 6—figure supplement 1 . Conditional Erbb3 deletion after glial specification of Schwann cell precursors does not affect bone marrow HSCs . ( A-A′ ) Whole-mount view of ventral ribcage of a Dhh-Cre;R26-Tomato neonate , showing labeling of glial cells along the intercostal peripheral nerves ( red ) . ( B-B′ ) Surface detail of Tomato+ Schwann cells between 2 ribs . ( C ) High magnification of ventral skull showing an intricate network of Tomato+ Schwann cells . ( D ) Frequency of hematopoietic lineage− c-Kit+ Sca-1+ ( LSK ) cells in neonatal bone marrow of Dhh-Cre;Erbb3f/f and control littermate mice ( mean ± SD , n = 4–5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 016 Despite the low frequency of HSCs and Nes-GFP+ cells , detailed immunofluorescence analysis showed significant proximity of HSCs to Nes-GFP+ cells in neonatal bone marrow ( Figure 7A , B ) , suggesting that nestin+ cells might attract circulating HSCs toward their final developmental destination in the bone marrow . To study the contribution of nestin+ cells to HSC migration from fetal liver to bone marrow , we used mice expressing the diphtheria toxin ( iDTA ) or its receptor ( iDTR ) in nestin+ cells . Depletion of nestin+ cells at E15 . 5 in Nes-CreERT2;iDTR mice caused an ∼4-fold reduction in fetal bone marrow HSC activity within 48 hr , inversely correlating with an ∼8-fold increase in fetal liver HSC activity ( Figure 7C , D ) . The cell cycle profile and apoptosis in hematopoietic progenitors were unchanged ( Figure 7—figure supplement 1A ) , but their numbers increased in fetal liver by 40% ( Figure 7E ) . Similar results were obtained by depleting nestin+ cells during the first postnatal week in Nes-CreERT2;iDTA mice , which otherwise showed normal bone marrow histology ( Figure 7F and Figure 7—figure supplement 1B ) . Developmental HSC migration to bone marrow proceeds until the second week after birth ( Dzierzak and Speck , 2008 ) , suggesting that the bone marrow environment might still mature during this period to accommodate HSCs . We therefore analyzed GFP+/− BMSCs from E18 . 5 and P7 Nes-Gfp bone marrow . The expression of HSC-supporting genes was markedly higher and progressively upregulated in GFP+ cells during the first postnatal week ( Figure 7G ) . These results suggest that perinatal maturation of nestin+ cells allows the colonization of bone marrow by circulating HSCs . 10 . 7554/eLife . 03696 . 017Figure 7 . CXCL12 produced by nestin+ MSCs contributes to the establishment of the HSC niche in the bone marrow . ( A and B ) HSCs are localized near Nes-GFP+ cells in neonatal bone marrow . Neonatal femoral sections from Nes-Gfp mice were immunostained with antibodies for mature hematopoietic lineages , CD48 ( blue ) and CD150 ( red ) . ( A ) Quantification of the distance of Lin− CD48− CD150+ HSC-enriched cells from Nes-GFP+ cells ( mean ± SEM , n = 41 ) . ( B ) Representative image of a putative HSC ( asterisk ) near a Nes-GFP+ cell . ( C–F ) Depletion of nestin+ cells compromises developmental HSC migration to bone marrow . ( C and D ) Long-term culture-initiating cell ( LTCIC ) assay from nestin-depleted fetal liver and bone-marrow cells . Nes-CreERT2;iDTR ( red dots ) and control iDTR ( black dots ) mice were exposed to tamoxifen at E14 . 5 and diphtheria toxin at E15 . 5 , and liver cells ( C ) and bone marrow cells ( D ) were isolated at E17 . 5 ( n = 5-6 ) . The percentage of culture dishes that failed to generate hematopoietic colony-forming units in culture ( CFU-C ) is plotted against five serial dilutions of ( C ) fetal liver Lin− Sca-1+ cells and ( D ) nucleated bone marrow cells . HSC frequencies and p values are indicated ( Pearson’s chi-squared test ) . ( E ) Frequency of Lin− Sca-1+ E17 . 5 liver cells in mice in ( C ) . ( F ) Bone marrow CFU-C content in 1-week old Nes-CreERT2;R26-DTA and control littermates treated with tamoxifen at birth ( n = 3–7 ) . ( G ) Expression of core HSC maintenance genes increases in perinatal Nes-GFP+ BMSCs . qPCR analysis of Cxcl12 , stem cell factor/kit ligand ( Kitl ) , angiopoietin-1 ( Angpt1 ) , and vascular cell adhesion molecule-1 ( Vcam1 ) mRNA in CD45- CD31- Ter119- GFP+/− cells isolated from E18 . 5 and P7 Nes-Gfp bone marrow . ( H ) Relative Cxcl12 mRNA expression levels in endothelial cells and Nes-GFP+/- BMSCs isolated from 1-week old mice ( qPCR; n = 2 ) . ( I and J ) Relative enrichment of Cxcl12 ( I ) and Nestin ( J ) mRNA expression in populations sorted from the bone marrow of P7 Nes-Gfp;Wnt1-Cre2;R26-Tomato compound transgenic mice . ( K ) Representative confocal image of a bone marrow section from a 1-week old ( P7 ) Nes-Gfp;Nes-CreERT2;R26-Tomato mouse treated with tamoxifen at birth . Both sinusoidal ( asterisk ) and arteriolar GFP+ cells express the Nes-CreERT2-derived Tomato ( red ) reporter ( yellow in overlaid picture , K′ ) . ( L and M ) Efficiency of perinatal Cxcl12 excision by the Nes-CreERT2 driver in CD45-Ter119−CD31- cells ( L ) and endothelial ( M ) cells isolated from P7 bone marrow; qPCR in CD45-Ter119−CD31- cells isolated from Cxcl12f/f;Nes-CreERT2 ( E ) and control ( C ) littermates treated with tamoxifen at birth ( n = 2-3 ) . ( N and O ) Bone marrow CFU-C ( N ) and long-term HSC ( O ) content in P7 Cxcl12f/f;Nes-CreERT2 and control littermates treated with tamoxifen at birth . ( O ) Lethally-irradiated mice ( CD45 . 1 ) were transplanted with 1 ×106 bone marrow cells from P7 Cxcl12f/f;Nes-CreER or Cxcl12f/f mice ( CD45 . 2 ) , together with 1 × 106 recipient bone marrow cells ( CD45 . 1 ) . Peripheral donor-derived blood chimerism after 16 weeks is shown ( n = 4 per group ) . ( E , L ) Each dot represents an individual mouse . ( F , H–J ) Mean ± SD . *p < 0 . 05 , unpaired two-tailed t test . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 01710 . 7554/eLife . 03696 . 018Figure 7—figure supplement 1 . Neural crest-derived cells direct developmental HSC migration to the bone marrow . ( A ) Cell cycle profile of Lin− Sca-1+ E17 . 5 liver cells isolated from Nes-CreERT2;iDTR embryos and control iDTR littermates 24 hr after tamoxifen administration and 48 hr after diphtheria toxin treatment ( mean ± SEM , n = 6 ) . ( B ) Representative femoral bone marrow section from a tamoxifen-treated Nes-CreERT2;R26-DTA mouse immunostained with collagen IV antibody ( red ) to reveal blood vessels ( B′ ) . Nuclei were counterstained with DAPI ( gray ) . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03696 . 018 HSC migration to fetal bone marrow is enhanced by Cxcl12 and stem cell factor ( Christensen et al . , 2004 ) , both of which are highly expressed and progressively upregulated in bone marrow nestin+ cells at perinatal stages ( Figure 7G ) . Cxcl12 is produced by several stromal cells and is required for developmental bone marrow colonization by HSCs ( Ara et al . , 2003 ) . It has been argued that Cxcl12 produced by endothelial cells and nestin− mesenchymal progenitors—but not by nestin+ cells—is necessary for adult HSC maintenance ( Ding and Morrison , 2013; Greenbaum et al . , 2013 ) . We found that , one week after birth , Cxcl12 mRNA levels in Nes-GFP+ BMSCs were >20-fold higher than in bone marrow endothelial cells and 80-fold higher than in Nes-GFP- BMSCs ( Figure 7H ) . Among neural crest-traced cells , Nes-GFP+ BMSCs were particularly enriched in the expression of Cxcl12 and endogenous Nestin ( Figure 7I , J ) . To conditionally delete Cxcl12 in nestin+ cells in the first postnatal week , we intercrossed Cxcl12 fl mice ( Tzeng et al . , 2010 ) with Nes-CreERT2 mice , which mostly label Nes-GFP+ cells during this period ( Figure 7K ) . Tamoxifen administration did not significantly alter Cxcl12 mRNA levels in bone marrow endothelial cells but decreased these levels by 5-fold in BMSCs ( Figure 7L–M ) . This was associated with an ∼30% reduction of bone marrow hematopoietic progenitors and HSCs measured by long-term competitive repopulation assays ( Figure 7N , O ) . These results demonstrate that Cxcl12 production by nestin+ MSCs contributes to the HSC niche formation in the developing bone marrow . The aim of this study was to investigate the ontogeny and specific functions of mesenchymal progenitors in the fetal bone marrow . We show that the developing bone marrow in axial and appendicular skeleton harbors different MSC populations with distinct origins and specialized roles . While mesoderm-derived nestin− MSCs give rise to bone and cartilage , the neuroectoderm provides an additional source of MSCs , marked by nestin expression , that are endowed with specific HSC niche functions . We therefore conclude that osteochondroprogenitor and stem cell niche functions are separate and non-overlapping during bone marrow ontogenesis . The neural crest thus gives rise to three regulators of adult HSC activity: sympathetic neurons , associated Schwann cells , and nestin+ MSCs . Although it is accepted that mammalian connective tissues , such as bone or skeletal muscle , are derived mainly from mesoderm , the precise origin of BMSCs has remained unclear . While the neural crest contributes to the craniofacial skeleton , the trunk neural crest is thought to generate mostly non-ectomesenchymal derivatives , including melanocytes , neurons , and glia of the peripheral nervous system . In the trunk skeleton , mesenchymal cells have thus been considered to be derived mostly from the mesoderm ( Olsen et al . , 2000 ) , but the neural crest is also a source of pericytes , mural cells , and fibroblasts ( Bergwerff et al . , 1998; Etchevers et al . , 2001; Joseph et al . , 2004 ) that can differentiate into mesenchymal lineages in vitro ( Morikawa et al . , 2009b; Glejzer et al . , 2011; John et al . , 2011; Komada et al . , 2012 ) . In addition , mature endothelial cells can generate mesenchymal cells through endothelial-to-mesenchymal transition . Genetic fate-mapping studies using the neuroepithelial marker Sox1 identified a neuroectodermal origin of the earliest trunk MSCs , but other MSCs are recruited from undefined sources at later stages ( Takashima et al . , 2007 ) . This picture raised the question of whether several MSCs might transiently coexist in the developing bone marrow and whether neural-crest-derived MSCs with specific functions might persist in the postnatal bone marrow . In this study , we show that most mesenchymal activity and chondrogenic capacity in the fetal bone marrow is associated with nestin− MSCs , but that these rapidly differentiate towards committed osteochondral lineages early in postnatal life . In contrast , slow-proliferating neural crest-derived nestin+ MSCs do not contribute to fetal endochondrogenesis but are instead required to establish the HSC niche in the same bones . Interestingly , we also found that nestin+ cells retained most of their steady-state MSC activity after the second postnatal week . Future studies will determine the potential contribution of this MSC population to adult skeletal turnover . Very recent data showed that Osterix-Cre-labeled cells in neonatal bone give rise to Nes-GFP+ MSCs ( Mizoguchi et al . , 2014; Ono et al . , 2014 ) ; however , the Osterix-Cre lines used in these studies have been shown previously to mark not only osteolineage cells but also a large variety of non-osteolineage cells , including adventitial reticular cells , vascular smooth muscle cells , adipocytes , and perineural cells ( Liu et al . , 2013 ) . It therefore remains possible that some Nes-GFP+ cells , which highly express Osterix mRNA ( but not the protein , as we show here ) , also express the Osterix-Cre transgene; however , this might not necessarily reflect a lineage relationship or osteoblastic commitment . Moreover , it is unclear whether this Osterix-Cre-traced population is uniformly mesodermal or neural crest-derived . We did not find Nes-GFP+ cells within cartilage , but some GFP+ cells were associated with the innermost part of the perichondrium , a region that contains MSCs ( Maes et al . , 2010; Yang et al . , 2013; Zaidi and Mendez-Ferrer , 2013 ) . Moreover , we found perichondrial Wnt1-Cre2-traced cells and neural crest-derived chondrocytes in the most superficial layers of articular cartilage , suggesting that the neural crest also contributes to these mesenchymal cells outside the marrow . Neural crest-derived skeletal embryonic precursors might be Pdgfrα+ Nes-GFP- BMSCs , since we have shown that most neural crest cells traced by Wnt1-Cre2 are Pdgfrα+ Nes-GFP+ cells , which do not seem to contribute to fetal osteochondral lineages . Previous studies suggested that only endochondral cells can form a hematopoietic microenvironment when implanted beneath the kidney capsule ( Chan et al . , 2008 , 2013 ) , although bones formed by intramembranous ossification , such as the skull , can also form hematopoietic marrow without progressing through a cartilage intermediate . Notably , two surface markers used to isolate osteochondrogenic precursors in these studies , CD105 and CD90/Thy1 , are particularly enriched in bone marrow Nes-GFP+ cells . Our genetic studies clearly show that neural crest-derived cells that are not yet committed to the Schwann cell lineage migrate to bone in association with developing nerve fibers and give rise to bone marrow nestin+ MSCs with specialized HSC niche functions . Constitutive deletion of the neuregulin-1 receptor ErbB3 , which does not directly affect the peripheral nerves but impairs perineural migration of neural crest-derived cells ( Riethmacher et al . , 1997 ) , reduces BMSC number and impairs developmental HSC migration from liver to bone marrow . In contrast , conditional deletion of ErbB3 in committed glial precursors severely reduced Schwann cell numbers ( Sheean et al . , 2014 ) but did not affect bone marrow HSCs in our study . HSC maintenance genes are highly enriched and progressively upregulated in Nes-GFP+ Pdgfrα+ MSCs , coincident with HSC bone marrow colonization ( Christensen et al . , 2004 ) . Perinatal deletion of nestin+ cells or blockade of their Cxcl12 production prevented HSC bone marrow seeding . Therefore , we are confident that neural crest-derived nestin+ MSCs are indeed required for developmental HSC migration and formation of the bone marrow HSC niche . Uncertainties remain about the possible overlap among different adult bone marrow mesenchymal cells with proposed HSC niche functions , which might also differ in fetal and adult bone marrow . Two populations of adult bone marrow Nes-GFP+ cells can be separated by fluorescence intensity by microscopy . Nes-GFPbright peri-arteriolar cells , unlike Nes-GFPdim peri-sinusoidal cells , were found highly enriched in CFU-F activity and expression of HSC maintenance genes ( Kunisaki et al . , 2013 ) . However , these features have been previously attributed to peri-sinusoidal cells ( Kiel et al . , 2005; Sacchetti et al . , 2007 ) , which , unlike arteriolar cells , express both Nes-GFP+ and leptin receptor ( Ding et al . , 2012 ) . Alternatively , an intermediate type of vessel that connects arterioles with sinusoids near the bone surface might be highly enriched in nestin+ MSCs . These vessels , different from arterioles and sinusoids , contain CD31hi Endomucinhi endothelial cells and have been associated with osteoprogenitor activity ( Kusumbe et al . , 2014 ) . We initially reported that adult bone marrow Nes-GFP+ cells were CD31− after using an intensive enzymatic digestion protocol to isolate the cells ( Mendez-Ferrer et al . , 2010 ) . In the present study , we have used milder conditions that better preserve antigen expression . In these conditions , we detect Nes-GFP+ CD31+ putative bone marrow endothelial cells progressively increasing in number with age , as reported by others ( Ono et al . , 2014 ) . We also find that multiple layers of Nes-GFP+ endothelial and perivascular cells likely make the arteriolar GFP signal appear brighter for GFP under the microscope . This raises the possibility that arteriolar Nes-GFP+ cells ( Kunisaki et al . , 2013 ) might not completely coincide with the brightest cells isolated by FACS . However , we also found that Cxcl12 is more abundantly produced by Nes-GFP+ BMSCs than by endothelial cells , and that Cxcl12 deletion in Nes-CreERT2 mice impairs developmental HSC migration by preferentially targeting BMSCs . Different bone marrow cells have been proposed as the main producers of Cxcl12 for HSCs . One report has argued that the only relevant source of Cxcl12 for adult HSC maintenance is nestin–leptin receptor–mesenchymal progenitors targeted by the Prx1-Cre driver ( Greenbaum et al . , 2013 ) . This conclusion is based on the stronger HSC depletion when Cxcl12 was deleted using the Prx1-Cre line—which targets somatic lateral plate mesoderm and its derivatives , including chondrogenic and osteogenic lineages—compared to deletion in more specific stromal populations . Also , the HSC defect was specifically attributed to CD45- Lin− Pdgfrα+ Sca-1+ Prx1-cre-tdtomatohi cells which were not enriched in the expression of nestin or leptin receptor but also showed no marked enrichment in the expression of Cxcl12 , Kitl or other mesenchymal markers . An alternate model is that endothelial cells are needed for Cxcl12-mediated adult HSC maintenance ( Ding et al . , 2012; Ding and Morrison , 2013 ) . Nonetheless , the Tek-Cre system used would also target endothelial cells in fetal hematopoietic organs , so some of the effects might have been exported to the bone marrow during development . The study , in essence , proposed that nestin-negative leptin receptor ( Lepr ) -Cre-traced mesenchymal progenitors are another key source of Cxcl12 for HSC maintenance . These studies , however , analyzed the adult bone marrow , whereas the focus of our study has been the fetal and perinatal period . Differences may also have arisen from distinct experimental settings , Cre drivers ( constitutive/inducible ) , Cre induction regimes and reporters used . It is also likely that constitutive Prx1-Cre and Lepr-Cre lines target multiple mesenchymal derivatives and that combined deletion of Cxcl12 in these populations would have a more pronounced effect than deletion in specific cell types; however , the responsible cell populations might not be clear yet . Conversely , the lower excision in Nes-Cre mice and inefficient bone marrow recombination in adult Nes-CreERT2 mice might not target all MSCs , or might result in compensatory actions by other Cxcl12-producing cells . In the present study , recombination efficiency in Nes-GFP+ MSCs was higher when tamoxifen was administered in neonatal Nes-CreERT2 mice than in adults . It has also been proposed that Nes-Gfp and Nes-CreERT2 lines might target different populations ( Ding et al . , 2012 ) . To directly address this , we generated Nes-Gfp;Nes-CreERT2;R26-Tomato triple transgenic mice that demonstrate consistent labeling using a different induction protocol . Bone marrow endothelial Cxcl12 expression levels are significantly lower in this model , and seem unaffected by Nes-CreERT2-driven excision , contrasting with the reduction in BMSCs , which was associated with decreased HSPC numbers in perinatal bone marrow . Our results thus clearly show that Cxcl12 produced by nestin+ MSCs is required for developmental HSC migration to bone marrow . We recently reported that sympathetic neuropathy of the HSC niche is required for the manifestation of myeloproliferative neoplasms , disorders previously considered to be autonomously driven by mutated HSCs and typically associated with excessive fibroblasts and osteoblasts in the bone marrow . During this pathogenesis , bone marrow nestin+ cells do not seem to differentiate into fibroblasts or osteoblasts , but instead activate the Schwann cell program as a consequence of the neuroglial damage caused in the bone marrow by mutated HSCs ( Arranz et al . , 2014 ) . These changes could be explained by a neural crest contribution found for HSC niche-forming MSCs and suggest the possible re-programming of these cells towards the closest ontogenically-related linages during the pathogenesis of these disorders . In summary , this study designates separate biologic functions to ontogenically distinct populations of MSCs , and demonstrates that not all MSCs are alike . In the appendicular skeleton , nestin− MSCs derived from the mesoderm have a primarily osteochondroprogenitor function . In contrast , a distinct population of neural crest-derived nestin+ MSCs contributes to directed HSC migration through the secretion of the chemokine Cxcl12 to ultimately establish the HSC niche in the neonatal bone marrow . These niche-forming MSCs share a common origin with sympathetic neurons and Schwann cells , an ontogenic relationship that underscores our earlier observations on the sympathetic control of HSC niche function ( Mendez-Ferrer et al . , 2008 , 2010; Arranz et al . , 2014 ) . Future studies will also determine whether tight regulation of other peripheral adult stem cell niches by the nervous system also builds upon an ontogenic relationship of their components . Mouse lines used in this study ( please see Supplementary file 1 for a detailed list of mouse strains used in this study ) included Nes-Gfp ( Mignone et al . , 2004 ) , Nes-CreERT2 ( Balordi and Fishell , 2007 ) , Sox10-CreERT2 ( Matsuoka et al . , 2005 ) , Col2 . 3-Cre ( Dacquin et al . , 2002 ) , Dhh-Cre ( Jaegle et al . , 2003 ) , RCE-loxP ( Sousa et al . , 2009 ) , LSL-KFP ( Dieguez-Hurtado et al . , 2011 ) , R26-DTA ( Brockschnieder et al . , 2006 ) , Cxcl12floxed ( Tzeng et al . , 2010 ) , Erbb3floxed ( Sheean et al . , 2014 ) , Erbb3-null ( Riethmacher et al . , 1997 ) , and 129S4 . Cg-Tg ( Wnt1-cre ) 2Sor/J , C57BL/6-Gt ( ROSA ) 26Sortm1 ( HBEGF ) Awai/J , B6 . Cg-Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze/J , wild-type CD1 and wild-type C57BL/6J ( Jackson Laboratories ) . Material and methods were approved by the Animal Care and Use Committees of the Spanish National Cardiovascular Research Center and Comunidad Autónoma de Madrid ( PA-47/11 and ES280790000176 ) . Embryos were dissected as previously described ( Isern et al . , 2008 ) . Briefly , selected intercrosses between mice carrying the alleles of interest were set and the morning of detection of the vaginal plug was considered as day 0 . 5 of gestation . We preferentially used paternal transgene transmission , by mating compound or simple transgenic males with females of wild-type background ( C57BL/6 or CD1 ) . Inducible lineage tracing studies were conducted as follows . Tamoxifen ( T5648; Sigma , St . Louis , MO ) was dissolved in corn oil at a final concentration of 20 mg/mL and given to pregnant dams by oral gavage ( 100-150 mg/kg ) on the morning of the indicated stages . For neonatal induction , mothers of newborn pups were given tamoxifen ( by oral gavage , 4 mg ) on days 1 and 3 after delivery . Dissected tissues for histology were fixed in 2% paraformaldehyde at 4°C , cryopreserved by consecutive equilibration in 15% and 30% sucrose , and snap frozen and embedded in OCT compound ( Tissue-Tek ) . In some cases , fixed frozen limbs or sterna were trimmed sequentially from both sides to expose the central medullar cavity and processed further for whole-mount fluorescence staining . Cryostat sections ( 15 μm ) were prepared and processed for immunostaining or regular hematoxylin–eosin staining . Oil red O staining was performed as described ( Isern et al . , 2013 ) . Cryostat sections were stained using standard procedures . Briefly , tissues were permeabilized for 5-10 min at room temperature ( RT ) with 0 . 1% Triton X-100 and blocked for 1 h at RT with TNB buffer ( 0 . 1 M Tris–HCl , pH 7 . 5 , 0 . 15 M NaCl , 0 . 5% blocking reagent , Perkin Elmer , Waltham , MA ) . Primary antibody incubations were conducted either for 1-2 h at RT or overnight ( o/n ) at 4°C . Secondary antibody incubations were conducted for 1 h at RT . Repetitive washes were performed with PBS + 0 . 05% Tween-20 . Stained tissue sections were counterstained for 5 min with 5 μM DAPI and rinsed with PBS . Slides were mounted in Vectashield Hardset mounting medium ( Vector Labs , Burlingame , CA ) and sealed with nail polish . For whole-mount staining of thick-sectioned tissue pieces , all the incubations , including permeabilization and blocking , were performed o/n at 4°C with gently agitation , and washing steps were extended . Specimens were mounted in glass bottom dishes ( Mat-Tek , Ashland , MA ) . Fetal bone marrow and fetal liver sections were stained following standard procedures . Antibodies used are indicated in the table below . SLAM staining was performed in bone marrow sections from neonate mice . Slides were first blocked in 20% goat serum in PBS for 45 min . Endogenous avidin and biotin were blocked with an Avidin/Biotin Blocking Kit ( Vector Labs ) for 30 min with each reagent , washing 3 times with PBS in between . Slides were then incubated with rat anti-mouse CD150 antibody ( Biolegend , San Diego , CA ) at 1:50 dilution in goat blocking buffer for 2 h , and after washes with goat anti-rat IgG conjugated to Alexa555 ( Molecular Probes , Eugene , OR ) at 1:200 dilution in 20% goat serum in PBS for 1 h . Slides were then blocked in 20% rat serum in PBS for 10 min and incubated for 1 h with hamster anti-mouse biotin-conjugated CD48 ( Abcam , UK ) and the Biotin Mouse Lineage Panel ( BD Pharmingen , San Jose , CA ) , which includes rat anti-mouse B220 , rat anti-mouse CD3 , rat anti-mouse Gr1 , rat anti-mouse Mac-1 , and rat anti-mouse Ter119 antibodies , each at 1:200 dilution in rat blocking buffer . Cy5-conjugated streptavidin ( Molecular Probes ) was added at 1:200 in rat blocking buffer for 30 min . Finally , slides were incubated with DAPI ( 1:1000 dilution of 5 mg/ml stock ) for 10 min at room temperature and mounted in Vectashield Mounting Medium ( Vector Labs ) . Antibodies used for Immunohistochemistry:NameTypeCompanyTHRabbit pAbMilliporeGFAPRabbit pAbDakoCD31Rat mAbBD PharmingenS100Rabbit pAbDakoCollagen type IVRabbit pAbMilliporeKi67Rabbit pAbAbcamAnti-KFPRabbit pAbEvrogenCD150Rat mAbBiolegendCD48ArHm mAbAbcamLineage-biotinRat mAbBD BiosciencesTuj1Mouse mAbPromegaAnti-GFPRabbit pAbAbcamc-kitgoat pAbR&DNestinRabbit pAbAbcamα-SMAMouse mAbSigma Confocal images of fluorescent staining were acquired with a laser scanning confocal microscope ( Zeiss LSM 700 , 10×/0 . 45 , 25×/0 . 85 ) or with a multi-photon Zeiss LSM 780 microscope ( 10×/0 . 7 , 20×/1 . 0 ) . Optical z-stack projections were generated with the Zen2011 software package ( Zeiss , Germany ) using a maximal intensity algorithm . Wide-field views of whole-mount specimens were imaged with a Leica MZFLIII stereomicroscope equipped with an Olympus DP71 color camera . Images were post-processed and quantified using ImageJ ( Schneider et al . , 2012 ) and Photoshop ( Adobe , San Jose , CA ) . Fetal skeletal elements were sub-dissected from fetuses , homogenized by cutting , and digested for 15-30 min at 37°C with shaking in 0 . 25% collagenase ( StemCell Technologies , Canada ) . Postnatal bone specimens were cleaned from surrounding tissue , crushed in a mortar with a pestle , and collagenase-digested for 45-60 min at 37°C , with constant agitation . After enzymatic treatment , skeletal preparations were filtered through a 40-μm cell strainer and undigested bone material was discarded . The resulting bone marrow-enriched cell suspensions were pelleted , washed twice , and resuspended in FACS staining buffer ( 2% FCS in PBS ) for further analysis . Dispersed bone marrow cell preparations were stained in FACS buffer for 15-30 min on ice with selected multicolor antibody cocktails ( see below ) , washed , and resuspended with streptavidin conjugates when necessary . Stained cells were pelleted and resuspended in buffer containing DAPI to exclude dead cells . Cell cycle was analyzed by first isolating defined stromal populations by FACS , and then acquiring the cell cycle profile after staining the sorted populations with Hoechst 33342 . Flow cytometry analysis and FACS were done in FACS CantoII or LSRFortessa machines ( BD Biosciences ) equipped with Diva Software ( BD Biosciences ) or in a FACS AriaII cell sorter ( BD Biosciences ) . Data were analyzed using Diva and FlowJo ( Tree Star , Inc , Eugene , OR ) . Antibodies used for cytometry:NameCloneCompanyCD45-APC/Cy7104BD BiosciencesCD45-APC104BD BiosciencesCD31-APCMEC 13 . 3BD BiosciencesTer119-APCTer119BD BiosciencesCD140a-biotinAPA5eBioscienceCD140a-APCAPA5BiolegendCD90 . 2-APC53-2 . 1eBioscienceLy6a-PEE13–161 . 7BD BiosciencesVcam1-PE429 ( MVCAM . A ) BiolegendStreptavidin-PE--BD BiosciencesLineage cocktail-biotinBD Biosciences For fibroblast colony-forming unit ( CFU-F ) assays , bone marrow cell suspensions were FACS sorted directly into 6-well plates at a cell density of 100–500 cells/cm2 and cultured in maintenance medium ( α-MEM/15% FCS with antibiotics ) . After 10-12 days in culture , adherent cells were fixed with 100% methanol and stained with Giemsa stain ( Sigma ) to reveal fibroblast clusters . Colonies with more than 50 cells were scored as CFU-Fs . For osteoblast colony-forming unit ( CFU-OB ) assays , plated cells were cultured in maintenance medium in the presence of 1 mM L-ascorbate-2-phosphate . All cultures were maintained with 5% CO2 in a water-jacketed incubator at 37°C , and medium was changed weekly . After 25 days in culture , cells were fixed and stained with alizarin red or alkaline phosphatase ( Isern et al . , 2013 ) . Single cell suspensions were prepared from bone marrow and mixed with methylcellulose-containing medium with cytokines ( Casanova-Acebes et al . , 2013 ) . Cells ( 5-7 . 5 × 104 ) were plated in duplicate 35 mm dishes ( Falcon , BD ) and incubated under 20% O2 and 5% CO2 in a water-jacketed incubator . Hematopoietic colonies ( CFU-Cs ) were scored after 6-7 days in culture . Long-term culture-initiating cell assay was performed as described ( Woehrer et al . , 2013 ) . Briefly , the feeder fetal stromal cell line AFT024 ( kindly provided by Dr . K . Moore ) was maintained as previously described ( Nolta et al . , 2002 ) . One week before use , the feeders were irradiated ( 15 Gy ) with a 137Cs irradiator and seeded in 96-well plates at confluency . After 7-10 days , five serial dilutions ( each with 16 replicates ) of sorted fetal liver Lin− Sca1+ cells and bone marrow nucleated cells were seeded on the irradiated feeders and cultured with Myelocult M5300 supplemented with 10−6 M hydrocortisone ( StemCell Technologies ) and 1% penicillin-streptomycin ( Invitrogen , Carlsbad , CA ) . Cultures were maintained for four weeks at 33°C under 20% O2 and 5% CO2 in a water-jacketed incubator . Medium was half-changed weekly . Each well was then trypsinized for 10 min , washed with PBS , and plated for the hematopoietic progenitor assay . Twelve days after plating , the percentage of culture dishes in each experimental group that failed to generate CFU-Cs was plotted against the number of test cells . The frequencies of long-term culture-initiating cells were calculated by the Newton–Raphson method of maximum likelihood and Poisson statistics ( using L-CalcTM software; StemCell Technologies ) as the reciprocal of the number of test wells that yielded a 37% negative response . Primary bone marrow cells were obtained from dissected bones using a mortar . All cultures were maintained at 37°C with 20% O2 , 5% CO2 in a water-jacketed incubator . To obtain CFU-Fs and CFU-OBs , 0 . 5 × 106 bone marrow nucleated cells were seeded in each well of a 12-well plate with α-MEM supplemented with 1% penicillin-streptomycin , 15% FBS ( Invitrogen ) , and 1 mM L-ascorbic acid 2-phosphate ( Sigma ) . Half medium was replaced every 5 days . The numbers of CFU-Fs and CFU-OBs were scored after 10 and 28 days in culture , respectively . CFU-F cultures were fixed in methanol for 10 min at room temperature . Cultures were stained with Giemsa diluted 1:10 in phosphate buffer , pH 6 . 8 , for 10 min at 37°C . CFU-F colonies ( those with more than 50 cells ) were counted the next day . CFU-OB cultures were fixed with 4% paraformaldehyde ( PFA ) for 5 min at room temperature . von Kossa staining was performed by adding 5% AgNO3 to the culture and exposing plates to UV radiation for 20 min . Cells were then incubated with 5% ( NH4 ) 2S2O3 in distilled water for 5 min and counterstained with 2% eosin . For Alizarin Red staining , cells were incubated with 2% alizarin red reagent ( Sigma ) in distilled H2O for 15 min . For alkaline phosphatase staining , Sigma Fast BCIP/NBT substrate ( Sigma ) was added to cell cultures followed by incubation in the dark for 15 min . RNA from CFU-Fs and osteoblast cultures was extracted with Trizol reagent ( Sigma-Aldrich ) and purified on RNeasy mini columns ( Qiagen , Netherlands ) . An on-column DNase digest ( Qiagen ) was performed before the clean-up step to eliminate residual genomic DNA . For osteoblast cultures , mRNA was extracted with the Dynabead mRNA DIRECT kit ( Invitrogen ) . cDNA was generated using High Capacity cDNA Reverse Transcription reagents ( Applied Biosystems , Waltham , MA ) . qPCR was performed in triplicate with SYBRgreen Universal PCR Master Mix ( Applied Biosystems ) , using primers optimized for each target gene . The expression level of each gene was determined by using the relative standard curve method . Briefly , a standard curve was performed by doing serial dilutions of a mouse reference total RNA ( Clontech , Palo Alto , CA ) . The expression level of each gene was calculated by interpolation from the standard curve . Relative quantifications of each transcript were obtained by normalizing against Gapdh transcript abundance , using the standard curve method . The sequences of oligonucleotides for qPCR are detailed below . Target geneSymbolForwardReverseAlkal . Phosphat . AlplCACAATATCAAGGATATCGACGTGAACATCAGTTCTGTTCTTCGGGTACAOsterixSp7ATGGCGTCCTCTCTGCTTGAGAAGGGTGGGTAGTCATTTGRunx2Runx2TTACCTACACCCCGCCAGTCTGCTGGTCTGGAAGGGTCCRank ligandRanklCAGCATCGCTCTGTTCCTGTACTGCGTTTTCATGGAGTCTCAGpnmbGpnmbCCCCAAGCACAGACTTTTGAGGCTTTCTGCATCTCCAGCCTOsteocalcinBglapGGGCAATAAGGTAGTGAACAGGCAGCACAGGTCCTAAATAGTOsteoglycinOgnACCATAACGACCTGGAATCTGTAACGAGTGTCATTAGCCTTGCRankRankTGCAGCTCAACAAGGATACGGAGCTGCAGACCACATCTGATRAPAcp5CAGCAGCCAAGGAGGACTACACATAGCCCACACCGTTCTCCathepsin kCtskGGCCTCTCTTGGCCATACCTTCCCACTCTGGGTAGMmp-9Mmp9CGTCGTGATCCCCACTTACTAACACACAGGGTTTGCCTTCPpar gammaPpargACCACTCGCATTCCTTTGACTGGGTCAGCTCTTGTGAATGAdiponectinAdipoqTGTTCCTCTTAATCCTGCCCACCAACCTGCACAAGTTCCCTTAdipsinCfdTGCATCAACTCAGAGTGTCAATCATGCGCAGATTGCAGGTTGTSox9Sox9GAACAGACTCACATCTCTGTGGCAAGTATTGGTCAACol2a1Col2a1GTGGAGCAGCAAGAGCAAGGACTTGCCCCACTTACCAGTGTGAggrecanAcanCACGCTACACCCTGGACTTTGCCATCTCCTCAGCGAAGCAGTCxcl12Cxcl12CGCCAAGGTCGTCGCCGTTGGCTCTGGCGATGTGGCKit ligandKitlCCCTGAAGACTCGGGCCTACAATTACAAGCGAAATGAGAGCCAngiopoietin 1Angpt1CTCGTCAGACATTCATCATCCAGCACCTTCTTTAGTGCAAAGGCTNestinNesGCTGGAACAGAGATTGGAAGGCCAGGATCTGAGCGATCTGAC For sphere formation , cells were plated at clonal density ( <1000 cells/cm2 ) in ultra-low adherent 35 mm dishes ( StemCell Technologies ) . The growth medium consisted of DMEM/F12 ( 1:1 ) mixed 1:2 with human endothelial serum-free medium ( Invitrogen ) and contained 15% chicken embryo extract , prepared as described ( Stemple and Anderson , 1992; Pajtler et al . , 2010 ) ; 0 . 1 mM ß-mercaptoethanol; 1% non-essential aminoacids ( Sigma ) ; 1% N2 and 2% B27 supplements ( Invitrogen ) ; recombinant human fibroblast growth factor ( FGF ) -basic; recombinant human epidermal growth factor ( EGF ) ; recombinant human platelet-derived growth factor ( PDGF-AB ) ; recombinant human oncostatin M ( 227 a . a . OSM ) ( 20 ng/ml ) ; and recombinant human insulin-like growth factor-1 ( IGF-1; 40 ng/ml ) ( Peprotech , Rocky Hill , NJ ) . The cultures were kept at 37°C under 5% CO2 , 20% O2 in a water-jacketed incubator , and were left untouched for one week to prevent cell aggregation in low density cultures . Medium was half-changed weekly . Mesenspheres were scored on days 10–14 . We used an adaptation of the original method ( Biernaskie et al . 2006 ) . Defined stromal populations were isolated based on GFP and Pdgfrα expression from collagenase-treated bone marrow of Nes-Gfp neonates . Sorted cells were plated onto laminin/polylysine-coated chamber slide dishes ( Labtek ) and allowed to attach and expand in SKP medium I . After 3 days , cells were changed to SKP medium II ( containing neuregulin-1 at 50 ng/mL ) and allowed to differentiate further for >10 days . In vitro-generated Schwann cells were defined by morphology as thin and elongated cells . After differentiation , cells were fixed in 4% PFA , gently permeabilized with Triton X-100 , and stained for immunofluorescence with anti-glial fibrillary acidic protein ( Gfap ) antibody ( Dako , Carpinteria , CA ) . Defined stromal populations were isolated based on GFP and Pdgfrα expression from collagenase-treated bone marrow of Nes-Gfp neonates and plated directly onto plastic dishes to allow attachment of fibroblasts . Adherent cells were cultured for 7-14 days in regular α-MEM supplemented with 15% FBS . In some cases , recombinant human PDGF was added at 20 ng/mL . At the end of the culture period , cells were fixed and stained with Oil red O to reveal adipocytes and counterstained with hematoxylin . For next-generation sequencing , total RNA was isolated using the Arcturus Picopure RNA isolation kit ( Life Technologies , Carlsbad , CA ) from small numbers of sorted cells ( 15 , 000-80 , 000 ) , obtained from neonatal Nes-Gfp bone marrow preparations ( two biological replicates ) . Each independent set of samples was obtained from pooled skeletal elements ( long bones and sterna ) from multiple littermates .	During the earliest phases of development , the embryo is formed by groups of stem cells that can develop into all the different types of tissue in the body—from bones to brain tissue . Later in life , small stockpiles of adult stem cells are found in various tissues and provide a reservoir of new cells available for replacing old or damaged cells . The most important source of blood stem cells is the bone marrow , which produces and stores cells that are capable of developing into blood and immune system cells . These processes are assisted by different bone marrow cells called stromal cells , which create a specialized local environment or ‘niche’ . But are the stromal stem cells that form the skeleton the same ones that form this niche during development ? Or do the various types of stromal stem cells develop from distinct groups of cells in the embryo ? Furthermore , it is unclear which cells guide blood stem cells towards the forming bones . Other types of cells , including some of the cells of the nervous system , can communicate with the stem cells in the adult marrow and influence their behavior . This led scientists to wonder whether the stem cells in the bone marrow niche and the cells that communicate with them developed from the same type of embryonic stem cell . Isern et al . tracked down the developmental origins of different types of bone marrow stromal stem cells by examining the bone marrow from the long bones ( for example , the bones in the leg ) of unborn and infant mice . It turns out that not all stromal stem cells in the developing bone marrow are alike . In fact , one pool of stromal stem cells forms the skeleton and loses stem cell activity in the process . In contrast , a different population of stromal stem cells develops from the same group of embryonic cells that gives rise to the cells of the nervous system . The stromal stem cells in this second group function as a niche to recruit and store the incoming blood stem cells and retain their stem cell activity throughout life . The findings of Isern et al . help to explain why the nervous system is able to communicate with stem cells in the adult marrow , and provide a model for understanding how stem cell niches in organs that contain nerve tissue are established .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2014	The neural crest is a source of mesenchymal stem cells with specialized hematopoietic stem cell niche function
Adipose tissue is a key determinant of whole body metabolism and energy homeostasis . Unraveling the regulatory mechanisms underlying adipogenesis is therefore highly relevant from a biomedical perspective . Our current understanding of fat cell differentiation is centered on the transcriptional cascades driven by the C/EBP protein family and the master regulator PPARγ . To elucidate further components of the adipogenic gene regulatory network , we performed a large-scale transcription factor ( TF ) screen overexpressing 734 TFs in mouse pre-adipocytes and probed their effect on differentiation . We identified 22 novel pro-adipogenic TFs and characterized the top ranking TF , ZEB1 , as being essential for adipogenesis both in vitro and in vivo . Moreover , its expression levels correlate with fat cell differentiation potential in humans . Genomic profiling further revealed that this TF directly targets and controls the expression of most early and late adipogenic regulators , identifying ZEB1 as a central transcriptional component of fat cell differentiation . Obesity and its associated diseases such as diabetes and hypertension affect a high percentage of the world population ( von Ruesten et al . , 2011 ) . Excess fat mass in obese individuals is driven by an increase in adipocyte cell size ( hypertrophy ) or number ( hyperplasia ) ( Stephens , 2012; Rosen and Spiegelman , 2014 ) . A comprehensive knowledge of adipocyte differentiation is therefore relevant and timely both from a basic and medical research perspective . Adipocytes arise from multipotent mesenchymal stem cells ( MSCs ) through an initial lineage commitment phase , followed by terminal differentiation with accumulation of lipid droplets ( Cawthorn et al . , 2012 ) . Although many of the key transcription factors ( TFs ) controlling this adipogenic program are known , a cohesive model of the underlying molecular events has yet to be revealed . Most of our current understanding of adipogenesis has been derived using the in vitro murine committed pre-adipocyte cell line 3T3-L1 ( Green and Kehinde , 1975 , 1976 ) , which to a certain extent recapitulates key features of adipogenesis ( Green and Kehinde , 1979 ) . The adipocyte differentiation program is accomplished by the expression and activity of a cascade of TFs ( reviewed in [Rosen and Spiegelman , 2000; Rosen et al . , 2000; Farmer , 2006; Rosen and MacDougald , 2006; Siersbæk and Mandrup , 2011] ) , most notably the early-expressed CCAAT/enhancer binding proteins beta ( C/EBPβ ) and delta ( C/EBPδ ) , which induce C/EBPα and the adipogenic master regulator nuclear hormone receptor peroxisome proliferator-activated receptor gamma ( PPARγ ) . C/EBPα and PPARγ cooperate to activate adipogenic genes and cross-regulate each other in a positive feedback loop to maintain the terminally differentiated state of mature adipocytes ( Wu et al . , 1999 ) . Several studies have revealed the importance of other TFs in regulating adipogenesis both in vitro and in vivo ( Soukas et al . , 2001 ) . These include members of the Krüppel-like family ( KLF4 [Birsoy et al . , 2008] , KLF5 [Oishi et al . , 2005] , and KLF15 [Mori et al . , 2005] ) , ADD1/SREBP1c ( Fajas et al . , 1997; Kim et al . , 1998 ) , the E2F ( Fajas et al . , 2002 ) , and the interferon regulatory factor ( IRF ) families ( Eguchi et al . , 2008 ) , STAT5A/B ( Floyd and Stephens , 2003 ) and GATA2/3 ( Tong et al . , 2000 ) . Efforts are therefore underway to integrate this information into comprehensive adipogenic regulatory networks ( aGRNs ) ( Rosen and MacDougald , 2006 , Siersbæk et al . , 2012 ) , to which new nodes such as the early regulators ZFP432 ( Gupta et al . , 2010 ) , TCF7L1 ( Cristancho et al . , 2011 ) , or EVI1 ( Ishibashi et al . , 2012 ) keep being added . This suggests that many important components of this aGRN remain to be discovered . To systematically identify novel aGRN members , we performed a large-scale overexpression screen with 734 full-length mouse TFs in 3T3-L1 cells , which led to the identification of 26 pro-adipogenic TFs . We found that the top ranking TF , ZEB1 , previously known for its role in epithelial to mesenchymal transition ( EMT ) and tumor metastasis ( Vandewalle et al . , 2009; Gheldof et al . , 2012 ) , is a critical mediator of in vitro and in vivo adipogenesis , as it directly controls the majority of aGRN genes . To systematically identify TFs enhancing adipogenesis in an unbiased manner , we tested the effect of overexpressing almost half ( 48% ) of all predicted mouse TFs on 3T3-L1 fat cell differentiation . We transferred 750 available mouse TF open reading frames ( ORFs ) ( Gubelmann et al . , 2013 ) into Tet-On Gateway-compatible inducible lentiviral expression vectors and obtained 734 clones ( Figure 1—figure supplement 1A ) . Using a robotic platform , lentiviral particles containing each TF ORF were produced and were then used to transduce 3T3-L1 cells with three replicates to ensure reproducibility . We overexpressed each of the 734 TFs in 3T3-L1 cells at confluence and during terminal adipocyte differentiation ( ‘Materials and methods’; Figure 1A and Figure 1—figure supplement 1A ) . After 7 days , we stained cells for lipids with the lipophilic , fluorescent dye BODIPY , nuclei with Hoechst , and complete cells with SYTO60 and calculated the percentage of mature adipocytes per total number of cells in each well using automated image analysis as described previously ( Meissburger et al . , 2011 ) ( Figure 1A and ‘Materials and methods’ ) . Using this approach , we were able to define the percentage of cells that had undergone differentiation ( percentage of differentiated cells , PDC ) , which was used to evaluate the impact of each tested TF on adipogenesis . 10 . 7554/eLife . 03346 . 003Figure 1 . A large-scale TF overexpression screen identifies novel positive regulators of adipogenesis . ( A ) Schematic overview of high-throughput screening illustrating how 3T3-L1 cells were transduced with 734 individual TFs in three replicates each , 3 days before induction of adipocyte differentiation ( ‘Materials and methods’ ) . The effect of TF overexpression was quantified at differentiation day 7 by lipid , nucleus and cellular staining and summarized as a percentage of differentiated cells ( PDC ) per TF . ( B ) Overview of fold-changes ( FC ) compared to control for all TFs showing a differentiation FC > 1 . TFs that significantly induced differentiation ( FC ≥ 1 . 5 , α = 0 . 05 ) are highlighted in red and PPARγ specifically in orange . ( C ) Effect of stably overexpressing eight putatively novel regulators of adipogenesis , PPARγ , or a control vector on 3T3-L1 differentiation as assessed by Oil Red O staining of lipid droplets at day 5 after induction . ( D ) Effect of knocking down ZEB1 or PPARγ ( as a positive control ) , or the negative control ( empty shRNA ) on 3T3-L1 differentiation as assessed by Oil Red O staining at day 6 after induction . In the shRNA pool of ZEB1 , shRNA2 was not used because the robustness of the cells after treatment was low . Examples of microscopic images illustrating the overexpression or knockdown ( KD ) effects on 3T3-L1 differentiation are shown in Figure 1—figure supplement 2 or Figure 1—figure supplement 3 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 00310 . 7554/eLife . 03346 . 004Figure 1—figure supplement 1 . Large-scale TF overexpression screen identifies novel positive regulators of adipogenesis . ( A ) Workflow for transferring mouse TF open reading frames ( ORFs ) into a lentiviral vector to overexpress HA-tagged TFs in 3T3-L1 cells . 750 fully sequence-verified entry TF clones were transferred using LR Gateway cloning into the Tet-On expression vector ( derived from the original TRE_GOI_rtTA_hPGK vector [Barde et al . , 2006] , ‘Materials and methods’ ) , during which the attL sites recombine with the attR sites . 734 ORF TFs were successfully transferred . ( B ) Barplots: percentage of expressed/transcribed TFs of all TFs that significantly enhance adipogenesis ( positive candidates ) in mouse 3T3-L1 cells based on microarray expression data in mouse 3T3-L1 ( mExpr ) and human hASC ( hExpr ) as well as POLII signal over genes ( mPolII ) and combined POLII signal and expression ( mAny ) in mouse 3T3-L1 cells ( Nielsen et al . , 2008; Mikkelsen et al . , 2010 ) ; table: positive candidates that are significantly up- or down-regulated in mouse adipose tissue compared to other probed tissues based on ArrayExpress Expression Atlas ( Kapushesky et al . , 2012 ) . ( C ) Protein levels of stably overexpressed HA-tagged TFs selected for follow-up in 3T3-L1 cells ( follow-up TFs ) . The expected molecular mass for each protein is indicated above the image . Note that especially for ZEB1 , several bands were detected which likely correspond to cryptic translation or specific protein degradation products given that they stem from the same open-reading frame construct and that they are all tagged by HA . ( D–F ) Relative ( to control ) : Pparg2 ( D ) , Cebpa ( E ) , and Adipoq ( F ) mRNA fold-changes ( FCs ) in 3T3-L1 cells stably overexpressing each follow-up TF , as measured by qPCR . To measure Pparg2 mRNA levels , primers were used that target the 5′ UTR of the endogenous transcript , allowing us to differentiate between the overexpression and endogenous Pparg transcripts . ( G ) Microarray-based ( Mikkelsen et al . , 2010 ) expression analysis of follow-up TFs during 3T3-L1 adipogenesis . In contrast to Pparg ( orange ) , most follow-up TFs are at their maximal expression level already prior to induction of differentiation ( days −2 and day 0 ) . ( H ) Relative ( to control; i . e . transduced with the shEmpty vector ) Zeb1 mRNA FCs in knockdown 3T3-L1 cells at four different time points during adipogenesis , as measured by qPCR . Error bars depict the standard error of the mean from three biological replicate experiments . *p ≤ 0 . 01 and 0 . 01 < p ≤ 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 00410 . 7554/eLife . 03346 . 005Figure 1—figure supplement 2 . Microscopic images of Oil Red O stained 3T3-L1 adipocytes after overexpression of candidate TFs . These images were acquired from the wells presented in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 00510 . 7554/eLife . 03346 . 006Figure 1—figure supplement 3 . Microscopic images of Oil Red O stained 3T3-L1 adipocytes after ZEB1 and PPARG KD using distinct KD constructs . These images were acquired from the wells presented in Figure 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 006 Specifically , we compared the PDC obtained for each TF to that of the negative control ( original lentiviral vector ) . We found 26 TFs that significantly enhance adipogenesis ( ≥ 1 . 5-fold relative to control ( FC ) , Bonferroni < 0 . 05 ) of which 22 have to our knowledge never been implicated in this process , with the TF ZEB1 showing the strongest effects ( Supplementary file 1A [Gubelmann et al . , 2014] and Figure 1B ) . Importantly , the master regulator PPARγ was among these 26 stringently selected top ranking candidates , serving as a positive control . Because our screen was blind to the expression and functional characteristics of the overexpressed TFs , we investigated which of our top enhancing TFs are also endogenously expressed in 3T3-L1 cells or other pre-adipocyte cell lines . For this purpose , we analyzed publicly available microarray expression data from mouse 3T3-L1 and primary human adipose stromal cells , as well as RNA polymerase II ( POLII ) binding data for expressed TFs ( Nielsen et al . , 2008; Mikkelsen et al . , 2010 ) . We found that 18 of the top 26 candidates ( 69% ) show high microarray signal or POLII gene body occupancy , suggesting that they are actively transcribed during 3T3-L1 differentiation ( Figure 1—figure supplement 1B , Supplementary file 1A [Gubelmann et al . , 2014] and ‘Materials and methods’ ) . Additionally , we found using Expression Atlas ( Kapushesky et al . , 2012 ) that nine of the positive candidates ( including PPARγ and the top candidate ZEB1 ) are expressed significantly higher in adipose tissue compared to their mean expression across tissues ( Figure 1—figure supplement 1B ) . The high abundance of these TFs may point to a regulatory function in adipocyte-specific processes . Moreover , 15 ( 83% ) of the candidates' human orthologs are also highly expressed in human adipose stromal cells ( Figure 1—figure supplement 1B ) , suggesting that their regulation is conserved . To validate the adipogenic activity of our newly identified pro-adipogenic TFs , we generated 10 stably transduced 3T3-L1 cell lines , including for eight of the top candidates that were not previously implicated in adipogenesis ( referred to as ‘follow-up TFs’ ) , as well as for PPARγ and the negative control vector , using puromycin selection . TFs were overexpressed for 2 days before inducing differentiation , and lipids were stained with Oil Red O 5 days after differentiation . Overexpression of the TFs was confirmed by Western blots ( Figure 1—figure supplement 1C ) . Compared with cells transduced with the control vector , overexpression of 6 out of these 8 TFs led to increased lipid content within the respective cells ( Figure 1C and Figure 1—figure supplement 2 ) and a strong induction of adipogenic gene expression as measured by quantitative real-time polymerase chain reaction ( qPCR ) ( Pparg2 , Cebpa , and Adipoq ) after differentiation ( Figure 1—figure supplement 1D–F ) . Interestingly , all but one of these six follow-up TFs were expressed at their maximal level in wild-type 3T3-L1 cells already at the onset of adipogenesis ( day −2 or day 0 ) , in contrast to Pparg , which is highly induced upon differentiation ( Tontonoz et al . , 1994 ) ( Figure 1—figure supplement 1G and Supplementary file 1A [Gubelmann et al . , 2014] ) . This suggests that these TFs may be important in defining the pre-adipocyte state and may act as early regulators of fat cell differentiation . To confirm the involvement of the top differentiation-enhancing TF ZEB1 in 3T3-L1 adipogenesis , we reduced its expression levels with three distinct shRNAs . In parallel , shRNAs targeting Pparg were used as a positive control . After transducing each shRNA into 3T3-L1 cells , we induced differentiation and stained for lipid accumulation at day 6 . The pooled knockdown ( KD ) reduced Zeb1 gene expression during adipogenesis by approximately 80–90% ( Figure 1—figure supplement 1H ) . Oil Red O staining revealed a dramatic reduction in differentiation for individual shRNAs and an almost completely abolished differentiation when the shRNA pool was used , an effect that mimicked that of PPARγ KD ( Figure 1D and Figure 1—figure supplement 3 ) . Thus , we identified several TFs that increase adipogenesis when transiently or stably overexpressed in 3T3-L1 cells ( Figure 1B–C ) . In addition , we revealed that KD of the top adipogenic candidate ZEB1 inhibits adipogenesis in 3T3-L1 cells ( Figure 1D ) , suggesting that this TF is a so far unrecognized , important mediator of 3T3-L1 fat cell differentiation . To explore the mechanism underlying ZEB1-induced stimulation of adipogenesis , we used 3T3-L1 cells . First , we quantified its expression level by qPCR at six adipogenic differentiation time points ( Figure 2—figure supplement 1A ) . Unlike Pparγ , whose expression is highly induced upon adipogenesis ( Tontonoz et al . , 1994 ) , Zeb1 mRNA levels were already high in pre-adipocytes and moderately but significantly decreased during terminal differentiation ( Figure 2—figure supplement 1A–B , p = 0 . 009 , Wilcoxon rank-sum test days −2 to 2 vs . day 4 ) . This result is consistent with data from previously published microarray-based gene expression during adipogenesis ( Mikkelsen et al . , 2010 ) as well as with data comparing pre-adipocyte to adipocyte gene expression ( Expression Atlas: [Kapushesky et al . , 2012] ) ( Figure 1—figure supplement 1G and Figure 2—figure supplement 1C ) . ZEB1 may thus already be active at early stages of adipogenesis , in line with the observation that it is among several genes that were highly upregulated immediately after adipogenic induction of mouse embryonic stem cells ( Billon et al . , 2010 ) . We next examined ZEB1 protein levels during differentiation using our recently developed quantitative proteomics assay ( Simicevic et al . , 2013 ) . We found that ZEB1 is expressed at comparable levels to the nuclear receptor RXRα at day 0 ( about 0 . 25 fmol/μg nuclear extract ) ( [Simicevic et al . , 2013] and Figure 2A ) . We observed a ZEB1 protein increase of about 1 . 4- to 2 . 1-fold at day 2 compared to day 0 after which ZEB1 decreased to intermediate levels ( Figure 2A and Figure 2—figure supplement 1D ) . These results indicate that , even though ZEB1 is already highly expressed in pre-adipocytes , its nuclear protein level tends to further increase over the course of differentiation , which appears consistent with the enhancing effect of ZEB1 upon overexpression . This effect may be explained through post-transcriptional regulation . 10 . 7554/eLife . 03346 . 007Figure 2 . ZEB1 knockdown perturbs the expression of adipogenic regulators . ( A ) Protein levels ( fmol/μg nuclear extract ) of ZEB1 during 3T3-L1 differentiation ( one representative biological replicate ) . ( B ) Pparg2 and Cebpa mRNA levels after ZEB1 knockdown and overexpression in un-induced 3T3-L1 pre-adipocytes as measured by qPCR . ( C ) Expression levels [ln ( FPKM ) , ‘Materials and methods’] of mouse genes in ZEB1 KD vs . control cells at day 0 and day 2 after differentiation induction as measured by RNA-seq . Significantly up-regulated genes ( FC ≥ 1 . 5 , padj ≤ 0 . 01 ) are highlighted in blue , down-regulated genes in orange ( FC ≤ 0 . 67 , padj ≤ 0 . 01 ) , significantly de-regulated follow-up TFs as well as adipogenic TFs such as PPARγ and C/EBPs are indicated in black . Bar plots represent the percentage of genes that are significantly up- or down-regulated . Representative enriched GeneGO pathway categories for up- or down-regulated genes are highlighted ( complete results in Supplementary file 1C [Gubelmann et al . , 2014] ) . ( D ) Number of significantly up- or down-regulated genes belonging to previously defined expression clusters ( High/7/4/3 ) ( Mikkelsen et al . , 2010 ) . The typical expression pattern of genes in each cluster as well as of representative members that are significantly down-regulated upon ZEB1 KD is sketched . Clusters are sorted by increasing enrichment of down-regulated genes and corresponding p-values ( chi-square test ) are listed . ( E ) Distribution of gene expression FCs at day 0 after ZEB1 KD for genes annotated as positive or negative regulators of adipogenesis ( Supplementary file 1B [Gubelmann et al . , 2014] ) . Error bars depict the standard error of the mean . *p ≤ 0 . 01 and 0 . 01 < p ≤ 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 00710 . 7554/eLife . 03346 . 008Figure 2—figure supplement 1 . ZEB1 knockdown perturbs the expression of adipogenic regulators . ( A ) Relative ( to day −2 ) Zeb1 mRNA levels in wild-type 3T3-L1 cells during differentiation , as measured by qPCR . ( B ) Raw Ct values for Pparg , Zeb1 as well as the housekeeping gene HPRT1 at days 0 and 4 of 3T3-L1 differentiation as measured by qPCR . ( C ) Pparg and Zeb1 mRNA levels in pre-adipocytes and adipocytes derived from publicly available data through ArrayExpress ( Rustici et al . , 2013 ) . ( D ) Protein levels ( fmol/μg nuclear extract ) of ZEB1 during 3T3-L1 differentiation ( biological replicate of data shown in Figure 2A ) . ( E ) Pparg2 and Cebpa mRNA levels after ZEB1 knockdown and overexpression at day 4 after adipogenic induction as measured by qPCR . ( F ) Fold-changes of expression levels of selected adipogenic factors in response to ZEB1 KD as measured by qPCR and RNA-seq at day 0 and day 2 of 3T3-L1 differentiation . The Pearson's correlation coefficient ( r ) is indicated . ( G ) Number of significantly up- and down-regulated genes after ZEB1 knockdown belonging to previously defined expression clusters ( 2/5/Low/1 ) ( Mikkelsen et al . , 2010 ) . The typical expression pattern of genes in each cluster is sketched . Clusters are sorted by decreasing enrichment of up-regulated genes and corresponding p-values ( chi-square test ) are listed . ( H ) Changes in mRNA levels and POLII binding over gene bodies after ZEB1 and SMRT knockdown , respectively ( Raghav et al . , 2012 ) . The Spearman's ρ is indicated , showing a significantly negative correlation . ( I ) Percent of genes with dynamic SMRT binding during adipogenesis ( red ) , of genes that lose/gain POLII upon SMRT KD ( red ) , and of random genes ( grey ) that are significantly differentially expressed upon ZEB1 KD . Error bars depict the standard error of the mean . *p ≤ 0 . 01 and 0 . 01< p ≤ 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 008 We next assessed whether the expression of key adipogenic transcriptional regulators is sensitive to nuclear ZEB1 levels . Indeed , ZEB1 overexpression increases Pparg2 and Cebpa levels already in pre-adipocytes and later after induction of differentiation at day 4 ( Figure 2B and Figure 2—figure supplement 1E ) . Conversely , reducing ZEB1 levels prevents Pparg2 and Cebpa induction as measured at day 4 , and significantly reduces their expression in pre-adipocytes ( Figure 2B and Figure 2—figure supplement 1E ) . To gain global insights into gene expression alterations upon ZEB1 KD , we performed replicate RNA-seq experiments in control and ZEB1 KD cells prior to differentiation ( day 0 ) and 2 days after the onset of differentiation ( ‘Materials and methods’ ) . As expected , Zeb1 mRNA levels were significantly reduced in both data sets ( Figure 2C , FC cutoff 1 . 5 and padj ≤ 0 . 01 ) . Further , the expression fold-changes of several adipogenic TFs and markers measured by qPCR and RNA-seq were highly correlated ( Pearson's r ≥ 0 . 95; Figure 2—figure supplement 1F ) , validating expression estimates obtained by RNA-seq . In total , 3 , 426 ( 17% of all expressed ) and 3 , 221 ( 16% of all expressed ) genes were significantly de-regulated in ZEB1 KD cells compared to control samples at day 0 and day 2 , respectively ( Figure 2C and Supplementary file 1B [Gubelmann et al . , 2014] ) . We observed no difference between the fractions of genes that are significantly up- or down-regulated after KD ( Figure 2C ) , consistent with a summarizing report indicating that the regulatory function of ZEB1 is versatile and may be context-dependent ( Gheldof et al . , 2012 ) , although indirect regulatory effects cannot be excluded . Genes down-regulated upon ZEB1 KD as measured at day 2 are enriched for fat-cell-specific pathways such as ‘Putative pathways for stimulation of fat cell differentiation by Bisphenol A’ , ‘Role of Diethylhexyl Phthalate and Tributyltin in fat cell differentiation’ and ‘RXR-dependent regulation of lipid metabolism via PPAR , RAR , and VDR’ ( Figure 2C and Supplementary file 1C [Gubelmann et al . , 2014] ) . Pathway enrichment analysis of genes whose expression changed already in pre-adipocytes revealed a clear distinction between up- and down-regulated genes . While the former enriched for cell cycle and cytoskeleton remodeling-related processes , the latter singled out various developmental pathways previously implicated in adipogenesis ( James , 2013 ) , including ‘TGF-beta-dependent induction of EMT via SMADs’ , ‘Ligand-dependent activation of the ESR1/AP1 pathway’ , ‘WNT signaling pathway Part 2’ , and ‘Notch Signaling Pathway’ ( Figure 2C and Supplementary file 1C [Gubelmann et al . , 2014] ) . Specifically , reducing ZEB1 expression in 3T3-L1 cells had a strong impact on the expression of the adipogenic master regulators Cebpa , Cebpd as well as the known adipogenic TFs Nr3c1 and Krox20 ( Supplementary file 1B [Gubelmann et al . , 2014] and Figure 2C and Figure 2—figure supplement 1F ) . Moreover , we found that the expression of four novel adipogenic candidates identified in the TF screen ( Zfp30 , Ebf3 , Msx1 , and Atoh8 ) is equally perturbed ( at day 0 ) by reduced ZEB1 levels , suggesting that they may belong to the same regulatory module ( Figure 2C and Supplementary file 1B [Gubelmann et al . , 2014] ) . To better understand the gene regulatory function of ZEB1 , we analyzed the proportion of up- and down-regulated genes in each of nine distinct gene clusters that were previously derived from genome-wide microarray-based expression data generated during 3T3-L1 terminal differentiation ( Mikkelsen et al . , 2010 ) . Strikingly , genes showing decreased expression upon confluence under normal conditions ( i . e . , expression at day −2 is higher than at day 0 ) were significantly up-regulated at day 0 in ZEB1 KD cells ( Figure 2—figure supplement 1G ) . The enrichment was strongest for genes in cluster 2 , which were previously associated with mitosis and cell cycle functions . This is consistent with the pathway analysis results reported above and suggests that ZEB1 may control the expression of genes involved in coordinated cell cycle arrest , an essential step in 3T3-L1 differentiation ( Tang et al . , 2003 ) . Conversely , we found that a striking majority of genes whose expression increases upon confluence ( day −2 expression lower than day 0 expression ) were downregulated at day 0 upon ZEB1 KD ( Figure 2D ) . The strongest signal was observed for cluster 3 genes , which were previously associated with cell adhesion and extracellular matrix functionality . We note that several other EMT factors such as Snai2 ( previously shown to be a promoter of adipogenesis [Pérez-Mancera et al . , 2007] ) , Twist1 , Id2 and Id3 fall in this cluster and were significantly down-regulated upon ZEB1 KD . To test how ZEB1 KD specifically impacts the expression of known adipogenic TFs in a statistical threshold-independent manner , we directly compared expression fold-changes of positive and negative regulators ( Supplementary file 1B [Gubelmann et al . , 2014] ) of fat cell differentiation ( Figure 2E ) . In pre-adipocytes , positive regulators tended to be down-regulated upon ZEB1 KD while negative regulators were up-regulated ( p = 0 . 007 , Wilcoxon rank-sum test comparing fold-changes ) , suggesting that ZEB1 acts primarily as a positive regulator in the context of the aGRN , consistent with observations from our phenotype-profiling experiments . We recently dissected the role of SMRT ( NCOR2 ) , a transcriptional co-repressor , in fat cell differentiation and showed how it maintains genes in a poised state until adipogenesis is induced ( Raghav et al . , 2012 ) . If ZEB1 is , as hypothesized , an activator of early adipogenic genes , which are partially repressed by SMRT , we expect ( 1 ) a high overlap between genes regulated by SMRT and ZEB1; and ( 2 ) SMRT and ZEB1 KD to have inverse effects on the expression of the genes that they control . Indeed , we found that the fold-changes induced in POLII binding upon SMRT KD and in gene expression upon ZEB1 KD were significantly negatively correlated ( Spearman's ρ = −0 . 36 , p < 10−16 ) ( Figure 2—figure supplement 1H ) . For example , while SMRT KD significantly increased POLII levels at the positive adipogenic regulators Cebpa and Id4 ( Murad et al . , 2010 ) , ZEB1 KD highly reduced their mRNA levels in pre-adipocytes . Overall , we found that over 40% ( compared to an expected 14% ) of both the genes that lose SMRT binding upon fat cell differentiation and those that experience a POLII binding change upon SMRT KD are significantly ( p < 0 . 004 Fisher's exact test and permutation test ) de-regulated upon ZEB1 KD ( Figure 2—figure supplement 1I ) . Collectively , these results provide molecular evidence that ZEB1 is required for the adipogenic transcriptional program by controlling mRNA levels of a broad range of genes during adipogenesis . Specifically , ZEB1 promotes the expression of established key adipogenic regulators such as PPARγ and C/EBPα , and balances against pathways that repress and in favor of those that mediate terminal differentiation . To delineate the direct transcriptional effects of ZEB1 from other indirect layers of regulation , we next performed ZEB1 ChIP-seq in 3T3-L1 pre-adipocytes ( ‘Materials and methods’ ) . Replicate experiments including a pull-down with anti-HA in ZEB1-HA overexpressing 3T3-L1 cells showed high ChIP enrichment compared to negative controls as well as high correlation of read counts ( Spearman's ρ ≥ 0 . 83 ) ( Figure 3A and Figure 3—figure supplement 1A–C ) . ZEB1 exhibited widespread DNA binding ( 27 , 854 target regions ) , including at the genomic locus of the adipogenic master regulator Pparg ( Figure 3A–B ) . We note that ZEB1 preferentially bound gene-proximal locations , with almost 25% of ZEB1 targets overlapping TSSs and the vast majority of regions located within 10 kb of an annotated gene ( Figure 3B and Figure 3—figure supplement 1D for comparison with random regions , POLII , and C/EBPβ distribution ) . The association of ZEB1 with promoters as well as with CpG islands and exonic regions is highly significant at a genome-wide scale ( p < 10−16 , Supplementary file 1D [Gubelmann et al . , 2014] ) . 10 . 7554/eLife . 03346 . 009Figure 3 . ZEB1 co-binds the genome with established adipogenic regulators such as C/EBPβ . ( A ) ZEB1 , C/EBPβ , POLII , and Control ( CTRL ) read density tracks at the Pparg locus . ( B ) Number of ZEB1-bound regions and of their proximal ( ≤ 10 kb ) genes in 3T3-L1 cells . Distribution of ZEB1 binding with respect to genomic annotation ( ‘Materials and methods’ ) . ( C ) De novo motif discovery using MEME and a 50 bp sequence centered on ZEB1 peak summits reveals the canonical ZEB1 motif ( p = 10−8 , ‘Materials and methods’ ) ( D ) Motif enrichment analysis in a 100 bp window around ZEB1 peak summits reveals 138 significantly enriched motifs ( complete results in Supplementary file 1E ) . Highlighted here are motif names of the known early adipogenic regulators C/EBPβ , NFI , and AP1 factors as well as RUNX and SMAD3 . ( E ) Peak overlap between ZEB1 and C/EBPβ ( day 0 ) as well as AP1 factors ( day 0 , 4 hr ) in 3T3-L1 cells . ( F ) Overview of ZEB1 , C/EBPβ , AP1 proteins ATF2 and ATF7 , POLII normalized ChIP-seq as well as DNase-seq ( DHS ) enrichments ( ‘Materials and methods’ ) in a 2 kb window around the summits of ZEB1 peaks that overlap C/EBPβ binding . Intervals are sorted based on decreasing ZEB1 enrichment . ( G ) Summarized results from mass spectrometry experiments of proteins that were identified when pulling down ZEB1 ( complete results in Supplementary file 1F [Gubelmann et al . , 2014] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 00910 . 7554/eLife . 03346 . 010Figure 3—figure supplement 1 . ZEB1 co-binds the genome with established adipogenic regulators such as C/EBPβ . ( A ) ZEB1 ChIP-qPCR validation of ChIP-seq data at 12 selected ZEB1 target sites and three negative control ( CTRL ) regions during 3T3-L1 adipogenesis . ( B ) Scatterplot and Spearman's ρ of ZEB1 ChIP-seq and ZEB1-HA ChIP-seq read counts inside genomic intervals defined by ZEB1 binding in pre-adipocytes . ( C ) Spearman correlations between read counts for replicate ZEB1 ChIP-seq ( including ZEB1-HA ) as well as publicly available POLII and DNase-seq data ( Siersbæk et al . , 2011; Raghav et al . , 2012 ) inside genomic intervals defined by ZEB1 binding in pre-adipocytes . ( D ) Distribution of randomly shifted ZEB1 , C/EBPβ , and POLII peaks with respect to genomic annotation ( ‘Materials and methods’ ) . ( E ) ZEB1 motif density at 800 bp centered on ZEB1 peak summits . ( F ) Fraction of ZEB1 and randomly shifted ZEB1 peaks ( to show background values ) that contain at least one or two , respectively , ZEB1 , CACCTG ( E-box ) , C/EBPβ , AP1 , NFIC and SMAD3 motif hits ( ‘Materials and methods’ ) . ( G ) Peak overlap between randomly shifted ZEB1 and C/EBPβ bound regions in 3T3-L1 pre-adipocytes . ( H ) Overview of ZEB1 , ZEB1-HA , C/EBPβ , AP1 factors ATF2 and ATF7 , POLII and H3K9AC normalized ChIP-seq as well as DNase-seq and control ( CTRL ) enrichments ( ‘Materials and methods’ ) in a 2 kb window around the summits of ZEB1 peaks . Intervals are sorted based on decreasing ZEB1 enrichment . ( I ) Mean C/EBPβ and AP1 complex proteins JUN and FOSL normalized ( to total read number ) ChIP-seq enrichments in human HepG2 and lymphoblastoid cell lines ( LCLs ) in a 8 kb window around the summits of ZEB1 peaks detected in LCLs . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 01010 . 7554/eLife . 03346 . 011Figure 3—figure supplement 2 . ZEB1 binding and expression is affected by C/EBPβ KD . ( A ) C/EBPβ expression ( mRNA level ) in stable C/EBPβ KD and control 3T3-L1 pre-adipocytes as measured by qPCR . ( B ) C/EBPβ ChIP-qPCR at 10 C/EBPβ-ZEB1 , 6 ZEB1-only and six negative control regions according to our ZEB1 D0 ChIP-seq data and publicly available C/EBPβ ChIP-seq data ( Siersbæk et al . , 2011 ) . 5 out of 6 ZEB1-only regions also show C/EBPβ ChIP enrichment . C/EBPβ-enriched regions ( C ) ZEB1 ChIP-qPCR at 9 C/EBPβ-ZEB1 regions as well as one ZEB1-only region ( 10 ) in C/EBPβ KD and control 3T3-L1 cells . regions showing changes in ZEB1 enrichment after C/EBPβ KD ( D ) Zeb1 expression ( mRNA level ) in stable C/EBPβ KD and control 3T3-L1 cells as measured by qPCR . ( E ) POLII , ZEB1 , and C/EBPβ read density tracks at the Zeb1 locus in 3T3-L1 pre-adipocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 011 De novo motif analysis using 50 bp around the ZEB1 peak summit ( Figure 3C ) revealed a top-scoring motif that strongly matched MA0103 . 2 ( TOMTOM p=10−8 ) , which was previously annotated as the ZEB1 position weight matrix ( PWM ) in Jaspar ( Gupta et al . , 2007 ) ( Figure 3C ) . Although the ZEB1 motif was preferentially located proximal to the point of highest ZEB1 ChIP enrichment ( Figure 3—figure supplement 1E ) , only a subset ( 24% ) of ZEB1-bound regions contained such a motif match ( Figure 3—figure supplement 1F ) . To discover additional enriched sequence patterns of ZEB1-bound regions , we extended the motif search to 100 bp around the peak summit . We obtained a large list of significantly enriched motifs ( Supplementary file 1E [Gubelmann et al . , 2014] ) , including AP1\|Jun-AP1 , NF1 , RUNX , C/EBPβ , and SMAD3 motifs ( p ≤ 10−12 ) ( Figure 3D and Figure 3—figure supplement 1F ) , of which several occurred multiple times within the same ZEB1-bound region ( Figure 3—figure supplement 1F ) . Together , these results suggest that , while there is sequence specificity to ZEB1 binding , the protein is most likely recruited to a broad number of regulatory regions , either by or in cooperation with other DNA binding proteins . To verify whether ZEB1 targets previously identified ( pre ) adipogenic regulatory regions , we used publicly available C/EBPβ and AP1 binding data in 3T3-L1 cells ( Siersbæk et al . , 2011; Siersbæk et al . , 2014 ) . We found that almost one third of ZEB1-bound regions were co-bound by C/EBPβ , which represents a significant enrichment over random expectation ( Fisher's exact test , p ≤ 10−12 ) ( Figure 3E ) . Both ZEB1 and C/EBPβ ChIP-seq signal intensities were high at these 5 , 032 genomic locations , which were also enriched for AP1 factor binding ( in particular ATF7 and ATF2 ) , high DNase I signal , and to a lesser extent POLII ( Figure 3F ) . Strikingly , 27% of regions bound by ZEB1 at day 0 were also bound by at least one of the five AP1 complex proteins JUNB , FOSL2 , CJUN , ATF7 , and ATF2 4 hr after induction of differentiation , again a highly significant co-localization ( Fisher's exact test , p ≤ 10–12 ) ( Figure 3E and Figure 3—figure supplement 1G ) . Globally , regions showing higher ZEB1 enrichment co-localized more with TF binding and with genomic marks of transcriptional activity such as POLII and H3K9AC as well as open chromatic regions as indicated by high DNase I signal intensity ( Figure 3—figure supplement 1H ) . Indeed , genome-wide correlations between ZEB1 and POLII ChIP-seq read counts were high ( Figure 3—figure supplement 1C ) . Finally , we also analyzed the overlap between ZEB1 , AP1 , and C/EBPβ in the human HepG2 as well as lymphoblastoid cell lines ( LCLs ) ( data from ENCODE [Gerstein et al . , 2012] ) . We observed consistent enrichment of C/EBPβ , JUN , and FOSL ( AP1 complex members ) around ZEB1 peak summits ( Figure 3—figure supplement 1I ) , providing support for a conserved molecular relationship between these TFs in mediating DNA binding . To test whether the observed genomic co-localization of ZEB1 , C/EBPβ , and AP1 factors relies on physical interactions , we assayed ZEB1 interaction partners in 3T3-L1 cells by mass spectrometry ( Figure 3G and Supplementary file 1F [Gubelmann et al . , 2014] ) . We first verified that we could recover the ZEB1 protein as well as its known interaction partners such as CtBP1/2 and HDAC1/2 ( Gheldof et al . , 2012 ) . The presence of these four known partners among the 89 stringently selected ( ‘Materials and methods’ ) and reproducibly pulled down proteins confirmed the validity of our approach . Importantly , we also detected C/EBPβ at similar stringency and additionally , RUNX2 , NFIX , and AP1-family members ATF2 and ATF7-specific peptides in one of the replicate experiments ( Figure 3G ) . These data suggest that , in 3T3-L1 cells , ZEB1 is located within the same protein complex as at least one of the major regulators of adipogenic gene expression ( C/EBPβ ) and potentially also cooperates with other adipogenic TFs such as ATF2 or ATF7 to mediate transcription . To further assess the functional relationship between ZEB1 and C/EBPβ binding , we performed ZEB1 ChIP experiments after stable C/EBPβ knockdown in 3T3-L1 cells ( Figure 3—figure supplement 2A ) , testing 10 ZEB1-C/EBPβ co-bound regions , six ZEB1-only regions , and six negative control regions ( Supplementary file 1I [Gubelmann et al . , 2014] ) based on our ZEB1 and publicly available C/EBPβ ChIP-seq data ( Siersbæk et al . , 2011 ) . We first validated these C/EBPβ-bound regions by performing C/EBPβ ChIP-qPCR in our ( control ) 3T3-L1 cells ( Figure 3—figure supplement 2B ) . We found that C/EBPβ was almost invariably enriched at ZEB1-bound regions , even at regions ( five out of six ) that did not show C/EBPβ enrichment in the publicly available ChIP-seq data and that were thus presumed to be bound by ZEB1 only ( Figure 3—figure supplement 2B ) . Thus , the genomic co-localization of the two proteins may be even more widespread than originally appreciated , strengthening their functional relationship . In stable C/EBPβ knockdown cells , we observed a decrease of both C/EBPβ and ZEB1 DNA binding at the majority of the tested regions with the exception of the single ZEB1-only bound region , suggesting that ZEB1 DNA-binding is dependent on C/EBPβ ( Figure 3—figure supplement 2C ) . However , when we measured Zeb1 mRNA levels in the C/EBPβ KD cells , we observed a two-fold decrease compared to control ( Figure 3—figure supplement 2D ) . This result implies that C/EBPβ mediates Zeb1 expression , which is substantiated by the fact that C/EBPβ directly targets the Zeb1 gene in 3T3-L1 pre-adipocytes ( Figure 3—figure supplement 2E ) . Thus , at this point , we cannot exclude that the observed decrease in ZEB1 DNA binding in C/EBPβ KD cells may in part be a consequence of decreased cellular ZEB1 levels . Further experiments will be required to examine putative DNA binding cooperativity effects between these two TFs . Collectively , these observations demonstrate that in committed pre-adipocytes , ZEB1 is bound to open/active regulatory regions already before the onset of adipogenesis . Many of these regions are targeted by first-wave adipogenic TFs such as C/EBPβ and AP1-family members , implying a functional relationship between these TFs and , given the provided evidence , especially between ZEB1 and C/EBPβ in early adipogenic regulatory events . To better understand the dynamic regulatory properties of ZEB1 , we profiled its DNA binding during 3T3-L1 fat cell differentiation by including days −2 , 2 , and 4 in addition to day 0 . The vast majority of ZEB1-bound regions ( 42 , 050 ) showed consistent enrichment across all time points ( Figure 4—figure supplement 1A ) , reflected in high correlations among samples ( Spearman's ρ ≥ 0 . 75 day 4 vs . any other day; Figure 4—figure supplement 1B ) . However , ZEB1 binding profiles prior to ( days −2 and 0 , subsequently referred to as ‘early’ ) and post ( days 2 and 4 , referred to as ‘late’ ) differentiation induction were more similar to one another , respectively , forming two distinct clusters ( Figure 4—figure supplement 1B ) . Globally , we detected 552 early-only and 803 late-only ZEB1-bound regions ( FC ≥ 2 , FDR 0 . 1 , Figure 4A–B and Figure 4—figure supplement 1A , C , ‘Materials and methods’ ) . Interestingly , genomic loci of several adipogenic regulators , including Pparg , Klf15 , and Zbtb16 ( Mikkelsen et al . , 2010; Asada et al . , 2011 ) contained regions with increased ZEB1 binding after differentiation induction ( Supplementary file 1B [Gubelmann et al . , 2014] , Figure 4A and Figure 4—figure supplement 1C ) . 10 . 7554/eLife . 03346 . 012Figure 4 . ZEB1 binding increases at adipogenic genes during 3T3-L1 differentiation . ( A ) ZEB1 and POLII read density tracks at the Klf15 locus during 3T3-L1 differentiation ( days −2 , 0 , 2 , and 4 ) . Late-only bound regions are highlighted . ( B ) ZEB1 , C/EBPβ , RXRα , PPARγ , POLII normalized ChIP ( ‘Materials and methods’ ) as well as DHS enrichments in a 2 kb window around the summits of late-only ( days 2 and 4 but not days −2 and 0; padj ≤ 0 . 1 , FC ≥ 2 ) ZEB1-bound regions during 3T3-L1 differentiation . ( C ) Differential motif discovery using MEME and a 50 bp sequence centered on summits of late-only vs . static ZEB1 peaks reveals adipogenic motifs: C/EBPα\|C/EBPβ , NFIC and PPARG::RXR ( p < 10−3 , ‘Materials and methods’ ) . ( D ) GREAT-based ( McLean et al . , 2010 ) Gene Ontology enrichment analysis of genes associated with late-only vs . static ZEB1 binding reveals terms associated with fat cell differentiation and function ( complete results in Supplementary file 1G [Gubelmann et al . , 2014] ) . ( E ) Fraction of genes associated with late-only ZEB1 binding and fraction of all genes significantly up ( blue ) and down ( orange ) -regulated after ZEB1 KD as measured at differentiation day 2 ( complete results in Figure 4—figure supplement 1G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 01210 . 7554/eLife . 03346 . 013Figure 4—figure supplement 1 . ZEB1 binding increases at adipogenic genes during differentiation . ( A ) Overview of ZEB1 , C/EBPβ , RXRα , PPARγ , POLII normalized ChIP-seq as well as DNase-seq enrichments ( ‘Materials and methods’ ) in a 2 kb window around the summits of static ( padj ≥ 0 . 1 or FC < 2 ) and early-only ( days-2 and 0 but not days 2 and 4; padj ≤ 0 . 1 , FC ≥ 2 ) ZEB1-bound regions during 3T3-L1 differentiation . ( B ) Spearman correlations between read counts for ZEB1 ChIP-seq data at distinct adipogenic time points ( days −2 , 0 , 2 , and 4 ) inside genomic intervals defined by ZEB1 binding at any of these time points . ( C ) ZEB1 and POLII read density tracks at the Zbtb16 and Pparg loci during 3T3-L1 differentiation ( days-2 , 0 , 2 , and 4 ) . Summarized genome-wide results are included in Supplementary file 1G . ( D ) Differential motif discovery using MEME and a 50 bp sequence centered on summits of early-only vs static ZEB1 peaks reveals non-adipogenic motifs: RUNX1/2 and TEAD1 ( p < 10−5 , ‘Materials and methods’ ) . ( E ) GREAT-based ( McLean et al . , 2010 ) Gene Ontology enrichment analysis of genes associated with early-only vs static ZEB1 binding reveals terms associated with chemokine secretion and non-adipogenic functions . Full results are displayed in Supplementary file 1G . ( F ) Number of significantly up- or down-regulated genes associated ( ≤10 kb ) with at least one early-only or late-only ZEB1 bound region , respectively , belonging to previously defined expression clusters ( 1/Low/7/5 ) ( Mikkelsen et al . , 2010 ) . The typical expression pattern of genes in each cluster is sketched . Clusters are sorted by increasing enrichment of late-only ZEB1-bound genes and corresponding p-values ( chi-square test ) are listed . Only clusters showing a highly significant p-value ( p < 10−10 ) are shown . ( G ) Fraction of genes associated with early-only , late-only , and static ZEB1 binding as well as the fraction of all genes significantly up ( blue ) and down ( orange ) -regulated after ZEB1 KD as measured at differentiation days 0 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 013 We next contrasted dynamically and statically ZEB1-bound regions in terms of their sequence properties , enrichment for other factors , as well as types of genes in their proximity . Early-only ZEB1-bound regions enriched for the non-adipogenic motifs RUNX1/2 ( p < 10−5 ) and TEAD1 ( p < 10−5 ) , while late-only binding regions for the sequence motifs C/EBPα/β ( p < 10−6 ) , NFIC ( p ≤ 10−3 ) , and PPARγ::RXR ( p < 10−6 ) ( ‘Materials and methods’; Figure 4C and Figure 4—figure supplement 1D ) , indicating that ZEB1 may in part relocate to adipogenic-specific regulatory regions after differentiation induction . Consistently , late-only ZEB1-bound regions showed strong C/EBPβ binding as well as high PPARγ and RXRα enrichments upon differentiation ( Figure 4B ) . Moreover , these same regions also featured high DHS and POLII signals at days 2 and 4 , respectively , suggesting that they become accessible and transcriptionally active during differentiation in contrast to early-only regions , which showed no to little DNase I hypersensitivity or POLII binding in 3T3-L1 pre-adipocytes ( Figure 4B and Figure 4—figure supplement 1A ) . As these regions are largely already bound by C/EBPβ in pre-adipocytes ( Figure 4B ) , it is possible that ZEB1 specifically relocates there upon differentiation induction to mediate gene activation . Functionally , late-only ZEB1 peaks were associated with genes annotated as important for fat cell differentiation , insulin response , and lipid storage among others ( Supplementary file 1G [Gubelmann et al . , 2014] , Figure 4D ) , a property distinct from early-only regions ( Supplementary file 1G , Figure 4—figure supplement 1E ) . Previously defined patterns of gene expression across 3T3-L1 differentiation ( Mikkelsen et al . , 2010 ) further support this emerging , functional distinction of early and late ZEB1 enrichment . Specifically , we found that genes marked by early-only ZEB1 binding ( ≤ 10 kb distance , ‘Materials and methods’ ) were much more likely to have low expression in pre-adipocytes or to be repressed upon differentiation ( clusters ‘1’ and ‘Low’ ) ( Figure 4—figure supplement 1F ) . On the other hand , late-only genes enriched for strong induction at either early of late time-points of adipogenesis ( expression clusters ‘7’ and ‘5’ ) ( Figure 4—figure supplement 1F ) . To further assess the functional impact of ZEB1 depletion on these late- vs . early-only bound genes , we integrated the expression data into our analyses ( ‘Materials and methods’ ) . The most notable observation was that a large fraction of late-only genes ( here defined as genes that contain at least one late-only bound region but no early-only one ) are down-regulated upon ZEB1 KD , with almost a quarter of them being significantly lower expressed at differentiation day 2 ( Figure 4E and Figure 4—figure supplement 1G ) , corresponding to an almost three-fold enrichment compared to statically bound genes ( p ≤ 10–12 , Fisher's exact test ) . Collectively , these results are consistent with the observation that an important fraction of genes that are induced during adipogenesis show increased proximal ZEB1 binding after addition of the differentiation cocktail . We conclude that ZEB1 DNA binding is largely static during adipogenesis and given the strong overlap with DNase I hypersensitive sites may therefore contribute to establishing the regulatory landscape in pre- and mature adipocytes . Nevertheless , a subset of ZEB1 target sites is clearly dynamically bound . Genes that acquire ZEB1 binding after day 0 tend to be also bound by C/EBPβ , up-regulated after differentiation induction , particularly sensitive to alternation of nuclear ZEB1 levels and overall enriched for adipogenic functions . This suggests that ZEB1 may be involved in mediating the regulatory switch in primed pre-adipocytes toward the terminal differentiation program . To gain a complete overview of all levels at which ZEB1 regulates adipogenesis , we specifically focused on the aGRN that we manually assembled and curated based on recent reviews as well as Wikipathways ( Rosen and MacDougald , 2006; Kelder et al . , 2012; Siersbæk et al . , 2012 ) . We displayed the network as well as fold-changes per gene after ZEB1 KD and information on ZEB1 and C/EBPβ binding using Pathvisio ( van Iersel et al . , 2008 ) ( Figure 5A ) . The vast majority of network members are directly targeted by ZEB1 and their expression significantly decreases upon ZEB1 KD ( p = 2 × 10−4 based on permutation , ‘Materials and methods’ ) . The number of bound and regulated genes is significantly greater than expected by chance alone , suggesting that ZEB1 is a central component of the aGRN . As expected based on the co-localization of ZEB1 and C/EBPβ , the majority of ZEB1 targets are also targeted by C/EBPβ , suggesting that the two factors cooperate to promote adipogenesis . 10 . 7554/eLife . 03346 . 014Figure 5 . ZEB1: a central component of the adipogenic regulatory network . ( A ) Effect of ZEB1 knockdown on the adipogenic gene regulatory network . The network was assembled on the ‘Adipogenesis’ Pathway scaffold in WikiPathways as well as reviews and most recent publications of novel adipogenic regulators ( Rosen and MacDougald , 2006; Kelder et al . , 2012; Siersbæk et al . , 2012 ) . ZEB1 and C/EBPβ-bound regions that are proximal ( within 500 bp ) to TSSs or genes in pre-adipocytes are highlighted . Significant ( padj ≤ 0 . 01 ) expression changes after ZEB1 KD at day 0 of 3T3-L1 differentiation . Other candidate adipogenic regulators identified by our high throughput screen are listed . ( B ) Expression changes of adipogenic commitment genes after ZEB1 KD in 3T3-L1 pre-adipocytes as measured by RNA-seq . Displayed genes are either part of the pre-adipocyte expression signature derived by Gupta et al . ( 2010 ) or of the list of pre-adipocyte commitment factors compiled by Cawthorn et al . ( 2012 ) . Lpl and Igfbp4 occur in both lists . Significant differences in expression ( padj ≤ 0 . 01 ) are marked in orange ( FC ≤ 0 . 67 ) and blue ( FC ≥ 1 . 5 ) . Black-grey squares depict ZEB1 binding to TSSs or gene bodies . ( C ) Effect of ZEB1 knockdown and overexpression on C3H10T1/2 adipogenesis as assessed by Oil Red O staining at day 7 and day 8 , respectively after induction . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 01410 . 7554/eLife . 03346 . 015Figure 5—figure supplement 1 . ZEB1 regulates adipogenic commitment factors . ( A ) Effect of ZEB1 knockdown on the expression [log2 ( FC ) mRNA] of adipogenic ( Adipoq , Cebpa , Ebf1 , Pparg2 ) and EMT ( Snai1 , Snai2 , Twist1 ) factors as measured by qPCR at day 8 after induction . ( B ) Effect of ZEB1 overexpression on the expression [log2 ( FC ) mRNA] of adipogenic ( Cebpa , Ebf1 , Pparg2 ) , pre-adipogenic ( Zfp423 , Zfp521 ) , and EMT ( Snai1 , Twist1 ) factors as measured by qPCR at day 0 after induction of differentiation . ( C ) Western Blot showing PPARγ induction upon ZEB1 overexpression ( visualized using anti-HA antibody ) in C3H10T1/2 cells using PCNA as a normalization control . R1-3 indicates biological replicates . Error bars depict the standard error of the mean . p ≤ 0 . 01 and 0 . 01 < p ≤ 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 015 One aspect emerging from the regulatory network display is that ZEB1 targets late onset master regulators such as C/EBPα and PPARγ , first wave regulators such as KLF4 and CREB , as well as early adipogenic commitment factors including ZFP423 , TCF7L1 , and EVI1 ( Figure 5A ) . We thus performed a more systematic survey of the extent of ZEB1 binding to genes previously involved in adipogenic commitment , as defined by ( 1 ) a specific 3T3 pre-adipocyte expression signature ( Gupta et al . , 2010 ) and ( 2 ) pre-adipocyte factors found to be regulated at the expression level ( recently compiled in Cawthorn et al . , 2012 ) . We found that the expression of about half ( 48% ) of these genes significantly changes after ZEB1 KD in 3T3-L1 cells , with a striking majority ( 87% ) of them being down-regulated . We note that this includes both transcriptional regulators such as the above-mentioned ZFP423 , ligands such as WNT6 and membrane receptors such as PDGFRs ( Figure 5B ) and that the majority of their encoding loci are directly targeted by ZEB1 in un-induced pre-adipocytes ( Figure 5B ) . Given ZEB1's involvement in mediating the expression of early adipogenic commitment factors , we next tested whether ZEB1 also regulates adipogenesis in uncommitted precursors such as the C3H10T1/2 MSCs . C3H10T1/2 cells can be committed to the adipocyte lineage by activation of the TGFβ pathway by BMP2/4 and these cells subsequently differentiate into mature adipocytes by the addition of the same hormonal cocktail as used for 3T3-L1 differentiation ( Pinney and Emerson , 1989; Tang et al . , 2004 ) . We found that upon shRNA-mediated reduction of ZEB1 levels , the differentiation of MSCs into adipocytes was significantly impaired despite addition of the commitment factor BMP-2 ( Figure 5C ) . In addition , central adipogenic TFs and markers , including Cebpa , Pparg2 , Ebf1 , and Adipoq , were significantly down-regulated upon ZEB1 KD ( Figure 5—figure supplement 1A ) . We also investigated the effect of ZEB1 overexpression in C3H10T1/2 cells , and observed a slight increase in differentiation potential compared to control cells ( Figure 5C ) . In addition , molecular analyses revealed a strong up-regulation of Pparg2 and Zfp423 mRNA levels ( Figure 5—figure supplement 1B ) as well as PPARγ protein levels ( Figure 5—figure supplement 1C ) , providing further evidence as to the importance of ZEB1 in controlling the expression of this adipogenic master regulator . Collectively , these results indicate that ZEB1 is important for adipogenesis of both 3T3-L1 cells and MSCs . Our findings using mouse pre-adipocyte and mesenchymal cell cultures support an active , regulatory role for ZEB1 throughout adipogenesis and speak for a context- and dose-dependent function of this versatile transcriptional regulator . To assess whether these results also translate to an in vivo context , we next investigated the impact of altered ZEB1 levels onto in vivo adipogenesis . Since Zeb1 knockout mice are not viable ( Higashi et al . , 1997 ) , we used a previously described method ( Kawaguchi et al . , 1998; Meissburger et al . , 2011 ) to probe the effect of ZEB1 overexpression or KD on in vivo adipose stromal-vascular fraction ( SVF ) cell differentiation . Specifically , GFP-expressing , Matrigel-embedded murine SVF cells that were transduced with lentiviral particles containing either shRNAs or constitutive overexpression constructs for ZEB1 or negative controls were transplanted into the subcutaneous layer of the mouse neck . Mice with transplants were then subjected to a high fat diet for 6 weeks after which the Matrigel pads were analyzed for adipocyte number ( Meissburger et al . , 2011 ) ( Figure 6A and Figure 6—figure supplement 1A ) . 10 . 7554/eLife . 03346 . 016Figure 6 . ZEB1 is required for adipogenesis in vivo in mouse and its expression levels correlate with adipogenic indicators in humans . ( A and B ) Adipocyte differentiation in stromal vascular fraction ( SVF ) transplants from different donor mice ( as indicated ) fed a high-fat diet for 6 weeks ( Meissburger et al . , 2011 ) . ( A ) Fat sections from representative samples of ZEB1-overexpressing and control SVF transplants stained with Hematoxylin ( blue ) and Eosin ( pink ) . ( B ) Fat cell content of the transplanted SVF cells containing ZEB1 and control overexpression or knockdown constructs . Error bars depict the standard error of the mean . p = 0 . 05 , one-sided Wilcoxon-rank sum test . ( C ) Zeb1 mRNA expression normalized to 36B4 in human subcutaneous SVF of obese subjects plotted against percent ex vivo differentiated adipocytes of human subcutaneous SVF , subject fat mass , and adiponectin levels . Spearman's ρ is indicated , p ≤ 0 . 01 and 0 . 01 < p ≤ 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 01610 . 7554/eLife . 03346 . 017Figure 6—figure supplement 1 . Analysis of the functional involvement of ZEB1 in mouse in vivo adipogenesis . ( A ) Hematoxylin ( blue ) and Eosin ( pink ) -stained fat sections from representative samples of mouse ZEB1 knockdown transplants as well as the corresponding control ( scrambled siRNA ) . ( B ) Number of nuclei per section ( average over three sections ) from ZEB1 overexpression , KD , or control SVF transplants . DOI: http://dx . doi . org/10 . 7554/eLife . 03346 . 017 Implanted SVF cells serving as controls produced a similar percentage of mature fat cells ( ∼12% ) , demonstrating high assay reproducibility . SVF cells overexpressing ZEB1 yielded a significantly ( p = 0 . 05 , Wilcoxon rank-sum test ) higher number of mature fat cells compared to control ( Figure 6B ) . Conversely , ZEB1 KD almost halved the formation of mature fat cells in the transplanted pad compared to the negative control ( Figure 6B ) . Thus , ZEB1 is also highly important for adipogenesis in vivo . To assess whether the observed ZEB1 overexpression or KD effects could be linked to proliferation changes of adipocyte precursor cells , we quantified the nuclei from the implant sections ( ‘Materials and methods’ ) . We did not observe any change in proliferation in the ZEB1 KD and overexpression samples compared to the respective controls , although ZEB1 KD tended to yield more variable data ( Figure 6—figure supplement 1B ) . These results suggest that the observed effect of ZEB1 on adipogenesis does not involve major changes in the extent of cell proliferation capacity , although more in-depth molecular studies will be required to substantiate these findings . Having demonstrated ZEB1's importance in mouse adipose biology , we asked whether its highly conserved ortholog would exert a similar function in humans . Interestingly , previous research linked sequence variation of the genomic locus in which Zeb1 is located to body fat distribution and obesity ( Hager et al . , 1998; Heid et al . , 2010 ) , supporting a possible role for ZEB1 in these processes . To investigate this , we used subcutaneous fat biopsies from a previously sampled cohort of 62 obese patients ( 41 females and 21 males ) with body mass indices ( BMI ) of 31–64 kg/m2 ( Meissburger et al . , 2011; Winkler et al . , 2013 ) . Specifically , we determined Zeb1 mRNA expression in the SVF from each patient's fat sample . We then correlated the resulting expression values to multiple adipose-relevant measures ( Supplementary file 1H [Gubelmann et al . , 2014] , ‘Materials and methods’ ) , including the ex vivo differentiation potential of human SVF from subcutaneous adipose tissue biopsies , as previously described ( Meissburger et al . , 2011 ) . Interestingly , we found strong positive correlations ( Spearman's ρ ≥ 0 . 40 , p ≤ 0 . 03 ) with adipocyte differentiation , total fat mass , as well as adiponectin levels ( Figure 6C and Supplementary file 1H [Gubelmann et al . , 2014] ) . In contrast , it was previously reported that Rorg gene expression levels show an inverse correlation with adipocyte differentiation potential , and no significant correlation with either fat mass or adiponectin ( Meissburger et al . , 2011 ) ( Supplementary file 1H [Gubelmann et al . , 2014] ) . Thus , Zeb1 SVF expression levels appear indicative of differentiation potential , fat mass , and adiponectin levels in humans , consistent with the positive effect of ZEB1 on adipogenesis observed in mice . Together , these results demonstrate the functional relevance of ZEB1 in the context of both mouse and human adipocyte biology . A comprehensive understanding of the regulatory mechanisms mediating adipocyte differentiation has great fundamental and medical value . Significant efforts have therefore previously been undertaken to uncover adipogenic regulators . These range from classical phenotype- and ‘guilt by association’-driven studies ( reviewed in Tang and Lane , 2012 ) , over enrichment analyses in adipogenic compared to non-adipogenic clonal lines ( Gupta et al . , 2010; Zhou et al . , 2013 ) , to chromatin state-based inference of key regulators ( Eguchi et al . , 2008; Mikkelsen et al . , 2010; Waki et al . , 2011 ) . Of late , more large-scale studies have started to emerge in both mouse and human systems ( Villanueva et al . , 2011; Söhle et al . , 2012 ) . Our study differs from these high-throughput screening approaches in that we set out to systematically identify novel transcriptional regulators of adipogenesis through an overexpression screen of TFs in 3T3-L1 pre-adipocytes . This was made possible due to the recent generation of a comprehensive library of mouse TF ORF clones ( Gubelmann et al . , 2013 ) , which allowed us to explore the completeness of the so-far assembled aGRN ( Figure 5A , [Tang and Lane , 2012] ) . We were able to identify several TFs having strong reproducible effects on fat cell differentiation . The majority of these have so far either never or only indirectly been linked to adipogenesis . For example , the only available information for the second ranking regulator ZFP30 is that it belongs to the family of KRAB zinc finger proteins , typically repressors with important roles in vertebrate development ( Urrutia , 2003 ) . Other examples include MSX1 and ATOH8 , which have been associated with differentiation processes in mesenchymal lineages but to our knowledge never with adipogenesis ( Cheng et al . , 2003; Jimenez et al . , 2007; Rawnsley et al . , 2013 ) . More generally , the identification of several novel adipogenic TFs suggests that , besides the well-characterized TF cascade including the C/EBP family of TFs and the canonical master regulator PPARγ , a substantial number of yet unconnected TFs also contribute to the adipogenic phenotype . This underscores the value of our screen , which provides a comprehensive resource that will accelerate the complete characterization of the core GRN underlying fat cell differentiation . The TF with the strongest effect on fat cell differentiation in our screen was ZEB1 . This protein ( also known as δEF1 , TCF8 , NIL2-a , BZP , AREB6 , MEB1 , ZFHX1a , and ZFHEP ) is a highly conserved and versatile TF that was originally identified as a repressor of the lens-specific δ1-crystalline enhancer in chicken ( Funahashi et al . , 1991 ) . It has subsequently been shown to be involved in a broad range of regulatory processes , including EMT , tumor metastasis , development , and differentiation ( reviewed in Gheldof et al . , 2012 ) , providing a rationale as to why homozygous Zeb1 null mice are not viable ( Higashi et al . , 1997 ) . Similar to the majority of the other TFs identified in our screen , little is known about the involvement of ZEB1 in fat cell differentiation . One report linked a ZEB1 gain-of-function mutation to increased adiposity ( Kurima et al . , 2011 ) , consistent with our findings; another found that female mice that are heterozygous for a targeted deletion of exon 1 of Zeb1 show increased adiposity ( Saykally et al . , 2009 ) , thus identifying ZEB1 as a repressor of adipogenesis . However , the latter study also reported that ZEB1 expression in parametrial fat increases as fat accumulates , in line with the results presented here . We therefore suspect that the phenotype observed in the studied ZEB1 haploinsufficient mice may not be related to adipocyte hyperplasia , but rather to hypertrophy , another well-established means of adipose tissue expansion . Thus , our findings consolidate and explain ZEB1's involvement in adipogenesis . Indeed , ( 1 ) the large effects that reduction of ZEB1 levels has on the adipogenic phenotype and on the expression of both early ( e . g . including ZFP423 , TCF7L1 and EVI1 ) and late ( e . g . PPARγ , C/EBPα , KLF15 ) members of the aGRN , ( 2 ) the fact that many of these TFs are also directly targeted by ZEB1 , and ( 3 ) the fact that ZEB1 operates at the crossroads of several pathways underlying adipogenesis collectively indicate that this TF is an integral part of the molecular identity of pre-adipocytes . This is further supported by earlier observations revealing that Zeb1 is expressed at high levels in adipose stem cells in vivo ( Kupershmidt et al . , 2010 ) , that it is significantly expressed higher in pre-adipocytes compared to adipocytes ( Kapushesky et al . , 2012 ) , and that it is one of relatively few genes that is highly up-regulated during transition of mouse embryonic stem cells to adipocytes ( Billon et al . , 2010 ) . Interestingly , we found that many of the newly identified TFs such as ZFP30 , MSX1 , and ATOH8 are also highly expressed in pre-adipocytes . In addition , they are all direct targets of ZEB1 and their expression is affected by ZEB1 KD , suggesting that they may also have a role in early adipogenesis . The identification of several , putatively novel early adipogenic regulators likely reflects the nature of our experimental set-up , which employs early induction of TF overexpression before the cells undergo terminal differentiation . The ability to detect such early regulators is valuable , because the uncharted gene regulatory territory is especially confined to early processes rather than late ones for adipogenesis ( Gupta et al . , 2010 ) . To provide a mechanistic understanding of how ZEB1 mediates adipogenesis and to further characterize its regulatory mode of action , we assessed genome-wide gene expression differences after ZEB1 KD at days 0 and 2 of fat cell differentiation . In addition , we performed ZEB1 ChIP-seq at distinct time points during adipogenesis , providing a first glance into the dynamic DNA binding properties of ZEB1 in any biological system . We found that ZEB1 DNA binding is widespread , to an extent similar to that of other essential regulators such as C/EBPβ , PU . 1 , or MYC ( Heinz et al . , 2010; Siersbæk et al . , 2011; Nie et al . , 2012 ) , preferentially promoter-proximal and largely static , consistent with its relatively stable expression profile during terminal fat cell differentiation . In terms of sequence specificity , we found that ZEB1-bound regions were not only enriched for E-boxes ( in particular ‘CACCTG’ ) as previously described , but also for a large number of DNA binding motifs of established adipogenic regulators , including C/EBPβ , AP1 , and NFI factors ( Waki et al . , 2011 ) . Together with its strong overlap with DNase hypersensitive sites , these results suggest that ZEB1 contributes to establishing the regulatory landscape in pre- adipocytes as well as mature adipocytes . Nevertheless , we found that a subset of ZEB1 target regions is dynamically bound . Genes that acquire ZEB1 binding upon differentiation induction tend to be co-bound by C/EBPβ , transcriptionally up-regulated , and enriched for adipogenic functions . These data suggest that ZEB1 may be involved in mediating the regulatory switch in primed pre-adipocytes toward the terminal differentiation program . Interestingly , a comparable mechanism has been uncovered in vascular smooth muscle cells , where ZEB1 has been shown to interact with SMAD3 and to synergistically activate the transcription of key differentiation genes ( Nishimura et al . , 2006 ) . In the context of adipocytes , we propose that ZEB1 forms a complex with C/EBPβ , as supported by our mass spectrometry data , and that these two TFs cooperate to control adipogenic gene expression . In contrast , we found that early-only bound regions were enriched for motifs of the osteogenic master regulator RUNX2 , raising the possibility that ZEB1 may bind to osteogenic regulatory regions ( likely in a repressive context ) , and may thus be involved in the bone-fat switch . This is consistent with ZEB1's previous implication in differentiation of various mesenchymal lineages , including osteogenesis , myogenesis , and chondrogenesis ( reviewed in [Gheldof et al . , 2012] ) . Importantly , murine ZEB1 shows strong conservation to its human ortholog and its gene was located within one of the top 16 regions of association in a recently performed meta-analysis of 32 body fat distribution genome-wide association studies ( Heid et al . , 2010 ) , consistent with an earlier report indicating a putative link between genomic variation in the ZEB1 locus and obesity ( Hager et al . , 1998 ) . Our data from obese patients underscore this importance . Specifically , we found that ZEB1 levels in the SVF correlate with differentiation potential , total fat mass , and circulating adiponectin levels . It is well accepted that adipocyte hyperplasia in obesity can lead to an increased fat cell number with smaller more insulin sensitive adipocytes ( Tilg and Moschen , 2006; Roberts et al . , 2009 ) . The positive correlation of ZEB1 expression with adiponectin levels , which confers insulin sensitivity in a wide variety of tissues , points towards such a scenario . Our study therefore breaks new ground with respect to investigating the functional significance of this association . 3T3-L1 mouse fibroblast cells and C3H10T1/2 mesenchymal stem cells , obtained from ATCC , were cultured in high-glucose Dulbecco's modified Eagle's medium ( DMEM , Life Technologies , Carlsbad , CA ) supplemented with 10% fetal calf serum ( FCS , Bioconcept , Switzerland ) , and 1x penicillin/streptomycin solution ( 1x Pen/Strep , Life Technologies ) in a 5% CO2 humidified atmosphere at 37°C and maintained at less than 80% confluence before passaging . Differentiation of 3T3-L1 cells was induced by exposing 2-day post-confluent cells ( day 0 ) to DMEM containing 10% FCS supplemented with 1 µM dexamethasone , 0 . 5 mM 3-isobutyl-1-methylxanthine , and 167 nM insulin ( Sigma , Saint-Louis , MO ) , a medium called MDI . Note that we did not add any rosiglitazone since the potent differentiation effect of this PPAR activator would have severely limited our ability to identify positive regulators of adipogenesis . After 2 days ( day 2 ) , cells were washed with Dulbecco's Phosphate-Buffered Saline 1x ( PBS , Bioconcept , Switzerland ) and were fed with DMEM containing 10% FCS and 167 nM insulin . 2 days later ( day 4 ) , the medium were changed to 10% FCS/DMEM . Full differentiation was usually achieved by day 6 . To identify the effect of ZEB1 overexpression or knockdown ( KD ) on differentiation of C3H10T1/2 cells into fat cells , 5 × 104 cells were plated per well of six well plates . When the cells were 80% confluent , BMP-2 ( 50 ng/ml , Life Technologies ) supplemented complete DMEM medium was added to the cells for commitment into pre-adipocytes . At confluence , day 0 , adipocyte differentiation induction medium ( MDI supplemented with 50 ng/ml BMP-2 ) was added to the cells . After 2 days the cells were washed once with 1x PBS and were fed with DMEM containing 10% FCS , 1x Pen/Strep solution containing 167 nM insulin and 50 ng/ml BMP-2 . This was followed by the addition of 500 nM rosiglitazone containing complete media at day 4 to the cells . Rosiglitazone containing medium was changed after 2 days and then fat-containing cells were stained with Oil Red O at day 8 . The generation of our 750 fully sequence-verified mouse TF ORF clones via Gateway cloning is explained in Gubelmann et al . ( 2013 ) . The 750 TF ORFs were subcloned from the pDONR221 entry clones into a Tet-On Gateway-compatible expression vector ( IRES-PURO , 3 HA tags and Gateway sites were added to the original TRE_GOI_rtTA_hPGK vector [Barde et al . , 2006] , by mixing 100 ng of TF entry clone with 100 ng of the expression vector and 0 . 5 µl of LR clonase II enzyme mix ( Life Technologies ) . After incubating 18 hr at 25°C , this mix was transformed into competent STBL3 cells . Successfully subcloned TFs ( 734 ) were miniprepped using the NucleoSpin 96 Plasmid miniprep kit ( Macherey–Nagel ) , typically yielding a concentration of approximately 150–300 ng/µl , from 2 ml overnight cultures in terrific broth II ( MP Biomedicals , Santa Ana , CA ) . We note that the adipogenic regulators C/EBPα and C/EBPβ were not successfully cloned . Lentivirus production was done using the TF ORF Tet-On Gateway-compactible expression vectors . Briefly , 293T packaging cells were seeded at 0 . 2 millions of cells/ml ( 100 µl per well ) in low-antibiotic growth medium ( DMEM , 10% iFBS ( HyClone ) , 0 . 1x Pen/Strep ) in a 96-well tissue culture plate ( Corning , Corning , NY ) . The cells were incubated until they reached 70% confluence . In a 96-well polypropylene storage plate ( Corning ) , 100 ng TF ORF Tet-On expression vector , 100 ng psPAX2 and 10 ng pMD2 . G plasmids were mixed with OPTI-MEM ( Life Technologies ) to a total volume of 25 µl per well . At the same time , lipofectamine 2000 Transfection Reagent ( Life technologies ) was mixed with OPTI-MEM in an eppendorf tube ( 0 . 5 µl lipofectamine with 24 . 5 µl OPTI-MEM per well ) and incubated for 5 min at room temperature . The transfection reagent mix was subsequently transferred to each well of the 96-well plasmid plate and the plate was incubated for 20 min at room temperature . Before transfection , the medium in the 96-well culture plate was changed with a fresh low-antibiotic growth medium , 100 µl per well , and the transfection mix ( 50 µl per well ) was carefully transferred to the packaging cells . The cells were incubated for 18 hr in a 5% CO2 humidified atmosphere at 37°C . The medium was then replaced with 170 µl high-BSA growth medium ( DMEM , 10% iFBS , 11 mg/ml microbiology-grade Bovine Serum Albumin ( VWR Radnor , PA ) , 1x Pen/Strep ) . After 24 hr of incubation ( 37°C , 5% CO2 ) , 150 µl per well was collected and stored at −20°C . The differentiation assay was done in a collagen-coated 384-well black tissue culture plate for the overexpression screen . 3T3-L1 cells were seeded at a density of 20 , 000 cells per well ( 100 µl ) and after 1 day , lentivirus was added to the medium ( ratio 1:1 or 1:2 with the medium ) with polybrene ( 8 µg/ml , Santa Cruz Biotechnology , Santa Cruz , CA ) . After 24 hr of incubation ( 37°C , 5% CO2 ) , the medium was replaced with complete DMEM ( DMEM , 10% FBS , 1% Pen/Strep ) to remove viruses . When confluence was reached , the medium was changed with complete DMEM , 1 µl/ml Doxycycline ( hyclate from Sigma ) and cells were incubated for 72 hr ( 37°C , 5% CO2 ) . Then , the cells were differentiated as described under ‘Cell culture and differentiation’ , with addition of doxycycline in the medium until day 2 . The cells were fixed and stained at day 7 of differentiation . The cells were washed with 100 µl 1x PBS per well and fixed by adding 1x PBS with 4% formaldehyde ( Sigma ) . The cells were incubated at room temperature for 1 hr or overnight at 4°C . The fixation solution was then removed , cells washed once with 1x PBS and 100 µl of 1x PBS was added . Plates were then either stored at 4°C or stained immediately . Before staining , the cells were washed two times with water ( 100 µl ) . The staining of lipids with the lipophilic , fluorescent dye BODIPY , nuclei with Hoechst , and complete cells with SYTO60 , the image acquisition ( CellWorx , Applied Precision ) and analyses were performed as described previously ( Meissburger et al . , 2011 ) . The percentage of differentiation ( PDC ) for each TF was obtained by dividing the number of cells having at least four lipid droplets by the total number of cells . Fold changes ( FCs ) were calculated by comparison to the differentiation percentages obtained with the original Tet-on expression vector ( 47 . 9% ( average ) of differentiation , Supplementary file 1A [Gubelmann et al . , 2014] ) . 307 TFs showing a FC > 1 are displayed in Figure 1B and those with a FC ≥ 1 . 5 ( i . e . , a 50% increase in the fraction of mature adipocytes ) in three technical replicates at a FWER of ≤ 0 . 05 ( each replicate normalized to negative controls using normal transform and normal two-sided p-values corrected for multiple testing with Bonferroni's correction ) were considered significant enhancers of adipogenesis , are referred to as ‘candidates’ in the manuscript , and are highlighted in red or orange in Figure 1B and in green in Supplementary file 1A ( Gubelmann et al . , 2014 ) . We assessed which of the 26 candidates were endogenously transcribed in 3T3-L1 cells as well as human adipose stromal cells using publicly available data ( Nielsen et al . , 2008; Mikkelsen et al . , 2010 ) : mRNA levels ( micro-array-based ) as well as RNA POLII genome-wide binding ( ChIP-seq ) . RNA POLII enrichment over gene bodies across all fat cell differentiation time points was used to determine whether genes were transcribed or not , based on a threshold adjusted by qPCR ( 10 sequence tags per 500 bp at the highest transcribed day; data not shown ) . We also used previously determined gene expression classes ( Nielsen et al . , 2008 ) based on mRNA levels during differentiation of both human and mouse pre-adipocytes . The three qualitative transcription measures are listed in Supplementary file 1A ( Gubelmann et al . , 2014 ) for each candidate and summarized in Figure 1—figure supplement 1B . We used Expression Atlas ( September 2013 version ) to compare the expression of the candidates in adipogenesis ( filters ‘adipose’ and ‘adipose tissue’ ) to their expression in other tissues . Nine and eight TFs , respectively , were significantly expressed higher and lower in adipose tissue compared to the average across all tissues ( p = 0 . 05 ) . This corresponds to a 1 . 5-fold enrichment of significantly higher expressed genes compared to random , an enrichment which is not significant ( p = 0 . 1 , 10 , 000 permutations ) . Viral particles containing Tet-on or shRNA expression plasmids were generated in 293T cells as described previously ( Barde et al . , 2010 ) , with slight modifications . Instead of five plates , one plate of 15 cm was transfected for each expression plasmid and the supernatant was harvested three times , every 8–12 hr . After the ultracentrifugation step ( 1 tube for each vector , 35 ml ) , the pellets were re-suspended in 35–45 µl of 1x PBS , centrifuged at maximum speed and stored at −80°C in 10 µl aliquots . For stable and selected lines , 3T3-L1 cells at a density of 5 × 104 cells were transduced with 10 µl viral particles in a six-well plate . After 12 hr , the medium was changed and puromycin ( Life Technologies , 2 μg/ml ) was added after 72 hr to select stably transduced cells . If the cell confluence reached 80% , the cells were split and transferred into larger dishes . After every 2 days , puromycin selection media was changed and the stably transduced cells were selected for 2 weeks before performing actual experiments . For the Tet-On Gateway-compatible expression vector , the original vector was used as a negative control . For the KD , the lentiviral mammalian vector pLKO . 1 containing specific shRNAs ( three shRNAs per target were used , listed below ) along with the control shRNA ( empty pLKO . 1 plasmid ) were obtained from Sigma . Finally , the shRNA 2 ZEB1 was not used in the shRNA pool of ZEB1 as the robustness of the cells after treatment was low . ShRNA: shRNA 1 PPARγ ( TRCN0000001657 ) , shRNA 2 PPARγ ( TRCN0000001658 ) , shRNA 3 PPARγ ( TRCN0000001660 ) , shRNA 1 ZEB1 ( TRCN0000235850 ) , shRNA 2 ZEB1 ( TRCN0000235851 ) , shRNA 3 ZEB1 ( TRCN0000235853 ) . As the KDs were not stable for ZEB1 in these generated cell lines , we directly infected 50 , 000 3T3-L1 cells/well in six-well plates ( three replicates per constructs ) . The medium was changed the following day , and when the cells reached confluence , we proceeded with the adipocyte differentiation protocol . Total RNA was isolated using a Qiagen RNAeasy plus mini kit according to the manufacturer's protocol by using either the RLT buffer lysis system or , for the RNA-seq samples , the TRIzol/Chlorophorm extraction procedure ( Sigma , Saint-Louis , MO ) . The RNA concentration and quality was determined using a nanodrop ( 1 . 8 ≤ A260/A280 ≤ 2 . 2 ) and by visual inspection of separated bands on agarose gels . 2 µg of total RNA was used for single strand cDNA synthesis using the SuperScript VILO cDNA Synthesis Kit ( Life Technologies ) . Then , cDNA was diluted 1:100 using nuclease free water and 1 . 5 µl was used for each qPCR reaction . Quantitative real-time PCR was performed in 384-well plates with three technical replicates on the ABI-7900HT Real-Time PCR System ( Applied Biosystems ) using the Power SYBR Green Master Mix ( Applied Biosystems ) using standard procedures . A Hamilton Liquid Handling Robotic System was used to assemble the 384-well plates . The qPCR primers were designed with in-house developed GETPrime software ( Gubelmann et al . , 2011 ) or taken from previous publications ( Supplementary file 1I [Gubelmann et al . , 2014] ) . They were checked for linearity and single product amplification . The in vivo experiments were performed as described in Meissburger et al . ( 2011 ) by using the shRNAs listed above and the ZEB1 as well as PPARγ ( positive control ) overexpression clones . In short: fat tissue was dissected , minced and incubated in collagenase type II for 1 hr at 37°C . Approximately 106 cells were treated with virus and resuspended in 100 μl of Matrigel ( BD Biosciences , San Jose , CA ) before injection subcutaneously into a skin fold of the neck . After 6 weeks , Matrigel pads were excised . From each pad pictures of three full sections were taken and adipocyte numbers as well as the number of nuclei were determined automatically using the Cell Profiler Software . To overcome the doxycycline effect of a Tet-on inducible lentiviral vector , the TF ORFs were subcloned from the pDONR221 entry clones into the pLenti6/UbC/V5-DEST expression vector ( Life Technologies , Carlsbad , CA ) . As negative controls , the original vector and the shEMPTY ( Sigma ) were used . All experiments were performed in three replicates and the significance of the observed changes was estimated using a one-sided Wilcoxon rank-sum test . Differentiation of the pre-adipocytes was induced as described previously ( Meissburger et al . , 2011 ) . RNA were extracted after day 8 of differentiation and analyzed by qPCR , as explained above . mRNA expression was normalized to 36B4 . For expression of Zeb1 , TaqMan gene expression assays ( Ambion Life Technologies , Carlsbad , CA ) were used according to the manufacturer's protocol . Differentiated cells were co-stained with adiponectin to verify differentiation . After isolation of RNA , as described in the section ‘RNA isolation and quantitative PCR’ , the Illumina Truseq RNA Sample Preparation kit v2 protocol ( Illumina , San Diego , CA ) was followed using 500 ng of RNA per sample as starting material . Half of the ligated reaction volume was used for PCR ( 14 cycles ) and the other half was kept at −80°C as a backup . Libraries were checked for quality and quantified using the Bioanalyzer 2100 ( Agilent , Santa Clara , CA ) , before being sequenced in barcoded pools of 16 samples on the Illumina Hiseq 2500 instrument ( 100 base paired-end sequencing , 4 lanes; Genomics Sequencing Facility , Nestle , Lausanne ) . Sequenced tags were aligned to Ensembl 70 gene annotation of the NCBI38/mm10 genome using Bowtie 1 . 0 . 0 and the parameters ‘-a --best --strata -S -m 100 --chunkmbs 256 -p 8’ . Expression levels per transcript and gene were estimated using mmseq-1 . 0 . 2 ( Turro et al . , 2011 ) with default parameters . Quantile normalized expression estimates were transformed into pseudo-counts by un-logging , un-standardising and multiplying with gene length . Expression differences between the samples were quantified with DESeq ( Anders and Huber , 2010 ) , FC ≥ 1 . 5 and padj ≤ 0 . 01 . Estimated FCs were validated by using qPCR-measured FCs at days 0 and 2 , revealing significant correlations with Pearson's r > 0 . 95 and p = 2 × 10−4 and p = 3 × 10−5 , respectively . 3T3-L1 stably selected cells were collected at days −2 , 0 , 2 and 4 for ZEB1 . The cells were fixed as described previously ( Raghav et al . , 2012 ) and stored at −80°C . Ten million cells were used for each immunoprecipitation ( IP ) . The ChIP experiment was performed as described previously in Raghav et al . ( 2012 ) , under ‘Chromatin Immunoprecipitation of SMRT’ . A ZEB1 antibody ( Santa Cruz , sc-25388 , 10 μg per IP ) and a rabbit isotype control IgG ( Santa Cruz Biotechnology , Santa Cruz , CA , sc-8994 , 10 μg per IP ) was used for each time point . The DNA was stored at −20°C until verification of ChIP enrichment by qPCR and ChIP-seq library preparation . For the ChIP of ZEB1-HA and C/EBPβ , the chromatin samples were incubated overnight at 4°C with an anti-HA antibody ( Abcam , UK , ab9110 , 5 μg per IP ) , an anti-C/EBPβ antibody ( Santa Cruz , sc-150 , 10 µg per IP ) respectively , or a rabbit control IgG ( Santa Cruz , SC-8994 ) coupled to magnetic beads ( sheep anti-rabbit IgG , Life Technologies , Carlsbad , CA Dynabeads ) , as described previously ( Kilpinen et al . , 2013 ) with few modifications . 50 μl of antibody-coupled beads were added to each 1 ml chromatin material instead of 100 μl . After incubation , we washed the beads five times with a LiCl wash buffer ( 100 mM Tris at pH 7 . 5 , 500 mM LiCl , 1% NP-40 , 1% sodium deoxycholate ) , mixed for 10 min at 4°C and removed remaining ions with a single wash with 1 ml of TE ( 10 mM Tris–HCl at pH 7 . 5 , 0 . 1 mM Na2EDTA ) at 4°C for 1 min mix . The beads were then resuspended in 200 μl IP Elution Buffer ( 1% SDS/0 . 1 M NaHCO3 ) , incubated in a 65°C shaker for 1 hr and placed on the magnet to recover the supernatant . The supernatant was incubated at 65°C overnight to complete the reversal of the formaldehyde cross-links . The next day , DNA was purified from the reverse-crosslinked chromatin by proteinase and RNase digestion followed by purification using Qiagen DNA purification columns . Multiplex libraries were prepared using barcoded adapters for each sample following the protocol described in Raghav and Deplancke ( 2012 ) with slight modifications . In brief , ChIP-DNA fragments were end-repaired using an End-IT DNA end repair kit ( Epicentre Technologies ) followed by the addition of an A-base and ligation of bar-coded adapters . After the ligation incubation , the DNA was cleaned up using Agencourt AMPure XP Beads ( Beckman Coulter , Fullerton , CA ) and eluted in 12 µl . These purified , ligated DNA fragments were separated on a 2% agarose gel to select 200–500 bp-sized DNA fragments and DNA was subsequently isolated from gel slices using a Qiagen gel extraction kit . The gel-extracted DNA was then amplified for 17 cycles by PCR using high fidelity Phusion hot start polymerase ( NEB , Ipswich , CA ) . The concentration and quality of purified , amplified DNA were estimated using respectively a Qubit dsDNA high sensitivity kit ( Life Technologies ) and a high sensitivity DNA assay Bioanalyzer 2100 ( Agilent ) . After quality confirmation , the DNA libraries were sequenced on an Illumina High Seq 2500 ( Genomics Sequencing Facility , Nestle , Lausanne; Genomic Technologies Facility , UNIL , Lausanne ) . Pass-filtered reads from the Illumina analysis pipeline were used for further analysis . For IP of ZEB1 protein complexes , 3T3-L1 cells stably overexpressing ZEB1 and control vectors were grown in 150-mm plates and ZEB1 expression was induced at day −2 ( the time of cell confluence ) using doxycycline as described before . At day 0 of differentiation , the cells were washed with 1x PBS and dissociated using trypsin-EDTA solution ( Life Technologies ) . The dissociated cells were collected in 15-ml falcon and centrifuged at 250 × g to pellet the cells . The cell pellet was washed once with 1x PBS containing 1 mm PMSF ( phenylmethylsulfonyl fluoride ) protease inhibitor and stored at −80°C . Before lysing the cells for IP , recombinant protein-A sepharose beads ( Life Technologies ) were prepared by washing with IP buffer ( 20 mM Tris-Cl , pH . 7 . 4 , 150 mM NaCl , 10% glycerol , 1% Triton X-100 , 1 mM EDTA , 1 mM DTT ) supplemented with protease and phosphatase inhibitors ( Roche ) . The anti-rabbit HA tag antibody ( Abcam , UK ) was conjugated with the protein-A sepharose beads by incubating antibody with the beads overnight at 4°C at rotation . The cells were then lysed in 1 ml IP buffer by douncing 20-25 times using 1 ml syringe . Lysates were cleared by centrifuging at maximum speed in a tabletop refrigerated centrifuge for 10 min . The tagged complexes were pulled down using 50 μl HA-antibody tagged beads for 4 hr at 4°C at rotation . The beads were then washed 5 times using IP buffer and the bound protein complexes were eluted by heating the beads in 60 μl 2x Laemmli sample buffer . To resolve the proteins present in pull down complexes , a 12% SDS-PAGE gel was used . Entire lanes of SDS-PAGE gels were then sliced into pieces . Samples were first washed twice for 20 min in 50% ethanol and 50 mM ammonium bicarbonate ( AB ) . Gel slices were dried down by vacuum centrifugation . All samples were reduced/alkylated using dithioerythritol and iodoacetamide . Gel pieces were dried again and re-hydrated using trypsin solution ( 12 . 5 ng/µl in 50 mM AB and 10 mM CaCl2 ) . Trypsin digestion was performed overnight and resulting peptides were extracted twice for 20 min in 70% ethanol and 5% Formic Acid ( FA ) . Samples were dried down and re-suspended in 2% acetonitrile and 0 . 1% FA for LC-MS/MS injections . One-dimensional liquid chromatography separation was performed on a Dionex Ultimate 3000 RSLC nano UPLC system ( Dionex , Sunnyvale , CA ) on-line connected with an Orbitrap Q Exactive Mass-Spectrometer ( Thermo Fischer Scientific ) . A self-made capillary pre-column ( Magic AQ C18; 3 μm-200 Å; 2 cm × 100 μm ID ) was used for sample trapping and cleaning . Analytical separation was then performed using a C18 capillary column ( Nikkyo Technos Co , Japan; Magic AQ C18; 3μm-100 Å; 15 cm × 75 μm ID ) at 250 nl/min . Separation of peptides was carried over an 85 min biphasic gradient . Mass spectrometric measurements were performed using a data-dependent top 20 method , with the full-MS scans acquired at 70 K resolution ( at m/z 200 ) and MS/MS scans acquired at 17 . 5K resolution ( at m/z 200 ) . A database search was performed using Mascot 2 . 3 ( Matrix Science ) and SEQUEST in Proteome Discoverer v . 1 . 3 against a mouse database ( UniProt release 2013_09 ) . All searches were performed with trypsin cleavage specificity , up to two missed cleavages were allowed and ion mass tolerance of 10 ppm for the precursor and 0 . 05 Da for the fragments . Carbamidomethylation was set as a fixed modification , whereas oxidation ( M ) , acetylation ( Protein N-term ) , phosphorylation ( STY ) were considered as variable modifications . Data were further processed and inspected in the Scaffold 4 software ( Proteome Software , Portland , OR ) . Sequenced ChIP tags were aligned to the NCBI38/mm10 genome with Bowtie2 ( Langmead and Salzberg , 2012 ) and the parameters ‘--very-sensitive -M 10 -p 8’ . Duplicates were removed , as well as the reads with a mapping quality under 10 . Regions showing significant ChIP enrichment ( peaks ) with CCAT 3 . 0 ( Xu et al . , 2010 ) ( ‘fragmentSize 200 slidingWinSize 100 movingStep 50 isStrandSensitiveMode 1 minCount 13 minScore 5 bootstrapPass 80’ , FDR 0 . 1 ) or Homer ( Heinz et al . , 2010 ) ( ‘-F 3 -L 4 -localSize 5000 -C 4 -fragLength 200 -minDist 500 -center’ ) using anti-HA ( in Zfp277-HA overexpressing cells , where Zfp277-HA serves as negative control as it is not imported into the nucleus , data not shown ) as control were merged into common ZEB1-bound regions . This resulted in a total of 43 , 405 ZEB1 bound regions ( 10 , 708 day −2 , 27 , 854 day 0 , 18 , 145 day 2 , 22 , 249 day 4 ) and 19 , 055 ZEB1-HA bound regions ( day 0 ) . Read counts contained in the genomic intervals defined by day 0 and merged ZEB1 binding , respectively , were correlated and the Spearman's ρ used to generate hierarchically clustered heatmaps ( Figure 3—figure supplement 1B–C and Figure 4—figure supplement 1B ) . As biological replicates were only available for days −2 and 0 , we compared the read counts contained in the genomic intervals defined by merged ZEB1 binding at early time points ( two replicates day −2 and high enrichment replicate day 0 treated as replicates ) to those at late time points ( single replicate day 2 and single replicate day 4 treated as replicates ) using DESeq 1 . 2 . 3 . We obtained 803 regions showing significantly ( padj ≤ 0 . 1 , FC ≥ 2 ) higher ZEB1 ChIP-seq read counts at days 2 and 4 vs . days −2 and 0 ( late-only ZEB1 binding ) and 552 regions showing significantly higher ZEB1 ChIP-seq read counts at days −2 and 0 vs . days 2 and 4 ( early-only ZEB1 binding ) . Genes overlapping at least one late-only bound region ( gene body +500 bp upstream of TSS ) and no early-only bound region were considered ‘late-only genes’; conversely , genes overlapping at least one early-only bound region ( gene body +500 bp upstream of TSS ) were considered ‘early-only genes’ . Genes overlapping only statically ZEB1-bound regions were considered ‘static-only genes’ . Motif discovery was performed with HOMER's findMotifsGenome ( Heinz et al . , 2010 ) ( “-size 100 -len 6 , 8 , 10 , 12 , 14 -local” ) on all ZEB1-bound regions at day 0 ( summary results are displayed in Figure 3D and complete results listed in Supplementary file 1E [Gubelmann et al . , 2014] ) and with MEME 4 . 9 . 1 ( Bailey et al . , 2006 ) ( ‘-nmotifs 10 -minsites 10 -minw 4 -maxw 20 -revcomp -maxsize 60 , 000 -dna’ ) using 50 bp centered on the summits of the most highest-scoring 1000 ZEB1-bound regions at day 0 ( the top enriched motif is displayed in Figure 3C ) . Motif matching to known motif databases was performed using TOMTOM 4 . 9 . 1 ( Bailey et al . , 2006 ) and ‘All vertebrates’ database . Motif scanning of bound regions ( 100 bp centered around the summit of ZEB1-bound regions at day 0 as well as randomly shifted regions for background comparison ) was performed with the PWMs available through Wang et al . ( 2012 ) and Jaspar ( Bryne et al . , 2008 ) for ZEB1 , CEBPB , AP1 , NFIC and SMAD3 using the package Biostrings 2 . 30 . 0 at a cutoff of 85% . Percentage of peaks showing at least one and at least two PWM matches are displayed in Figure 3—figure supplement 1F . ZEB1 PWM matches in an 800 bp region centered on ZEB1 peak summits ( at day 0 ) were displayed as a motif density plot in Figure 3—figure supplement 1E . Differential motif discovery contrasted early-only and late-only ZEB1-bound regions with static ( padj > 0 . 1 , FC < 2 ) ones . Early-only ZEB1 binding enriched for RUNX ( MEME E = 7 × 10−29 , matching RUNX2/MA0511 . 1 , TOMTOM p = 2 × 10−6 and RUNX1/MA0002 . 2 , TOMTOM p = 7 × 10−6 ) and TEAD1 ( MEME E = 2 × 10−7 , matching TEAD1/MA0090 . 1 , TOMTOM p = 5 × 10−6 ) motifs , while late-only binding enriched for C/EBP ( MEME E = 9 × 10−179 , matching C/EBPα /MA0102 . 3 , TOMTOM p = 1 × 10−12 and C/EBPβ/MA0466 . 1 , TOMTOM p = 2 × 10−7 ) , NFI ( MEME E = 4 × 10−12 , matching NFIC/MA0161 . 1 , TOMTOM p = 9 × 10−5 and TLX::NFIC/MA0119 . 1 , TOMTOM p = 1 × 10−3 ) and PPARG::RXR ( MEME E = 3 × 10−6 , matching PPARG::RXR/MA0065 . 2 , TOMTOM p = 4 × 10−7 ) motifs , as displayed in Figure 4C and Figure 4—figure supplement 1D . ZEB1 bound regions and regulated genes were annotated using Ensembl 70 . Peaks were assigned ( in this order ) to either TSS ( ± 500 bp of annotated TSS ) , exonic regions ( but not TSS ) , intronic regions ( but not TSS or exons ) , gene proximity ( 10 kb distance to a gene , but not TSS or gene ) , or gene-distal regions ( none of the above ) and the fraction belonging to these categories displayed as pie charts in Figure 3B . For comparison , we performed the same analysis for randomly shifted ZEB1 , C/EBPβ ( day 0 ) , and POLII ( day 0 ) bound regions ( Figure 3—figure supplement 1D ) . Additionally , we used HOMER's annotatePeaks with default options to assign each peak to its closest gene and report the p-values obtained using the hypergeometric test in the manuscript as indicative of a significant association of ZEB1 binding with promoters , CpG islands and exons . Complete results are included in Supplementary file 1D ( Gubelmann et al . , 2014 ) . GO term enrichment analysis of ZEB1-regulated genes was performed with GeneGO MetaCore ( https://portal . genego . com/ ) , representative summaries are displayed in Figure 2C and complete results included in Supplementary file 1C . GO term enrichment analysis on ZEB1-bound regions was performed with GREAT using default parameters ( McLean et al . , 2010 ) , contrasting late-only and early-only regions with static ones . Selected results are displayed in Figure 4D and Figure 4—figure supplement 1E and complete ones are included in Supplementary file 1G ( Gubelmann et al . , 2014 ) . We used the following publicly available data: ( 1 ) in 3T3-L1 cells: microarray-based gene expression , ChIP-seq using antibodies against JUNB , FOSL , CJUN , ATF7 , ATF2 , C/EBPβ , PPARγ , POLII , H3K9AC , RXRα , DNAse-seq ( Nielsen et al . , 2008; Steger et al . , 2010; Siersbæk et al . , 2011; Raghav et al . , 2012; Siersbæk et al . , 2014 ) ; ( 2 ) pre-adipocyte vs . adipocyte data and meta-analysis of mouse tissue expression data available through Array Express ( September 2013 ) ( Rustici et al . , 2013 ) ; ( 3 ) Human ENCODE HepG2 and lymphoblastoid cell lines ( LCLs ) : ChIP-seq with antibodies against C/EBPβ , FOSL , JUN and ZEB1 ( ENCODE Project Consortium , 2012 ) . All ChIP-seq and DNase-seq data were reanalyzed analogous to the in-house generated data . The human data was aligned to the GRCh37 ( hg19 ) genome . Overlaps between regions bound by ZEB1 ( or randomly shifted ZEB1 ) and C/EBPβ at day 0 as well as AP1 factors ( JUNB , FOSL , CJUN , ATF7 and ATF2 ) at 4 hr ( day 0 ) are displayed as Venn diagrams using the R package VennDiagrams 1 . 6 . 5 in Figure 3E and Figure 3—figure supplement 1G . Read counts were divided by normalization factors estimated using DESeq2 and shifted ZEB1 peaks as genomic intervals and extended to 200 bp each . Counts were then summed across 400 windows of 5 bp each ( total 2 kb ) centered around ZEB1 peak summits and log2 transformed . These transformed values ( referred to as ‘norm ChIP’ in Figures 3F and 4B , Figure 3—figure supplement 1H , and Figure 4—figure supplement 1A ) were displayed as heatmaps using the R package pheatmap 0 . 7 . 7 . We note that only a subset ( top , mid and bottom 4000 peaks sorted by mean ZEB1 day 0 ChIP enrichment ) is displayed in Figure 3—figure supplement 1H and only the two AP1 factors showing the highest colocalization ( data not shown ) with ZEB1—ATF2 and ATF7—are displayed in Figure 3F and Figure 3—figure supplement 1H . Similarly , counts for the human ChIP-seq datasets were normalized to the library size , extended to 200 bp and log2 transformed . Mean values in a 8 kb region centered on ZEB1 peak summits in lymphoblastoid cell lines ( LCLs ) were then plotted in Figure 3—figure supplement 1I . Genomic loci plots were visualized in the UCSC Genome Browser based on bedGraph files obtained using HOMER's makeUCSCfile and the parameters ‘-o auto -res 1 -fsize 5e7' . The scale used for each individual track is displayed in Figures 3A and 4A , Figure 3—figure supplement 2E and Figure 4—figure supplement 1C . We used Wikipathways , as well as recent reviews to manually compile an adipogenic transcriptional regulatory network ( Kelder et al . , 2012; Rosen and MacDougald , 2006; Siersbæk et al . , 2012 ) and displayed it using Pathvisio ( van Iersel et al . , 2008 ) . We then superimposed the expression information ( log2 FCs and multiple-testing corrected p-values after ZEB1 KD at day 0 ) as well as ZEB1 and C/EBPβ binding information ( at day 0: overlapping TSS or overlapping the gene ) on the network in Figure 5A . All mouse experiments were conducted in strict accordance with Swiss law and all experiments were approved by the ethics commission of the state veterinary office ( 60/2012 , 43/2011 ) . The work on obese subjects was approved by the ethics committee at the University Hospital of Heidelberg and is conforming to the ethical guidelines of the 2000 Helsinki declaration . All participants provided witnessed written informed consent prior to entering the study ( S-365/2007 ) . The trial was registered as NCT00773565 . ChIP-seq data are available in the ArrayExpress database ( www . ebi . ac . uk/arrayexpress ) under accession number E-MTAB-2537 . RNA-seq data are available in the ArrayExpress database under accession number E-MTAB-2538 . Results of the TF overexpression screen , processed RNA-seq data , mass spectrometry results and the clinical data ( Supplementary file 1A , Supplementary file 1B , Supplementary file 1F and Supplementary file 1H ) have additionally been deposited in the Dryad data repository under doi: 10 . 5061/dryad . j966f ( see Gubelmann et al . , 2014 ) .	The growing rates of obesity and related metabolic diseases are a major public health concern worldwide . People who are overweight or obese are at increased risk for a range of diseases including diabetes and heart disease , which may reduce their quality of life and shorten their lifespans . Obese people have more , larger fat cells than individuals of healthy weight , and so understanding how the body creates fat cells may provide new insights into ways to prevent or treat obesity . Fat cells arise from a population of stem cells that can also give rise to bone cells and cartilage . Some of the proteins—called transcription factors—that work together to turn on the expression of genes needed for a stem cell to become a fat cell have been identified . However , the exact regulatory processes that cause an unspecialized cell to develop into a fat cell remain unclear . Gubelmann et al . set out to identify more of the transcription factors that cause stem cells to become fat cells . A high-throughput , automated process was used to test the effect of over-expressing each of 734 transcription factors in mouse cells that are primed to become fat cells . Twenty-six transcription factors were found to increase the number of these primed cells that became mature fat cells—most of which had not previously been shown to affect how fat cells develop . The most powerful driver of fat cell development was ZEB1: a transcription factor that has previously been implicated in many other biological processes . Most notably , ZEB1 was linked to a transition during development that allows cells to migrate to the correct location in the embryo , but also to a mechanism that allows cancerous cells to spread to new tissues . Using studies of mouse cells and live mice , computational analyses , and biopsies from obese patients , Gubelmann et al . show that ZEB1 regulates numerous other transcription factors that promote the development of fat cells . These include factors that initially set an unspecialized cell on the path to becoming a fat cell and those that guide the changes that occur as the fat cell matures . Further studies will be required to show whether the ZEB1 protein itself is needed to prime stem cells to start becoming fat cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	Identification of the transcription factor ZEB1 as a central component of the adipogenic gene regulatory network
Cell-type specification through differential genome regulation is a hallmark of complex multicellularity . However , it remains unclear how this process evolved during the transition from unicellular to multicellular organisms . To address this question , we investigated transcriptional dynamics in the ichthyosporean Creolimax fragrantissima , a relative of animals that undergoes coenocytic development . We find that Creolimax utilizes dynamic regulation of alternative splicing , long inter-genic non-coding RNAs and co-regulated gene modules associated with animal multicellularity in a cell-type specific manner . Moreover , our study suggests that the different cell types of the three closest animal relatives ( ichthyosporeans , filastereans and choanoflagellates ) are the product of lineage-specific innovations . Additionally , a proteomic survey of the secretome reveals adaptations to a fungal-like lifestyle . In summary , the diversity of cell types among protistan relatives of animals and their complex genome regulation demonstrates that the last unicellular ancestor of animals was already capable of elaborate specification of cell types . The process by which multicellular animals develop from a unicellular zygote is believed to mirror the first evolutionary steps that led to the origins of animal multicellularity from a unicellular species ( King , 2004 ) . As the development of complex multicellularity is dependent upon differential genome regulation , the evolutionary onset of animal multicellularity must likewise have involved the appearance of differential genome regulatory capacities leading to distinct cell types . Many of the genes involved in the control of animal development and cell type identity , including signaling pathways and transcription factors ( TFs ) , pre-date animal origins ( Sebe-Pedros et al . , 2011; Sebé-Pedrós et al . , 2012; Suga et al . , 2012; Richter and King , 2013 ) . As these genes are known to be present in the genomes of the protistan relatives of animals , the unicellular holozoans ( King et al . , 2008; Suga et al . , 2013; Fairclough et al . , 2013 ) , the evolution of complex multicellularity , with cell type-specific transcriptional programs , must have involved changes in gene regulation . Therefore , a key step in understanding the evolution of multicellularity will be to infer the regulatory complexity of the last common ancestor of all living animals . To address ancestral regulatory complexity , it will be necessary to elucidate the molecular control of cell differentiation through development in the unicellular relatives of animals . Three distinct developmental modes that lead to transient simple multicellular forms have been described in the protistan relatives of animals ( Figure 1 ) ( Sebé-Pedrós et al . , 2013; Suga and Ruiz-Trillo , 2013; Dayel et al . , 2011 ) . Colonial clonal development has been shown to involve differential regulation of a few multicellularity-related genes in the choanoflagellate Salpingoeca rosetta ( Fairclough et al . , 2013 ) . On the other hand , the filasterean amoeba Capsaspora owczarzaki exhibits up-regulation of adhesion-related genes in its aggregative stage ( Sebé-Pedrós et al . , 2013 ) . To date , however , there has been no molecular characterization of coenocytic development , a third and completely distinct mode of development observed in the ichthyosporeans . Ichthyosporeans are the earliest branching holozoan lineage ( Torruella et al . , 2012; Paps et al . , 2013; Torruella et al . , 2015 ) , and coenocytic development is a shared feature within the group ( Glockling et al . , 2013; Mendoza et al . , 2002 ) . This developmental mode comprises a growth stage in which nuclei divide synchronously within a common cytoplasm before undergoing cellularization , followed by release of motile ameboid zoospores ( Suga and Ruiz-Trillo , 2013; Marshall et al . , 2008 ) . These stages have distinct physiological and structural characteristics , the amoeboid stage is mono-nucleated , non-diving , and motile ( Figure 1B ) , while the multinucleate stage has a cell wall , a big central vacuole , and does not move ( Figure 1C ) . Despite being quite distinct from canonic animal development , coenocytic development and multinucleate cell types are found in some animal lineages , such as in Drosophila syncytial blastoderm ( Suga and Ruiz-Trillo , 2013 ) . Thus , in order to infer ancestral regulatory complexity by comparison of premetazoan developmental modes , it will be essential to obtain a detailed molecular characterization of ichthyosporean coenocytic development . 10 . 7554/eLife . 08904 . 003Figure 1 . Evolution of developmental and feeding modes across holozoans . ( A ) The cladogram represents known phylogenetic relationships among holozoans ( Torruella et al . , 2012; Torruella et al . , 2015 ) . Each lineage is represented by the species proposed as a model system with a schema of its developmental mode on the right . The evolution of specialized osmotrophy is shown as a blue triangle in the cladogram , while the putative ancestral phagotrophic feeding mode of opisthokonta is shown as an orange circle ( Cavalier-Smith , 2012 ) . Divergence times of the lineages shown in this figure range between 700 Mya ( considered the latest estimates of animal origins ) and 1200 Mya ( earliest estimates of Opisthokont origins ) ( Sharpe* et al . , 2015 ) . Micrographs depicting the ( B ) amoeboid stage and ( C ) multinucleate stage of Creolimax fragrantissima are shown . Scale bars = 10 μm . Choanoflagellate adapted from Mark Dayel ( CC BY-SA 3 . 0 ) www . dayel . com/blog/2010/10/07/ choanoflagellate-illustration . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 00310 . 7554/eLife . 08904 . 004Figure 1—figure supplement 1 . Creolimax synchronized stages . ( A ) Culture after 5 μm filtering . ( B ) Culture grown for 24 hr after 5 μm filtering . Scale bars = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 004 Here we describe the transcriptome dynamics of Creolimax fragrantissima , which has been proposed as a model system for ichthyosporeans ( Suga and Ruiz-Trillo , 2013; Marshall et al . , 2008 ) . We show that Creolimax employs complex gene regulation including alternative splicing and the use of long intergenic non-coding RNAs ( lincRNA ) . Through analysis of the Creolimax secretome in silico and by proteomics , we also provide evidence of secondary adaptation to a specialized osmotrophic feeding mode through lateral gene transfer ( LGT ) and gene duplication . Taken together , our results suggest that the last common unicellular ancestor of animals was already capable of implementing elaborate , cell type-specific differentiation programs . As a reference genome for mapping RNA sequencing ( RNA-seq ) data , we assembled the 45 Mb draft genome of Creolimax from a combination of 454 reads and mate-paired end Illumina reads corresponding to a 75× coverage . The draft assembly comprises 82 scaffolds , with an N50 of 1 . 5 Mb ( see Materials and methods ) . The genome is twice as large as that of Capsaspora but in line with sequenced choanoflagellate genomes . We annotated 8695 genes , 92% of which are supported by transcriptional evidence . A diverse array of annotation pipelines rendered functional information on 78% of the predicted protein coding genes ( see Material and methods ) . Among those genes , many belong to gene families involved in multicellularity and development in animals , such as TFs and signaling pathways , as previously reported in targeted gene family analysis ( Suga et al . , 2014; de Mendoza et al . , 2013; de Mendoza et al . , 2014 ) . We isolated two different stages of the Creolimax lifecycle by taking advantage of the cell size difference between the amoeboid stage and the multinucleate growth stages ( Marshall et al . , 2008 ) . After filtering with a 5 μm mesh , we obtained a highly enriched culture of amoebae . The amoebae encysted and grew for at least 24 hr ( Figure 1—figure supplement 1 ) , and then multinucleate coenocytic cysts matured asynchronously , releasing new amoeboid zoospores with different cell sizes to form a heterogeneous culture . Given the drastic morphological difference between the non-motile mitotic multinucleate stage pre-dominant in the 24 hr culture and the motile non-dividing amoeba isolated after filtration , we decided to investigate their transcriptomic differences through RNA-seq ( see Material and methods ) . The amoeboid and the multinucleate stage showed distinct transcriptional profiles , consistent among replicates ( Figure 2A ) , from which we identified 956 genes as significantly differentially expressed . Functional enrichment analyses of Gene Ontologies ( GOs ) and PFAM domains ( p<0 . 01 , Fisher’s exact test ) revealed that the multinucleate stage shows up-regulation of genes associated with cell growth , including ribosome , translation , DNA replication , amino acid and RNA metabolism activities ( Figure 2 , Figure 2—figure supplement 1 ) . Conversely , in the amoeboid stage , we found enrichment for signaling activities ( GTPases and kinases ) and an up-regulation of the actin cytoskeleton , most likely involved in the motile behavior of the amoeba ( see raw GOs in Figure 2—source data 1 ) . Strikingly , single-celled amoebas also showed up-regulation of extracellular matrix ( ECM ) adhesion , including the up-regulation of the integrin pathway components . 10 . 7554/eLife . 08904 . 005Figure 2 . Differential gene expression in Creolimax . ( A ) Diagram of the amoeboid and multinucleate stages , and heatmap showing the significantly differentially expressed genes across biological replicates in the pair-wise stage comparison . ( B ) Gene set enrichment analysis for the two stages . Orange represents enrichment in the amoeboid stage and blue represents enrichment in the multinucleate stage , color intensity depicts level of significance ( p value ) . Node size represents the total number of genes in each GO , and edge width represents the total number of genes shared between each enriched GO category . Functionally related GOs are manually circled in gray shade according to functional and genic redundancy established by network connectivity . Complete list of GOs and inclusive groupings are found in Figure 2—source data 1 . GOs , Gene Ontologies . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 00510 . 7554/eLife . 08904 . 006Figure 2—source data 1 . GOs enrichments and groupings from Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 00610 . 7554/eLife . 08904 . 007Figure 2—figure supplement 1 . Pfam domain set enrichments in differentially expressed genes . Orange represents domains enriched in the amoeboid stage and blue represents domains enriched in the multinucleate stage , color intensity depicts level of significance ( p value , Fisher’s exact test ) . Node size represents the total number of genes containing a Pfam domain and edge width represents the total number of genes sharing two distinct Pfam domains . Functionally related Pfams are manually circled in gray shade , primarily based on the information gathered in the Pfam database for each domain ( including Pfam2Go annotations ) . Additional criteria to include a given domain in a functionally related category included: checking the list of GOs of the statistically differentially expressed domain-containing genes in a given stage and using a network connectivity redundancy between GO and Pfam categories in a mixed network ( including both Pfam and GO annotation ) done with Enrichment Map plugin in Cytoscape ( Merico et al . , 2010 ) . GO , Gene Ontologies . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 00710 . 7554/eLife . 08904 . 008Figure 2—figure supplement 2 . Differential TFome expression . ( A ) Distribution of expression values ( FPKM ) for all TFs in the genome in the amoeboid and the multinucleate stages ( p value , Wilcoxon signed rank test ) . ( B ) Heatmap showing the expression levels of all the TFs with a two-fold change in expression level between stages . Those with the gene ID in bold have statistically significant differential expression according to at least three different differential expression pipelines ( see Material and methods ) . To the right , the domain architectures of the TFs show the DNA binding domain in dark blue and orange according to the stage where they are up-regulated . In the multinucleate stage: CSD ( PF00313 ) , Myb ( PF00249 ) , CBFB_NFYA ( PF02045 ) and TPR ( PF13414 ) . In the amoeboid stage: bZIP_1 ( PF00170 ) , HLH ( PF00010 ) , T-box ( PF00907 ) , RFX_DNA_binding ( PF02257 ) , Runt ( PF00853 ) . CSD: Cold shock protein; FPKM , fragments per kilobase of exon per million fragments mapped; HLH: helix-loop-helix ( protein structure ) ; TFs , transcription factorsDOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 008 Besides signaling and adhesion activities , regulation through TFs is also considered to be a defining characteristic of animal development ( de Mendoza et al . , 2013; Levine and Tjian , 2003 ) . To address whether that is the case for the multinucleate stage in Creolimax , we compared the stage-specific expression level of all the TFs encoded in the genome . Contrary to our expectation , the comparison revealed that TF expression is higher in the amoeboid stage ( p<0 . 05 , Wilcoxon signed rank test , Figure 2—figure supplement 2A ) . Among the 127 putative TFs found in the genome , only 5 are significantly up-regulated in the multinucleate stage . These five genes have DNA binding domains with ambiguous TF activity and include genes containing Myb_DNA-binding and cold-shock domains ( Figure 2—figure supplement 2B ) . In the amoeboid stage , we found significant up-regulation of 11 TFs , including T-box , Runx and hemophagocytic lymphohistiocytosis TFs , and TF classes that have important roles in animal development ( Sebe-Pedros et al . , 2011 ) . Altogether , these data indicate that , unlike in animal development , the multinucleate stage of Creolimax is not characterized by a tight regulation of adhesive , signaling , and TF activities . Alternative splicing is a widespread mechanism for post-transcriptional gene regulation in animals and many other eukaryotes . To evaluate the extent to which alternative splicing expands transcriptome complexity in the intron-rich genome of Creolimax ( 6 . 5 introns per gene ) , we undertook a comprehensive analysis of intron retention and exon skipping ( ES ) events , and quantified their inclusion levels using previously developed pipelines ( Braunschweig et al . , 2014; Irimia et al . , 2014 ) ( see Material and methods ) . We observed 3927 intron-retention events that affected 2172 genes . In contrast , we found only 211 genes affected by ES . Despite having a two-fold higher intron density and longer introns compared to Capsaspora , both holozoan species show comparable alternative splicing profiles that are highly dominated by intron retention ( Sebé-Pedrós et al . , 2013 ) . This provides further support for the hypothesis that ES-dominated alternative splicing is a unique feature of animal transcriptomes ( Sebé-Pedrós et al . , 2013; Mcguire et al . , 2008; Irimia and Roy , 2014 ) . Next , we evaluated the extent to which intron retention is differentially regulated across Creolimax stages . We found 865 introns with differential retention between stages ( differential percent intron retention ( ΔPIR ) >15 ) , the majority of which were more retained in the amoeboid stage ( Figure 3A ) . Reverse transcription-polymerase chain reaction ( RT-PCR ) assays confirmed differential retention for all 10 tested introns ( Figure 3D and Figure 3—figure supplement 2 ) . Despite these marked differences in intron retention between stages , the size distribution of retained introns was similar in both stages , and comparable to that of constitutively spliced introns ( PIR<2% in all samples ) ( Figure 3—figure supplement 1A , B ) . GO enrichment analysis of genes with amoeboid-specific intron retention revealed enrichment in spindle pole formation and other mitosis-related activities ( Figure 3C ) , supporting an active role of IR in down-regulating these functions in this stage . Indeed , consistent with recent reports in vertebrates ( Braunschweig et al . , 2014 ) , genes with amoeboid-specific intron retention showed significantly lower steady-state mRNA levels compared to the multinucleate stage ( Figure 3—figure supplement 1C; p<0 . 0001 , Wilcoxon signed rank test; the converse was true for genes with multinucleate-specific intron retention ) . Therefore , intron retention is a conserved mechanism for reducing transcript levels in pathway-specific genes from unicellular holozoans to vertebrates . 10 . 7554/eLife . 08904 . 009Figure 3 . Regulated alternative splicing modes in Creolimax . ( A ) Heatmap showing PIR inclusion levels of differentially retained introns . ( B ) Heatmap showing the PSI levels of differentially skipped exons . ( C ) GO enrichment activities of the genes showing differential IR . Bar length indicates the significance of the enrichment , orange indicates those with higher inclusion levels in the amoeboid stage and blue those with higher inclusion levels in the multinucleate stage . ( D–E ) RT-PCR validations of selected IR and ES events . The values correspond to relative intensity of the alternative isoform ( retained intron or skipped exon ) bands in the RT-PCR and the proportions observed for the inclusion values in the RNA-seq . ( F ) GOs enrichment of genes with differential ES , in blue those with higher exon inclusion levels in the multinucleate stage . ES , exon skipping; GO , Gene Ontology; IR , intron retention; PIR , percent intron retention; PSI , percent spliced in; RT-PCR , reverse transcription-polymerase chain reaction; RNA-seq , RNA sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 00910 . 7554/eLife . 08904 . 010Figure 3—figure supplement 1 . Intron size and transcriptional levels of differentially retained introns . ( A ) Relationship between intron length and retention level . The barplot shows the percentage of retained introns ( PSI >20 ) among the total number introns of a given size . ( B ) Cumulative frequency plot showing that the intron size distribution is the same for constitutive introns ( PSI<2 ) , highly retained introns ( PSI >50 both stages ) and differentially retained introns . ( C ) Distribution of expression values ( FPKM ) in the amoeboid and the multinucleate stage for genes differentially retained introns . FPKM , fragments per kilobase of exon per million fragments mapped; PSI , percent spliced in . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 01010 . 7554/eLife . 08904 . 011Figure 3—figure supplement 2 . Validations of intron retention and ES events . The RT-PCR gels show the different splice variants for each gene in the amoeboid stage ( left ) and multinucleate stage ( right ) . RT-PCR values indicate the levels of inclusion of the alternative isoform ( retained intron or skipped exon ) compared to the canonically spliced form obtained with ImageJ; RNA-seq values are based on read coverage for each event . The scatter plot shows the differences between the RT-PCR measures and the RNA-seq-based values for all the validated examples , overall showing a high correlation . ES , exon skipping; RT-PCR , reverse transcription-polymerase chain reaction; RNA-seq , RNA sequencingDOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 011 In the case of ES , we found 63 exons with differential inclusion levels across stages ( Figure 3B ) . RT-PCR assays were used to validate the differential inclusion for all 12 tested cases ( Figure 3E and Figure 3—figure supplement 2 ) . Only a minority of these exons ( Torruella et al . , 2015 ) keep the reading frame upon skipping , and only two of these overlapped with functional Pfam domains , suggesting that , similar to intron retention , ES in Creolimax predominantly contributed to functionally down-regulating specific genes . GO enrichment analysis revealed that differentially spliced exons belong to genes involved in various biological processes , including channel activity and histone modifications ( Figure 3F ) . Another layer of transcriptional complexity is provided by long non-coding RNAs , which are increasingly recognized as important players in animal development and cell type-specific genome regulation ( Ulitsky and Bartel , 2013; Morris and Mattick , 2014; Sauvageau et al . , 2013 ) . To characterize the repertoire of long non-coding RNAs in Creolimax , we assembled a de novo transcriptome from the RNA-seq data . We filtered out transcripts shorter than 200 bp and mapped the rest to the genome . We then applied a pipeline to identify putative non-coding RNAs , including searches for coding potential , homology , untranslated region ( UTR ) mis-annotation , and low expression ( see Material and methods ) . Restricting our search to transcripts that did not overlap with protein-coding genes , known as long-intergenic RNAs ( lincRNAs ) , we identified 692 putative lincRNA loci in Creolimax . In comparison to protein-coding transcripts , lincRNAs in Creolimax were shorter in length , harbored fewer exons , had longer exons , and had a lower GC content ( Figure 4—figure supplement 1 ) . Moreover , overall transcription levels of lincRNAs were significantly lower than those of protein-coding genes ( p<0 . 01 , Wilcoxon rank sum test ) . Interestingly , all those characteristics have been reported for animal lincRNAs ( Hezroni et al . , 2015; Pauli et al . , 2012; Gaiti et al . , 2015 ) . To infer possible functions of Creolimax lincRNAs , we first looked at the functional annotations of the closest neighboring protein-coding genes , as animal lincRNAs are enriched near developmental genes and TFs ( Ulitsky and Bartel , 2013 ) . In Creolimax , the only significantly enriched GOs of lincRNA closest neighboring genes were metabolic activities ( p<0 . 01 ) . However , when we analyzed the vicinity of all the TFs , we found that 23 . 6% had at least one neighboring lincRNA , a significant enrichment compared to the rest of the genome ( 14 . 5% , p=0 . 0007 , Fisher’s exact test ) . Next , we searched for putative homologs of the lincRNAs in the transcriptomes of closely related species and other unicellular holozoans , but we did not retrieve any positive hits . This pattern of rapid evolutionary sequence turnover of lincRNAs has also been described for animals ( Gaiti et al . , 2015; Kapusta and Feschotte , 2014 ) , where homology detection based on sequence similarity is restricted to short evolutionary distances . In animals , expression of lincRNAs is generally restricted to specific tissues and organs ( Ulitsky and Bartel , 2013; Necsulea and Kaessmann , 2014 ) . In Creolimax we detected 51 lincRNA loci that were differentially expressed between the amoeba and the multinucleate stage ( Figure 4 ) . Overall , only 7% of the total detected lincRNAs are differentially expressed , compared to 10% of coding genes ( p=0 . 0059 , Fisher’s exact test ) . Thus , in stark contrast to animals , the lincRNAs in Creolimax appear to be less cell type-specific than coding genes . 10 . 7554/eLife . 08904 . 012Figure 4 . Transcriptional and post-transcriptional regulation of lincRNAs in Creolimax . ( A ) Heatmap showing transcriptional levels of significantly differentially expressed lincRNAs across biological replicates of amoeboid and multinucleate stages . ( B ) Example of genomic region where two lincRNA loci are found in tail-to-tail orientation surrounded by two protein-coding genes . ( C ) RT-PCR validations of the lincRNA loci . ( D ) Barplot depicting the average gene expression of those lincRNA in each stage . ( E ) Alternative splicing isoforms of lincRNAs showing various degrees of IR . IR , intron retention; lincRNA , long intergenic non-coding RNAs; RT-PCR , reverse transcription-polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 01210 . 7554/eLife . 08904 . 013Figure 4—figure supplement 1 . Genomic architecture of lincRNAs compared to protein-coding genes . ( A ) Kernel density plot showing transcript length distribution . Protein-coding genes are shown in blue , lincRNA in red . ( B ) Density plot showing exon number distribution . ( C ) Kernel density plot showing exon length distribution; multi-exonic lincRNAs are shown in green . ( D ) Kernel density plot showing GC content distribution . The dashed line indicates the total genome GC content . ( E ) Cumulative frequency plot of expression levels obtained from log10 ( FPKM ) . FPKM , fragments per kilobase of exon per million fragments mapped; lincRNA , long intergenic non-coding RNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 01310 . 7554/eLife . 08904 . 014Figure 4—figure supplement 2 . Gene orientation and transcriptional co-regulation of neighboring genes . Distribution of Pearson correlation values between a gene and its upstream neighbor subdivided in four categories . Head-to-head oriented neighbors tend to be more co-expressed than head-to-tail genes , independently of coding potential . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 014 To test whether the transcription of lincRNAs is linked to their upstream genes , we first subdivided lincRNAs into two categories: head-to-head , where the lincRNA and the adjacent protein-coding gene have opposite orientation and may share the same promoter , and head-to-tail , where the lincRNA promoter is downstream of the adjacent protein-coding gene ( Figure 4—figure supplement 2 ) . Head-to-head orientation was significantly over-represented in the genome ( 485/692 , 70% of the total vs the 50% expected by chance , p<3 . 6e-14 , χ2 test ) . Furthermore , pairs in this head-to-head orientation showed a higher proportion of positively correlated expression levels , although this was similar to pairs of protein-coding genes in the head-to-tail orientation . Moreover , 80% of differentially regulated lincRNAs are in head-to-tail orientation or display uncoupled transcriptional profiles in relation to their neighbor genes ( p<0 . 05 Pearson’s correlation coefficient ) . Although the possibility that bidirectional promoters are the main source of lincRNAs in Creolimax cannot be excluded , differential regulation of lincRNA is largely independent of their neighboring protein-coding genes . Finally , we analyzed the impact of alternative splicing in lincRNAs . Although the majority of lincRNAs have a single exon , 132 ( 19 . 1% ) loci have two or more exons . Among those multi-exon transcripts , we detected 183 intron-retention events ( Figure 4E ) . Of these , eight events differed between samples ( PIR>15 ) , suggesting differential regulation of lincRNAs also at the level of splicing . Therefore , both transcriptional regulation and alternative splicing may play a role in the active regulation of lincRNAs in Creolimax . Direct comparison of steady-state transcript abundance is useful to reveal cross-species molecular homologies in animals , as homologous tissues in different species show more similarity in the transcriptional profiles than do non-homologous tissues in a single species ( Barbosa-Morais et al . , 2012; Chan et al . , 2009; Brawand et al . , 2011 ) . Using a similar rationale , we sought to compare the different cell stages of unicellular holozoans . From normalized transcriptional levels of a set of 2177 one-to-one protein-coding orthologs between Salpingoeca , Capsaspora , and Creolimax , we calculated pairwise Spearman correlation distances , which were then used to perform hierarchical clustering and neighbor-joining tree reconstructions ( Figure 5 ) . Both approaches showed consistent trees with species-specific clustering for the different stages . The signal obtained from those 2177 genes was enough to cluster samples consistent with their whole transcriptome patterns ( except one sample from Salpingoeca ) . 10 . 7554/eLife . 08904 . 015Figure 5 . Holozoan cross-species comparison of transcriptional profiles . ( A ) Symmetrical heatmap of the pair-wise Spearman correlation coefficients for the gene expression profiles of each cell stage . For each sample , log2 ( cRPKM+1 ) expression levels were obtained for 2177 one-to-one orthologs in the three species analyzed ( see Materials and methods ) . Dashed-line squares highlight the direct comparisons for 1 ) Cfra multinucleate stage replicates against Cowc cystic stage replicates and 2 ) Cfra amoeboid stage replicates against Cowc aggregate and filopodial stage replicates . ( B ) Neighbor-joining tree of the species cell stages based on the aforementioned Spearman correlation distances matrix . Filled circles represent >95% bootstrap replicate nodal support . ( C ) The cell types plotted according to the values of the principal components 2 and 3 from a PCA of a dataset of 3030 1-to-1 orthologs between Capsaspora and Creolimax . ( D ) The significant GO enrichments for the top positive loading genes ( >0 . 03 ) of the principal component 2 and 3 . Cfra , Creolimax fragrantissima; Cowc , Capsaspora owczarzaki; GO , Gene Ontology; Sros , Salpingoeca rosetta; PCA , principal component analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 015 Direct cell type comparisons between species revealed that Creolimax multinucleate stage and Capsaspora cystic stage are the most dissimilar among all samples , whereas the amoeboid stage in Creolimax has a stronger correlation with the filopodial and aggregative stages of Capsaspora ( Figure 5A ) . In order to characterize gene sets responsible for those cross-species similarities , we used a principal component analysis ( PCA ) to distinguish between the species-specific factors and the underlying biological similarities . Species-specific variability was captured by PC1 ( 47 . 96% ) ; however , when we plotted PC2 and PC3 we obtained different groupings of cell types independent of their species of origin ( Figure 5C ) . PC2 grouped the multinucleate stages of Creolimax with the aggregative and the filopodial stages of Capsaspora . This grouping was explained by top-loading genes involved in DNA replication and spliceosomal activities ( Figure 5D ) , a pattern consistent with the down-regulation of mitotic activity and cell growth in the Capsaspora cystic stage ( Sebé-Pedrós et al . , 2013 ) . On the other hand , PC3 grouped the amoeboid stage of Creolimax and the filopodial and aggregative stages of Capsaspora . PC3 was loaded with genes involved in the integrin adhesome and amoeboid actin-based motility and signaling activities ( Figure 5D ) . Therefore , despite the vast phylogenetic divergence among holozoan species , some functional patterns can be recovered through direct comparison of transcriptional profiles . We used the same approach to compare RNA-seq data from a wide range of human cell types and tissues with those of Creolimax ( Figure 6A ) . Certain human cell types showed differential grades of similarity in the expression profiles with each of the Creolimax stages . The human cell types with a higher positive correlation with the multinucleate stage were those with high proliferation rates , including embryonic stem cells , induced pluripotent stem cells and transformed cell lines ( 293T , HeLA , K562 ) ( Figure 6B ) . Similar patterns were obtained using PCA on the normalized steady-state transcriptional levels from both species . PC1 explained most of the variability due to species-specific transcriptomic profiles , whereas PC2 distinguished between highly proliferative cell types and the rest in both species ( Figure 6C ) . Consistent with this observation , GO enrichment analysis of the top 125 genes that most contributed to PC2 showed highly significant enrichment for genes involved in genome replication and proliferation ( Figure 6D ) . These results suggest that a signal from the evolutionarily conserved machinery for cell proliferation in eukaryotes ( Harashima et al . , 2013; Cross et al . , 2011 ) can be detected from direct expression pattern comparisons between the multinucleate stage of Creolimax and highly proliferative human cell types . 10 . 7554/eLife . 08904 . 016Figure 6 . Comparison of human and Creolimax cell types and tissues . ( A ) Symmetrical heatmap of the pair-wise Spearman correlation coefficients for the gene expression profiles of each cell type or tissue . For each sample log2 ( FPKM+1 ) expression levels were obtained for 2272 one-to-one orthologs between Creolimax and human ( see Materials and methods ) . ( B ) The human cell types sorted by the difference of Spearman correlation between the amoeboid and the multinucleate cell stages . Highlighted in gray are those that displayed the major differences ( >0 . 05 ) . ( C ) The cell types plotted according to values of the principal components 1 and 2 from a PCA of the same transcriptional dataset of 2272 orthologs . The dots in gray are the human cell lines highlighted in the previous section . ( D ) The significant GO enrichments for the top positive loading genes of the principal component 2 ( >0 . 04 ) . Sampled human cell types described in Figure 6—source data 1 . GO , Gene Ontology; PCA , principal component analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 01610 . 7554/eLife . 08904 . 017Figure 6—source data 1 . Human RNA-seq datasets used in this analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 017 To obtain more specific insights into the evolution of co-regulated gene programs across holozoans , we investigated several gene modules with key roles in animal multicellularity . Despite gene co-regulation providing indirect evidence for shared functionality in a given species ( Gerstein et al . , 2014; Stuart , 2003 ) , comparative analysis of transcriptional co-regulated gene modules in different species may offer additional insights into the functional evolution of the animal multicellular toolkit . First , we focused on the integrin adhesome , which is crucial for ECM adhesion in animals . The core components of the integrin adhesome have been identified in Capsaspora ( Sebe-Pedros et al . , 2010 ) . Interestingly , these components are significantly up-regulated in the aggregative stage ( Sebé-Pedrós et al . , 2013 ) . In contrast , most integrin adhesome components have been lost in choanoflagellates ( Sebe-Pedros et al . , 2010 ) . In the case of Creolimax , we found a nearly complete repertoire of the integrin adhesome components ( Figure 7D ) , and most of them are significantly up-regulated in the amoeboid stage ( Figure 7A ) . To test whether the integrin adhesome components constitute a co-regulated gene module in the three unicellular holozoans ( ichthyosporeans , filastereans , and choanoflagellates ) , we calculated the transcriptional Pearson correlation coefficients between all of the genes encoding the integrin adhesome for each species ( Figure 7A–C ) . In the case of Creolimax , we observed a remarkable co-regulation among all of the components . The only exception is the Src tyrosine kinase ( Figure 7A ) , suggesting that tyrosine kinase signaling is not connected to the integrin signaling pathway in Creolimax , consistent with the absence of a focal adhesion kinase in the ichthyosporean lineage ( Suga et al . , 2014 ) . In the filasterean Capsaspora , we identified a more complex pattern of co-regulation , grouped in two submodules that are associated with specific paralogs of both alpha and beta integrins ( Figure 7B ) . This complex pattern of co-regulation could be affected by the number of Capsaspora’s samples , higher than those analysed in Creolimax . In contrast , we did not detect any co-regulation among the few conserved integrin adhesome components in the choanoflagellate Salpingoeca ( Figure 7C ) , suggesting that gene loss in the choanoflagellate lineage was accompanied by dismantling of this ancient co-regulatory module . 10 . 7554/eLife . 08904 . 018Figure 7 . Co-regulation of the integrin adhesome in holozoans . Heatmaps depicting expression levels of integrin adhesome orthologs ( red–green ) and their pair-wise Pearson correlation coefficients ( white–blue ) obtained from genome-wide FPKM transcriptional levels in the ichthyosporean Creolimax ( A ) , the filasterean Capsaspora ( B ) , and the choanoflagellate Salpingoeca ( C ) . ( D ) Diagram of integrin adhesome components , those in green are found in Creolimax and those in white are absent . In gray , a tyrosine kinase receptor with extracellular EGF domains encoded Creolimax genome that could be interacting with an ECM component . ( E ) Repertoire of animal ECM domains in the three unicellular holozoan genomes; green = presence , white = absence . ( F ) Pfam domain architectures of fibronectin-domain containing genes in Creolimax . ECM , extracellular matrix; FPKM , fragments per kilobase of exon per million fragments mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 01810 . 7554/eLife . 08904 . 019Figure 7—figure supplement 1 . Co-regulation of the filopodial molecular toolkit genes in holozoans . ( A ) Diagram of the filopodial machinery components as described in Sebé-Pedrós et al 2013 ( Sebé-Pedrós et al . , 2013 ) ; those in green are the genes found in unicellular holozoans . Heatmaps depicting expression levels of filopodial component orthologs ( red–green ) and their pair-wise Pearson correlation coefficients ( white–blue ) obtained from genome-wide FPKM transcriptional levels in the ichthyosporean Creolimax ( B ) , the filasterean Capsaspora ( C ) and the choanoflagellate Salpingoeca ( D ) . Highlighted in red are those genes that appear outside the module in at least two lineages . FPKM , fragments per kilobase of exon per million fragments mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 01910 . 7554/eLife . 08904 . 020Figure 7—figure supplement 2 . Co-regulation of the pre- and post- synaptic genes in holozoans . ( A ) Diagram of the pre-synaptic ( paler green ) and post-synaptic ( darker green ) found in unicellular holozoans . Heatmaps depicting expression levels of pre- and post-synaptic orthologs ( red–green ) and their pair-wise Pearson correlation coefficients ( white–blue ) obtained from genome-wide FPKM transcriptional levels from the ichthyosporean Creolimax ( B ) , the filasterean Capsaspora , ( C ) and the choanoflagellate Salpingoeca ( D ) . Flotillin genes , despite not being directly related to the post-synpatic scaffold , have been shown to interact with Homer in choanoflagellates and animals ( Burkhardt et al . , 2014 ) . FPKM , fragments per kilobase of exon per million fragments mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 020 In animals , integrins mediate cell-adhesion through binding to ECM proteins , such as laminins and fibronectins ( Cromar et al . , 2014 ) . In Capsaspora , laminin-containing genes with predicted secretion signals are up-regulated in the aggregative stage ( Sebé-Pedrós et al . , 2013 ) . However , despite showing a co-regulated integrin adhesome , we did not find any laminin-containing genes in Creolimax ( Figure 7E ) . The only ECM-related domains that we found in the genome were fibronectins , but those are fused to metabolic domains unrelated to the ECM and none had a secretion signal peptide ( Figure 7F ) . Therefore , it seems unlikely that integrins are involved in adhesion to an endogenous animal-like ECM in Creolimax . Next , we investigated the filopodial machinery ( Sebé-Pedrós et al . , 2013 ) . We observed strong positive co-regulation of a core set of components in all unicellular holozoans ( Figure 7—figure supplement 1 ) , consistent with the use of an ancestral cellular machinery for filopodial formation . In stark contrast to the situation for cell replication machinery , we did not observe a single co-regulated module in any unicellular species for genes involved in animal neuronal pre-synaptic and post-synaptic processes , including most pairs of genes that are known to directly interact in animals ( with the exception of syntaxin , synaptobrevin , and synaptogryin , which are involved in secretory vesicle formation ( Burkhardt et al . , 2011 ) ; Figure 7—figure supplement 2 ) . In fact , a lack of interaction has been shown for some of the core proteins involved in the post-synaptic scaffold in Salpingoeca ( Burkhardt et al . , 2014 ) . Thus , our results suggest that some molecular complexes directly involved in cell morphology and behavior already formed co-regulated gene modules in unicellular holozoans , whereas other complexes involved in unique animal cell types were assembled later , despite having conserved orthologs . Ichthyosporeans are an interesting case of convergent evolution of fungal-like traits; in fact , they were once thought to be fungi based on their lifestyle and morphology ( Glockling et al . , 2013; Mendoza et al . , 2002 ) . For instance , both groups have a cell wall , similar parasitic lifestyles , and a specialized osmotrophic feeding mode , unlike any of the other holozoan lineages ( Figure 1 ) . Specialized osmotrophs are characterized by their highly adapted secretomes that are key to the external digestion of complex compounds . Therefore , we analyzed the secretome of Creolimax to investigate the convergent evolution of this specialized osmotrophic feeding mode in the ichthyosporeans . In addition , secretome analysis should provide further insights into the production and modification of ECM and signaling ligands . To characterize the secretome , we performed a combination of in vivo and in silico approaches . First , using high-throughput proteomics of an in vivo secretome sample in culture conditions , we identified 91 proteins . Next , we applied an in silico approach to predict 453 proteins that are likely to be secreted ( see Material and methods ) . Interestingly , only 48 proteins were common to both datasets , indicating that proteins without a canonical signal peptide or with additional transmembrane domains could also be found in the in vivo secretome ( Figure 8A ) . Among the 43 non-canonically secreted proteins , we detected some with transmembrane domains that could be the product of shedding , as proposed for other species ( Meijer et al . , 2014 ) . For instance , we found peptides corresponding to the extracellular region of both alpha and beta integrins . Although shedding of integrins has been observed during the inflammatory response in animals ( Gomez et al . , 2012 ) , the functional implications of integrin shedding in Creolimax remain elusive . 10 . 7554/eLife . 08904 . 021Figure 8 . Functional enrichments of Creolimax secretome . ( A ) Venn diagram showing the number of genes identified in the Creolimax secretome through an in silico approach ( see Material and methods ) and an in vivo proteomics approach . A circle diagram describes the features of genes only identified in the in vivo approach , lacking a signalP or having TM domains . ( B ) GO categories and ( C ) PFAM domains enriched in the secretome; in dark blue are those enriched in the in vivo dataset; in pale blue are those enriched in the in silico dataset; in gray are the total amount of PFAM-domain containing genes in the genome . GO , Gene Ontology . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 02110 . 7554/eLife . 08904 . 022Figure 8—source data 1 . In vivo proteomics of Creolimax secretome . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 022 Both secretome datasets were highly enriched in peptidase and proteolytic activities ( Figure 8B ) , whereas only the in silico dataset showed further enrichments in other catabolic functions such as lipase and hydrolase activities . Analysis of Pfam-domain enrichment consistently revealed proteolytic and peptidase domains in both datasets ( Figure 8C ) . In contrast , an in-depth search for candidate ECM-related proteins and diffusible ligands based on Pfam domains ( using known animal domains as EGF , DSL , or laminins ) retrieved no positive hits . Thus , the proteolytic activity of the Creolimax secretome does not seem to be related to modifying the endogenous ECM , but possibly to the adaptation to a specialized osmotrophic feeding mode . Osmotrophic lifestyles are characterized by the external digestion of complex polymers ( Talbot , 2013 ) . Therefore , the enrichment in proteolytic activities observed in the secretome strongly suggests that proteins and peptides are the main food source of Creolimax , at least under culture conditions . Such external digestion requires a coupled mechanism for nutrient uptake ( Talbot , 2013 ) . Consistent with this requirement , we found 38 genes with four distinct Pfam domains ( PF00324 , PF01490 , PF03169 and PF00854 ) that are predicted to be involved in the amino acid and oligopeptide transporter activity . The most-abundant gene family in Creolimax secretome was the trypsins ( 15 proteins in the in vivo proteomic dataset out of the 31 genes found in the genome ) . Trypsins are usually found in multiple copies in animal genomes , and they have important roles in the digestive system as serine proteases and also as ECM remodellers . To further complement these observations , we profiled a wide range of eukaryotes and found independent expansions of trypsins in other osmotrophic lineages ( Figure 9A ) . Among these , two fungal species ( Coemansia reversa and Conidiolobus coronatus ) and an oomycete ( Saprolegnia parasitica ) are also animal-dwelling parasites , thus sharing a similar lifestyle to the known ichthyosporeans . Phylogenetic analyses of Creolimax trypsins ( Figure 9B ) revealed that most are the product of rapid lineage-specific gene duplications , a common source of molecular adaptation ( Kondrashov , 2012 ) . Moreover , we found very distinct patterns of transcript and protein abundance ( measured by the number of unique peptides identified ) across the trypsin paralogs ( Figure 9B ) , suggesting a rapid process of functional diversification . Thus , trypsins seem to be a recurrently used effector gene family in osmotrophic eukaryotes , and have evolved rapidly in Creolimax . 10 . 7554/eLife . 08904 . 023Figure 9 . Trypsin evolution . ( A ) Barplot showing the total number of Trypsin proteins ( PF00089 ) found in the genomes of diverse eukaryotes . Branches are color coded according to the taxonomy shown in the legend . ( B ) Maximum-likelihood phylogenetic tree based on the amino acid sequence of the Trypsin domain from Creolimax fragrantissima and Sphaeroforma arctica . Expression levels obtained from genome-wide FPKM calculation are shown . Number of unique peptides obtained from the in vivo secretome proteomic dataset is also shown . In red are those genes that do not present a signal peptide according to SignalP . FPKM , fragments per kilobase of exon per million fragments mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 023 LGT is a rare but recurrent mechanism for the acquisition of osmotrophy-related genes in both fungi and oomycetes ( Talbot , 2013 ) . We therefore investigated LGT in Creolimax by building automatic taxon-rich phylogenies . Manual inspection of these trees yielded a list of 163 genes that confidently branched within bacterial clades , supporting bacterial origins . Of these , 35 ( 21 . 5% ) were found only in Creolimax among all eukaryotes sampled , 50 ( 30 . 7% ) were shared with the close relative Sphaeroforma arctica , and 71 ( 43 . 6% ) were shared with other ichthyosporeans ( Figure 10—figure supplement 2 ) . Although most LGT genes ( 102 out of 163 ) were intron-less , the vast majority ( 143 , 87 . 7% ) showed transcriptional support from our polyA-selected RNA-seq , minimizing the possibility of bacterial contaminations . LGT from bacteria accounts for 1 . 8% of the total proteome , which is slightly higher than in other eukaryotic genomes ( Alsmark et al . , 2013 ) . Importantly , we found 6 genes acquired by LGT in the in vivo secretome and 18 more in the in silico secretome ( Figure 10—figure supplement 2 ) . Therefore , similar to other eukaryotic lineages ( Richards et al . , 2011; Richards et al . , 2011 ) , ichthyosporeans have enriched their osmotrophy-related gene complement through prokaryotic LGT . Among the six genes of prokaryotic origin that we found in the in vivo secretome , three belonged to the spore coat homology ( CotH ) family ( PF08757 ) . CotH proteins were first described as being fundamental for spore coat formation in the bacteria Bacillus subtilis ( Naclerio et al . , 1996 ) , but they have recently been characterized as critical factors for host invasion in the fungus Rhizopus oryzae ( Gebremariam et al . , 2014 ) . Our phylogenetic reconstruction of CotH family evolution revealed that the presence of CotH homologs in R . oryzae and other fungi ( only belonging to the class mucorales ) originated from an LGT event that was independent of those found in ichthyosporeans ( Figure 10 and Figure 10—figure supplement 1 ) . Interestingly , both lineages have expanded their CotH family members after LGT acquisition . Moreover , CotH genes were also found in other eukaryotic and archaeal lineages , suggesting a complex history of interdomain LGT similar to that recently described for other gene families ( Funkhouser-Jones et al . , 2014; Chou et al . , 2014 ) . Active gene duplication and domain shuffling characterized the evolutionary history of CotH in the ichthyosporea , where we could observe variable transcriptional levels and peptides among distinct paralogs , as well as the acquisition of an N-terminal LTD domain ( PF00932 ) ( Figure 10B ) . Active transcription and secretion of Creolimax CotH genes in axenic culture conditions underscore their putative role in host invasion and highlight the importance of LGT in effector gene acquisition across different osmotrophic lineages . 10 . 7554/eLife . 08904 . 024Figure 10 . CotH evolution . ( A ) Maximum-likelihood phylogenetic tree of the CotH domain ( PF08757 ) . Nodal support is shown in key branches ( 100 maximum likelihood replicates bootstrap values and Bayesian posterior probabilities ) . Color code indicates taxon distribution in each clade as depicted in the legend; for a detailed tree , see Figure 10—figure supplement 1 . ( B ) Detail of the phylogenetic tree depicting ichthyosporean CotH sequences , covering Creolimax , Sphaeroforma arctica , and Amoebidium parasiticum . Expression levels obtained from genome-wide FPKM calculation and the number of unique peptides obtained from the in vivo secretome proteomic dataset are shown . Domain configurations obtained from a PfamScan analysis . Gene identifiers in red are those that do not present a signal peptide according to SignalP . FPKM , fragments per kilobase of exon per million fragments mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 02410 . 7554/eLife . 08904 . 025Figure 10—figure supplement 1 . CotH extended phylogeny . Maximum-likelihood phylogenetic tree of the CotH domain ( PF08757 ) . Nodal support is shown in key branches ( 100 maximum likelihood replicates bootstrap values and Bayesian posterior probabilities ) . Color code indicates taxon distribution in each clade as depicted in the legend . Domain configurations obtained from a PfamScan analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 02510 . 7554/eLife . 08904 . 026Figure 10—figure supplement 2 . Features of prokaryotic LGT . ( A ) Phylostratigraphy depicting origins of LGT events found in Creolimax . Each node represents the total number of horizontally acquired genes in Creolimax found in any of the remaining ichthyosporean transcriptomes/genomes . Phylogenetic relationships among ichthyosporeans and out-groups obtained from ( Torruella et al . , 2015 ) . ( B ) Boxplot showing intron number distribution according to LGT phylostratigraphic age . ( C ) Venn diagram of horizontally acquired genes found in the in silico and the in vivo secretome datasets . LGT , lateral gene transfer . DOI: http://dx . doi . org/10 . 7554/eLife . 08904 . 026 In this study , we showed how different layers of genome regulation shape the coenocytic development and lifecycle of the ichthyosporean Creolimax , a unicellular relative of animals . These regulatory layers include complex mechanisms that also play a role in animal development and cellular differentiation . In Creolimax , differential genome regulation is not only limited to the transcriptional control of protein coding genes , but also includes cell-stage specific alternative splicing and lincRNA expression . Despite some global similarities to animal regulation , we observed significant differences in how these regulatory layers are deployed in Creolimax and animals . For example , the population of lincRNAs that we described in Creolimax does not show a major enrichment near developmental genes nor a highly specific cell type-dependent expression , both hallmarks of metazoan lincRNAs ( Ulitsky and Bartel , 2013; Gaiti et al . , 2015 ) . The differential processing of the intron-rich genes during alternative splicing in Creolimax is dominated by IR and not ES . Although this is similar to most other eukaryotic groups including sponges ( Fernandez-Valverde et al . , 2015 ) , it contrasts with alternative splicing in more complex animals , which predominantly involves ES ( Sebé-Pedrós et al . , 2013; Mcguire et al . , 2008 ) . Therefore , while alternative splicing in Creolimax is likely to contribute to the regulation of gene expression , it does not provide a greater expansion of the proteome by generating multiple protein isoforms from individual genes as in animals ( Irimia and Roy , 2014 ) . Intriguingly , while this complex genome regulation in Creolimax often involves similar gene toolkits to those employed in animal development , it is the unicellular amoeboid stage and not the multinucleate stage that is defined by multicellularity-related activities . This is the case for most differentially expressed TFs , signaling pathways , and adhesion molecules , which are characteristically associated with animal development and multicellularity . In fact , our results indicate that the transcriptional profile of the multinucleate stage of Creolimax is more similar to those of highly proliferative cell types in humans , despite the obvious differences in cell morphology and cell division strategy ( cell divison vs . coenocytic nuclear division ) . Thus , we consider that the multinucleate stage of Creolimax can be regarded as a highly specialized proliferative cell type . This cell type can be considered to function in a manner analogous to stem cells , with the undifferentiated nuclei dividing in the multinucleate cell before differentiation into amoeboid cells closes the lifecycle . This would suggest that separation of functions as crucial as self-renewal and differentiation can occur in a unicellular context in a temporal manner , which pre-dates the exclusive ability of multicellular organisms to engulf both functionally distinct cell types within a single entity ( Arendt , 2008; Hemmrich et al . , 2012 ) . Drastic differences in cell structure , morphology , function , and molecular signatures found between stages in protistan relatives of animals indicate that cell stages can be considered cell types according to established definition ( 57 and see author’s response section for an in-depth discussion on this topic ) . In addition to assessing the impact of genome regulation in Creolimax , our multiple genome-wide approaches reveal other novel aspects of ichthyosporean biology . For example , we show that Creolimax has undergone secondary adaptation to a specialized osmotrophic feeding mode , shaping its secretome and genome through both gene duplication and acquisition of bacterial genes by LGT . These observations highlight the uniqueness of the different holozoan lineages , each representing derived specializations from an ancestral state . However , these specializations of unicellular holozoans are achieved through a largely common genetic toolkit shared with animals , including several signaling pathways , TFs and adhesion molecules ( Richter and King , 2013; Suga et al . , 2013 ) . Moreover , many of the components of this toolkit are assembled in co-regulated gene modules preserved since the common ancestor of all holozoans , suggesting that recurrent recruitment of full co-regulated gene programs underlies the evolution of lineage-specific cell types and developmental modes ( Newman , 2012 ) . Additional complementary insights on the evolution of developmental modes will be provided by studying the immediate out-group of the Holozoa and Holomycota ( fungi + Discicristoidea ) ( Torruella et al . , 2015 ) . The Holomycota show a wide variety of developmental modes , ranging from aggregative fruiting body formation to several modes of coenocytic development ( Stajich et al . , 2009; Brown et al . , 2009 ) . We have shown that the diversity of cellular behaviors and morphologies observed in holozoan lifecycles is likely to have evolved from lineage-specific specializations . A widely accepted hypothesis states that the origin of animals involved a clonally dividing organism , similar to choanoflagellate colonies , that subsequently evolved specialized cellular differentiation ( King , 2004; Nielsen , 2008 ) . A competing hypothesis , however , suggests that pre-existing cell types and their associated molecular mechanisms were integrated in the spatiotemporal developmental dynamics of the last common ancestor of all animals ( Mikhailov et al . , 2009 ) . Our data suggests a third mixed model , in which the capacity to build differentiated cell types and transient multicellular entities was not a rare feature in pre-metazoan evolution . Nevertheless , the rise of animal multicellularity is not directly homologous to any ancestral developmental mode , and it may be seen as another derived specialization involving the integration of ancestral molecular modules and their associated cell behaviors into one single multicellular entity . Consequently , our results reveal the importance of obtaining complementary data from multiple lineages before significant insights can be gained into the organism that took the first steps on the road towards complex animal multicellularity . Creolimax fragrantissima cells ( available from the Canadian Centre for the Culture of Microorganisms under accession numbers CCCM 101 – 107 ) were grown axenically in liquid medium ( marine broth Difco 2216 ) at 12ºC . To obtain biological replicates for the RNA extraction , three independent cell lines were isolated from distinct colonies grown on a marine agar plate ( marine agar Difco 2216 ) . Those initial isolates were then grown in liquid medium ( marine broth Difco 2216 ) . After one pass , new 1/10 subcultures ( 10 ml ) were initiated and grown statically for 5 days in 25 ml flasks . When the cells became confluent on the fifth day , they were scratched and passed into 50 ml flasks with an additional 25 ml of fresh medium . These 50 ml flasks were then grown for 48 hr with gentle agitation ( 150 rpm ) , allowing the mature coenocytes to form clumps . Then , the 50 ml flasks were filtered using a 5 . 0 μm Isopore membrane filter ( Millipore ) and collected into a 50 ml Falcon tube . As only amoebas pass through the 5 μm filter step , filtered cells were then immediately pelleted by centrifugation at 1500 rcf for 3 min and harvested to get the RNA from the amoeboid stage . For the multinucleate-stage RNA , filtered cells were re-cultured in a new 50 ml flask , grown for 24 hr and harvested by centrifugation at 3000 rcf for 3 min . For all cell lines and stages , the RNA was extracted using Trizol reagent ( Life Technologies , Carlsbad , CA ) with a further step of DNAseI ( Roche ) to avoid gDNA contamination , and then purified using RNeasy columns ( Qiagen ) . We generated 1 . 7 million 454 single reads and 34 million Illumina 5kb mate-pair reads ( both after trimming , totaling 3 . 4G bp ) . Those were combined and preassembled with a Newbler 2 . 7 assembler ( Roche ) . The mitochondrial DNA sequence was removed before the assembly . Using the pairing information of the Illumina mate-pair reads , the 846 pre-scaffolds were broken at unreliable positions found by REAPR 1 ( Hunt et al . , 2013 ) and re-assembled by SSPACE 2 ( Boetzer et al . , 2011 ) . Some of the N-stretches within the scaffolds were filled by Gapfiller ( Boetzer and Pirovano , 2012 ) . This array of improvement tools assembled the pre-scaffolds into 196 sequence pieces . We then used the pairing information of the mate-pair reads for further assembly improvement , breaking and re-connecting the scaffolds . Finally , we obtained an assembly with 82 final scaffolds , of which 29 were short ( <1000 bp ) fragments . To predict the protein-coding genes from the whole genome sequence , we used Augustus 2 . 7 ( Stanke et al . , 2008 ) combined with RNA-seq data ( see details on this data in ‘RNA-seq and differential expression analysis’ section ) . We followed the protocol described here: http://bioinf . uni-greifswald . de/bioinf/wiki/pmwiki . php ? n=IncorporatingRNAseq . Tophat . Briefly , we pooled all the RNA-seq samples and mapped them to the genome using Tophat2 ( Kim et al . , 2013 ) , using the resulting introns to train Augustus ab initio predictions in an iterative process . The resulting predictions were manually screened in a genome browser and compared to the spliced-aligned reads resulting from Tophat2 . We further validated our predicted annotation , comparing the data to a set of genes that we had previously cloned by RT-PCR and rapid amplification of cDNA ends PCR , including highly expressed genes ( e . g . Histone 2B , Tubulin beta ) and lowly expressed genes ( e . g . Myc , Grainyhead , p53 , Src Tyrosine Kinase ) . Moreover , we used the mapped transcriptome to perform a genome-guided Trinity assembly ( Haas et al . , 2013 ) . Those transcripts were then used to annotate the UTRs of the protein-coding genes resulting from the Augustus annotation step . The elongation of the UTRs was done using PASA ( Haas , 2003 ) . The transcripts that did not overlap with Augustus annotations were then searched against the NCBI non-redundant protein database using tBlastX . Those that retrieved significant hits ( e-value <10e-3 ) and had clear open reading frames were then manually annotated as protein-coding transcripts . The resulting annotation retrieved 8695 genes , which are available here: http://dx . doi . org/10 . 6084/m9 . figshare . 1403592 . To functionally annotate the genes , we used Blast2GO ( Gotz et al . , 2008 ) , searching the protein sequences against the NCBI non-redundant database using a BLASTP threshold of 10e-6 and the default InterProScan settings . We also performed a PfamScan search with PFAM A database version 26 using the default gathering threshold parameters ( Punta et al . , 2012 ) . As a result , 6814 genes were functionally annotated . 100-base paired-end libraries were constructed using the TruSeq Stranded mRNA Sample Prep kit ( Illumina , San Diego , CA , USA ) . The libraries were sequenced in two lanes of an Illumina HiSeq2000 instrument at the CRG genomics unit ( Barcelona , Spain ) . We obtained 417 million reads that were then mapped to the genome using Tophat2 ( Kim et al . , 2013 ) , resulting in an average mapping of 82% . Raw gene counts and FPKM ( fragments per kilobase of exon per million fragments mapped ) values were obtained using Cufflinks2 ( Trapnell et al . , 2013 ) . Differential expression analysis was performed by comparing the three replicates from each stage using DEseq ( Anders and Huber , 2010 ) ( threshold 5e-5 ) , EdgeR ( Robinson et al . , 2010 ) ( threshold 5e-5 ) , Cuffdiff2 ( Trapnell et al . , 2013 ) ( threshold 5e-5 ) , and NOIseq ( Tarazona et al . , 2011 ) ( threshold 0 . 8 ) . Only the genes that were identified as differentially expressed with at least three methods were taken , resulting in 956 genes . Data can be downloaded from GEO GSE68616 . GO enrichments were obtained using the Topology-Weighted method in Ontologizer ( Bauer et al . , 2008 ) taking a P value lower than 0 . 01 as a threshold ( see full list in Figure 2—source data 1 ) . The resulting Gene Ontology ( GO ) enrichments were then visualized as a network in Cytoscape using the Enrichment map plug-in ( Merico et al . , 2010 ) . Enrichment map plug-in connects GO terms according to gene annotation; therefore connected GOs belong to the same set of genes with multiple associated GOs . We used network connectivity between enriched GO terms as a criterion to collapse GO redundancy , shading general GO groups in more inclusive categories as seen in Figure 2 . Inclusive categories complementary relied on the functional similarity between GOs based on GO definitions ( e . g . distinct unconnected aminoacid metabolic pathways are collapsed in ‘Aminoacid metabolism’ inclusive category ) . Additionally , PFAM domain enrichments were calculated using a Fisher’s exact test implemented in R , taking a threshold of 0 . 01 . PFAM enrichments were also visualized as a network using the Enrichment map plug-in , where connected nodes reflect domain presence in the same genes . Identification and quantification of ES ( including events with single or multiple cassette exons and microexons of 3–15 nucleotides [nt] ) and IR were performed as previously described ( Braunschweig et al . , 2014; Irimia et al . , 2014 ) . For ES , we used two complementary approaches . First , we implemented a ‘splice site-based module’ , which utilizes the joining of all hypothetically possible exon–exon junction ( EEJ ) forward combinations from annotated and de novo splice sites ( as described in [Han et al . , 2013] ) . To identify splice sites de novo , for each annotated splice site donor/acceptor , we scanned two downstream/upstream introns for potential splice-site acceptors/donors that would constitute a novel EEJ . Next , we mapped our RNA-seq data to this library of all potential EEJs , and considered ‘novel splice sites’ those supported by at least five reads mapped to multiple positions of the EEJ . Then , we implemented our recently described ‘microexon module’ ( Irimia et al . , 2014 ) , which also includes de novo searching of pairs of donor and acceptor splice sites in intronic sequences to detect novel , very short ( 3–15 nt ) microexons . For IR , we used our recently described pipeline ( Braunschweig et al . , 2014 ) , which employs a comprehensive set of reference sequences comprising exon–intron junctions ( EIJs ) , intron sequences ( if introns were longer than 200 nt , only a mid-intron segment of 200 nt was used ) , and EEJs formed by intron removal . Introns were classified as retained when there was a balanced accumulation of reads mapping to 5´ and 3´ EIJs and the intron body sequence , relative to the EEJ sequence . The level of retention was calculated based on PIR , which is the percentage of transcripts from a given gene in which the intron sequence is present . In all modules , quantification of alternative sequence inclusion in the transcripts is derived only from junction reads ( either EEJs or EIJs ) . To increase the fraction of mapping junction reads within each RNA-Seq sample , each read was first split into 50 nt read groups using a sliding window of 25 nt . Therefore , each 100 nt ( replicates a and b ) and 125 nt ( replicate c ) read produces 3 and 4 overlapping reads , respectively . In addition , both read mates from the paired-end sequencing were pooled . These 50 nt split reads were then mapped to the genome using Bowtie ( Langmead et al . , 2009 ) with –m 1 –v 2 parameters ( unique mapping with no more than two mismatches ) . Reads that mapped to the genome were discarded for ES quantifications . For quantification , only one random count per read group ( i . e . all sub-reads coming from the same original read ) was considered to avoid multiple counting of the same original sequenced molecule . In addition , for all modules and alternative splicing types , final read counts were corrected for the number mappable positions in each EEJ or EIJ following the formula: Corrected_EEJcount=EEJcount*MaximummappabilityEEJmappability where EEJcount is the number of read groups mapped to the EEJ , Maximummappability the maximum number of mapping positions that any EEJ can have for reads of length 50 nt ( i . e . 35 positions ) , and EEJmappability the number of positions that can be mapped uniquely to the EEJ using specific bowtie parameters ( –m 1 –v 2 ) , and thus EEJmappability ≤ Maximummappability ( see [Barbosa-Morais et al . , 2012; Han et al . , 2013] for details ) . The different modules to detect and quantify AS have been integrated into vast-tools ( https://github . com/vastgroup/vast-tools; species key “Cfr” ) . Associated files can be downloaded at http://vastdb . crg . eu/libs/vastdb . cfr . 31 . 1 . 15 . tar . gz . We used a threshold of ≥20 PIR in at least one stage for positive intron retention events . As a threshold for ES events we used skipping rates below 90% ( measured using the metric Percent Spliced In , PSI ) in at least one stage . In both cases , the minimum coverage allowed was 20 reads per splice junction . To evaluate differential alternative splicing , we used differences over 15 PSI or PIR between stages , allowing a standard deviation of <10 between replicates . From the genome-guided Trinity assembly ( see Genome sequencing , assembly and annotation ) we further analyzed the transcripts that did not retrieve any significant TBLASTX hit against the NCBI non-redundant database ( e-value > e-3 ) and were more than 200 bp long . To avoid lineage-specific protein-coding genes , we performed an additional TBLASTX search against six frame translations of the de novo assembled transcriptomes of several closely related species ( S . arctica , Ichthyophonus hoferi , Pirum gemmata , Amoebidium parasiticum , Abeoforma whisleri , Corallochytrium limacisporum , C . owczarzaki , S . rosetta , Monosiga brevicollis ) and the protein coding genes of Creolimax , filtering out the positive hits ( e-value < e-3 ) . From the remaining transcripts , we performed a RfamScan_2 search against RFAM 11 . 0 ( Burge et al . , 2013 ) to annotate all known ncRNAs ( e . g . U6 , 18S , 28S ) . Additionally , we used the coding potential calculator ( Kong et al . , 2007 ) to discard all those transcripts with putative coding potential ( coding potential score < −0 . 5 ) . With the final list of transcripts , we selected those that did not overlap with gene+UTRs annotations or were close to uncertain assembly regions ( multi-N stretches ) . For all those transcripts that were in a head-to-tail orientation regarding protein-coding genes , we manually inspected those that had an intergenic distance shorter than 1000 bp to the nearby gene to filter out misannotated UTRs . Additionally , we discarded transcripts overlapping repetitive regions of the genome . Finally , we collapsed all the remaining transcripts into single loci and quantified their expression level using Cuffcompare and Cufflinks ( Trapnell et al . , 2013 ) . From the resulting 2661 loci , to avoid noisy transcription , we filtered out all those that did not have at least 5 FPKM in at least one sample , and over 1 FPKM in any other sample . To detect putative homology , lincRNAs were searched using BLASTN against the same list of closely related species described above . Differential expression analysis of the lincRNAs was done using the same parameters as for the coding genes ( see RNA-seq and differential expression analysis ) . Consequently , we only accepted lincRNA loci as differentially expressed when they were identified by at least 3 out of 4 methods . We validated 6 out of 6 lincRNA using RT-PCR ( see below ) . Finally , we used the same pipeline to detetect alternative splicing events in coding genes , using a minimum coverage of 10 reads for each splice junction . To validate lincRNAs and alternative splicing events , RNA samples obtained as described in Cell culture , gDNA and RNA extraction were reverse transcribed to cDNA using SMARTer cDNA kit ( CloneTech ) . For both stages , the same amount of initial purified RNA was used ( 1 μg ) . Pairs of primers with melting temperatures close to 60ºC were designed to capture the lincRNA and the alternative splicing events , and the PCR was performed using Expand high-fidelity Taq polymerase ( Roche ) . Validations of IR and ES events were preformed using primer pairs spanning the neighboring constitutive exons . Quantification of alternative sequence inclusion levels from gel band intensity was done using ImageJ software ( Schneider et al . , 2012 ) . For the cross-species comparison , we first identified one-to-one orthologs in the proteomes of Creolimax , Capsaspora , and Salpingoeca using the Multiparanoid pipeline ( Alexeyenko et al . , 2006 ) . We trimmed all RNA-seq datasets into the same length ( 50 bp ) and only mapped the left reads when paired-end data was available . We then obtained cRPKM ( corrected by mappability ) values only for the subset of orthologs , transformed them to log2 ( cRPKM +1 ) and further normalized the expression data using quantile normalization . Hierarchical clustering ( ‘complete’ method ) of samples was obtained by comparing pairwise distances based on Spearman correlation coefficients in R . To obtain the neighbor-joining trees and bootstrap supports across the samples , we used the ‘ape’ package in R ( Paradis et al . , 2004 ) . For the Creolimax/human comparison we followed the same methodology , using the data detailed in Figure 6—source data 1 . The PCA of the expression data was performed as implemented in R ‘prcomp’ function . GO enrichments were obtained as described in RNA-seq and differential expression analysis section . To obtain the secretome sample for proteomics , we cultured Creolimax cells in liquid medium ( marine broth Difco 2216 ) at 12ºC for 5 days and allowed the cells to attach to the bottom of the flask . The medium was then replaced with artificial seawater to avoid excessive protein contamination , and the culture was incubated for another 24 hr . Then , we collected the medium by gently tilting the flask to avoid collecting attached cells . The medium was immediately centrifuged at 10 , 000 rcf for 2 min , and the supernatant was collected and filtered twice through a 0 . 2 μm filter . The filtered medium was concentrated by ultrafiltration using a molecular weight cut-off membrane ( Vivaspin 6 , 3000 MW; Sartorius , Gottingen , Germany ) and quantified using the BCA Protein Quantification Kit ( Thermo Fisher Scientific , San Jose , CA ) . The resulting protein extract was digested with 5 µg of trypsin ( cat # V5113 , Promega ) ( overnight , 37ºC ) . Finally , 2 μg of the sample was analyzed using an LTQ-Orbitrap XL mass spectrometer ( Thermo Fisher Scientific ) coupled to an EasyLC ( Thermo Fisher Scientific ( Proxeon ) , Odense , Denmark ) at the CRG proteomics unit ( Barcelona , Spain ) . All data were acquired with Xcalibur software v2 . 2 . Proteome Discoverer software suite ( v1 . 4 , Thermo Fisher Scientific ) and the Mascot search engine ( v2 . 5 , Matrix Science ( Perkins et al . , 1999 ) ) were used for peptide identification and quantification . The data were searched against a database containing Creolimax proteome , a list of common contaminants , and all the corresponding decoy entries . Resulting data files were filtered for false discovery rate ( FDR ) <0 . 05 . Finally , we discarded the contaminants and all the proteins that were identified by less than two unique peptides , resulting in a list of 91 proteins . To obtain the list of in silico secretome components , we performed an initial search step using SignalP 3 . 0 ( Dyrløv Bendtsen et al . , 2004 ) ( D-cutoff = 0 . 450 ) to identify all the proteins with a canonical signal peptide . We then performed a search step with TMHMM v . 2 . 0 ( Krogh et al . , 2001 ) to discard all those proteins with a transmembrane domain downstream of the first 60 amino acids . We also filtered out proteins tagged to the mitochondria using TargetP v . 1 . 1 ( Emanuelsson et al . , 2007 ) , and proteins with endoplasmic reticulum retention signal with c-terminal motifs KDEL or HDE[LF] and GPI-anchored proteins using PredGPI ( Pierleoni et al . , 2008 ) . The resulting list comprised 453 proteins . PFAM and GO enrichment analyses were performed as described in RNA-seq and differential expression analysis . In order to obtain orthology assignments for the various gene families analyzed in this study , we used a phylogeny-based pipeline . First , we used PFAM domain information to obtain all the members of a gene family across a database comprising 108 eukaryotic proteomes . Then , proteins were aligned using MAFFT software with L-INS-i parameters ( Katoh and Standley , 2013 ) . The resulting alignments were automatically trimmed using trimAl v1 . 2 ( Capella-Gutierrez et al . , 2009 ) ( -gt 0 . 7 ) and phylogenies were obtained using RAxML v8 . 0 ( Stamatakis , 2014 ) ( LG model , gamma distribution , 100 bootstrap supports ) and Phylobayes 3 ( Lartillot et al . , 2009 ) ( LG model , ran until two chains converged ) . Tree visualization and annotation was performed using iTol v2 ( Letunic and Bork , 2011 ) . This pipeline was slightly modified to detect LGT cases . Instead of using PFAM , we gathered close orthologs of all the proteins in the genome by performing a BLASTP search against the NCBI non-redundant database plus the 108 eukaryotic genomes . Only those proteins that retrieved lower e-values for bacterial/archaeal hits were selected for downstream analysis . Those proteins were then separately searched using BLASTP against all bacteria in nr , all Archaea in nr , and the 108 eukaryotic proteomes . We selected those sequences that had at least 25 hits with an e-value under e-10 , selecting a maximum of 50 proteins for bacteria , and 25 for archaea and eukaryotes . We got rid of redundancy using CD-HIT ( Li and Godzik , 2006 ) by filtering for 0 . 95 identity and then performed alignment , trimming , and phylogeny as described in the general pipeline above . The resulting trees were analyzed manually , taking LGT positives when Creolimax ( and other ichthyosporean ) sequences branched within bacterial clades with nodal bootstrap supports over 70% . Finally , we manually checked that the resulting LGT genes were found in distinct parts of the genome and not in genomic singletons . For those LGT candidates without introns and not found in any other ichthyosporean ( therefore , Creolimax specific ) we further checked if they were located in scaffolds with other genes containing introns . To further discard bacterial contaminations , the neighboring genes to LGT candidates were blasted against NCBI nr to verify their eukaryotic origins . When the immediate neighbor did not retrieve any hit , the following gene was searched .	All living animals are descended from a single-celled ancestor , and understanding how these ancestors became the first multicellular animals remains a major challenge in the field of evolutionary biology . An early breakthrough towards this goal was the realization that , even though they’re mostly single-celled organisms , the closest living relatives of animals share most of the basic gene toolkit that animals use to support their multicellular lifestyles . This shared toolkit also includes the genes that allow each specialized cell type in an animal ( for example , a skin cell or liver cell ) to express the subset of genes that it needs to fulfil its specific role . Discovering how the single-celled relatives of animals regulate these and other “multicellularity-related” genes during their life cycles is the next crucial step towards understanding how animals became multicellular . Creolimax fragrantissima is a single-celled relative of animals . One stage in this organism’s life cycle involves its nucleus ( which contains its genetic material ) replicating multiple times without the cell itself dividing . After this stage of development , new cells are formed , each receiving with a single nucleus , and released to live freely in the environment . Characterizing how C . fragrantissima regulates which genes are expressed during these two very different stages of development could shed new light on how multicellular animals evolved to regulate their genes in specific cell types . However , little is known about these processes in C . fragrantissima . Now , de Mendoza et al . have both sequenced C . fragrantissima’s genome and analysed which genes are expressed during the stages of its life cycle . This analysis reveals that this organism regulates its gene expression in several ways that are more commonly associated with gene regulation in multicellular animals . Furthermore , when compared to two other living relatives of animals that have brief multicellular stages in their life cycles , de Mendoza et al . found that the three organisms expressed similar genes during these similar life cycle stages . Furthermore , like fungi , C . fragrantissima digests its food externally and then absorbs the nutrients . Using a range of techniques , de Mendoza et al . identified the proteins involved in these processes and discovered that many had evolved independently from their counterparts in fungi . Furthermore , in some cases , the genes for these proteins had actually been acquired from bacteria via a process called lateral gene transfer . Together these findings suggest that it was likely that the last single-celled ancestor of multicellular animals already had the biological ability to create different cell types . Understanding if the cell types found in single-celled species resemble cell types from simple animals , such as sponges and comb jellies , at a molecular level is the next step towards determining what the ancestor of living animals looked like .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2015	Complex transcriptional regulation and independent evolution of fungal-like traits in a relative of animals
The medial subnucleus of the amygdala ( MeA ) plays a central role in processing sensory cues required for innate behaviors . However , whether there is a link between developmental programs and the emergence of inborn behaviors remains unknown . Our previous studies revealed that the telencephalic preoptic area ( POA ) embryonic niche is a novel source of MeA destined progenitors . Here , we show that the POA is comprised of distinct progenitor pools complementarily marked by the transcription factors Dbx1 and Foxp2 . As determined by molecular and electrophysiological criteria this embryonic parcellation predicts postnatal MeA inhibitory neuronal subtype identity . We further find that Dbx1-derived and Foxp2+ cells in the MeA are differentially activated in response to innate behavioral cues in a sex-specific manner . Thus , developmental transcription factor expression is predictive of MeA neuronal identity and sex-specific neuronal responses , providing a potential developmental logic for how innate behaviors could be processed by different MeA neuronal subtypes . One of the major functions of the limbic system is to integrate conspecific and non-conspecific environmental cues with social and survival salience to generate appropriate behavioral responses ( Sokolowski and Corbin , 2012; Stowers et al . , 2013 ) . The medial subnucleus of the amygdala ( MeA ) serves as a hub in this function , residing only two synapses away from sensory neurons in the vomeronasal organ ( Dulac and Wagner , 2006; Sokolowski and Corbin , 2012 ) . The MeA along with the bed nucleus of the stria terminalis ( BNST ) and multiple nuclei of the hypothalamus including the ventromedial hypothalamus , form a core limbic circuit largely dedicated to processing innate behaviors ( Dulac and Wagner , 2006; Gross and Canteras , 2012; Choi et al . , 2005 ) . Classical studies investigating patterns of neuronal activation in response to behavioral or olfactory cues ( Kollack and Newman , 1992; Erskine , 1993 ) , lesioning studies ( Vochteloo and Koolhaas , 1987; Takahashi and Gladstone , 1988; Kondo , 1992 ) and more recent optogenetic approaches ( Hong et al . , 2014 ) have revealed a central role for the MeA in the regulation of innate behaviors such as aggression , mating and predator avoidance . In addition to the processing of innate cues , the MeA is one of many known sexually dimorphic regions of the brain , with differences in numbers of neurons , glia , and synaptic organization between males and females ( Cooke and Woolley , 2005; Johnson et al . , 2008; McCarthy and Arnold , 2011 ) . The critical role that the MeA plays in regulating sex-specific behaviors is reflected in the high expression levels of steroid hormone pathway proteins such as aromatase , estrogen receptor and androgen receptor ( Wu et al . , 2009; Juntti et al . , 2010; Unger et al . , 2015 ) . MeA neuronal subpopulations expressing different combinations of these proteins have been shown to regulate aggression or mating behaviors in male and female mice ( Juntti et al . , 2010; Hong et al . , 2014; Unger et al . , 2015 ) . Nonetheless , understanding how developmental programs are linked to behavioral processing in the MeA remains unknown . As unlearned behaviors are largely inborn , we reasoned that there must be embryonic developmental programs that guide the formation of sub-circuitry dedicated for different innate behaviors . Previous studies of MeA development revealed that progenitors located at the telencephalic-diencephalic border are a major source of MeA neuronal populations ( Zhao et al . , 2008; Hirata et al . , 2009; Soma et al . , 2009; García-Moreno et al . , 2010 ) . Our previous work revealed that one of these progenitor populations is defined by the transient expression of the developmentally regulated transcription factor , Dbx1 , which in turn generates a subclass of MeA putative inhibitory projection neurons ( Hirata et al . , 2009 ) . However , the MeA is also comprised of diverse populations of local interneurons and both excitatory and inhibitory output neurons ( Bian , 2013; Keshavarzi et al . , 2014 ) . This suggests the contribution of other progenitor subpopulations , perhaps also originating from the POA to MeA neuronal diversity , populations which may in turn play different roles in innate behavioral processing . Here , we demonstrate that in addition to Dbx1 expression , a subset of MeA embryonic progenitors are complementarily marked by expression of Foxp2 , a forkhead transcription factor implicated in the development and function of neurons and required in the motor coordinating centers of the brain for the appropriate production of speech ( French and Fisher , 2014 ) . We find this embryonic parcellation interestingly persists into postnatal stages where Dbx1-derived and Foxp2+ MeA neurons are separate , non-overlapping inhibitory output neuronal subpopulations . Strikingly , both subpopulations are activated during specific innate behaviors in a sex-specific manner . Thus , our findings link developmental patterning to innate behavioral processing and further provide an embryonic developmental framework for how these behaviors may emerge . Our previous work along with the work of others revealed that the telencephalic-diencephalic border is a major source of neurons that will populate the MeA ( Zhao et al . , 2008; Hirata et al . , 2009; Soma et al . , 2009; García-Moreno et al . , 2010 ) . Our previous studies ( Hirata et al . , 2009 ) revealed that the preoptic area ( POA ) , which lies on the telencephalic side of this border ( Flames et al . , 2007 ) , is a source of Dbx1+ progenitors fated to generate a subpopulation of MeA inhibitory output neurons . Our previous studies further revealed that progenitors arising from ventral telencephalic Shh+ and Nkx2 . 1+ domains also contributed to diverse neuronal subpopulations of the MeA ( Carney et al . , 2010 ) . Thus , while a molecular map of MeA embryonic niche diversity is beginning to emerge , the diversity of mature neurons derived from this niche and whether there is a link between embryonic identity , mature identity and function remains unknown . Moreover , as these previously identified subpopulations only generate a subset of MeA neurons , we reasoned that there must be other transcription factor marked progenitor populations within the telencephalic-diencephalic niche . Here , in addition to Dbx1+ progenitors , we observed a progenitor population comprised of Foxp2+ cells , residing primarily in the putative subventricular zone ( SVZ ) of the POA ( Figure 1a–f , s ) . Interestingly , during embryogenesis , Dbx1-derived and Foxp2+ progenitor populations were non-overlapping ( Figure 1a–l ) . Both populations were also generally distinct from OTP+ progenitors ( Figure 1—figure supplement 1a–i ) , a population previously shown to define a subset of MeA-fated progenitors ( García-Moreno et al . , 2010 ) . We next investigated whether Foxp2+ progenitors overlapped with ventral telencephalic populations derived from Shh or Nkx6 . 2 lineages , which also encompass the POA ( Carney et al . , 2010; Fogarty et al . , 2007 ) . We found embryonic Foxp2+ cells were not derived from either lineage ( ~5% overlap ) ( Figure 1—figure supplement 2 ) further expanding our knowledge of the molecular diversity of the MeA niche . 10 . 7554/eLife . 21012 . 003Figure 1 . Embryonic and postnatal segregation of Dbx1-derived and Foxp2+ cells . Minimal co-localization of Dbx1-derived ( green ) and Foxp2+ ( red ) labeled cells in coronal sections at the level of the POA at E11 . 5 ( a–f ) and E13 . 5 ( g–l ) . As shown at E11 . 5 ( s ) Foxp2+ precursors in the VZ ( arrows ) are observed putatively migrating to the SVZ . The segregation of Dbx1-derived and Foxp2+ cells persists into adulthood ( m–r ) . Summary schematic of spatial segregation of Dbx1-derived and Foxp2+ cells in the embryonic POA and postnatal MeA ( t ) . Venn diagram depicting the overlap of the total number of Dbx1-derived and Foxp2+ cells in 2–3 sections/embryo or 5–8 sections/adult brain . The scale bars represent 200 μm ( a–c , g–i , s ) , 100 μm ( m–o ) and 25 μm ( d–f , j–l , p–r ) . Abbreviations: MeA , medial amygdala; MePD , medial amygdala-posterior dorsal; MePV , medial amygdala posterior ventral; POA , preoptic area; SVZ , subventricular zone; V , ventricle; VZ , ventricular zone . n = 3 embryonic brains; n = 5 postnatal brains . Data are mean ± s . e . m . n is the number of animals . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 00310 . 7554/eLife . 21012 . 004Figure 1—figure supplement 1 . Embryonic and postnatal distribution of OTP+ cells . Dual immunostaining of coronal sections at the telencephalic-diencephalic border at E11 . 5 shows very little overlap between Dbx1-derived ( green ) ( n = 3 ) and the Foxp2+ ( red ) ( n = 3 ) progenitor populations with the OTP+ population ( white ) ( a–f ) . Quantification of overlap of OTP with the Dbx1-derived and Foxp2+ populations ( YFP+ and OTP+/ total YFP+ population for Dbx1-derived cells; Foxp2+ OTP+/ total Foxp2+ population for Foxp2+ cells ) ( g ) . Venn diagrams depicting the overlap of the total number of Dbx1-derived ( h ) and Foxp2+ cells ( i ) with the OTP+ population ( 2–3 sections/embryo ) . Dual immunostaining of postnatal coronal sections also revealed very little overlap of Dbx1-derived population ( n = 3 ) or Foxp2+ populations ( n = 4 ) with OTP ( j–u ) . Quantification of overlap of OTP with the Dbx1-derived and Foxp2+ populations ( YFP+ and OTP+/ total YFP+ population for Dbx1-derived cells; Foxp2+ OTP+/ total Foxp2+ population for Foxp2+ cells ) ( v ) . Venn diagram depicting the overlap of the total number of Dbx1-derived ( w ) and Foxp2+ cells ( x ) with the marker OTP+ ( average of MeA A-P extent across 5–8 sections ) . p<0 . 05 ( * ) . The scale bar represents 100 μm ( a–f , j–l , p–r ) and 100 μm ( m–o , s–u ) . Data are mean ± s . e . m . n is the number of animals . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 00410 . 7554/eLife . 21012 . 005Figure 1—figure supplement 2 . Foxp2+ cells are not derived from the Shh- or Nkx6 . 2 lineages . Dual immunostaining of the amygdala primordium in horizontal sections at E11 . 5 reveals minor contribution of either the Nkx6 . 2-lineage ( n = 3 ) ( white ) ( a–d , i ) or the Shh-lineage ( n = 3 ) ( white ) ( e–h , i ) to the Foxp2+ population ( red ) . Data are mean ± s . e . m . n is the number of animals . p<0 . 05 ( * ) . The scale bar represents 200 μm ( a , e ) and 25 μm ( b–d , f–h ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 00510 . 7554/eLife . 21012 . 006Figure 1—figure supplement 3 . Localization of Dbx1-derived and Foxp2+ cells in the adult MeA . Double immunofluorescence staining of adult coronal sections at the level of the MeA shows patterns of localization of Dbx1-derived ( green ) and Foxp2+ ( red ) cells across the anterior to the posterior extent of the MeA ( a–i ) . Bregma levels correspond to the K . Franklin and G . Paxinos Mouse Brain in Stereotaxic Coordinates Atlas . The scale bar represents 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 006 Interestingly , this embryonic molecular parcellation persisted into adulthood as Dbx1-derived and Foxp2+ cells remained non-overlapping across the rostro-caudal extent of the postnatal MeA ( Figure 1m–r; Figure 1—figure supplement 3 ) . Similarly , postnatally , the Dbx1-derived and Foxp2+ neurons remained distinct from OTP+ cells ( Figure 1—figure supplement 1j–x ) . Taken together , these findings reveal that Dbx1-derived and Foxp2+ populations , although appearing to derive from the same embryonic niche , remain distinct subpopulations from embryonic development to adulthood ( Figure 1t ) , a novel finding that we hypothesize has implications for later subtype identity and function , explored in the next sets of experiments . Our previous work revealed that MeA Dbx1-derived neurons are a subclass of inhibitory neurons , likely projection as opposed to local interneurons ( Hirata et al . , 2009 ) . However , the identity of MeA Foxp2+ neurons remains unknown . Therefore , we next examined whether adult MeA Foxp2+ cells were neurons or glia . Our analysis revealed that 81% ± 2 . 6 of Foxp2+ cells expressed NeuN , a pan neuronal marker ( Mullen et al . , 1992 ) ( Figure 2a–d ) , with none co-expressing the oligodendrocyte marker , CC1 ( Koenning et al . , 2012 ) ( Figure 2e–h ) . We next wanted to determine if Foxp2+ neurons were excitatory or inhibitory . We found that only 3% ± 1 . 2 of Foxp2+ cells were derived from the Emx1-lineage , a broad marker of excitatory neurons ( Gorski et al . , 2002 ) ( Figure 2i–l ) . We further found that 22% ± 6 . 6 of Foxp2+ cells expressed the inhibitory marker Calbindin ( Figure 2m–p ) , with a smaller percentage of Foxp2+ cells ( 15% ± 2 . 5 ) expressing nNOS ( Figure 2q–t ) , or somatostatin ( 5% ± 0 . 6 ) ( Figure 2u–x ) , inhibitory markers that mark a subset of MeA output neurons ( Tanaka et al . , 1997 ) and interneurons ( Ascoli et al . , 2008 ) , respectively . Collectively , although we did not fully assess all putative inhibitory markers , these data reveal that Foxp2+ MeA are not excitatory and are likely inhibitory ( Figure 2y ) . 10 . 7554/eLife . 21012 . 007Figure 2 . Identity of Foxp2+ medial amygdala neurons . Dual immunofluorescence of Foxp2 with NeuN ( n = 3 mice ) ( a–d ) , CC1 ( n = 3 mice ) ( e–h ) , YFP ( in Emx1cre;RYFP mice ) ( n = 5 mice ) ( i–l ) , calbindin ( n = 3 mice ) ( m–p ) , nNOS ( n = 6 mice ) ( q–t ) , and somatostatin ( n = 4 mice ) ( u–x ) ( arrows show double-labeled cells ) . Quantification of co-localization with each marker ( y ) . Data are mean ± s . e . m . n is the number of animals . The scale bar represents 200 μm ( a , e , i , m , q , u ) and 25 μm ( b–d , f–h , j–l , n–p , r–t , v–x ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 00710 . 7554/eLife . 21012 . 008Figure 2—figure supplement 1 . Identity of OTP+ medial amygdala neurons . Dual immunofluorescence of Foxp2 with YFP ( in Emx1cre;RYFP mice ) ( n = 3 ) ( a–d ) , CAMKIIα ( n = 3 ) ( e–h ) , calbindin ( n = 3 ) ( i–l ) and somatostatin ( n = 3 mice ) ( m–p ) ( arrows show double-labeled cells ) . Quantification reveals that most Foxp2+ are not Emx1-derived , nor express somatostatin , express to some extent CAMKIIα , but are mainly Calbindin+ ( q ) Data are mean ± s . e . m . n is the number of animals . The scale bar represents 100 μm ( a , e , i , m ) and 25 μm ( b–d , f–h , j–l , n–p ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 008 As neither Dbx1-derived nor Foxp2+ cells expressed OTP+ , we therefore investigated the identity of this population . We found that 1% ± 0 . 4 of OTP+ cells were derived from the Emx1-lineage , and 13 . 5% ± 9 . 1 co-expressed CAMKIIα ( Jones et al . , 1994 ) , both excitatory markers . In contrast , we observed 56% ± 18 . 9 of OTP+ cells co-expressed calbindin , while none ( 0% ± 0 . 1 ) co-expressed somatostatin ( Figure 2—figure supplement 1 ) . Therefore , similar to Dbx1-derived and Foxp2+ neurons , the majority of OTP+ cells appear to be inhibitory . To determine if the Dbx1-derived and the Foxp2+ populations are functionally distinct subclasses , we next examined their electrophysiological properties . Previous studies ( Bian , 2013; Keshavarzi et al . , 2014 ) revealed a significant diversity in intrinsic electrophysiological properties of MeA local and projection neurons . Here , we found that the majority ( 19/28 ) of Dbx1-derived neurons were characterized by a regular , tonic spiking pattern with 3–4 spikes at rheobase ( Figure 3a ) . In contrast , the majority ( 15/23 ) of Foxp2+ neurons were distinguished by a phasic firing pattern and displayed a single or double spike upon repolarization after hyperpolarization , a profile characteristic of inhibitory neurons ( Llinás , 1988 ) ( Figure 3b ) . Dbx1-derived and Foxp2-derived neurons ( confirmed Foxp2+ by immunohistochemistry ) also displayed significant differences in resting membrane potential , input resistance , capacitance , and action potential frequency but not in rheobase ( Figure 3c–g ) . In addition , the presence of spines in Foxp2+ neurons ( Figure 3—figure supplement 1 ) suggested that similar to Dbx1-derived neurons , Foxp2+ neurons are projection neurons . This reveals that despite both populations being inhibitory , the Dbx1-derived and the Foxp2+ populations possess distinct firing patterns . 10 . 7554/eLife . 21012 . 009Figure 3 . Dbx1-derived and Foxp2+ MeA neurons possess distinct electrophysiological properties . Typical firing patterns of Dbx1-derived ( a ) and Foxp2+ ( b ) neurons with current injections at −60 pA , +20 pA and +60 pA . Significant differences across populations in resting membrane potential ( Rm ) ( c ) , voltage at rest ( Vrest ) ( d ) , capacitance ( e ) , and action potential ( AP ) firing patterns ( f ) but not rheobase ( g ) are observed ( two-tailed t-test , ( c ) p=0 . 0006 , t = 3 . 676 , df = 49; ( d ) p=0 . 036 , t = 2 . 159 , df = 49; ( e ) p=0 . 002 , t = 3 . 1987 , df = 49; ( f ) p=0 . 016 , t = 3 . 1988 , df = 49; ( g ) p=0 . 610 , t = 0 . 514 , df = 49; n = 28 Dbx1-derived cells and n = 23 Foxp2+ cells ) . Spontaneous excitatory post-synaptic currents ( sEPSCs ) are observed in both Dbx1-derived ( h ) and Foxp2+ ( i ) neurons , with significant differences in sEPSC frequency ( j ) , amplitude ( k ) and decay ( l ) ( two-tailed t-test , ( j ) p=<0 . 0001 , t = 3 . 041 , df = 34; ( k ) p=0 . 006 , t = 2 . 949 , df = 34; ( l ) p=0 . 005 , t = 3 . 041 , df = 34; n = 28 Dbx1-derived neurons , n = 23 Foxp2+ neurons ) . Data are mean ± s . e . m . n is the number of cells . p<0 . 05 ( * ) , p<0 . 01 ( ) , and p<0 . 001 ( * ) , n . s . ; not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 00910 . 7554/eLife . 21012 . 010Figure 3—figure supplement 1 . Foxp2+ neurons possess projection neuronal morphology . Biocytin filling depicting the morphology of a typical Foxp2+ neuron possessing dendritic spines ( a-b , arrow ) . Representative 3-D reconstruction of several Foxp2+ neurons ( c–e ) . The scale bars represent 100 μm ( a ) and 10 μm ( b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 010 We further analyzed spontaneous excitatory post-synaptic currents ( sEPSCs ) , a measure of excitatory inputs . Dbx1-derived neurons received significantly more frequent and greater amplitude of sEPSCs than Foxp2+ neurons ( Figure 3h–l ) . This suggests that Dbx1-derived MeA neurons receive a greater number and/or stronger excitatory inputs than Foxp2+ neurons . In summary , a combination of neuronal marker expression ( Figure 2 ) and electrophysiological ( Figure 3 ) analyses , combined with our previous analysis ( Hirata et al . , 2009 ) revealed that Dbx1-derived and Foxp2+ neurons are distinct subclasses of inhibitory , and are likely projection neurons . Based on the above analyses revealing that Dbx1-derived and Foxp2+ neurons are separate subclasses , we next wanted to determine whether these two populations express different combinations of steroid pathway proteins previously associated with MeA function such as estrogen receptor-alpha ( ERα ) , aromatase and androgen receptor ( AR ) ( Wu et al . , 2009; Juntti et al . , 2010; Unger et al . , 2015 ) . As the MeA is a sexually dimorphic nucleus ( Cooke and Woolley , 2005; McCarthy and Arnold , 2011; Johnson et al . , 2008 ) , we characterized the expression of these markers in both male and female mice ( Figure 4 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . We found that Dbx1-derived and Foxp2+ cells in males expressed ERα to the same extent ( 28 . 4% ± 4 . 8 in Dbx1-derived cells; 24 . 0% ± 7 . 2 in Foxp2+ cells ) . However , Dbx1-derived and Foxp2+ cells in females showed significant differences in ERα expression ( 45% ± 3 . 4 in Dbx1-derived cells; 24 . 8% ± 5 . 8 in Foxp2+ cells ) ( Figure 4a–g ) . The majority of Dbx1-derived cells expressed aromatase both in males ( 61 . 7% ± 7 . 6 ) and females ( 52 . 4% ± 5 . 2 ) , which was at a significantly higher percentage than in Foxp2+ cells in males ( 0 . 12% ± 0 ) and females ( 7 . 2% ± 6 . 0 ) ( Figure 4h–n ) . There were no subpopulation differences in AR expression as both Dbx1-derived and Foxp2+ neurons in both males ( 26 . 8% ± 4 . 1 in Dbx1-derived cells; 16 . 2% ± 3 . 2 in Foxp2+ cells ) and females ( 8 . 4% ± 3 . 6 in Dbx1-derived cells; 7 . 0% ± 1 . 8 in Foxp2+ cells ) co-expressed AR at the same levels ( Figure 4o–u ) . However , there were sex-specific differences observed as a greater percentage of Dbx1-derived cells in males ( 26 . 8% ± 4 . 1 ) expressed AR than Dbx1-derived cells in females ( 8 . 4% ± 3 . 6 ) . 10 . 7554/eLife . 21012 . 011Figure 4 . Expression of sex hormone pathway markers in Dbx1-derived and Foxp2+ cells . Dual immunofluorescence for YFP ( green ) or Foxp2 ( red ) with the sex steroid hormone pathway markers ( white ) : estrogen receptor α ( ERα ) ( a–g ) , aromatase ( h–n ) or androgen receptor ( AR ) ( o–u ) ( arrows show double-labeled cells ) . Subpopulations of both Dbx1-derived ( a–c , g ) and Foxp2+ cells express ERα ( d–g ) . Quantification reveals a greater percentage of Dbx1-derived cells expressing ERα compared to Foxp2+ cells in females ( two-way ANOVA with no interaction between subpopulation and sex , but with main effect for subpopulation p=0 . 0433 , F ( 1 , 10 ) =5 . 352; n = 4 and x = 45 . 38 for Dbx1-derived cells in female mice , n = 4 and x = 24 . 77 for Foxp2+ cells in female mice , n = 3 and x = 28 . 43 for Dbx1-derived cells in male mice , n = 3 and x = 24 for Foxp2+ cells in male mice ) . The majority of Dbx1-derived cells express aromatase in both males and females ( h–j , n ) . In contrast , only a small percentage of Foxp2+ cells in both males and females express aromatase ( two quantile regression analysis for non-normal distributions shows no interaction for sex and subpopulation but a main effect for subpopulation p=0 . 000 with a 95% confidence interval of 31 . 27 and 69 . 53 , n = 3 for Dbx1-derived cells in male mice , n = 3 and for Foxp2+ cells in male mice , n = 3 for mice for Dbx1-derived cells in female mice , n = 3 and for Foxp2+ female mice ) . A greater percentage of Dbx1-derived neurons in males express AR in comparison to Dbx1-derived cells in female mice ( o–q , u ) . No differences in percentages of Foxp2+ male or female subpopulations expressing AR ( r–u ) nor differences across subpopulations were observed ( two-way ANOVA with no interaction between subpopulation and sex , but with main effect for sex p=0 . 0166 , F ( 1 , 8 ) =9 . 118; n = 3 and x = 10 . 07 for Dbx1-derived cells in female mice , n = 3 and x = 7 . 133 for Foxp2+ cells in female mice , n = 3 and x = 24 . 3 for Dbx1-derived cells in male mice , n = 3 and x = 16 . 27 for Foxp2+ cells in male mice ) . Data are mean ± s . e . m . n is the number of animals and x is the mean . The scale bar represents 50 μm . p<0 . 05 ( * ) and p<0 . 01 ( ** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 01110 . 7554/eLife . 21012 . 012Figure 4—figure supplement 1 . Patterns of MeA expression of sex steroid hormone markers . Schematic of adult brain shows MeA ( red box ) ( a ) . Immunostaining for ERα ( b ) , aromatase ( c ) or AR ( d ) shows patterns of localization in the MeA . The scale bar represents 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 01210 . 7554/eLife . 21012 . 013Figure 4—figure supplement 2 . Percent contribution of Dbx1-derived and Foxp2+ cells to sex steroid hormone marker populations . The percent contribution of Dbx1-derived cells and Foxp2+ cells in male and female brains to the ERα+ , Aromatase+ and AR+ populations are as follows: ERα+ population: Dbx1-derived contribution: 18% ± 4 . 2 ( male n = 3 , x = 18 . 37 ) , 21% ± 3 . 59 ( female n = 4 , x = 20 . 75 ) . Foxp2+ contribution: 11% ± 2 . 72 ( male n = 3 , x = 10 . 89 ) ; 21% + 3 . 0 ( female n = 4 , x = 19 . 48 ) . No subpopulation or sex differences were observed ( two-way ANOVA with no interaction between subpopulation and sex and no main effects ( F ( 1 , 10 ) =0 . 8261 p=0 . 3848 ) ( a ) . Aromatase+ population: Dbx1-derived contribution: 37% ± 9 . 89 ( male n = 3 , x = 37 ) , 34% + 6 . 05 ( female n = 3 , x = 33 . 6 ) . Foxp2+ contribution: 0% ± 0 . 3 ( male n = 0 . 3 ) , 10% + 5 . 49 ( female n = 3 , x = 9 . 667 ) ( b ) . This difference in expression was significant across subpopulations ( two-way ANOVA with no interaction between subpopulation and sex , with a main effect for subpopulation p=0 . 0015 , F ( 1 , 8 ) =22 . 33 ) . AR+ population: Dbx1-derived contribution: 36 . 7% + 16 . 71 ( male n = 3 , x = 31 . 81 ) , 3% + 2 . 0 ( female n = 3 , x = 2 . 11 ) , Foxp2+ contribution: 27% ± 4 . 05 ( male n = 3 , x = 26 . 45 ) , 23% ± 0 . 85 ( female n = 3 , x = 22 . 69 ) ( c ) . A two-way ANOVA revealed no interaction between subpopulation and sex; however the value was trending ( p=0 . 0660 ) , with a main effect for sex within the Dbx1-derived subpopulation ( p=0 . 0107 , F ( 1 , 8 ) =10 . 95 ) . As the interaction was trending , we also analyzed subpopulation differences and found significant increases in the percent of AR+ expressing Foxp2+ cells in comparison to the Dbx1-derived subpopulation in females ( p=0 . 0308 , F ( 1 , 8 ) =6 . 85 ) ( c ) . n is the number of animals and x is the mean . p<0 . 05 ( * ) , p<0 . 01 ( ) , and p<0 . 001 ( * ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21012 . 013 We also analyzed the contribution of the Dbx1-derived and Foxp2+ populations to the total ERα , Aromatase and AR MeA populations ( Figure 4—figure supplement 2 ) . We observed that both Dbx1-derived and Foxp2+ cells comprised between ~10% to 22% of the total ERα population in male and female mice . The Dbx1-derived population contributed between ~30–40% of the total aromatase population in males and females , which was significantly greater than the contribution of the Foxp2+ population in both males and females . Dbx1-derived ( males only ) and Foxp2+ cells ( males and females ) comprised ~20–38% of the total AR+ population . In contrast , the Dbx1-derived population contributed to only 3% of the total AR+ population in females , which was significantly less than the Foxp2+ contribution in females and less than the contribution of Dbx1-derived cells in males . Collectively , these data reveal that Dbx1-derived and Foxp2+ cells contributed differentially to the aromatase and AR populations , but not to the ERα population . The MeA receives direct inputs from the accessory olfactory bulb ( AOB ) ( Scalia and Winans , 1975; Martel and Baum , 2009; Bergan et al . , 2014 ) and integrates this chemosensory information to regulate innate behaviors including territorial aggression , maternal aggression , mating , and predator avoidance ( Dulac and Wagner , 2006; Kim et al . , 2015 ) . Previous data revealed that at least aggressive and mating behaviors are controlled by MeA GABAergic neurons ( Choi et al . , 2005; Hong et al . , 2014 ) . However , whether different subsets of inhibitory neurons are activated during these behaviors , or if neuronal subtype activation is generalizable across behaviors remains unknown . To directly test these possibilities we performed well characterized aggression , mating and predator odor avoidance behavioral tests in both male and female resident mice and examined the patterns of activation of Dbx1-derived and Foxp2+ cells using c-fos as a readout of neuronal activity in resident mice ( Figure 5—figure supplement 1 ) . The generation of neuronal diversity across amygdala subnuclei has been posited to occur in a compartmentalized manner with amygdala inhibitory neurons generated in the subpallial ganglionic eminences and excitatory neurons arising from the cortical pallial region ( Swanson and Petrovich , 1998 ) . In this model , the amygdala and cerebral cortex develop by a similar mechanism with neurons in both structures originating in shared progenitor domains . However , more recent studies have revealed that the generation of amygdala neuronal diversity is more complex with large populations of neurons originating in progenitor niches dedicated for limbic structures ( Remedios et al . , 2007; Hirata et al . , 2009; Soma et al . , 2009; Waclaw et al . , 2010; García-Moreno et al . , 2010 ) . One of these major niches encompasses the region at the telencephalic-diencephalic border , an origin of MeA output neurons ( Hirata et al . , 2009; García-Moreno et al . , 2010 ) . Our previous studies revealed that the homedomain encoding transcription factor , Dbx1 , marks a subpopulation of progenitors within the POA , which will later generate a subset of MeA inhibitory output neurons ( Hirata et al . , 2009 ) . Here , we significantly extend this work by revealing the presence of a complementary population of progenitors within this niche marked by expression of Foxp2 . Thus , our findings , combined with previous work , suggest a model of MeA development in which distinct progenitor populations at the telencephalic-diencephalic border defined by differential transcription factor expression ( e . gs . Dbx1 , Foxp2 , OTP ) are a major source for MeA neuronal diversity . The function of combinatorial sets of transcription factors in neural progenitors has been shown across the neuraxis as the mechanism for the generation and specification of distinct subclasses of neurons ( Kepecs and Fishell , 2014; Stepien et al . , 2010; Shirasaki and Pfaff , 2002 ) . In addition to specification of neuronal subtype identity , recent studies in the spinal cord and globus pallidus ( Dodson et al . , 2015; Bikoff et al . , 2016 ) have revealed that different combinatorial codes in progenitor pools predict neuronal subtype connectivity patterns , neuronal firing properties and in the case of the globus pallidus , distinct functions in regulating voluntary movements ( Dodson et al . , 2015 ) . Thus , transcription factor expression at the earliest stages of neuronal development likely represents the beginning of an instructive continuum for the establishment of not only neuronal identity , but also development of sub-circuitry regulating different components of motor behaviors . Here , we show that in the MeA , complementary transcription factor expression marks subsets of progenitors and predicts neuronal subtype identity as defined by molecular and electrophysiological signatures . At the molecular level , Dbx1-derived and Foxp2+ neurons express different combinations of the sex steroid hormone pathway protein aromatase and estrogen receptor alpha ( ERα ) . At the electrophysiological level , these two populations possess distinct intrinsic electrophysiological profiles . Thus , our study generates a novel cell-specific transcription factor-based means to predict postnatal MeA neuronal identity . The central importance of the MeA for processing innate behaviors such as aggression , mating and predator avoidance is well-established ( Dulac and Wagner , 2006; Sokolowski and Corbin , 2012 ) . Despite this knowledge , the question of which MeA neuronal subtypes encode instinctive behavioral information has only recently begun to be addressed . Recent optogenetic manipulation of the MeA revealed that glutamatergic neurons mediate repetitive self-grooming behaviors while in contrast GABA-ergic neurons regulate either aggressive or mating behaviors , depending on the level of neuronal activity driven by light stimulation ( Hong et al . , 2014 ) . Moreover , MeA neuronal subclasses expressing different components of the stress response system control appropriate behavioral responses to social cues ( Shemesh et al . , 2016 ) . Here , we contribute to the understanding of amygdala cell-specific regulation of behavior by generating a transcription factor based map of MeA subtype responsiveness to innate-behavioral cues . Thus , our findings provide a developmental molecular context in which to further dissect neuronal subtype control of MeA-driven behaviors . In addition to playing a central role in processing sensory information required for instinctive behaviors , the MeA is one of the known sexually dimorphic structures of the brain ( Cooke and Woolley , 2005; McCarthy and Arnold , 2011; Johnson et al . , 2008 ) . Recent studies employing in vivo recording techniques revealed that a significant number of MeA neurons are dedicated to processing olfactory cues from the opposite sex rather than the same-sex , thus providing a direct demonstration of sex-specific differences in sensory processing ( Bergan et al . , 2014 ) . However , the identity of MeA neurons in males and females that differentially process olfactory-based sensory information has not been delineated . Here , we reveal that while Dbx1-derived and Foxp2+ neurons are broadly activated by mating , aggressive and predator cues , we found stark differences in Dbx1-derived , Foxp2+ and OTP+ cell-specific responses in the male and female brain to mating and predator odor cues . A similar sex-specific control of innate behavior has been directly demonstrated in the ventromedial hypothalamus ( VMH ) , where progesterone receptor-expressing neurons while required for male aggressive and mating behaviors , are only required for female mating behavior ( Yang et al . , 2013 ) . Collectively , our studies in combination with previous studies point to a larger picture in which there are neuronal subpopulations in the MeA and VMH that are involved in the regulation of different innate behaviors in a sex-specific manner . Although our data do not reveal the neuronal and/or circuit mechanisms underlying our observation of sex-specific subpopulation responses to mating behavior and predator odor presentation , some of our findings do provide potential insight as to how this differential processing may occur . The two most straightforward and non-exclusive potential mechanisms are: 1 ) with regard to mating , intrinsic differences in Dbx1-derived and Foxp2+ neurons and/or 2 ) subpopulation specific patterns of local and/or long-range connectivity . Regarding the first potential mechanism , sex steroid hormones and receptors such as aromatase , ERα and AR have been extensively characterized as critical for the output of distinct components of male and female aggressive and mating behaviors ( Yang and Shah , 2014 ) . For example , deletion of AR resulted in alterations in attack duration during territorial aggression and initiation of male mating ( Juntti et al . , 2010 ) , while ablation of aromatase neurons led to impairments in the production of distinct components of aggression in male and female mice ( Unger et al . , 2015 ) . Consistent with the critical role that these factors play in components of innate behaviors , we found that aromatase is expressed solely in the Dbx1-derived lineage . Across species , aromatase has a masculinizing effect ( Wu et al . , 2009; Balthazart et al . , 2011 ) . Thus , it will be interesting to explore how Dbx1-derived aromatase expressing neurons may control male behavioral displays such as mounting and territorial aggression . Furthermore , both Dbx1-derived and Foxp2+ neurons possess distinct electrophysiological profiles , another potential mechanism to control different components of behaviors . Previous work in both vertebrates and invertebrates has revealed that the timing of AP spiking is directly linked to specific behavioral actions . For example , in vertebrates timescale firing differences are associated with dopamine ( DA ) release for the determination of reward behaviors ( Schultz , 2007; Zhang et al . , 2009 ) . It will therefore prove interesting to explore if and how firing patterns of Dbx1-derived and Foxp2+ MeA neurons may control different components of innate behaviors across sexes . The second potential and perhaps more intriguing mechanism that may account for Dbx1-derived and Foxp2+ subtype specific male versus female patterns of neuronal activation during mating are sex specific local and/or long-range patterns of connectivity . Although currently not yet observed in a brain circuit in mammals , such a mechanism has recently been uncovered in c . elegans in which shared male and female circuits show differences in connectivity that is established during wiring ( Oren-Suissa et al . , 2016 ) . Consistent with this , there is some suggestion of sex-specific differences in olfactory-MeA projections in rodents ( Kang et al . , 2011 ) . Although we did not differentiate according to the sex of the animal , we found differences in both the amplitude and frequency of EPSCs between lineages , indicating differences in the strength and/or number of inputs between Dbx1-derived and Foxp2+ MeA neurons . Although the source of input cannot be determined from our analysis , there are direct excitatory inputs to the MeA emanating from the mitral/tufted neurons of the accessory olfactory bulb ( AOB ) ( Martel and Baum , 2009; Bergan et al . , 2014 ) . Thus , determination of input/output wiring patterns of male and female Dbx1-derived and Foxp2+ MeA neurons will be highly informative . In summary , although the precise instructive developmental mechanisms programming innate behaviors remain to be elucidated , we reveal that differential transcription factor expression during development is predictive of neuronal identity based on molecular and electrophysiological criteria and sex-specific patterns of neuronal activation during innate behaviors . Mice were housed in the temperature and light-controlled ( 12 hr light-dark cycle ) Children’s National Medical Center animal care facility and given food and water ad libitum . All animal procedures were approved by the Children’s National Medical Center’s Institutional Animal Care and Use Committee ( IACUC ) and conformed to NIH Guidelines for animal use . Mice used were Dbx1cre+/- ( kindly provided by A . Peirani , Institut Jacques Monod , Paris ) , Shhcre+/- ( Jackson Labs strain B6 . Cg-Shhtm1 ( EGFP/cre ) Cjt/J ) , Emx1cre+/- ( Jackson Labs strain B6 . 129S2-Emx1tm1 ( cre ) Krj/J ) , Nkx6 . 2cre+/- ( Jackson Lab strain Tg Nkx6-2-icre ) 1Kess/SshiJ ) , Foxp2cre+/- ( kindly provided by R . Palmiter , University of Washington ) ( Rousso et al . , 2016 ) , all crossed to Rosa26YFP+/+mice ( Jackson lab strain R26R-EYFP ) . For analysis of aromatase expression , we used Aromatase LacZ reporter mice ( kindly provided by N . Shah , University of California-San Francisco ) ( Wu et al . , 2009 ) . Mice were genotyped by Transnetyx Inc . Genotyping Services . Adult mice were considered between 3–7 months of age . The sample size was based on previous experiments and published data . No statistical methods were used to determine sample sizes . Postnatal mice were transcardially perfused with 4% paraformaldehyde ( PFA ) and brains post fixed overnight at 4°C , embedded in 4% agarose ( Invitrogen ) and sectioned at 50 μm with a vibrating microtome ( Leica VT1000S ) . Embryos were fixed in 4% PFA overnight at 4°C degrees , cryoprotected in 30% sucrose , embedded in O . C . T . compound ( Tekka ) and sectioned at 20 μm on a cryostat ( Leica CM1850 ) . For IHC , tissue sections were incubated overnight with primary antibody , then washed with PBST and 10% normal donkey serum and incubated for 4 hr with the corresponding secondary antibodies , and mounted with DAPI Fluoromount ( SouthernBiotech 0100–20 , Birmingham , AL ) . Primary antibodies used were rat anti-GFP ( to detect YFP expression , ( 1:1000 , Nacalai 04404–84 , San Diego , CA ) , goat anti-Foxp2 ( 1:200; Santa Cruz sc-21069 , Dallas , TX ) , rabbit anti-Foxp2 ( 1:500; abcam ab16046 , Cambridge , UK ) , rabbit anti-OTP ( 1:2000; kind provided by F . Vaccarino , Yale University ) , rabbit anti-androgen receptor ( 1:750; Epitomics AC-0071 , Cambridge , UK ) , rabbit anti-estrogen receptor α ( 1:6000; Millipore 06–935 ) ; mouse anti-NeuN ( 1:200; Millipore MAB-377 ) , mouse anti-CC1 ( 1:250; Calbiochem OP80-100 ) , goat anti-calbindin ( 1:200; Santa Cruz sc-7691 ) , rat anti-somatostatin ( 1:100; Millipore MAB354 , Billerica , MA ) , rabbit anti-cfos ( 1:500; Santa Cruz sc-52 , Dallas , TX ) , goat anti-cfos ( 1:300; Santa Cruz sc-52G , Dallas , TX ) , anti-rabbit nNOS ( 1:8000; ImmunoStar 24287 , Hudson , WI ) , mouse anti-CAMKIIα ( 1:500 Biomol ARG22260 . 50 , Farmingdale , NY ) and chicken anti-βGal ( 1:2000; abcam 9361 , Cambridge , UK ) . Secondary antisera used were donkey anti-rat or anti-goat Alexa 488 ( 1:200; Life Technologies , Waltham , MA ) , anti-rabbit or anti-goat Cy5 ( 1:1000; Jackson ImmunoResearch , Westgrove , PA ) , anti-rabbit or anti-mouse Cy3 ( 1:1000; Jackson ImmunoResearch , Westgrove , PA ) , anti-mouse dylight 649 ( 1:500; Jackson ImmunoResearch , Westgrove , PA ) , or anti-chicken dylight 549 ( 1:500; Jackson ImmunoResearch , Westgrove , PA ) . Fluorescent photographs were taken using an Olympus FX1000 Fluoview Laser Scanning Confocal Microscope ( 1 um optical thickness ) . Unless otherwise stated , data were analyzed using GraphPad six statistical software . We first tested the distribution of the data with the Shapiro-Wallis test for normality . Data that were normally distributed was analyzed using an unpaired two-tailed t-test for analysis of experiments involving two groups ( Figure 1—figure supplement 1; Figure 1—figure supplement 2; Figure 5x; Figure 6c females , k , m , l; Figure 7c , g males , k , l , m Foxp2+ subpopulation; Figure 7—figure supplement 1c , d , h , i ) and a one-way ANOVA ( Figure 5o , w , y Foxp2+ subpopulation ) followed by Tukey-Kramer multiple comparison test was used for analysis of experiments involving three or more groups . Data with a non-normal distribution were analyzed by using the non-parametric test Mann-Whitney when comparing two groups ( Figure 5c , g , k , y Dbx1-derived subpopulation; Figure 6c males , g; Figure 7g females , m Dbx1-derived subpopulation; Figure 7—figure supplement 1j , k ) and Kruskal-Wallis with Dunn’s post-hoc corrections for data with three groups ( Figure 5—figure supplement 5s ) . For the analysis of data shown in Figure 4 and Figure 4—figure supplement 1 we performed the following statistical analysis: when the data met the normality assumption or could be transformed to meet the normality assumption , generally two-way analysis of variance models were implemented to evaluate the evidence of differences in mean effects of the two experimental factors on cell activation ( Figure 4g , u; Figure 4—figure supplement 2a , b ) . In the situation where no data transformation could be found to achieve an acceptable level of normality , quantile regression was performed , which does not require the normality assumption , to evaluate comparable differences in median effects ( Figure 4n; Figure 4—figure supplement 2c ) . In each case , the initial models included a cross-products term to assess evidence of effect modification or interaction . When there was no interaction , it was taken as evidence of the absence of effect modification and the cross-products term was removed leaving only a model that assessed independent effects of the each factor separately , while holding constant any effects of the other factor . Depending on the underlying model , either mean or median effects ± 95% confidence intervals were derived to reflect the differences that were consistent with statistically meaningful differences in the final model . Under consideration of protecting the experiment-wise error rate , as long as the evaluation of differences focused only on identifying the nature of effects deemed statistically meaningful in the final model , there was no correction made for multiple comparisons . Analysis of data meeting the normality assumption was based on GraphPad Prism six and analysis based on quantile regression was implemented in Stata 13 . As mice had to be sacrificed after each behavioral assay in order to conduct c-fos immuno-analysis , technical repeats were not available . Measurements from different mice were considered biological repeats to determine sample size . Data points were considered outliers and excluded if they were two standard deviations away from the mean . Mice ( P25-40 ) were anaesthetized with isoflurane and sacrificed . Brains were removed and immediately immersed in ice-cold oxygenated ( 95% O2/5% CO2 ) sucrose solution ( 234 mM sucrose , 11 mM glucose , 26 mM NaHCO3 , 2 . 5 mM KCl , 1 . 25 mM NaH2PO4 . H2O , 10 mM MgSO4 and 0 . 5 mM CaCl2 ) . Coronal slices of 300 µm in thickness were cut . Slices with amygdala were collected and placed in oxygen-equilibrated artificial cerebral spinal fluid ( ACSF ) as previously described ( Hirata et al . , 2009 ) . Either Dbx1-derived or Foxp2-derived neurons were then visualized using a fluorescent lamp ( Nikon ) with a 450-490λ filter . Whole-cell patch-clamp recordings from YFP-positive fluorescent cells were performed at room temperature with continuous perfusion of ACSF ( Multiclamp 700A , DigiDATA1322 , Molecular Devices ) . Intracellular solution ( in mM ) : 130 Kgluconate , 10 KCl , 2 MgCl2 , 10 HEPES , 10 EGTA , 2 Na2-ATP , 0 . 5 Na2-GTP . All measurements of intrinsic and synaptic properties were analyzed off-line using Clampfit Software ( V . 10 . 2 , Molecular Devices ) and graphing software ( OriginPro 9 . 1 ) . At the end of each recording , biocytin ( 1% ) was injected with the depolarizing current ( 1nA ) for post-hoc morphology analysis . All slices were then fixed with paraformaldehyde overnight at 4°C and processed for Fluorescein-conjugated Avidin-D ( 1:200 , Vector Laboratories ) , YFP IHC ( Dbx1-derived and Foxp2+ recordings ) or Foxp2 IHC ( for Foxp2+ recordings ) as described above . Neurons were filled with biocytin and imaged using an Olympus FX1000 Fluoview Laser Scanning Confocal Microscope ( 0 . 5 um optical thickness ) . VIAS software was used to align confocal images taken at 40x and 60x in the same plane ( x , y , z ) . Neurons were then traced using neuTube software , which uses fixed radii small tubes to estimate the dendritic branches length and thickness ( Feng et al . , 2015 ) . Dbx1cre+/-;Rosa26YFP+/+ male and female mice 3–7 months old were used for the behavioral assays . One week prior to testing , animals were single housed . Testing was performed between the hours of 18:00 and 20:00 corresponding to the beginning of the dark cycle for all assays except the maternal aggression assay which took place from 13:00 to 15:00pm corresponding to the light cycle .	Within the brain , a set of interconnected structures called the limbic system is involved in emotion , motivation and memory . This system – and in particular a structure called the medial amygdala – also contributes to behavioral drives that help an animal to survive and reproduce . These include the drive to avoid predators , to defend territory , and to find a mate . Such behaviors are thought to be inborn or innate . This means that animals display them instinctively whenever specific triggers are present , without the need to learn them beforehand . However , just as a computer must be programmed to perform specific tasks , these innate behavioral responses must also be programmed into the brain . Given that animals do not learn these behaviors , Lischinsky et al . reasoned that specific events during the development of the brain must provide the animal’s brain with the necessary instructions . To test this idea , they studied how the development of the medial amygdala in mouse embryos may give rise to differences in innate mating behavior seen between male and female mice . The medial amygdala contains many subtypes of neurons , which show different responses to sex hormones such as estrogen and androgen . Lischinsky et al . show that two sets of cells give rise to some of the different neurons of the adult medial amygdala . One set of these precursor cells makes a protein called Dbx1 and the other makes a protein called Foxp2 . These two sets of precursors generate medial amygdala neurons with different arrays of sex hormone receptors in male and female mice . Moreover , while the two sets of medial amygdala neurons are activated during aggressive encounters , they show different patterns of activation in male and female animals during mating . These findings suggest that the development of Dbx1-derived and Foxp2+ neurons in the medial amygdala helps program innate reproductive and aggressive behaviors into the brain . The new findings also provide insights into why these behaviors differ in male and female mice . The next challenge is to identify the inputs and outputs of these two distinct subpopulations of medial amygdala neurons . This should make it possible to work out exactly how these populations of cells control innate behaviors in male and female animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Embryonic transcription factor expression in mice predicts medial amygdala neuronal identity and sex-specific responses to innate behavioral cues
Since 2015 , the World Health Organisation ( WHO ) recommends immediate initiation of antiretroviral therapy ( ART ) for all HIV-positive patients . Epidemiological evidence points to important health benefits of immediate ART initiation; however , the policy’s impact on the economic aspects of patients' lives remains unknown . We conducted a stepped-wedge cluster-randomised controlled trial in Eswatini to determine the causal impact of immediate ART initiation on patients’ individual- and household-level economic outcomes . Fourteen healthcare facilities were non-randomly matched into pairs and then randomly allocated to transition from the standard of care ( ART eligibility at CD4 counts of <350 cells/mm3 until September 2016 and <500 cells/mm3 thereafter ) to the ‘Early Initiation of ART for All’ ( EAAA ) intervention at one of seven timepoints . Patients , healthcare personnel , and outcome assessors remained unblinded . Data were collected via standardised paper-based surveys with HIV-positive adults who were neither pregnant nor breastfeeding . Outcomes were patients’ time use , employment status , household expenditures , and household living standards . A total sample of 3019 participants were interviewed over the duration of the study . The mean number of participants approached at each facility per time step varied from 4 to 112 participants . Using mixed-effects negative binomial regressions accounting for time trends and clustering at the level of the healthcare facility , we found no significant difference between study arms for any economic outcome . Specifically , the EAAA intervention had no significant effect on non-resting time use ( RR = 1 . 00 [CI: 0 . 96 , 1 . 05 , p=0 . 93] ) or income-generating time use ( RR = 0 . 94 , [CI: 0 . 73 , 1 . 20 , p=0 . 61] ) . Employment and household expenditures decreased slightly but not significantly in the EAAA group , with risk ratios of 0 . 93 [CI: 0 . 82 , 1 . 04 , p=0 . 21] and 0 . 92 [CI: 0 . 79 , 1 . 06 , p=0 . 26] , respectively . We also found no significant treatment effect on households’ asset ownership and living standards ( RR = 0 . 96 , [CI 0 . 92 , 1 . 00 , p=0 . 253] ) . Lastly , there was no evidence of heterogeneity in effect estimates by patients’ sex , age , education , timing of HIV diagnosis and ART initiation . Our findings do not provide evidence that should discourage further investments into scaling up immediate ART for all HIV patients . Funded by the Dutch Postcode Lottery in the Netherlands , Alexander von Humboldt-Stiftung ( Humboldt-Stiftung ) , the Embassy of the Kingdom of the Netherlands in South Africa/Mozambique , British Columbia Centre of Excellence in Canada , Doctors Without Borders ( MSF USA ) , National Center for Advancing Translational Sciences of the National Institutes of Health and Joachim Herz Foundation . NCT02909218 and NCT03789448 . Recent trials have pointed to substantial health benefits of immediate antiretroviral therapy ( ART ) initiation for all HIV-positive patients compared to initiating ART based on a CD4-cell count threshold . Benefits include reduced HIV-related mortality and morbidity and decreased transmission risk to HIV-negative sexual partners ( Danel et al . , 2015; Cohen et al . , 2016; Lundgren et al . , 2015; Hayes et al . , 2019; Ford et al . , 2018 ) . In line with this epidemiological evidence , the World Health Organization ( WHO ) has updated its consolidated guidelines on the use of antiretrovirals in 2015 , now advocating for immediate ART initiation ( or ‘universal test and treat’ ) for all HIV-positive adults , adolescents , and children ( WHO , 2019 ) . In view of these major changes in ART provision , it is crucial for health policy makers to understand the implications of immediate ART initiation for HIV patients’ economic outcomes . At high CD4-count levels , we would expect the majority of patients to be relatively healthy , and thus have productivity levels and out-of-pocket health expenditures that are similar to those of HIV-negative patients ( Thirumurthy et al . , 2013 ) . While ART may still improve economic welfare among these patients through an improvement in health status , it may also decrease these patients’ economic welfare through , for example , the side effects of antiretroviral drugs , increased frequency of ( ART ) clinic visits or stigma from taking ART ( daCosta DiBonaventura et al . , 2012; Unge et al . , 2008 ) . The economic consequences of early ART initiation for this specific patient group are therefore unclear and have to date not been investigated experimentally . Previous studies have assessed labour market outcomes and overall financial wellbeing of patients on ART , relative to patients not yet on ART . Of these , several studies have highlighted beneficial economic impacts of ART initiation , which are primarily based on the positive labour market effects of improved health . Accordingly , empirical evidence has pointed to higher work performance and productivity , ( Bor et al . , 2012; Larson et al . , 2008; Beard et al . , 2009; Thirumurthy et al . , 2008 ) lower absenteeism , ( Larson et al . , 2008 ) increases in savings rates , ( Baranov and Kohler , 2018 ) as well as increased educational expenditures and attainment ( Baranov and Kohler , 2018; Lucas et al . , 2019 ) following ART initiation . Conversely , other studies have documented detrimental economic effects of ART initiation ( even under universal access to ART schemes ) , largely driven by three suggested mechanisms: first , by increased patient-borne healthcare expenditures associated with travel to ART clinics , clinic and hospital fees , and income foregone; ( Rosen et al . , 2007; Chimbindi et al . , 2015; Leive and Xu , 2008 ) second , by elevated levels of food insecurity due to a treatment-induced increase in appetite and fewer financial resources to absorb the higher food expenditures; ( Patenaude et al . , 2018 ) and third , by reduced productivity as a result of short-term adverse and toxic effects linked to antiretroviral drugs ( Danel et al . , 2015; Rosen et al . , 2007 ) . However , these previous studies provide only little insights on the anticipated economic effects of immediate ART initiation because they are based on outdated treatment guidelines , thus comparing HIV-patients above and below a certain CD4 cell count level ( e . g . 500 cells/mm3 ) that determines ART eligibility . Given that HIV-patients who are not yet on ART but have a relatively low CD4 count may be more susceptible to opportunistic infections and adverse events than those with higher CD4 counts , this comparison group is inadequate for assessing the economic consequences of the current WHO-endorsed ART initiation strategy that is independent of patients’ CD4 counts . To decide whether and how much governments and international organisations should invest in scaling up immediate ART initiation for all HIV-patients , it is crucial to understand the impact of immediate ART initiation not only on health but also on HIV patients’ economic outcomes . This is the first randomised trial aimed at answering this question . Specifically , we conducted a stepped-wedge cluster-randomised controlled trial of the ‘Early Initiation of ART for All’ ( EAAA ) intervention for HIV-patients in Eswatini to test the causal impact of immediate ART initiation on a range of economic outcomes , including patients’ time use , employment , household expenditures , and household living standards . Fourteen healthcare facilities ( ‘clusters’ ) were consecutively enrolled into the Maximising ART for Better Health and Zero New HIV Infections ( MaxART ) stepped-wedge trial and 3019 participants were interviewed over the duration of the study . The mean number of participants approached at each facility and time step varied from 3 . 5 to 112 participants ( see Figure 1 ) . Table 1 summarises sociodemographic information separately for two study samples . The full sample was composed of 3019 participants , sampled across 14 healthcare facilities . Participants enrolled into the EAAA intervention arm were on average aged 38 . 3 years ( range: 18–85 years ) , 71 . 0% were female , 53 . 5% were married , and 56 . 0% had completed at least some secondary schooling . Participants in the standard of care group had similar characteristics: 74 . 3% were female , 56 . 6% married , and 56 . 0% had completed at least some secondary education . The random subset of participants with household-level data on household expenditures and living standards was composed of 1485 patients who were also sampled across all 14 healthcare facilities . Overall , sociodemographic characteristics were very similar to those of the full sample . Overall , causal random forests did not identify subgroups with effects that diverged significantly from the average treatment effect . Across outcomes , most heterogeneity was found along the variables ( i ) patients’ time on ART , ( ii ) number of months passed since patients’ HIV diagnosis , ( iii ) years of education completed , and ( iv ) age , whereas the importance metric for patients’ sex was very small , possibly due to an over-representation of women in our sample . The plots presented in Figure 2—figure supplement 5–9 depict heterogeneity in treatment effects along these four moderating variables . It appears that the program’s effect on most economic outcomes was slightly higher for patients with shorter rather than longer time on ART . However , it has to be cautioned that heterogeneity was not statistically significant for any of the four economic outcomes . We present the first causal evaluation of the effect of immediate ART for all HIV patients on wider economic outcomes . Based on our primary results and several robustness checks , we are able to conclude that large harmful effects are very unlikely . More specifically , we found that neither patients’ time use nor their employment status and living standards were positively or negatively affected by the EAAA intervention . Although we found a reduction in monthly household expenditures among patients in the EAAA group , the magnitude was small in size ( −126 . 17 SZL , corresponding to 3% of the average monthly household expenditures in Eswatini ) ( The World Bank , 2019 ) and not statistically significant . Lastly , in machine-learning-supported heterogeneity analyses , we also did not find any patient subgroup for which the EAAA intervention either significantly improved or deteriorated individual- and household-level economic outcomes . Two previous publications based on the same trial have assessed how early ART initiation affected patients’ health , revealing a 6% higher retention in care rate in the EAAA group but no significant differences with regard to all-cause , disease-related , and HIV-related mortality between the EAAA and the standard of care group ( Chao et al . , 2020; Perriat et al . , 2018 ) . While we were unable to link responses from this survey to patients’ clinical data , we may still infer that more substantial health impacts would have been necessary to significantly affect patients’ economic welfare . It is also possible that both the health and economic benefits of early ART initiation only materialise after a longer follow-up time , beyond the 36 months observation period covered in this trial ( May et al . , 2016 ) . A potential alternative explanation for the absence of strong and beneficial treatment effects could relate to the broader socioeconomic conditions of the study region . Hence , if income generation opportunities are generally constrained due to given economic circumstances , people living with HIV may be unable to find work , irrespective of whether they are healthy or not . If patients’ health status does not substantially impact their earning potential , other welfare indicators such as household expenditures and living standards are also unlikely to change . However , we partly alleviate this problem by adopting a broad definition of employment by including informal and short-term work and should therefore be able to capture even small changes in participants’ income generation activities . In contrast to several prior studies ( Beard et al . , 2009; Thirumurthy et al . , 2008; Habyarimana et al . , 2010; Baranov and Kohler , 2018 ) , our findings did not exhibit any substantial detrimental financial and economic consequences of ART initiation . At the very least , our results suggest that ART-related adverse events were not substantial enough to provoke significant drops in patients’ productivity levels ( Nansseu and Bigna , 2017 ) . We therefore add important new empirical evidence from the perspective of patients’ economic wellbeing , which – given that EAAA does not appear to have large detrimental effects on patients’ economic outcomes – support the 2015 WHO recommendation to offer immediate ART initiation to all HIV-patients . Our study has several key strengths . First , we examined a comprehensive set of outcome variables and are thus able to gain nuanced insights into participants’ overall economic situation . Although the different outcomes are likely correlated , time use and employment are patient-level variables , whereas expenditures and living standards are captured at the household level . The latter two variables could thus be differently affected by the EAAA intervention depending on whether the patient is the household’s main breadwinner or not . Household savings could have been another possible welfare-related aspect to assess . However , the general savings rate in Eswatini is low ( World Bank Group , 2017 ) and savings are often mainly used to smooth consumption , and thus likely highly correlated with overall household expenditures . Second , we have collected very detailed information on patients’ time use . Time use is a measure that is presumably highly sensitive to potential short-term changes in economic productivity and , given that we asked about the previous 24 hr , less prone to measurement error or bias from a long recall period . In view of the precise null effects for the time use outcome , we can more confidently conclude that immediate ART initiation had no harmful effects on patients’ overall productivity levels . Third , and arguably most importantly , this is the first randomised study - and thus the first study to allow for causal inference under no untestable assumptions – of the impact of immediate ART initiation on indicators of patients’ economic outcomes . This study has six main limitations . First , biological data on patients’ CD4 count levels and viral loads were not collected . It was thus not possible to assess whether the effects of EAAA on patients’ economic outcomes were different among those patients who had a CD4 count close to the treatment threshold at the time of ART initiation . Second , participant recruitment was implemented within healthcare facilities and it is therefore possible that patients who generally attend healthcare services more regularly and reliably were overrepresented in the study sample . Third , participants were not followed-up on longitudinally , which implies that for each individual , we either had a measurement of the pre- or the post-intervention phase ( but never for both ) . Our effect estimates are based on the comparison of patients in the standard of care phase with patients in the EAAA phase , and would turn invalid if there was significant imbalance in baseline characteristics between these two groups . However , this is unlikely in view of the sufficiently large sample size and the random selection of interview dates for each facility . Fourth , data on household expenditures and household living standards were only collected from a random subsample of 50% of patients . Given the wide confidence interval of the effect estimates for household expenditures , it is possible that we would have been able to find a significant effect for this outcome of a size that would still be meaningful to health policy makers if we had had a larger sample size . Fifth , data was based on patients’ self-report . Especially with regard to household expenditures and time use , this limitation is likely to have led to some degree of measurement error due to recall problems ( Filmer and Pritchett , 2001; Sahn and Stifel , 2000 ) . In addition , while monthly expenses were summarised into 20 distinct expenditure categories to reduce interview length and cost , this may have led to further measurement imprecisions , for instance through adding up expenses for numerous individual food items into an overall category of ‘total shopping for food and groceries’ . Yet , we expect that these measurement errors and reporting biases occurred – on average – to an equal degree in the EAAA and standard of care group and are therefore unlikely to systematically bias our point estimates of the causal intervention effect . Lastly , the employment rate in our study sample diverged from the national employment rate during the same period . This discrepancy could be explained by ( i ) the composition of our sample , which consisted of 75% female patients and is therefore not representative for the population as a whole , ( ii ) the temporal disaggregation into tertials , which might reflect some seasonal fluctuations in our data , and ( iii ) the lack of regional labour force data for the general population in the Hhohho region , rather than the aggregated national data that we have used as a reference . This study provides the first causal evidence on the effect of immediate ART initiation on individual- and household-level economic outcomes . EAAA is unlikely to have detectable , harmful economic repercussions for HIV patients in Eswatini . This is an important finding for health policy making in that it buttresses the WHO recommendation to discard eligibility thresholds for ART from the perspective of patients’ economic wellbeing – a perspective that is often ignored in the setting of clinical recommendations , yet important to those who are directly affected by these recommendations ( Govindasamy et al . , 2012; Fox et al . , 2010 ) . The study was implemented in North-Western Eswatini ( formerly ‘Swaziland’ ) . 27 . 0% of the general population in Eswatini are HIV-positive; the highest HIV prevalence worldwide ( Government of the Kingdom of Eswatini , 2017 ) . The trial enrolled 14 government-managed health facilities located in the Hhohho region ( see Figure 4 ) . At the study’s outset in 2014 , all health facilities provided comprehensive HIV care and treatment according to the national adult HIV treatment guidelines effective at the time , thus initiating ART according to prescribed CD4 count levels . According to the Annual HIV Program Report of 2014 ( Kingdom of Swaziland , Ministry of Health , 2014 ) , almost 60% of HIV-patients in the Hhohho region had been initiated on ART in the year prior to the trial roll-out . Health facilities were allocated non-randomly into seven pairs based on their geographic proximity to avoid possible contamination and based on their facility catchment size to ensure that group sizes were roughly equal . Over the course of 3 years , each of the seven pairs was randomly assigned to one of seven sequences , which determined the point in time at which each facility shifted from the standard of care ( control condition ) to the Early Access to ART for All ( EAAA ) intervention ( treatment condition ) ( see Table 2 ) . Hence , in the first period , all facilities adhered to the national standard of care while in the last period , all facilities had adopted EAAA . The randomisation was carried out by the trial statisticians . No stratification was used . This was an open-label trial in which healthcare providers and patients were unblinded to the intervention itself . However , the timing of the transition was only revealed to healthcare providers six to four weeks prior to the start of EAAA implementation . In the standard of care phase , following national treatment guidelines effective at the time , ART eligibility was restricted to patients with CD4-cell counts of <350 cells/mm3 in the first 1 . 5 years of the study . In October 2016 , the eligibility threshold was raised to CD4-cell counts < 500 cells/mm3 . Eligible patients were typically initiated on Eswatini’s first-line ART regimen ( Tenofovir ( TDF ) + Lamivudine ( 3TC ) + Efavirenz ( EFV ) ) . Those with contraindications to this regimen were initiated on alternative regimens , including TDF + 3TC + Nevirapine ( NVP ) or Zidovudine ( AZT ) + 3TC + NVP ( when EFV could not be used ) ; Abacavir ( ABC ) + 3TC + EFV or AZT + 3TC + EFV ( when TDF could not be used ) ; ABC + 3TC + EFV or Stavudine ( D4T ) + 3TC + EFV ( when AZT could not be used ) . Patients attended one private and one group counselling session prior to initiation . While same-day ART initiation was allowed according to the national Integrated HIV Management Guidelines ( Ministry of Health , Kingdom of Swaziland , 2015 ) , HIV diagnosis and ART initiation in the respective facilities were typically a few days apart . During the EAAA intervention phase , all patients who tested HIV-positive as well as patients enrolled in pre-ART care were offered immediate ART initiation , independent of their CD4-cell count . They received one counselling session and ART initiation on the same day and further monthly counselling after initiation . As in the standard of care , patients in the EAAA programme were initiated on Eswatini’s first-line treatment regimen or , if contraindicated , on the same alternative regimens detailed above . Data were collected via standardised paper-based questionnaires over eight time periods ( baseline and seven transitions ) . In every period , a sample of all HIV-care patients in each of the enrolled healthcare facilities was randomly selected . Eligibility was constrained to patients who were HIV-positive and over the age of 18 years , and who were neither pregnant nor breastfeeding . Patients were eligible irrespective of whether ART initiation could take place on the same day of HIV diagnosis or a few days thereafter . For each facility , the study team randomly selected data collection days . On these days , the study team adopted the sampling strategy of selecting the next patient entering the consultation room . This strategy yields a representative sample if the sample size is sufficiently large and the order with which patients are seen by a clinician is random . Monte Carlo simulations have shown that this sampling strategy also tends to be more efficient and unbiased compared to simple and systematic random sampling , and does not underrepresent potentially healthier patients with shorter consultations as is the case when sampling those exiting the consultation room ( Geldsetzer et al . , 2018 ) . Respondents gave verbal and written consent before completing the interview and were informed about their right to decline or withdraw their participation at any point in time . No prior sample size calculations were performed . We assessed the impact of the EAAA intervention on four economic outcomes . First , patients’ time use during the day prior to the interview was measured by collecting detailed information on hourly activities for a cycle of 24 hr . For our analysis , we specified two outcomes that are indicative of patients’ productivity levels: ( i ) ‘non-resting time’ to capture the total hours spent on activities other than sleeping and resting , and ( ii ) ‘income-generating time’ to capture the total hours spent on any income generation activities , which comprised formal employment , primary production activities in the informal sector , subsistence farming , and income generated from own businesses ( i . e . from the sale of goods ) . The second outcome was patients’ current labour market participation , categorised as ‘employed’ if patients were working or engaged in subsistence farming ( either part- or full-time ) , and categorised as ‘not employed’ if patients were unemployed , retired or taking sick or other leave . The third outcome was patients’ total past-month household expenditures on food- and non-food items , which was measured by asking each participant how much their household spends on 20 common expenditure items in a normal month ( or , if the respondent preferred , in the past year ) as well as on ‘other usual expenses’ and ‘large purchases or expenses in the last 12 months’ that were not mentioned in the list of common expenditure items . We opted for expenditure rather than income data because it is less affected by possible seasonal fluctuations in earnings and therefore better reflects a welfare level that households can maintain through consumption smoothing and informal borrowing ( Sahn and Stifel , 2003; Filmer and Pritchett , 2001 ) . The last outcome was household living standards , measured as an additive index counting the total number of realised housing quality indicators ( 12 items , e . g . , drinking water inside the house , concrete walls , flush toilet , etc . ) and assets owned ( 30 items , e . g . , refrigerator , phone , TV , animals , etc . ) . In line with economic literature ( Sahn and Stifel , 2003; Filmer and Pritchett , 2001 ) , we also computed a principal-component-weighted index from the answers to these housing quality indicators and owned assets as an alternative metric to the additive index , reported in Supplementary file 1H . Information on time use and employment was captured for the full sample . In order to reduce the length of the survey , questions on household expenditures and household living standards were asked to every second participant who was interviewed . We estimated the intent-to-treat effect ( ITT ) by comparing patients interviewed in the standard of care phase to patients interviewed in the EAAA phase ( see EXHIBIT 2 ) . We used mixed-effects negative binomial regressions ( showing the resulting risk ratios ) to account for the skewed distribution of some outcome variables ( income-generating time and household expenditures ) . For normally distributed outcome variables ( non-resting time and living standards ) , we additionally provide results from mixed-effects linear regressions in supplementary tables . For the binary employment outcome , we also estimated risk ratios for ease of interpretation by utilising a modified poisson regression model with a robust error structure . Following the conventional Hussey and Hughes approach , regression models included a binary indicator ( ‘fixed effect’ ) for each time period and a clinic-level random effect to account for clustering by clinic ( Hussey and Hughes , 2007 ) . While clinic-level random effects help to partly adjust for varying cluster size by assigning higher weights to larger clusters , we additionally included a permutation test to project more conservative p-values that correct for ( i ) the varying cluster sizes , ( ii ) the relatively small number of clusters , and ( iii ) potential violations in asymptotic properties of the regression models ( Athey and Imbens , 2016 ) . Specifically , for each of the main outcome models ( Hussey and Hughes model with control variables ) , we used a permutation test ( implemented in the ‘swpermute’ package in Stata Thompson , 2019 ) with 1000 repetitions to test for the statistical significance of the treatment effect point estimates . In supplemental results , we present a second , more flexible , model that allows for potentially heterogeneous time trends across healthcare facilities by including a random slope for time period ( Thompson et al . , 2017 ) . Each of the two models was estimated without and with control variables , consisting of patients’ age ( continuous ) , sex ( binary ) , marital status ( binary ) , and their level of education ( continuous , specifying the highest grade completed ) . While adjustment for these variables is not needed to obtain unbiased effect estimates , their inclusion in the regressions might correct for small sample biases and improve precision . 180 participants ( 12% of the complete random subsample ) did not respond to the household expenditure questions . For this outcome , we therefore ran two regression specifications , one based on the incomplete sample ( i . e . a complete case analysis ) and one based on a complete sample after imputing missing observations using multivariate imputation by chained equations ( MICE ) ( Azur et al . , 2011 ) . The imputation model was implemented using the ‘mice’ package in Stata ( Royston and White , 2011 ) . We implemented the imputation with 1000 repetitions and included all variables used in the main outcome analysis as well as additional ‘auxiliary’ variables , which were current employment , education level , household living standards , and patients’ sociodemographic characteristics . Assuming that the likelihood of a missing value is only a function of observed characteristics , the MICE procedure iteratively estimates missing values based on Markov Chain Monte Carlo techniques . It creates 1000 complete datasets to estimate missing values , which are then averaged across all datasets ( Yu et al . , 2007; Azur et al . , 2011 ) . Lastly , we estimated for each of the five outcomes whether there was heterogeneity in treatment effects between different groups of patients . For this purpose , we utilised a machine learning approach in the form of a non-parametric causal forest algorithm ( Athey et al . , 2019; Athey and Wager , 2019; Wager and Athey , 2018 ) . This approach has advantages over other subgroup tests ( Lee , 2009; Crump et al . , 2008 ) in that it ( i ) does not require an a priori hypothesis on the potential differential effects , ( ii ) increases statistical power , ( iii ) and yields treatment effect estimates that are asymptotically normal ( Athey and Wager , 2019; Wager and Athey , 2018 ) . In this analysis , we first assessed whether treatment effects for any subgroup were significantly different from the average treatment effect . In a second step , we explored the nature of potential heterogeneity through ordering moderating variables by their importance . The random forest heterogeneity analysis was implemented in R 3 . 6 . 2 . All other analyses were conducted in Stata 15 .	Human immunodeficiency virus ( HIV ) is an incurable virus that attacks the immune system and affects around 39 million people worldwide . Once diagnosed , HIV can be treated with antiretroviral therapy ( ART ) to limit its effects and stop it spreading to other people . HIV rates vary across the world , but the African country of Eswatini has the highest prevalence with more than one in four ( 27% ) people classed as HIV-positive . Until 2015 , people living with HIV were typically only treated with ART once their immune system weakened . Recent studies found that starting treatment earlier enhances the positive effects of ART . This caused the World Health Organization ( WHO ) to change their guidelines and advise people living with HIV to begin ART as soon as they are diagnosed . While antiretroviral drugs are usually provided to patients free of charge , accessing care can be expensive for patients because of high transport costs or lost time from income-generating activities . This means starting treatment earlier and , thus , having more frequent healthcare visits , may result in a greater cost to the patient . The economic impact of this change is unclear , and for patients living in poverty , these added costs can affect their decision on whether to continue treatment . Steinert et al . interviewed 3 , 019 HIV-patients from 14 health facilities in Eswatini who began treatment with ART either immediately after diagnosis or after their immune system became suppressed . Patients were asked about their time spent being active to generate income , employment status , monthly household expenditures , and household living standards . On average , beginning ART earlier appears to have had no large negative effects on the economic wellbeing of patients . The same results were found for patient groups defined by sex , education , age , and time spent taking ART . These findings suggest that starting ART for HIV as soon as possible offers medical benefits and seems to have no large economic consequences for patients in the short term , even for poorer communities . This adds weight to the WHO advice on HIV treatment and supports the need to continue to deliver effective treatments to countries like Eswatini that have a high rate of HIV infection .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "medicine" ]	2020	A stepped-wedge randomised trial on the impact of early ART initiation on HIV-patients’ economic outcomes in Eswatini
Microtubule dynamics and polarity stem from the polymerization of αβ-tubulin heterodimers . Five conserved tubulin cofactors/chaperones and the Arl2 GTPase regulate α- and β-tubulin assembly into heterodimers and maintain the soluble tubulin pool in the cytoplasm , but their physical mechanisms are unknown . Here , we reconstitute a core tubulin chaperone consisting of tubulin cofactors TBCD , TBCE , and Arl2 , and reveal a cage-like structure for regulating αβ-tubulin . Biochemical assays and electron microscopy structures of multiple intermediates show the sequential binding of αβ-tubulin dimer followed by tubulin cofactor TBCC onto this chaperone , forming a ternary complex in which Arl2 GTP hydrolysis is activated to alter αβ-tubulin conformation . A GTP-state locked Arl2 mutant inhibits ternary complex dissociation in vitro and causes severe defects in microtubule dynamics in vivo . Our studies suggest a revised paradigm for tubulin cofactors and Arl2 functions as a catalytic chaperone that regulates soluble αβ-tubulin assembly and maintenance to support microtubule dynamics . Microtubules ( MTs ) are dynamic polymers that modulate fundamental cellular processes through dynamic αβ-tubulin polymerization and depolymerization at their ends , and serve as polarized tracks for molecular motor proteins ( Akhmanova and Steinmetz , 2008 ) . Polarity and dynamic instability are fundamental features of the MT polymer , originating from the head-to-tail polymerization of αβ-tubulin heterodimers ( Nogales et al . , 1999; Alushin et al . , 2014 ) . The αβ-tubulin dimer contains two GTP-binding sites: an inactive non-exchangeable site ( N-site ) on α-tubulin , which is suggested to stabilize αβ-tubulin dimers during their biogenesis , and an active exchangeable site ( E-site ) on β-tubulin , which is stimulated to hydrolyze GTP upon αβ-tubulin incorporation into MT lattices at the plus ends ( Nogales et al . , 1999; Alushin et al . , 2014 ) . GTP hydrolysis at the E-site leads to dynamic instability ( catastrophe ) at MT plus ends , due to the strain induced by the curvature of individual protofilaments ( Alushin et al . , 2014; Brouhard and Rice , 2014 ) . Intracellular MT dynamics critically relies on a tightly controlled pool of soluble αβ-tubulin dimers in the cytoplasm . Despite their importance , the mechanisms for biogenesis , maintenance , and degradation of soluble αβ-tubulin dimers remain poorly understood ( Tian and Cowan , 2013 ) . αβ-tubulin is maintained at a high concentration ( ∼6 μM ) in the cytoplasm through regulation of translation from tubulin mRNAs ( Cleveland et al . , 1978; Cleveland , 1989 ) . α- and β-tubulin are translated and folded as monomers in the type II chaperonin TRIC/CCT ( Lewis et al . , 1997 ) . Biogenesis and degradation of the αβ-tubulin heterodimer are non-spontaneous processes that rely on five highly conserved tubulin cofactor ( TBC ) proteins: TBCA , TBCB , TBCC , TBCD , and TBCE ( described in Figure 1A; Lewis et al . , 1997; Lundin et al . , 2010 ) . Orthologs of these proteins have been identified in all eukaryotes studied to date ( Lewis et al . , 1997; Lundin et al . , 2010 ) . The maintenance of a concentrated pool of tubulin dimers by the TBC proteins is essential for proper MT dynamics in eukaryotic cells ( Tian et al . , 1996; Lewis et al . , 1997; Lundin et al . , 2010 ) . The TBC proteins' functions are finely balanced: their loss or their overexpression are both lethal in most eukaryotes , stemming from a complete loss of the MT cytoskeleton ( Steinborn et al . , 2002; Lacefield et al . , 2006; Jin et al . , 2009 ) . In budding yeast , the first identified chromosomal instability ( CIN ) phenotypes , showing severe mitotic spindle defects due to loss of MTs , were ultimately traced to loss of TBC proteins ( Hoyt et al . , 1990 , 1997; Antoshechkin and Han , 2002; Steinborn et al . , 2002; Lacefield et al . , 2006; Jin et al . , 2009 ) . In humans , missense mutations in TBCE and TBCB are linked to hypo-parathyroidism facial dysmorphism ( also termed Kenny-Caffey syndrome ) and giant axonal neuropathy , in which developmental defects are observed due to impairment of MT cytoskeleton function ( Parvari et al . , 2002; Wang et al . , 2005 ) . In addition to the five conserved TBC proteins , the small Arl2 GTPase ( ADP Ribosylation Factor-Like-2 ) regulates the function of TBC proteins in αβ-tubulin biogenesis/degradation through an unknown mechanism ( Figure 1A ) . Although Arl2 is not considered a tubulin cofactor , its loss causes nearly identical defects to those observed with TBCC , TBCD , or TBCE loss ( Hoyt et al . , 1997; Radcliffe et al . , 2000; Mori and Toda , 2013 ) . 10 . 7554/eLife . 08811 . 003Figure 1 . Tubulin cofactors and Arl2 GTPase: domain organization and paradigm for function . ( A ) Tubulin cofactors A–E , Arl2 GTPase masses , and domain organization . TBCA and TBCB co-expression is not required for TBC-DEG expression . Red arrowheads mark domains required for forming TBC-DEG complex assembly . Blue arrowheads mark domains not required for TBC-DEG complex assembly . ( B ) Initial paradigm for tubulin cofactors and Arl2 activities based on previous studies . Each of the molecules is suggested to be monomeric , and only assemble into complexes to drive αβ-tubulin biogenesis or degradation , via interactions regulated by dynamic equilibria . TBCA binds nascent β-tubulin and TBCB binds nascent α-tubulin . TBCA and TBCB are replaced by TBCD and TBCE , respectively . TBCC drives TBCE-α-tubulin and TBCD-β-tubulin to form a supercomplex . GTP hydrolysis in Arl2 is activated by TBCC in a parallel pathway to tubulin assembly . Tubulin biogenesis and degradation intermediate bind and form tubulin dimers , a process that requires Arl2 and tubulin to undergo GTP hydrolysis as an energy source . ( Adopted from Lewis et al . , 1997 . ) DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 003 A stepwise αβ-tubulin biogenesis/degradation paradigm has been proposed based on genetic and biochemical studies ( Tian et al . , 1996; Lewis et al . , 1997; Lundin et al . , 2010; shown in Figure 1B ) , in which TBC proteins form dynamic assemblies to dimerize αβ-tubulin , as follows: ( 1 ) TBCA and TBCB bind β-tubulin and α-tubulin monomers , respectively , after their folding; ( 2 ) TBCA hands off β-tubulin to TBCD , and TBCB hands off α-tubulin to TBCE; ( 3 ) TBCC drives association of TBCD and TBCE with their bound α- and β-tubulin monomers , to form a ‘super-complex’ that forms and activates the αβ-tubulin dimer ( Tian and Cowan , 2013 ) ; and ( 4 ) Arl2 is simulated to hydrolyze GTP through the GTPase activating protein ( GAP ) function of TBCC . The role of Arl2 GTP hydrolysis in this pathway remains unknown ( Bhamidipati et al . , 2000 ) ; Arl2 and its activation by TBCC have been suggested to operate in parallel to the TBC pathway ( Figure 1B ) . However , the roles for TBCC and the Arl2 GTPase remain poorly understood ( Tian et al . , 1999; Mori and Toda , 2013 ) . Overexpression of TBC proteins results in one of two unique phenotypes: TBCA or TBCB overexpression in budding or fission yeast suppresses defects induced by overexpression of α- or β- tubulin , but does not otherwise affect MT dynamics . In contrast , overexpression of TBCC , TBCD , TBCE , or Arl2 leads to rapid MT loss ( Archer et al . , 1998; Feierbach et al . , 1999; Radcliffe et al . , 1999; Lacefield et al . , 2006 ) . Here , we show that TBCD , TBCE , and Arl2 assemble into a stable heterotrimeric chaperone ( TBC-DEG ) with a cage-like structure . This chaperone binds αβ-tubulin and TBCC sequentially , serving as a catalytic platform powered by the Arl2 GTPase for αβ-tubulin assembly and activation . A soluble αβ-tubulin dimer binds TBC-DEG and primes Arl2 , followed by TBCC binding and GTP hydrolysis activation . We show that TBCC is a unique GAP for which affinity depends on αβ-tubulin binding onto TBC-DEG . TBCC promotes GTP hydrolysis through its C-terminal β-helix domain , which interfaces with both Arl2 and αβ-tubulin in a ternary complex . We further find that in Saccharomyces cerevisiae cells , a mutation locking the Arl2 GTPase into a GTP-bound state profoundly affects MT dynamics . Overall , our studies reveal a new role for tubulin cofactors TBCD , TBCE , and Arl2 , which together assemble a GTP-hydrolyzing tubulin chaperone critical for the biogenesis , maintenance , and degradation of soluble αβ-tubulin , defects in which have a profound effect on MT dynamics in vivo . The finding that αβ-tubulin is assembled on a multi-subunit platform establishes a new paradigm for the mechanisms of the TBC proteins in tubulin biogenesis , maintenance , and degradation ( Figure 1B ) . To gain insight into the molecular mechanisms of tubulin cofactors and Arl2 , we expressed the S . cerevisiae orthologs of TBCA , TBCB , TBCC , TBCD , TBCE , and Arl2 ( named Rbl2 , Alf1 , Cin1p , Pac2p , Cin2p , and Cin4p , and referred to hereafter as TBCA , TBCB , TBCC , TBCD , TBCE , and Arl2 [Figure 1A] ) both individually and in combinations , with the goal of reconstituting relevant complexes . TBCA and TBCB are small proteins ( 12 and 28 kDa in S . cerevisiae ) that have been suggested to sequester monomeric β- and α-tubulin , respectively , while TBCC , TBCD , TBCE , and Arl2 regulate αβ-tubulin dimer biogenesis and degradation through unknown mechanisms ( Archer et al . , 1998; Feierbach et al . , 1999; Lundin et al . , 2010 ) . Sequence alignments and structure predictions identify conserved domains within each protein ( Figure 1A ) , but the molecular functions of these domains remain unknown . We found that TBCA , TBCB , and TBCC are each soluble when expressed on their own in Escherichia coli , while TBCD , TBCE , and Arl2 are insoluble on their own ( see ‘Materials and methods’ ) . Co-expression of these three proteins , however , results in a stable and homogenous TBCD-TBCE-Arl2 GTPase complex that we term TBC-DEG ( Figure 2A ) . When we coexpressed TBCA , TBCB , or TBCC with TBC-DEG , we observed no interaction with TBCA or TBCB , and an unstable , transient interaction with TBCC ( as determined by mass spectrometry; Table 1 ) . Size exclusion chromatography with multi-angle light scattering ( SEC-MALS ) demonstrates that TBC-DEG is a 205 kDa heterotrimer with one copy of each protein ( Figure 2A , C , D; Table 2 ) . Similar analysis shows that TBCC is a 32 kDa monomer , and porcine brain αβ-tubulin is a 100 kDa heterodimer , as shown previously ( Figure 2A , C , D; Table 2 ) . Monomeric TBCD , TBCE , or Arl2 subunits were not observed in vitro at any concentration and the TBC-DEG complex behaves as a single biochemical entity ( Figure 2A , C , D; Table 2 ) . At high ionic strength , TBC-DEG complexes precipitate , presumably due to dissociation and insolubility of individual subunits ( data not shown ) . A recent study suggests human TBCE is soluble and forms complexes with TBCB ( Serna et al . , 2015 ) . We do not observe TBCE-TBCB complexes using our TBC protein bacterial expression system . TBCE is insoluble without TBCD and Arl2 coexpression in bacteria . We believe that TBCE solubility maybe due to its expression in a eukaryotic system , where assembly and co-purification with TBCD and Arl2 is possible . 10 . 7554/eLife . 08811 . 004Figure 2 . Hierarchical assembly of TBCC with TBC-DEG and soluble αβ-tubulin dimer binding in the GDP·Pi state . ( A ) Size exclusion chromatography ( SEC ) intensity traces of TBC-DEG ( black ) , TBC-DEG:αβ-tubulin ( cyan ) , αβ-tubulin ( red ) , and TBCC ( purple ) . ( B ) SEC intensity traces of TBC-DEG+TBCC+αβ-tubulin-GDP·ALFx ( green ) , TBC-DEG+TBCC+αβ-tubulin-GTP ( gray ) , TBC-DEG+TBCC-GTP-ALFx ( black ) , TBCC+αβ-tubulin ( blue ) , and αβ-tubulin+TBCC ( blue ) . Additional states are described in Figure 2—figure supplement 1C , D . ( C ) Composition of SEC fractions shown in A and B using SDS-PAGE . Panel I , TBC-DEG; panel II , TBC-DEG:αβ-tubulin; panel III , TBC-DEG+TBCC-GDP·ALFx; panel IV , TBCC+αβ-tubulin; panel V , TBC-DEG+TBCC+αβ-tubulin-GTP; and panel VI , TBC-DEG+TBCC+αβ-tubulin-GDP·ALFx . TBC-DEG forms an active heterotrimeric complex , and TBCC forms a complex that co-migrates with TBC-DEG upon αβ-tubulin binding in the presence of GDP·ALFx ( panel IV ) . The protein standard is shown on the left and proteins are marked on the right . TBC-DEG complexes interact weakly with the resin media leading to wide elution SEC profiles in most conditions . ( D ) Molecular masses of TBC-DEG , αβ-tubulin , TBCC , and their complexes measured using size exclusion chromatography with multi-angle light scattering ( SEC-MALS ) . Solid lines represent SEC intensity traces on an intensity scale shown on the right y-axis , and dotted lines represent masses calculated on the mass scale shown on the left y-axis; TBC-DEG ( black ) , αβ-tubulin ( red ) , TBCC ( purple ) , TBC-DEG:αβ-tubulin ( cyan ) , and TBC-DEG:αβ-tubulin:TBCC-GDP·ALFx ( green ) . Masses and elution volumes are detailed in Table 2 . ( E ) Scheme for the hierarchical assembly of TBC-DEG with TBCC and αβ-tubulin and the role of nucleotide . TBCD , TBCE , and Arl2 form TBC-DEG complexes ( TBC-DEG ) and bind a single αβ-tubulin dimer ( αβ-tub ) to form TBC-DEG:αβ-tubulin ( TBC-DEG:αβ-tub ) , which recruits TBCC in the GTP-like state to form TBC-DEG:αβ-tubulin:TBCC ( TBC-DEG:αβ-tub:TBCC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 00410 . 7554/eLife . 08811 . 005Figure 2—figure supplement 1 . Tubulin cofactor-Arl2 co-expression and biochemical studies on TBC-DEG constructs . ( A ) Domain structures of tubulin cofactors and Arl2 . Top , TBCD ( gray ) composed of HEAT repeats . Second , TBCE , composed of Cap-Gly ( blue ) , LRR ( cyan ) , and ubiqutin-like ( light blue ) domains . Third , the Arl2-GTPase composed of ARF-like G protein fold ( red ) and outer unique termini ( orange ) . Fourth , TBCC composed of an N-terminal spectrin homology domain ( light green ) , and a β-sheet domain ( dark green ) . ( B ) Summary of co-expression experiments , TBC , and Arl2 proteins . The masses of each of the proteins are shown on the left and correspond to the order shown in A . The effects of deletion ( Δ ) or addition of GFP ( GFP ) or 6Xhis-tags ( his ) at the N-termini and C-termini of each of the TBC and Arl2 proteins on TBC-DEG complexes are described , where check marks describe no effect on TBC-DEG expression , while a cross mark describes loss of TBC-DEG expression . ( C ) Size exclusion chromatography ( SEC ) intensity traces of TBC-DEG:αβ-tubulin ( cyan ) , TBC-DEG:αβ-tubulin 1:2 molar ratio ( blue ) , αβ-tubulin ( red ) , and TBCC ( purple ) . ( D ) Composition of SEC fractions shown in C using SDS-PAGE . Panel I , αβ-tubulin; panel II , TBCC; panel III , TBC-DEG+αβ-tubulin 2:1 molar ratio; and panel IV , TBC-DEG+αβ-tubulin+TBCC+GTPγS . The protein standard is shown on the left and proteins are marked on the right . ( E ) Size exclusion chromatography ( SEC ) intensity traces of TBC-DE ( N-GFP ) G ( black ) . ( F ) Composition of SEC fractions shown in E using SDS-PAGE for TBC-DE ( N-GFP ) G . The protein standard is shown on the left and protein positions are marked on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 00510 . 7554/eLife . 08811 . 006Table 1 . Identification of tubulin cofactor subunits* using nano-LC-MS/MSDOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 006Protein nameMolecular masspIPeptide coverageYeast TBCD ( cin1p ) 116 , 647 . 8 Da8 . 5382 . 1%Yeast TBCE ( pac2p ) 59 , 257 . 6 Da8 . 7779 . 3%Yeast Arl2 ( cin4p ) 22 , 066 . 6 Da5 . 7085 . 9%Yeast TBCC ( cin2p ) 34 , 045 . 4 Da7 . 059 . 0%Identified from His-TBCD TBCE , TBCC , TBCB , TBCA , and Arl2 co-expression . 10 . 7554/eLife . 08811 . 007Table 2 . Size exclusion chromatography ( SEC ) and SEC with multi-angle light scattering ( SEC-MALS ) parameters for tubulin cofactor and αβ-tubulin complexesDOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 007Protein complexElution volumePredictedApparentSEC-MALStokes RTBC-DEG11 . 5 ml198 kDa213 kDa215 ± 10 kDa∼50 ÅTBC-DEG:αβ-tub10 . 7 ml308 kDa322 kDa310 ± 10 kDa∼56 ÅTBC-DEG:C:αβ-tub-GDP . ALFx10 . 2 ml342 kDa376 kDa335 ± 10 kDa∼59 ÅTBC-DEG-Q73L10 . 6 ml198 kDa218 kDaN/A∼50 ÅTBC-DEG-Q73L:αβ-tub10 . 7 ml308 kDa322 kDaN/A∼56 ÅTBC-DEGQ73L:C:αβ-tub10 . 2 ml342 kDa376 kDa340 ± 10 kDa∼59 Åαβ-tubulin dimer12 . 9 ml100 kDa103 kDa100 ± 5 kDa∼39 ÅTBCC15 . 0 ml34 kDa35 kDa30 ± 5 kDa∼27 Å To understand the role of conserved domains within the TBC-DEG complex , we systematically deleted predicted domains in each subunit ( see ‘Materials and methods’; Figure 2—figure supplement 1A , B ) . Deletion of either the N- or C-terminal domains of both TBCD and Arl2 , or the C-terminal ubiquitin-like domain of TBCE , leads to insoluble TBC-DEG that cannot be purified from E . coli . In contrast , deleting the N-terminal Cap-Gly domain of TBCE , predicted to bind the C-terminal tail of α-tubulin , did not affect assembly of soluble TBC-DEG complexes . We next determined the effect of inserting small ( 6xHis ) or large ( GFP , green fluorescent protein ) tags on TBC-DEG assembly . Consistent with the deletion analysis , large or small tags were not tolerated on either end of Arl2 or at the C-termini of TBCD or TBCE ( Figure 2—figure supplement 1A , B ) . Both 6xHis and GFP tags were tolerated at the N-termini of TBCD and TBCE ( Figure 2—figure supplement 1A , B ) . These data suggest that the conserved domains of TBCD , TBCE , and Arl2 are required to assemble a TBC-DEG complex , in which the N-termini of TBCD and TBCE are exposed , while both termini of Arl2 , and the C-termini of TBCD and TBCE , are buried and do not tolerate insertions . We next sought to test the idea that TBC-DEG serves as a platform for soluble αβ-tubulin dimer assembly , and to examine the role of the Arl2 GTPase in this assembly . TBC-DEG binds soluble αβ-tubulin dimers with high affinity , forming stable complexes with a measured mass of 308 kDa ( Figure 2A , C , D; Table 2 ) , indicating that a single TBC-DEG ( 200 kDa ) binds a single αβ-tubulin dimer ( 110 kDa ) . The TBC-DEG:αβ-tubulin complex is likely to be the ∼300 kDa tubulin biogenesis intermediate identified two decades ago by Paciucci ( 1994 ) . We next determined the conditions for TBCC binding to TBC-DEG and αβ-tubulin . TBCC does not bind either αβ-tubulin or TBC-DEG in isolation , but strongly interacts with the TBC-DEG:αβ-tubulin complex ( Figure 2A–C ) . This interaction strongly depends on the GTP nucleotide present during complex assembly . We observed TBCC binding to the TBC-DEG:αβ-tubulin complex when incubated with the non-hydrolysable GTP analog GTPγS or the transition state analog GDP·ALFx , but no binding in the presence of GTP or GDP ( Figure 2B–C , Figure 2—figure supplement 1 B–E ) . We measured a 340 kDa mass for the TBC-DEG:αβ-tubulin:TBCC ternary complex , indicating that the TBC-DEG:αβ-tubulin complex ( 310 kDa ) associates with a single molecule of TBCC ( 34 kDa ) ( Table 2; Figure 2B–D ) . To determine if the Arl2 GTPase is responsible for increasing the TBCC binding affinity to TBC-DEG , we next generated a Gln73Leu ( Q73L ) mutation in Arl2 , which inhibits GTP hydrolysis and results in a ‘GTP-locked’ state ( Veltel et al . , 2008 ) . Bacterial expression of recombinant Arl2-Q73L shows that it assembles with TBCD and TBCE into a TBC-DEG-Q73L complex . In contrast to TBC-DEG , TBC-DEG-Q73L interacts with TBCC in the absence of αβ-tubulin ( Figure 3A , B , panel I ) , and assembles with αβ-tubulin and TBCC to form a stable and fully saturated ternary complex ( Figure 3A , B , panel II; mass of 335 kDa by SEC-MALS; Table 2 ) in the presence of GTP . Thus , our biochemical reconstitutions indicate that TBCC binding to TBC-DEG is promoted by both αβ-tubulin binding to TBC-DEG and the GTP-bound state of Arl2 ( Figure 2D ) . These findings support a proposed role for TBCC as a GAP for Arl2 , whose association with the TBC-DEG chaperone is responsive to αβ-tubulin binding . 10 . 7554/eLife . 08811 . 008Figure 3 . TBCC activates dual GTP hydrolyses in Arl2 and αβ-tubulin on TBC-DEG: αβ-tubulin complexes . ( A ) Size exclusion chromatography ( SEC ) intensity traces of TBC-DEG-Arl2-Q73L ( TBC-DEG-Q73L ) assembly with TBCC and αβ-tubulin; TBC-DEG-Q73L+TBCC ( black ) , TBC-DEG-Q73L+αβ-tubulin+TBCC ( green ) , αβ-tubulin ( red ) , and TBCC ( purple ) . ( B ) Analysis of SEC fractions described in A by SDS-PAGE . Panel I , TBC-DEG-Q73L+TBCC+GTP; panel II , TBC-DEG-Q73L+TBCC+αβ-tubulin-GTP . ( C ) Scheme for GTP hydrolysis by TBC-DEG and the effect of αβ-tubulin binding and TBCC on the GTP hydrolysis pathway . ( D ) Steady-state GTP hydrolysis assays of different 1 μM TBC-DEG , αβ-tubulin , and TBCC assemblies . TBC-DEG ( red ) and TBC-DEG+αβ-tubulin ( orange ) hydrolyze GTP very slowly . TBCC+αβ-tubulin ( black ) hydrolyzes negligible amounts of GTP . TBC-DEG+αβ-tubulin+TBCC hydrolyzes GTP ( blue; 1 . 8 min−1 ) at a rate roughly twofold higher than TBC-DEG+TBCC ( green; 0 . 8 min−1 ) . Km and kcat values are reported in Table 3 . ( E ) The effect of αβ-tubulin binding on TBC-DEG GTP hydrolysis . Top panel , scheme for GTP hydrolysis by TBC-DEG and the effect of limiting or varying the αβ-tubulin concentration on GTP hydrolysis . Bottom panel , titrating αβ-tubulin concentrations ( 0–3 . 0 μM ) to 1 μM TBC-DEG and 1 μM TBCC . The curves are labeled with the concentration at the plateau point for each curve . ( F ) The effect of TBCC concentration on TBC-DEG GTP hydrolysis . Top panel , scheme for GTP hydrolysis by TBC-DEG and the effect of limiting or varying the TBCC concentration on GTP hydrolysis . Bottom panel , titrating TBCC concentration ( 0 . 12–1 . 0 μM ) to 1 μM TBC-DEG and 1 μM αβ-tubulin . The curves are labeled with the concentration at the plateau point for each curve . ( G ) The effect of Arl2-Q73L on TBC-DEG GTP hydrolysis . Top panel , scheme for GTP hydrolysis by TBC-DEG-Q73L and the effect of αβ-tubulin binding and TBCC on the GTP hydrolysis reaction . Bottom panel , steady-state GTP hydrolysis assays of 1 μM TBC-DEG+αβ-tubulin+TBCC ( blue ) compared to TBC-DEG-Q73L+αβ-tubulin+TBCC ( purple ) . Km and kcat values are reported in Table 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 00810 . 7554/eLife . 08811 . 009Figure 3—figure supplement 1 . Summary of the catalytic GTP hydrolysis rates for different reconstitutions of TBC-DEG with TBCC and αβ-tubulin and the effects of Arl2 and TBCC mutations . ( A ) The TBC-DEG , αβ-tubulin , and TBCC assembly reaction and activation of GTP hydrolysis . ( B ) TBCC activates TBC-DEG GTP hydrolysis in a soluble tubulin dependent manner . ( C ) The binding of αβ-tubulin to TBC-DEG leads to low but robust GTP hydrolysis . ( D ) A GTP locked Arl2 mutant in TBC-DEG-Q73L has very poor GTP hydrolysis . ( E ) The TBCC arginine finger R186A mutant has robust GTP hydrolysis . ( F ) The TBCC C-terminal β-helix domain activates TBC-DEG GTP hydrolysis with loss of the αβ-tubulin independent phase . ( G ) The TBCC Δ233-245 loop deleted mutant displays low but robust GTP hydrolysis . ( H ) The effect of the αβ-tubulin to TBC-DEG molar ratio on GTP hydrolysis ( kcat ) . ( I ) The effect of the TBCC to TBC-DEG molar ratio on GTP hydrolysis ( kcat ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 009 Next , we studied the GTP hydrolysis activity of TBC-DEG and the effect of αβ-tubulin and TBCC binding , using a free-phosphate detection assay ( Figure 3C; Table 3 ) . In the absence of other factors , TBC-DEG hydrolyzes GTP extremely slowly ( Figure 3C; Table 3 ) . Addition of equimolar αβ-tubulin , which alone does not show detectable GTP hydrolysis in this assay , stimulates a modest level of GTP hydrolysis activity in TBC-DEG ( kcat = ∼0 . 40 min−1; Figure 3D , Figure 3—figure supplement 1C; Table 3 ) . In contrast , the addition of equimolar TBCC to TBC-DEG activates a substantially higher rate of GTP hydrolysis ( kcat = ∼0 . 77 min−1 ) , consistent with its proposed function as a GAP for Arl2 ( Tian et al . , 1997; Bhamidipati et al . , 2000; Mori and Toda , 2013; Newman et al . , 2014 ) . When equimolar amounts of both αβ-tubulin and TBCC are added to TBC-DEG , GTP hydrolysis was stimulated twofold more than in the presence of TBCC alone ( kcat = 1 . 85 min−1; Figure 3—figure supplement 1A ) . This increase may be due either to an increase in the affinity of TBCC for Arl2 in the presence of αβ-tubulin , or activation of GTP hydrolysis within the bound αβ-tubulin itself . To distinguish these models , we next assayed GTP hydrolysis of TBC-DEG-Q73L in the presence of equimolar αβ-tubulin and TBCC . This complex shows low GTP hydrolysis activity ( kcat = 0 . 5 min−1 ) with a high Km ( 387 μM ) , supporting the idea that within the ternary complex , αβ-tubulin contributes only a small fraction of the total GTP hydrolysis activity . Taken together , our data provide a new context to explain extensive prior genetic and biochemical data on the role of Arl2 in regulating tubulin cofactor activity ( Tian et al . , 1997; Bhamidipati et al . , 2000; Mori and Toda , 2013; Newman et al . , 2014 ) . Our studies reveal that TBCC is a novel αβ-tubulin-responsive GAP that activates Arl2 in the context of the TBC-DEG chaperone . 10 . 7554/eLife . 08811 . 010Table 3 . Steady-state GTP hydrolysis parameters for TBC-DEG , TBCC , and αβ-tubulinDOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 010GTP hydrolysis reactionsKm ( GTP ) kcat ( min−1/μM ) 1 μM TBC-DEG400 ± 30 μM0 . 06 ± 0 . 011 μM TBCC:1 μM αβ-tub20 ± 15 μM0 . 00 ± 0 . 011 μM TBC-DEG:1 μM αβ-tub69 ± 12 μM0 . 40 ± 0 . 011 μM TBC-DEG:1 μM TBCC94 ± 10 μM0 . 77 ± 0 . 041 μM TBC-DEG:1 μM TBCC:1 μM αβ-tub99 ± 10 μM1 . 88 ± 0 . 031 μM TBC-DEG-Q73L:1 μM TBCC:1 μM αβ-tub371 ± 20 μM0 . 50 ± 0 . 051 μM TBC-DEG:1 μM TBCC:0 . 25 μM αβ-tub87 ± 5 μM1 . 10 ± 0 . 061 μM TBC-DEG:1 μM TBCC:0 . 5 μM αβ-tub112 ± 4 μM1 . 41 ± 0 . 051 μM TBC-DEG:1 μM TBCC:1 . 5 μM αβ-tub98 ± 8 μM2 . 38 ± 0 . 081 μM TBC-DEG:1 μM TBCC:3 . 0 μM αβ-tub88 ± 9 μM2 . 77 ± 0 . 111 μM TBC-DEG:0 . 12 μM TBCC:1 μM αβ-tub149 ± 7 μM1 . 53 ± 0 . 041 μM TBC-DEG:0 . 25 μM TBCC:1 μM αβ-tub129 ± 3 μM1 . 69 ± 0 . 021 μM TBC-DEG:0 . 50 μM TBCC:1 μM αβ-tub108 ± 5 μM1 . 74 ± 0 . 041 μM TBC-DEG:1 μM TBCC-R186A:1 μM αβ-tub40 ± 10 μM0 . 54 ± 0 . 041 μM TBC-DEG:1 μM TBCC-Δ233-245:1 μM αβ-tub35 ± 8 μM0 . 34 ± 0 . 031 μM TBC-DEG:1 μM TBCC-Cterm:1 μM αβ-tub56 ± 7 μM1 . 13 ± 0 . 07 Next , we explored how the TBCC and αβ-tubulin concentration influences GTP hydrolysis by TBC-DEG . We measured the steady-state GTP hydrolysis of TBC-DEG titrated with a range of TBCC and αβ-tubulin concentrations . Decreasing the molar ratio of TBCC to TBC-DEG , while maintaining a stoichiometric amount of αβ-tubulin , increases the apparent Km for GTP hydrolysis , further supporting the idea that TBCC is a true GAP ( Figure 3F , Figure 3—figure supplement 1H; Table 3; Veltel et al . , 2008 ) . In contrast , increasing the ratio of αβ-tubulin to TBC-DEG ( 0–3 μM ) , while maintaining a stoichiometric amount of TBCC , stimulates a step-wise increase in the maximal rate of GTP hydrolysis ( Figure 3E; Table 3; Figure 3—figure supplement 3I ) . At 3 μM αβ-tubulin and at a 1:3:1 ratio of TBCC:αβ-tubulin:TBC-DEG , we observe the highest GTP hydrolysis rate ( Table 3: kcat = 3 . 0 min−1 ) , suggesting that each TBC-DEG chaperone can undergo multiple rounds of GTP hydrolysis , upon binding and releasing multiple αβ-tubulin dimers during each experiment ( Figure 3E ) . We were unable to test αβ-tubulin concentrations higher than 3 μM in this assay , as at 6 μM αβ-tubulin or higher we expect αβ-tubulin polymerization into MTs to significantly contribute to the overall GTP hydrolysis observed . Within the tested αβ-tubulin concentration range ( 0–3 μM ) , the TBC-DEG GTP hydrolysis rate ( kcat ) is proportional to the αβ-tubulin concentration , starting at ∼0 . 8 min−1 in the absence of αβ-tubulin ( Figure 3E , Figure 3—figure supplement 3H ) and climbing and then plateauing at 3 . 0 min−1 at 3 μM αβ-tubulin . Thus , TBCC is an αβ-tubulin dependent GAP that activates TBC-DEG GTP hydrolysis in a cyclic manner where the degree of GAP activity depends on the soluble αβ-tubulin concentration . To determine the 3D structure of the TBC-DEG-Q73L chaperone , we used electron microscopy ( EM ) and single-particle image analysis . While cryo-EM imaging of TBC-DEG was not possible due to solubility defects and aggregation in vitreous ice , we were able to collect negative-stain EM data and generate a robust medium-resolution 3D reconstruction . We collected a total of 20 , 000 particle images from 160 initial wide-field images , which showed a globular ∼100 × 100 × 100 Å particle with significant internal structure ( Figure 4A , Figure 4—figure supplement 1A ) . Particle orientations were well distributed , and reference-free classification showed homogeneous class averages representing a large range of views ( Figure 4—figure supplement 1B ) . We generated a starting model using a common-lines approach based on prominent classes and then used projection matching and angular reconstitution , and refined a 3D map to 24 Å resolution ( see ‘Materials and methods’; Figure 4A , Figure 4—figure supplement 1C , Table 4 ) . 2D projections generated from the refined 3D map matched well to the reference-free class averages ( Figure 4—figure supplement 1D ) . We obtained matching reconstructions using a variety of low-resolution starting models , which converged during angular refinement to the 3D reconstructions described below . 10 . 7554/eLife . 08811 . 011Figure 4 . TBC-DEG complexes are compact cage-like chaperone assemblies with hollow cores . ( A ) Left panel , an expanded negative-stain image of TBC-DEG Q73L showing the cage-like assemblies . Middle panel , higher magnification view of the TBC-DEG-Q73L . Right panel , reference-free class averages ( from Figure 4—figure supplement 1C ) of TBC-DEG Q73L showing the variety of views . ( B ) A refined 24 Å TBC-DEG-Q73L 3D map shown in three rotated views . The floor , bow , trunk , pillar , and thumb regions are marked in each view . ( C ) Segmented 24 Å TBC-DEG map with all unique segmented domains based on tagging assignment of the TBCE N-termini ( Figure 4—figure supplement 2B ) . The bow region ( blue ) includes two globular ends: the ubiquitin domain ( cyan ) and the Cap-Gly domain ( deep blue ) . Three interfaces stabilize the TBC-DEG cage: the bow pillar , the pillar floor , and the bow floor via the trunk . Video 1 shows the A and B views . ( D ) A TBC-DEG subunit domain map shown to length scale . TBCD ( pink , top panel ) is predicted to consist of HEAT repeats . TBCE ( middle panel ) consists of an N-terminal Cap-Gly domain ( dark blue ) , a leucine rich repeat ( LRR ) domain ( blue ) , and a C-terminal ubiquitin-like domain ( cyan ) . Arl2 ( bottom panel ) consists of a GTPase fold ( orange ) . Colors correspond to subunits shown in D–I . ( E ) Pseudo-atomic TBC-DEG cage model showing the TBCD , TBCE , and Arl2 domain organization in assembling the cage structure . Each 3D map region is shown in a glossy color , and x-ray structures for orthologs fitted are shown as ribbons in the same color . The floor and thumb segments ( pink ) were fitted by the Cse1 crystal structure . The bow segment ( blue ) was fitted by the TLR4 LRR domain x-ray structure . The pillar segment ( orange ) was fitted by the x-ray structure of Arl2 ( orange ) . The positions N-GFP-TBCE ( dark green ) and N-GFP-TBCD ( light green ) are shown . The trunk region ( purple ) was not fit with any atomic model . ( F ) A 90° vertically rotated view of that shown in D . ( G ) A 90° horizontally rotated view of that shown in D . ( H ) A central slice view of a 90° counterclockwise horizontally rotated view of that shown in D . Video 1 shows the C–F views . ( I ) Cartoon view of TBC-DEG domain organization comparable to the view shown in F . ( J ) Cartoon view of TBC-DEG domain organization comparable to the view shown in G . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 01110 . 7554/eLife . 08811 . 012Figure 4—figure supplement 1 . Electron microscopy and 3D reconstruction of the TBC-DEG-Q73L cage-like chaperones . ( A ) Left panel , expanded view of a raw negative-stain EM image showing TBC-DEG Q73L globular particles with hollow cores . Right panel , higher magnification view of TBC-DEG particles showing the variety of orientations . ( B ) Multivariant statistical analysis ( MSA ) reference-free class averages of TBC-DEG-Q73L show a variety of commonly observed particle views . Few class averages appear off-center due the large mask size , which has no effect on classification or further analyses . ( C ) 3D reconstruction for the TBC-DEG-Q73L complex is initiated with a 50 Å resolution starting model ( left ) , then iterative projecting matching ( middle ) , followed by refinement ( right ) . ( D ) Comparison of the reference-free class averages to 2D projections of the refined structure . Each panel shows a comparison between two images through projection matching: reference-free class averages ( MSA , on top ) and 2D projection from a 3D map of TBC-DEG Q73L:αβ-tubulin ( 2D Prj match ) . ( E ) Plot for phi and theta angular distribution for each individual TBC-DEG Q73L image used in the final reconstruction; the plot is using Angplot_dp . ( F ) Fourier shell correlation analysis of the TBC-DEG-Q73L reconstruction . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 01210 . 7554/eLife . 08811 . 013Figure 4—figure supplement 2 . Mapping the TBCE N-terminal Cap-Gly domain using TBC-DE ( N-GFP ) fusion and 3D reconstructions . ( A ) Multivariant statistical analysis based reference-free class averages for TBC-DE ( NGFP ) ( middle panels ) compared to TBC-DEG-Q73L ( left panels ) ; positions of GFPs are highlight in green in the TBC-DE ( N-GFP ) G class averages . Six comparable views are shown to describe the added density of GFP , marked by blue arrows . Some views , such as those on the bottom right , show GFP density on the TBC-DEG cage mass . ( B ) The TBC-DE ( N-GFP ) G raw map ( green ) in three different orientations rotated by 90° compared to the native TBC-DEG-Q73L map ( gray ) showing the N-GFP position . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 01310 . 7554/eLife . 08811 . 014Figure 4—figure supplement 3 . Cumulative docking of atomic models for TBCD , TBCE paralogs , and Arl2 without segmentation using low-resolution model filtering . ( A ) The TBCD paralog ( Cse1p ) was docked into the TBC-DEG EM map . Panels ( left to right ) : two orthogonal views of the TBCD paralog ( Cse1 ) atomic model , 24 Å resolution filtered model , 24 Å TBC-DEG Q73L map , and TBCD model positioned into the TBC-DEG Q73L map . ( B ) The TBCE LLR domain paralog ( TLR4 ) was docked into the TBC-DEG EM map . Panels ( left to right ) : two orthogonal views of the TBCE LLR domain paralog ( TLR4 ) atomic model , 24 Å resolution filtered model , 24 Å TBC-DEG Q73L map , and TBCE model positioned into the TBC-DEG Q73L map after TBCD paralog docking . Arrows point to the locations of globular densities that likely represent TBCE N and C-terminal domains . ( C ) Arl2 GTPase ( Arl2 ) was docked into the TBC-DEG EM map . Panels ( left to right ) : two orthogonal views of the Arl2 GTPase ( Arl2 ) atomic model , 24 Å resolution filtered model , 24 Å TBC-DEG Q73L map , and Arl2 model positioned into the TBC-DEG Q73L map after TBCD and TBCE paralog docking . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 01410 . 7554/eLife . 08811 . 032Table 4 . Electron microscopy Fourier shell correlation ( FSC ) resolution analysesDOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 032ComplexParticle imagesResolution ( Å ) TBC-DEG-Q73L16 , 00024 . 0TBC-DEG-Q73L-αβ-tub19 , 00024 . 0TBC-DEG-Q73L-αβ-tub:TBCC18 , 00024 . 0TBC-DE ( N-GFP ) G15 , 00024 . 0Resolution cross-correlation criterion cut-off set at 0 . 5 . 10 . 7554/eLife . 08811 . 015Video 1 . The video shows a 360° rotation of the raw TBC-DEG-Q37L map ( Figure 4A ) and a 360° rotation of the segmented TBC-DEG-Q73L map ( accompanies Figure 4 ) . The color scheme is described in Figure 3B . This is followed by a 360° rotation of the raw TBC-DEG-Q37L segmented and coordinate fitted map ( Figure 4D–G ) , followed by a clipping view slicing across a segmented and fitted TBC-DEG-Q73L map ( Figure 4G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 015 The 3D reconstruction of TBC-DEG-Q73L shows a compact cage-like structure with a hollow core , and overall dimensions of 120 × 100 × 90 Å ( Figure 4B ) . TBC-DEG consists of a circular ‘floor’ with a large vertical ‘thumb’ extension ( Figure 4B , pink ) , facing a ‘bow’ density with two globular ends ( Figure 4B , blue ) . The bow ( Figure 4B , blue ) is attached at its center to the floor ( Figure 4B , pink ) via a ‘trunk’ density and binds a ‘pillar’ density ( Figure 4B , orange ) via one of its globular ends ( Figure 4B , cyan ) . Three interfaces form the TBC-DEG cage structure: ( 1 ) bow to floor interface via the trunk; ( 2 ) bow to pillar interface ( cyan ) ; and ( 3 ) floor to pillar interface . To determine the locations of individual TBC-DEG subunits within this structure , we imaged a complex with N-terminally GFP-fused TBCE ( TBC-DE ( N-GFP ) G ) using negative-stain EM ( Figure 2—figure supplement 1G ) . TBC-DE ( N-GFP ) G class averages show the addition of ordered density when compared to equivalent class averages of TBC-DEG ( Figure 4—figure supplement 2A ) . We determined a 24 Å 3D structure for TBC-DE ( N-GFP ) G using a 50 Å resolution filtered TBC-DEG as a starting model ( Figure 4—figure supplement 2B ) . We located the TBCE N-terminus near one of the two globular domains at the end of the bow , suggesting that this density is the N-terminal Cap-Gly domain of TBCE ( Figure 4—figure supplement 2B ) . We built a pseudo-atomic model for TBC-DEG using the experimentally verified position for the TBCE Cap-Gly domain , followed by semi-automated docking of Arl2 and structural models for conserved domains of TBCD and TBCE , into the TBC-DEG map ( Figure 4E–G ) . We used the structure of Cse1p as a model for TBCD , as they share over 47% sequence identity and contain a similar number of HEAT repeats ( Cook et al . , 2005b ) . For the TBCE LRR domain , we used the structure of TLR4 , which shares 40% sequence similarity with TBCE and contains 14 LLR repeats ( Park et al . , 2009b ) . We used the Cap-Gly ( Fleming et al . , 2013c ) and ubiquitin-like ( Fleming et al . , 2013b ) domains of TBCB as models for the equivalent domains of TBCE , and the human Arl2 structure as a model for yeast Arl2 ( Hanzal-Bayer et al . , 2002b ) . We used two approaches to dock these subunit models into the TBC-DEG map , leading to similar results ( see ‘Materials and methods’ ) . First , we segmented the TBC-DEG-Q73L map , and docked the atomic models into segments using the ‘fit to segment’ feature in UCSF Chimera . Second , we used low-resolution filtered subunit models and successively fit them into an unsegmented map , using the ‘fit in map’ feature of UCSF Chimera , starting with the largest subunit ( TBCD ) and ending with the smallest subunit ( Arl2 ) ( Figure 4—figure supplement 3A–C ) . The Cse1 model representing TBCD fit well to the floor and thumb segments ( UCSF Chimera correlation coefficient 0 . 71 ) and its distinct circular-ring and rod shape allowed an unambiguous fit directly into the TBC-DEG map without segmentation . TLR4 , representing the central LRR domain of TBCE , fit well to the bow segment of the TBC-DEG map ( UCSF Chimera correlation coefficient 0 . 80 ) after placement of TBCD ( Figure 4—figure supplement 3B ) . The TBCB Cap-Gly structure fit well into the globular end of the bow closest to the TBCE N-GFP density from TBC-DE ( N-GFP ) G ( Figure 4—figure supplement 2 ) , while the TBCB ubiquitin-like domain fit well into the other globular end of the bow ( UCSF Chimera correlation coefficient 0 . 82 ) . The overall bow-shaped organization of TBCE LLR with two globular domains at its ends is similar to the TBCE organization in a recent study ( Serna et al . , 2015 ) . Finally , Arl2 fit well to the pillar density located in between the TBCE and TBCD subunits ( Figure 4—figure supplement 3C ) . The only region of the TBC-DEG map that was not accounted for in our modeling is the trunk , which likely includes the four to five HEAT repeats in TBCD that are missing from the Cse1 model according to sequence alignment comparison . Thus , our low-resolution structure and model for TBC-DEG support our domain deletion/insertion analysis of TBCD , TBCE , and Arl2 ( Figure 2—figure supplement 1B ) , indicating a non-linear assembly of these subunits into a cage-like complex , and demonstrating critical roles for N and C-terminal domains of TBCD , Arl2 , and the C-terminus of TBCE in TBC-DEG assembly ( Figure 4H , I ) . To examine the structural basis for TBC-DEG association with αβ-tubulin , we determined a 3D reconstruction for the TBC-DEG-Q73L:αβ-tubulin complex using the approach described above ( see ‘Materials and methods’ ) . Raw images and reference-free classification show TBC-DEG-Q73L:αβ-tubulin particles adopt a variety of orientations , with a moderate degree of preferred views ( Figure 5—figure supplement 1 ) . We generated a de novo starting model using common class averages ( Figure 5—figure supplement 1B , left panel ) . Using a low-resolution filtered starting model , we then carried out projection matching and angular reconstitution cycles ( see ‘Materials and methods’; Figure 5—figure supplement 1C , Table 4 ) and then refined a 24 Å TBC-DEG:αβ-tubulin map . The TBC-DEG:αβ-tubulin model projections match the class averages ( Figure 5—figure supplement 1D ) . When compared to TBC-DEG-Q73L , the TBC-DEG-Q73L:αβ-tubulin map shows an additional dual-lobed mass on top of the cage , with dimensions of approximately 80 × 40 × 40 Å , which we assigned as αβ-tubulin ( Figure 5A , Figure 5—figure supplement 1D ) . Although the orientation of the tubulin dimer cannot be determined solely from the electron density map , the TBCE Cap-Gly domain is known to bind the disordered α-tubulin C-terminus ( Akhmanova and Steinmetz , 2008 ) . This additional information allows us to unambiguously assign the orientation of the bound αβ-tubulin , and fitting of the atomic coordinates of the αβ-tubulin dimer into the dual-lobed density ( UCSF Chimera correlation coefficient 0 . 85 ) results in a pseudo-atomic model of the full TBC-DEG-Q73L:αβ-tubulin complex ( Figure 5D ) . 10 . 7554/eLife . 08811 . 016Figure 5 . TBC-DEG platforms engage the αβ-tubulin dimer asymmetrically , placing it in contact with Arl2 GTPase above the hollow core . ( A ) A refined TBC-DEG-Q73L:αβ-tubulin 3D map shown in three rotated views . The map shows the presence of dual regions at the top of the TBC-DEG cage density . ( B ) A segmented TBC-DEG-Q73L:αβ-tubulin map shown in three rotated views . A dual lobed density ( red ) assigned to αβ-tubulin is bound by domains at the top side of the TBC-DEG-Q73L cage . Video 3 shows the A and B views . ( C ) A TBC-DEG-αβ-tubulin linear domain map shown to length scale . TBCD ( pink , top panel ) is composed of HEAT repeats . TBCE ( second panel ) includes a Cap-Gly domain ( dark blue ) , a leucine rich repeat ( LRR ) domain ( blue ) , and a ubiquitin-like domain ( cyan ) . Arl2 ( third panel ) consists of a G-domain or GTPase fold ( orange ) . αβ-tubulin ( red ) is shown in the bottom panel . Colors correspond to subunits shown in D–I . ( D ) A Pseudo-atomic model of the TBC-DEG-Q73L:αβ-tubulin complex showing the interfaces of TBCD , TBCE , and Arl2 engaging the intact αβ-tubulin asymmetrically . The model is built by fitting the densities of TBC-DEG segments as described in Figure 4 in addition to αβ-tubulin structure into the dual lobed density . ( E ) A 90° vertically rotated view of that shown in D . ( F ) A 90° horizontally rotated view of that shown in D . Video 2 shows the C–F views . ( G ) A central slice view of 90° counterclockwise horizontally rotated view of that shown in D . ( H ) Cartoon view of TBC-DEG domain organization comparable to the view shown in F ( I ) Cartoon view of TBC-DEG domain organization comparable to the view shown in G . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 01610 . 7554/eLife . 08811 . 017Figure 5—figure supplement 1 . Electron microscopy and 3D reconstruction of the TBC-DEG-Q73L:αβ-tubulin complex . ( A ) Left panel , expanded view of a raw negative-stain EM image showing the TBC-DEG Q73L:αβ-tubulin complex . Right panel , higher magnification view of TBC-DEG particles showing the variety of orientations . ( B ) Multivariant statistical analysis ( MSA ) reference-free class averages of TBC-DEG-Q73L:αβ-tubulin show a variety of commonly observed particle views . ( C ) 3D reconstruction for TBC-DEG:αβ-tubulin is initiated with a 50 Å resolution starting model ( left ) , then iterative projecting matching ( middle ) , followed by refinement ( right ) . ( D ) Overlay of the TBC-DEG-Q73L:αβ-tubulin map ( red ) over the TBC-Q73L map ( transparent blue ) . ( E ) Comparison of the reference-free class averages to 2D projections of the refined reconstruction . Each panel shows a comparison between two images through projection matching: reference-free class averages ( MSA , top ) and 2D projection from a 3D map of TBC-DEG Q73L:αβ-tubulin ( 2D Prj match ) . ( F ) Phi and theta angular distribution plot for each individual TBC-DEG Q73L:αβ-tubulin image used in the final reconstruction; the plot is using Angplot_dp . ( G ) Fourier shell correlation analysis of the TBC-DEG-Q73L:αβ-tubulin reconstruction . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 01710 . 7554/eLife . 08811 . 018Video 2 . The video shows a 360° rotation of the raw TBC-DEG-Q37L:αβ-tubulin map ( Figure 5A ) and a 360° rotation of the segmented TBC-DEG-Q73L:αβ-tubulin map ( accompanies Figure 5 ) . This is followed by 360° rotation of the overlaid TBC-DEG-Q73L ( blue ) and TBC-DEG-Q73L:αβ-tubulin maps ( red ) as shown in Figure 5—figure supplement 1D . This is followed by 360° rotation of the raw TBC-DEG-Q37L:αβ-tubulin segmented and coordinate fitted map ( Figure 5D–G ) , followed by a clipping view slicing across the segmented and fitted TBC-DEG-Q73L:αβ-tubulin map . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 018 The pseudo-atomic model of TBC-DEG-Q73L:αβ-tubulin shows the αβ-tubulin dimer embraced by the TBCD thumb and the TBCE LRR domain on its two lateral MT-forming interfaces . In addition , the TBCE Cap-Gly domain is positioned above α-tubulin , and Arl2 contacts β-tubulin from below . The TBCE Cap-Gly domain moves ∼20 Å from its position in TBC-DEG-Q73L , shifting closer to the TBCD thumb , while interfacing with the density we assign as α-tubulin . Upon tubulin binding , the TBCD thumb , TBCE LRR , and Arl2 GTPase domains each move ∼10 Å closer to the center of the TBC-DEG cage ( Figure 5—figure supplement 1C ) . TBC-DEG thus engages three sides of the αβ-tubulin dimer , intimately embracing the individual monomers above its hollow core . The domain connecting the TBCE bow to the Cap-Gly domain becomes disordered upon binding , suggesting that TBCE undergoes a conformational change ( Figure 5D–G , dark blue ) . The TBCE ubiquitin domain could not be assigned in this map ( Figure 5D–G ) . Importantly , our TBC-DEG:αβ-tubulin model shows that both longitudinal interfaces used in MT protofilament assembly are accessible while tubulin is engaged by TBC-DEG ( Figure 5D–G ) . Thus , our structural analysis suggests that the TBC-DEG chaperone organization is likely critical for recognizing α- and β-tubulin monomers during the biogenesis or degradation of αβ-tubulin dimer ( Figure 5H , I ) . Sequence alignments suggest that TBCC is a two-domain protein ( Figure 6—figure supplement 1E ) , with an N-terminal spectrin-like domain ( residues 1–99 of 267; Garcia-Mayoral et al . , 2011 ) , and a C-terminal domain predicted to consist of β-sheets ( residues 100–267 ) that is likely to be responsible for TBCC GAP activity ( Figure 6—figure supplement 1F; Kuhnel et al . , 2006; Mori and Toda , 2013 ) . To better understand TBCC's interactions with TBC-DEG , we sought to determine TBCC's crystal structure . We crystallized full-length S . cerevisiae TBCC and determined a 2 . 0 Å resolution structure encompassing residues 100–267 ( Figure 6—figure supplement 1A; see ‘Materials and methods’; Table 5 ) . Electron density for the TBCC N-terminal domain was absent , indicating it is either disordered or proteolyzed during crystallization . The TBCC C-terminal domain adopts a β-helix fold composed of 13 β-strands arranged in a helical staircase in the shape of a narrow triangular wedge ( Figure 6A–C ) . TBCC shows structural homology to retinitis pigmentosa-2 ( RP-2 ) protein ( RMSD 1 . 7 Å; Figure 6—figure supplement 1C ) , a well-studied GAP for the Arl2 paralog Arl3 ( Kuhnel et al . , 2006 ) . In RP2 , the β-helix domain binds Arl3 and inserts an ‘arginine finger’ into the Arl3 active site to stimulate GTP hydrolysis ( Veltel et al . , 2008 ) . TBCC possesses a conserved arginine ( Arg186 ) in the same position ( Figure 6C , Figure 6—figure supplement 1D ) , which in our structure projects outward from a highly conserved surface ( Figure 6C , D ) . In addition , TBCC includes two conserved features: ( 1 ) two additional β-strands with an intervening 15-residue loop ( residues 220–245 ) projecting above the β-helix; and ( 2 ) a short C-terminal α-helix that folds onto the TBCC β-helix domain ( Figure 5A ) . The TBCC loop is rich in conserved hydrophobic and acidic residues , including Phe233 , Phe237 , Glu240 , Glu241 , Glu243 , and Asp244 ( Figure 6B ) . We generated an Arl2:TBCC interface model by superimposing the TBCC and Arl2 structures onto the RP2:Arl3 co-crystal structure ( Figure 5E; Veltel et al . , 2008 ) . This model ( detailed in Figure 6—figure supplement 1D ) predicts that TBCC inserts Arg186 into the Arl2 active site to catalyze GTP hydrolysis , while Phe233 and Phe237 in the TBCC loop bind Arl2 hydrophobic residues , and the TBCC acidic residues 240 , 241 , 243 , and 244 project above the Arl2-TBCC interface . 10 . 7554/eLife . 08811 . 020Table 5 . Crystallographic statistics table for TBCC structure determinationDOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 020TBCC nativeTBCC Pt-peakTBCC Pt-inflectionTBCC Pt-remoteData collection Resolution range ( Å ) 34 . 90–2 . 00 ( 2 . 07–2 . 00 ) 41 . 80–2 . 18 ( 2 . 30–2 . 18 ) 35 . 01–2 . 18 ( 2 . 30–2 . 18 ) 31 . 94–2 . 18 ( 2 . 30–2 . 18 ) * Space groupP 43P 43P 43P 43 Wavelength ( Å ) 0 . 97951 . 07151 . 07190 . 9537 Unit cell ( Å ) : a , b , c69 . 79 , 69 . 79 , 78 . 1870 . 03 , 70 . 03 , 77 . 9570 . 03 , 70 . 03 , 77 . 9570 . 03 , 70 . 03 , 77 . 95 Total reflections193 , 620198 , 772198 , 980199 , 576 Unique reflections25 , 377 ( 2504 ) 19 , 716 ( 2871 ) 19 , 752 ( 2879 ) 19 , 748 ( 2877 ) Average mosaicity0 . 290 . 400 . 400 . 42 Anomalous multiplicity–5 . 1 ( 5 . 0 ) 5 . 1 ( 5 . 0 ) 5 . 2 ( 5 . 1 ) * Multiplicity7 . 6 ( 7 . 6 ) 10 . 1 ( 10 . 0 ) 10 . 1 ( 9 . 9 ) 10 . 1 ( 10 . 1 ) Anomalous completeness ( % ) –100 . 0 ( 100 . 0 ) 100 . 0 ( 100 . 0 ) 100 . 0 ( 100 . 0 ) * Completeness ( % ) 100 . 0 ( 100 . 0 ) 100 . 0 ( 100 . 0 ) 100 . 0 ( 100 . 0 ) 100 . 0 ( 100 . 0 ) <I/σ ( I ) >13 . 4 ( 2 . 7 ) 28 . 6 ( 9 . 0 ) 28 . 4 ( 8 . 8 ) 28 . 1 ( 8 . 5 ) Rmerge†0 . 082 ( 0 . 72 ) 0 . 046 ( 0 . 23 ) 0 . 046 ( 0 . 23 ) 0 . 046 ( 0 . 25 ) f′–17 . 4023 . 234 . 70 f′′–15 . 6710 . 2911 . 30Structure refinement Rwork0 . 20 ( 0 . 26 ) ––– Rfree0 . 24 ( 0 . 28 ) ––– Molecules per asymmetric unit2––– Number of atoms2744––– Protein residues329––– Number of water molecules93––– RMS bond lengths ( Å ) 0 . 007––– RMS bond angles ( ° ) 1 . 00––– Ramachandran favored ( % ) 96 . 0––– Ramachandran allowed ( % ) 0 . 0––– Ramachandran outliers ( % ) 0 . 0––– Clashscore4 . 6––– Mean B values ( Å2 ) Overall50 . 4––– Main-chain atoms46 . 2––– Side-chain atoms54 . 6––– Solvent49 . 4–––*Numbers represent the highest-resolution shell . †Rmerge = ΣhklΣi\|Ii ( hkl ) − Iav ( hkl ) \|/ΣhklΣiIi ( hkl ) . 10 . 7554/eLife . 08811 . 021Figure 6 . TBCC catalytic C-terminal domain x-ray structure suggests a TBCC-Arl2 binding interface to dissect the Arl2 contribution TBC-DEG GTP hydrolysis . ( A ) The 2 . 0 Å x-ray structure of the TBCC C-terminal β-helix domain ( 100–267 ) in two rotated views . β-sheets ( red ) form a narrow helical structure in which turns ( green ) lie at the ends and a large conserved and structured loop ( purple ) is presented on top of the structure . ( B ) A close-up view of the TBCC conserved residues in the structured loop showing the hydrophobic ( Phe233 , 237 ) and acidic ( Glu240 , 241 , 243 and Asp244 ) residues . ( C ) A close-up view of the Arl2 catalytic interface showing Glu184 , Arg186 , and Phe164 . ( D ) TBCC β-helix surface conservation showing the high conservation of the Arl2 catalytic site on one side of the TBCC β-helix domain . The left panel , front view , and 90° rotated view are shown . A color key gradient describing the conservation is shown below , with purple denoting highest and cyan denoting lowest conservation . ( E ) An interface model for TBCC-β-helix domain-Arl2 , based on a superimposition onto the RP2-Arl2 structures , which is described in detail in Figure 6—figure supplement 1 , panel D . The model shows Arg186 to be the arginine finger activating GTP hydrolysis in Arl2 . The Arl2 Gln73 interacts with a water molecule required for GTP nucleotide ( shown in blue ) during hydrolysis . The model shows the TBCC loop resides above Arl2 during the catalytic interface . ( F ) Steady-state GTP hydrolysis activity of TBC-DEG+αβ-tubulin+TBCC β-helix C-terminal domain 100–267 , TBCC-C ( DEG+TBCC-C+αβ-tub , brown ) compared to TBC-DEG+αβ+tubulin+wild type TBCC ( DEG+wtTBCC+αβ-tub , blue ) , showing the TBCC C-terminal domain is sufficient for GTP hydrolysis . TBC-DEG alone ( DEG , shown in red ) has very low basal GTP hydrolysis activity . Parameters are described in Table 3 . ( G ) Steady-state GTP hydrolysis activity of TBC-DEG+αβ-tubulin+TBCC Arg186Ala ( DEG+R186+αβ-tub , orange ) compares well to wild type TBCC+TBC-DEG ( DEG+wtTBCC; green ) , suggesting similar GTP hydrolysis parameters . TBC-DEG+αβ+tubulin+TBCC ( DEG+Δ233-245+αβ-tub , pink ) shows a similar defect in GTP hydrolysis . TBC-DEG+TBCC+αβ-tubulin ( DEG+wtTBC+αβ-tub; blue ) has a twofold higher GTP hydrolysis rate . Parameters are described in Table 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 02110 . 7554/eLife . 08811 . 022Figure 6—figure supplement 1 . TBCC C-terminal domain is a β-helix structure with dual interfaces . ( A ) Refined electron density map ( 1 . 0 σ ) for the TBCC extended loop showing the hydrophobic and acidic residues . ( B ) Two TBCC C-terminal domain structures in the asymmetric crystallographic unit of space group P43 . The structure described here is shown in red , while the second copy is shown in gray . ( C ) Structural overlay of the β-helix domain of TBCC ( red ) C-terminal domain with the RP2 domain ( blue ) . ( D ) Detailed TBCC-Arl2 interface model showing catalytic role of Arg186 ( Arg finger ) and selectivity role of Phe233 and Phe235 . Gln73 is required to catalyze the water residue that is necessary for catalysis . Acidic residues Glu241 , Asn241 , and Asp245 likely interact with other molecules in the TBC-DEG:αβ-tubulin complex . ( E ) Conformational changes in the Arl2 N-terminal domain upon nucleotide hydrolysis ( Valtel et al . , 2008 ) leading to αβ-tubulin catalysis activities observed in TBC-DEG:αβ-tubulin:TBCC ternary complexes . ( F ) Sequence alignments of TBCC β-helix domain showing the conserved residues . TBCC is predicted to consist of an N-terminal α-helical domain ( yellow ) and a C-terminal β-sheet domain ( green ) with its unique loop highlighted in red and its C-terminal helix ( blue ) . The alignment includes TBC and Arl2 orthologs from Saccharomyces cerevisae ( SC ) , Saccharomyces kluyveri ( SK ) , Schizosacchromyces pombe ( SP ) , Kluyveromyces lactis ( Kl ) , Drosophila melanogaster ( DM ) , Caenorhabdits elegans ( CE ) , and Human ( HS ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 022 To determine the significance of the unique structural features of TBCC , we measured the effect of their mutation on GTP hydrolysis activity in TBC-DEG . We first removed the TBCC N-terminal spectrin domain to generate TBCC-C ( residues 100–267 ) ; this mutant showed a 38% decrease in kcat when compared to wild type TBCC ( Table 3; Figure 6F ) , and lost the robust αβ-tubulin independent activation of GTP hydrolysis ( Figure 3—figure supplement 1F ) . This suggests that TBCC's N-terminal domain likely regulates the αβ-tubulin independent affinity of TBCC for TBC-DEG , while the β-helix domain is sufficient for αβ-tubulin dependent GAP activity . Mutation of the TBCC putative arginine finger , Arg186 ( R186A ) , decreased the GTP hydrolysis rate ( kcat ) of TBC-DEG:αβ-tubulin by slightly more than 70% ( kcat = 0 . 53 min−1 compared to 1 . 85 min−1; Figure 6G , Figure 3—figure supplement 1E ) . As removal of the arginine finger is expected to eliminate the GAP activity of TBCC ( Veltel et al . , 2008 ) , the substantial remaining GTP hydrolysis activity with TBCC R186A supports the idea that GTP hydrolysis observed in TBC-DEG:αβ-tubulin:TBCC complexes arises from a combination of Arl2 and αβ-tubulin ( Figure 6G , Figure 3—figure supplement 1G ) . A TBCC loop-deleted mutant ( Δ233-245; residues 233–245 replaced with a six-residue Ser-Gly linker ) reduces GTP hydrolysis activity by 82% ( kcat = 0 . 34 min−1 ) , to a low yet still robust level of activity , similar to the rate of GTP hydrolysis for TBC-DEG:αβ-tubulin without TBCC ( Figure 3—figure supplement 1B ) . The TBCC loop deletion is expected to interfere with Arl2 recognition ( Figure 6B , Figure 2—figure supplement 1G ) . Our structural and biochemical analyses of TBCC suggest its β-helix domain is a non-classical αβ-tubulin dependent GAP that activates Arl2 GTP hydrolysis , and may activate αβ-tubulin to hydrolyze GTP through an unknown mechanism . Residual GTP hydrolysis after specific inactivation of Arl2 GAP activity through the R186A and Δ233-245 TBCC mutants suggests that a secondary GTPase remains robustly active in the TBC-DEG:αβ-tubulin:TBCC complex . This GTPase is likely αβ-tubulin itself; however , which GTPase site ( N or E-site ) is becoming activated , and which mechanism is behind its activation , both remain to be determined . To gain insight into the TBCC GAP mechanism , we determined 3D reconstructions for TBC-DEG-Q73L:αβ-tubulin:TBCC ternary complexes using negative-stain EM and single particle image analysis ( see ‘Materials and methods’ ) . We used the TBC-DEG-Q73L complex to guarantee 100% stoichiometric TBCC binding in the TBC-DEG-Q73L:αβ-tubulin:TBCC ternary complex to ensure its full occupancy in the structural studies . Raw images and reference-free classification indicate that the hollow core of the cage becomes largely occupied in the ternary complex . This conformation is distinct in appearance from the previous conformations ( Figure 7—figure supplement 1B ) . We used angular reconstitution and refinement to generate a 24 Å TBC-DEG-Q73L:αβ-tubulin:TBCC map ( Figure 7A , Table 4 ) , and the model projections match well to the reference-free class averages ( Figure 7—figure supplement 1E ) . The ternary complex map was then interpreted with respect to the TBC-DEG-Q73L and TBC-DEG-Q73L:αβ-tubulin maps ( Figure 7B ) . The ternary complex map shows an additional wedge-shaped density inside the hollow core of the TBC-DEG cage , which we assign to the TBCC β-helix domain . A second density is observed engaging the αβ-tubulin at its intra-dimer interface , located in proximity to the TBCE Cap-Gly domain densities in previous maps ( Figure 7A , B ) . The TBCC C-terminal domain-Arl2 GTPase interface lies directly below the intra-dimer interface of tubulin . We generated a pseudo-atomic model for TBC-DEG-Q73L:αβ-tubulin:TBCC by fitting atomic coordinates for TBCC-N spectrin in the additional density along the TBC-DEG floor , and fit the C-terminal β-helix domain to the wedge density ( Figure 7C ) . The TBCC β-helix docked well to the wedge density engaging the Arl2 GTPase ( UCSF Chimera correlation coefficient 0 . 85 ) , and this fit matches the conformation of our homology model for the TBCC-Arl2 binary complex ( Figure 6E ) . Strikingly , despite the low resolution of our maps , we find that the αβ-tubulin dimer adopts a unique conformation in the ternary complex . The intact αβ-tubulin dimer did not fit into this density in the ternary complex map . Therefore , α and β-tubulin models were fit individually ( UCSF Chimera correlation coefficients 0 . 55 and 0 . 67 for α and β-tubulin , respectively; Figure 7H ) . The αβ-tubulin intra-dimer interface is wedged open by a 20 Å-wide globular density , which we assigned to be the TBCE Cap-Gly domain due to its physical proximity in the TBC-DEG and TBC-DEG:αβ-tubulin maps . We suggest that the TBCE Cap-Gly domain is repositioned by 10 Å in the ternary complex , wedging between α- and β-tubulin . TBCC C-terminal β-helix and its loop lie directly below the intra-dimer interface . At the current resolution it remains unclear how the αβ-tubulin dimer is modified and we require higher resolution studies to understand its conformation and the positioning of TBC-DEG domains in the ternary complex . Our structural analysis supports the biochemical finding that TBCC is an αβ-tubulin dependent non-classical GAP for Arl2 . Our ternary complex map suggests that TBCC Arg186 activates the Arl2 GTPase while engaging αβ-tubulin at its intra-dimer interface via the extended loop . Arl2 GTP hydrolysis leads to a conformational change that involves a well-documented rotation of its conserved N-terminal helix ( Figure 6—figure supplement 1E; Veltel et al . , 2008 ) ; this conformational change may reposition the associated TBCE Cap-Gly domain to deform αβ-tubulin or activate αβ-tubulin GTP hydrolysis at its N-site ( Figure 6—figure supplement 1E ) . 10 . 7554/eLife . 08811 . 023Figure 7 . A TBC-DEG:αβ-tubulin:TBCC ternary complex structure shows TBCC engages both Arl2 and αβ-tubulin dimer , deforming its intra-dimer interface . ( A ) A refined 24 Å TBC-DEG-Q73L:αβ-tubulin:TBCC 3D map shown in three rotated views . The map shows conformational changes in tubulin density and the presence of new densities in the hollow core of the cage . ( B ) A segmented 24 Å TBC-DEG-Q73L:αβ-tubulin:TBCC map shown in three rotated views . The tubulin dimer density ( red ) is deformed by two new densities: a TBCC wedge shaped density engages the Arl2 interface ( green ) , and a second density ( cyan ) is wedging between the two αβ-tubulin dimer lobes ( red ) . ( C ) A TBC-DEG:αβ-tubulin:TBCC linear domain map shown to length scale . TBCD ( pink , top panel ) is composed of HEAT repeats . TBCE ( second panel ) includes a Cap-Gly domain ( dark blue ) , a leucine rich repeat ( LRR ) domain ( blue ) , and a ubiquitin-like domain ( cyan ) . TBCC consists of a spectrin domain ( yellow ) and a C-terminal β-helix domain ( green ) ( described in Figure 5 ) , and Arl2 ( third panel ) consists of a GTPase fold ( orange ) . αβ-tubulin ( red ) is shown in the bottom panel . Colors correspond to subunits shown in D–I . ( D ) A pseudo-atomic model of the TBC-DEG-Q73L:αβ-tubulin:TBCC ternary complex showing the interfaces of the TBCC β-helix catalytic domain ( described in Figure 5 , green ) engaging Arl2 ( orange ) on top of TBCD ( pink ) while bound by the TBCE LRR bow ( blue ) , while the α and β-tubulins are wedged by the ubiquitin domain ( cyan ) . The α and β-tubulin coordinates were fit individually due to deformation in the tubulin intra-dimer interface in this map . The TBCC N-terminal spectrin domain ( yellow ) was fit into a density added to the floor segment . ( E ) A 90° vertically rotated view of that shown in D . ( F ) A 90° horizontally rotated view of that shown in D . ( G ) A central slice view of a 90° counterclockwise horizontally rotated view of that shown in D . The TBCC C-terminal catalytic domain engages Arl2 and binds the αβ-tubulin at the deformed intra-dimer interface with its unique loop ( pink ribbon ) . Video 3 shows the C–F views . ( H ) Comparison of αβ-tubulin conformation based on αβ-tubulin coordinates fit into the αβ-tubulin density from the TBC-DEG-Q73L:αβ-tubulin map shown in the left panel ( −TBCC ) compared to the αβ-tubulin coordinates fit into the αβ-tubulin density in the TBC-DEG-Q73L:αβ-tubulin:TBCC map shown on the right ( +TBCC ) , showing the conformational change at its intra-dimer interface that is associated with TBCC binding . ( I ) Cartoon view of TBC-DEG domain organization comparable to the view shown in E . ( J ) Cartoon view of TBC-DEG domain organization comparable to the view shown in F . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 02310 . 7554/eLife . 08811 . 024Figure 7—figure supplement 1 . Electron microscopy and 3D reconstruction of the TBC-DEG-Q73L:αβ-tubulin:TBCC complex . ( A ) Left panel , expanded view of a raw negative-stain EM image showing TBC-DEG Q73L:αβ-tubulin:TBCC complex . Right panel , higher magnification view of TBC-DEG-Q73L:αβ-tubulin:TBCC particles showing the variety of orientations . ( B ) Multivariant statistical analysis ( MSA ) reference-free class averages of TBC-DEG-Q73L:αβ-tubulin show a variety of commonly observed particle views . ( C ) 3D reconstruction for the TBC-DEG complex is initiated with a 50 Å resolution starting model ( left ) , then iterative projecting matching ( middle ) , followed by refinement ( right ) . ( D ) Overlay of the TBC-DEG-Q73L:αβ-tubulin map ( red ) over the TBC-Q73L map ( transparent blue ) . ( E ) Comparison of the reference-free class averages to 2D projections of the refined structure . Each panel shows a comparison between two images through projection matching: reference-free class averages ( MSA , on top ) and 2D projection from the 3D map of TBC-DEG Q73L:αβ-tubulin:TBCC ( 2D Prj match ) . ( F ) Phi and theta angular distribution plot for each individual TBC-DEG Q73L:αβ-tubulin:TBCC image used in the final reconstruction; the plot is using Angplot_dp . ( G ) Fourier shell correlation analysis of the TBC-DEG-Q73L:αβ-tubulin:TBCC reconstruction . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 02410 . 7554/eLife . 08811 . 019Video 3 . The video shows a 360° rotation of the raw TBC-DEG-Q37L:αβ-tubulin:TBCC map ( Figure 7A ) followed by a 360° rotation of the overlaid TBC-DEG-Q73L:αβ-tubulin:TBCC map ( blue ) with the TBC-DEG-Q73L:αβ-tubulin map ( red ) as shown in Figure 7—figure supplement 1D ( accompanies Figure 7 ) . This is next followed by a 360° rotation of the segmented TBC-DEG-Q73L:αβ-tubulin:TBCC map ( accompanies in Figure 7 ) , followed by a 360° rotation of the TBC-DEG-Q37L:αβ-tubulin:TBCC segmented and fitted map ( Figure 7D–G ) , and followed by a clipping view slicing across the segmented and fitted TBC-DEG-Q73L:αβ-tubulin:TBCC map . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 019 To determine the roles of the TBC-DEG chaperone in regulating MT dynamics and function , we introduced the Q73L mutation into the Arl2 ortholog in budding yeast , Cin4 ( cin4-Q73L ) , and observed its effects on MT function and dynamics . First , we tested whether cin4-Q73L sensitizes cells to the MT depolymerizing drug benomyl . Wild type and Arl2-deleted ( cin4∆ ) yeast cells were transformed with plasmids containing the cin4-Q73L mutant , wild type CIN4 , or no protein ( empty vector ) under a galactose-inducible promoter . Consistent with previous studies , cin4∆ null mutants exhibit hypersensitivity to benomyl that is rescued by expression of wild type CIN4 ( Stearns , 1990; Figure 8A ) . In contrast , cin4-Q73L expression elicits dominant hypersensitivity to benomyl; ectopic expression of cin4-Q73L sensitized both wild type and cin4∆ cells to benomyl ( Figure 8B ) . Our genetic rescue experiments suggest that dominant MT function defects are induced when Arl2/cin4-Q73L is overexpressed in native or Arl2-deleted cells . 10 . 7554/eLife . 08811 . 025Figure 8 . Introducing the Arl2 GTP-locked Q73L mutation induced pausing of dynamic MTs in vivo . ( A ) Expression of cin4-Q73L elicits dominant benomyl sensitivity and MT polymerization defects . Wild type or cin4∆ mutant cells transformed with plasmids expressing the indicated genes under galactose-inducible promoters were plated on inducing media without benomyl or with 10 µg/ml benomyl , and grown at 30°C for 4 days . ( B ) Expression of cin4-Q73L interferes with MT dynamics and activates pausing . Right panel , wild type cells expressing MT labeled with Tub1-GFP . Strain: yJM0562 . Second panel , wild type yeast cells expressing Tub1-GFP transformed with a cin4-Q73L expression plasmid and treated with galactose to induce expression for 90 min before imaging . Third panel , a separate population of wild type cells transformed with a cin4-Q73L expression plasmid and induced for 180 min before imaging . Right panel , cells constitutively expressing cin4-Q73L from the chromosomal locus . Images are 2D projections of 13 Z planes separated by 400 nm . ( C ) Representative raw fields of yeast cells in each condition with MTs labeled with GFP-tub . Videos 4–7 accompany these panels . ( D ) Representative lifeplots of astral MT dynamics in wild type cells ( top panel ) , wild type cells expressing cin4-Q73L for 90 min ( middle panel ) , and cin4-Q73L mutants ( bottom panel ) . Astral MT length was measured over time by plotting the distance between the spindle and the distal end of the MT . Strains: wild type , yJM0562; cin4-Q73L , yJM1375 . ( E ) MT rescue frequencies from time-lapse imaging of wild type ( n = 5 ) , wild type expressing cin4-Q73L for 90 min ( n = 7 ) , and constitutive cin4-Q73L mutant ( n = 10 ) cells . Asterisks indicate statistical significance ( p<0 . 01 ) determined by t-test , compared to wild type . Strains: wild type , yJM0562; cin4-Q73L , yJM1375 . ( F ) Durations of pause events from time-lapse imaging of wild type ( black ) , wild type expressing cin4-Q73L for 90 min , and constitutive cin4-Q73L mutant cells ( shown in pink ) . Asterisks indicate statistical significance ( p<0 . 01 ) determined by t-test , compared to wild type . Strains: wild type , yJM0562; cin4-Q73L , yJM1375 . ( G ) Average MT disassembly rates from time-lapse imaging of wild type ( black ) , wild type expressing cin4-Q73L for 90 min , and constitutive cin4-Q73L mutant cells ( shown in pink ) . Asterisks indicate statistical significance ( p<0 . 01 ) determined by t-test , compared to wild type . Strains: wild type , yJM0562; cin4-Q73L , yJM1375 . ( H ) Average length of astral MTs ( aMT ) per yeast cell measured , for wild type ( black ) , wild type expressing cin4-Q73L for 90 min , and constitutive cin4-Q73L mutant cells ( shown in pink ) . Asterisks indicate statistical significance ( p<0 . 01 ) determined by t-test , compared to wild type . Strains: wild type , yJM0562; cin4-Q73L , yJM1375 . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 02510 . 7554/eLife . 08811 . 026Video 4 . Microtubule dynamics in wild type cells ( accompanies Figure 8C ) . Time-lapse images of wild type cells expressing Tub1-GFP were captured on a spinning disk confocal microscope at 4 s intervals for 10 min . Each image represents a composite of 13 planes separated by 400 nm . Video plays at 60 times real time . Strain yJM0562 . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 02610 . 7554/eLife . 08811 . 029Video 5 . Microtubule dynamics after 90 min of cin4-Q73L expression ( accompanies Figure 8C ) . Time-lapse images of wild type cells expressing Tub1-GFP containing a cin4-Q73L expression plasmid after 90 min of induction . Images were captured on a spinning disk confocal microscope at 5 s intervals for 10 min . Each image represents a composite of 15 planes separated by 400 nm . Video plays at 60 times real time . Strain yJM0562 , with plasmid pJM0231 . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 02910 . 7554/eLife . 08811 . 030Video 6 . Microtubule dynamics after 180 min of cin4-Q73L expression ( accompanies Figure 8C ) . Time-lapse images of wild type cells expressing Tub1-GFP containing a cin4-Q73L expression plasmid after 180 min of induction . Images were captured on a spinning disk confocal microscope at 5 s intervals for 10 min . Each image represents a composite of 15 planes separated by 400 nm . Video plays at 60 times real time . Strain yJM0562 , with plasmid pJM0231 . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 03010 . 7554/eLife . 08811 . 027Video 7 . Microtubule dynamics in cin4-Q73L mutant cells ( accompanies Figure 8C ) . Time-lapse images of cells constitutively expressing cin4-Q73L from the chromosomal locus and Tub1-GFP were captured on a spinning disk confocal microscope at 4 s intervals for 10 min . Each image represents a composite of 13 planes separated by 400 nm . Video plays at 60 times real time . Strain yJM1375 . DOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 027 Next , we examined the dynamics of GFP-labeled MTs in yeast cells . We compared the effects of transient cin4-Q73L expression in wild type cells to mutant cells with cin4-Q73L constitutively expressed from the chromosomal locus ( Figure 8C ) . Arl2/cin4-Q73L expression decreased the number of MTs per cell , and this effect scaled with duration of cin4-Q73L expression ( Figure 8H; Video 4 , Video 7 ) . A 90 min pulse of ectopic cin4-Q73L expression decreased the number of astral MTs ( aMTs ) in wild type cells . This effect was exacerbated after 180 min of expression ( Figure 8C , H ) . Mutant cells constitutively expressing cin4-Q73L exhibited the strongest loss of MTs ( Figure 8C , H ) . Analysis of individual aMT lengths in time-lapse imaging revealed striking changes in MT dynamics . After a 90 min pulse of ectopic cin4-Q73L , aMTs were longer and exhibited slower disassembly , decreased rescue frequency , and increased pauses compared to those observed in wild type yeast ( Figure 8D–G; Table 6; Videos 5 , 6 ) . Cells expressing constitutive cin4-Q73L also exhibited decreased rescue frequency and increased pauses . However , aMTs were slightly but significantly shorter than those in wild type cells; in contrast , MT disassembly rates were not significantly different . In conclusion , our studies suggest that the previously well-characterized phenotypes of TBC protein inactivation ( Hoyt et al . , 1990; Radcliffe et al . , 1999; Antoshechkin and Han , 2002; Jin et al . , 2009 ) may be a result of soluble αβ-tubulin regulation defects leading to aberrant MT dynamics . 10 . 7554/eLife . 08811 . 028Table 6 . Microtubule ( MT ) dynamics in cin4-Q73L mutant yeast cellsDOI: http://dx . doi . org/10 . 7554/eLife . 08811 . 028MT length ( µm ) Assembly rate ( µm/min ) Assembly duration ( s ) Disassembly rate ( µm/min ) Disassembly duration ( s ) Catastrophes ( min−1 ) Rescues ( min−1 ) Pause duration ( s ) Wild type0 . 90 ± 0 . 021 . 4 ± 0 . 0646 ± 33 . 06 ± 0 . 2427 ± 20 . 90 ± 0 . 121 . 5 ± 0 . 1818 ± 3Wild type + cin4-Q73L ( 90 min ) 2 . 69 ± 0 . 051 . 4 ± 0 . 1858 ± 121 . 56 ± 0 . 1253 ± 100 . 66 ± 0 . 120 . 48 ± 0 . 1236 ± 11cin4-Q73L0 . 71 ± 0 . 021 . 5 ± 0 . 0641 ± 43 . 00 ± 0 . 3624 ± 40 . 78 ± 0 . 060 . 66 ± 0 . 1237 ± 4Values are mean ± SEM of measurements from at least five cells imaged for 600 s . Values in boldface are significantly different from wild type ( p<0 . 05 ) , determined by t-test . We provide a revised paradigm for the assembly , biochemical activity , and organization of the well-conserved tubulin cofactors and Arl2 GTPase as a cage-like chaperone that catalytically alters tubulin dimers in the cytoplasm , powered by GTP hydrolysis . The GTPase activity of Arl2 is central to power and gate these chaperones , while tubulin cofactors TBCD and TBCE mediate molecular recognition of α- and β-tubulin in the heterodimer ( Lewis et al . , 1997; Tian and Cowan , 2013 ) . The concept that tubulin cofactors and Arl2 function together as a catalytic chaperone is consistent with long-standing genetics and cell biology studies indicating that their concentration is critical for proper MT dynamics and MT homeostasis . These chaperones represent a new MT regulatory pathway that may enhance MT dynamics by improving the activities of individual soluble αβ-tubulin dimers in the cytoplasmic pool . This regulation is likely critical for the homeostasis of the MT cytoskeleton in eukaryotes , which is underscored by human disorders related to tubulin cofactor mutations . Full length S . cerevisiae TBCC , TBCD , TBCE , and Arl2 cDNAs ( also named Cin2 , Cin1 , Pac2 , and Cin4 , respectively ) were amplified by PCR using oligonucleotides and inserted in two polycistronic bacterial expression vectors using isothermal assembly and confirmed by DNA sequencing . Each vector contains a single T7 promoter , individual ribosomal binding sites before each insert , and a single T7 terminator ( Tan et al . , 2005 ) . To determine the accessibility of unique N- or C-termini of different TBC proteins , 6xHis or 6xHis-EGFP tags were inserted at either the 5′ or 3′ ends of TBCD , TBCE , or Arl2 cDNAs in different polycistronic expression vectors ( as described ‘Results’ and shown in Figure 2—figure supplement 1A , B ) and were tested for expression and purification , as described below . We determined the composition of TBC-DEG complexes purified from a TBCA , TBCB , TBCC , TBCD , TBCE , and Arl2 co-expression system using a nano LC-MS/MS approach , showing TBCD-E-Arl2 complexes ( TBC-DEG ) as described in Table 1 . We focused on the study of TBC-DEG using two polycistronic vectors . We constructed modified TBC-DEG expression constructs , including a TBC-DEG-Arl2 Gln73Leu mutant , and EGFP inserted at the N-terminus of TBCE for further studies . TBCD , TBCE , and Arl2 deletion polycistronic constructs ( described in Figure 2—figure supplement 1A ) were assembled through PCR by using inserts where cDNA sequences coding for either TBCD N-terminus ( 1–116 residues ) , TBCD C-terminus ( 866–1016 residues ) , TBCE N-terminal Cap-Gly domain ( 1–70 residues ) , and TBCE C-terminal ubiquitin domain ( 420–518 residues ) , Arl2 N-terminal ( 1–50 residues ) and Arl2 C-terminus ( 90–125 residues ) were deleted . Recombinant TBC-DEG is purified as follows: polycistronic constructs were co-transformed into a bacterial expression strain at 37°C and then induced with 0 . 5 mM isothio-beta-glucopyranoside ( IPTG ) overnight at 20°C . Cells were pelleted and then lysed in 150 mM KCl , 50 mM HEPES , 1 mM MgCl2 , 3 mM β-mercaptoethanol , and 50 μM GTP with protease inhibitors including 1 mM PMSF , 1 μg/ml leupeptin , 20 μg/ml benzamidine , and 40 μg/ml aprotinin ( RPI ) . The lysate was clarified by centrifugation at 18 , 000 rpm for 30 min at 4°C . Ni-NTA affinity ( Macherey-Nagel , Bethlehem , PA , USA ) was used to purify TBC-DEG complexes . NI-NTA purified TBC-DEG complexes were diluted with low salt buffer ( 100 mM KCl , 50 mM HEPES , 1 mM MgCl2 ) , bound to HiTrap SP FF ( GE Healthcare , Pittburgh , PA , USA ) anion exchange and then eluted with a five column volume gradient using high salt buffer ( 500 mM KCl , 50 mM HEPES , 1 mM MgCl2 ) . The TBC-DEG containing fractions were concentrated using Amicon concentrators and then loaded on a HiLoad 16/600 Superdex-200 gel filtration column ( GE Healthcare ) . TBC-DEG was then used in subsequent studies as described below , without freezing . Recombinant TBCC , its deletion and point mutants were expressed in bacteria . TBCC constructs were assembled using point mutagenesis and isothermal assembly , expressed in bacteria , and purified using the approach described above with few modifications . Briefly , bacterial cells overexpressing TBCC were resuspended in 50 mM MES , 100 mM KCl , and 3 mM β-mercaptoethanol , cells were lysed , and then the lysate was clarified by centrifugation as described above . TBCC was bound to Hitrap-SP FF and then eluted with a five column volumes gradient of 50 mM MES , 100 to 500 mM KCl , pH 6 . 0 , and 3 mM β-mercaptoethanol . TBCC containing fractions were concentrated using Amicon concentrators and loaded on a Superdex 200 HiLoad 10/16 gel filtration column , analyzed by SDS-PAGE , and then frozen in liquid nitrogen . Recombinant purified TBC-DEG ( 5–10 μmol ) was diluted in 50 mM HEPES , 100 mM KCl , pH 7 . 0 , and 3 mM β-mercaptoethanol including either GTP , GDP . ALFx , or GTPγS nucleotide analogs , and then mixed with equimolar double-cycled porcine brain αβ-tubulin and/or TBCC . TBC-DEG , αβ-tubulin , and TBCC and their complexes were purified by size exclusion chromatography ( SEC ) using a Superdex 200 10/300 gel filtration column using an AKTA purifier ( GE Healthcare ) system , and 0 . 5 ml fractions were collected and analyzed using a Bis-Tris based XT criterion SDS-PAGE system ( Bio-Rad , Hercules , CA , USA ) . The molecular masses of TBC-DEG , αβ-tubulin , TBCC , and their complexes were measured using SEC-MALS , proteins were separated on a WTC-03S5 size exclusion column ( Wyatt Technologies , Santa Barbara , CA , USA ) , while UV absorbance ( detected by Agilent 1100 Series HPLC ) , light scattering ( Wyatt Technology miniDAWN TREOS ) , and refractive index ( Wyatt Technology Optilab T-rEX ) were measured and the concentration-weighted molecular weights of each peak were calculated using ASTRA V . 6 software ( Wyatt Technologies ) ( Tarazona and Saiz , 2003 ) . Steady-state GTP hydrolysis activity was measured using a malachite green free-phosphate detection assay as previously described ( Leonard et al . , 2005 ) , with the following modifications: purified 10 μM recombinant TBC-DEG , αβ-tubulin , and TBCC were desalted using reaction buffer ( 50 mM HEPES , 100 mM KCl ) , combined in 96-well plates , in the presence of 0–800 μM GTP ( Jena Biosciences , Jena , Germany ) , and incubated for 90 min at 30°C . The GTP hydrolysis reactions were quenched by the addition of 0 . 1 mM EDTA , followed by 1 mM malachite green . Phosphate-malachite green complex concentration was measured at 621 nm in a 96-well plate format using an Amersham plate reader ( GE Healthcare ) . The phosphate concentration was determined using a 0–5 μM linear phosphate standard curve treated the same way as the reaction conditions . Km and Vmax were measured using a Michaelis–Menten curve fit . Vmax was used to calculate kcat values based on a 1 μmol TBC-DEG enzyme concentration and fit against a range of GTP substrate concentrations . Fresh SEC purified 0 . 5 mg/ml TBC-DEG-Q73L , NGFP-TBCE-DEG-Q73L , TBC-DEG-Q73L:αβ-tubulin , and TBC-DEG-Q73L:αβ-tubulin:TBCC complexes in 50 mM HEPES , 100 mM KCl , 0 . 1 mM GTP , and 3 mM β-mercaptoethanol were each incubated on carbon coated grids , briefly washed , and then stained with 1% uranyl formate . Electron microscopy was performed using a JEOL-2100 FF operating at 50 , 000 nominal magnification , and approximately 80–100 EM images were recorded on S0163 film ( Kodak ) for each condition , focusing mostly on areas of thick stain where particles are less likely to be flattened . Film images for each data set were scanned using a D8200 PrimeScan Heidelberg drum densitometer at 5 . 0 μm/pixel leading to 1 Å/pixel on the specimen . Images were normalized and binned two-fold ( 2 Å/pixel ) using the EMAN2 . 1 software package ( Tang et al . , 2007 ) . Roughly 18 , 000–20 , 000 globular cage-like individual TBC-DEG particles for each group were picked semi-automatically using e2boxer . py . Particle stacks were generated , then contrast transfer function ( CTF ) corrected with e2ctf . py using the phase flipping function . The image stacks were then subjected to iterative reference-free classification using e2refine2d . py generating 400 class averages , to remove roughly 10–30% of deformed , rare , or broken particle images . Additional rounds of classification were performed and the resulting 80-100 class averages show unique orientations suggesting a moderate degree of preferred orientations on the grids representing different , yet commonly observed views , as judged by their representation in the class averages ( Figure 4—figure supplement 1B , Figure 5—figure supplement 1B , Figure 7—figure supplement 1B ) . For each data set , prominent class averages were then used to generate a starting model using a common lines strategy ( Tang et al . , 2007 ) . These starting models were then filtered to 50 Å resolution and used in cycles of 3D projection matching and angular reconstitution using the SAMUEL program utilities running the SPIDER program ( https://sites . google . com/site/maofuliao/samuel ) . Projection matching and angular reconstitution of the starting model were initiated at 30° increments and then decreased successively by 5° increments down to 5° ( Shaikh et al . , 2008 ) ( Figure 4—figure supplement 1C , Figure 5—figure supplement 1C , Figure 7—figure supplement 1C ) . The resulting volume was filtered to 35 Å and the individual angular assignments were then further refined in multiple cycles in the program FREALIGN ( Grigorieff , 2007; Lyumkis et al . , 2013 ) . Refinement convergence was determined from changes in phase residuals , angular assignment changes based on the program angplot_dp , and by comparing model projections to reference-free class averages with a global search using FREALIGN ( Figure 4—figure supplement 1D , Figure 5—figure supplement 1D , Figure 6—figure supplement 1D ) . Fourier shell correlation ( FSC ) calculations from two half data sets indicate 25 Å resolution for all maps based on the 0 . 5 cutoffs ( Table 4; Figures 4—figure supplement 1E , Figure 5—figure supplement 1E , Figure 7—figure supplement 1E ) . Final maps were aligned using the program XMIPP ( Sorzano et al . , 2004 ) , were resolution truncated to nominal resolution based on FSC cutoffs , and were visualized using the program UCSF Chimera ( Pettersen et al . , 2004 ) . The individual subunits were docked into the maps using two approaches that led to similar results . First , the maps were segmented using the segment map utility , and x-ray crystal structures for paralogs were fit using the fit-to-segment utility ( Pettersen et al . , 2004 ) . Second , x-ray models were filtered to 24 Å resolution and then docked using the fit-in-map feature without segmentation , starting with the largest down to the smallest subunit , and each time cumulatively excluding regions of the map fit by the previous subunit . The floor and thumb regions ( shown in Figures 4E–H , Figure 5E–H , and Figure 7E–H in pink ) were fit with the Cse1 paralog x-ray structure ( PDB-ID 1Z3H; Cook et al . , 2005a ) , the TBCE bow region ( shown in blue ) was fit with the TLR4 LLR structure ( PDB-ID 3FXI; Park et al . , 2009a ) , its two globular end segments ( shown in dark blue and cyan ) were fit with the TBCB Cap-Gly structure ( PDB-ID 4B6M; Fleming et al . , 2013a ) and the TBCB ubiquitin domain structure ( PDB-ID 4B6W; Fleming et al . , 2013b ) , respectively , the Arl2 pillar region ( shown in orange ) was fit with the human Arl2 structure ( PDB-ID 1KSJ; Hanzal-Bayer et al . , 2002a , 2002b ) , αβ-tubulin dimer dual density ( shown in red ) was fit with the αβ-tubulin dimer ( PDB-ID 1JFF; Lowe et al . , 2001 ) , the TBCC N-terminal domain segment ( shown in yellow ) was fit by the TBCC spectrin N-terminal domain structure ( PDB ID 2L3L ) , and the wedge segment in the ternary complex map ( Figure 7E–H , shown in green ) was fit by the TBCC-C-terminal domain structure ( determined here , PDB ID 5CYA ) . The final resolution truncated maps were deposited into the EMD database under accession numbers EMD-6393 , EMD-6392 , EMD-6391 , and EMD-6390 . Purified budding yeast TBCC was screened for crystallization in 96-well format using a mosquito crystallization robot ( TTPlabtech , Oxford , UK ) , using a combination of home-made or commercial screens ( Qiagen , Valencia , CA , USA ) . TBCC crystals formed in 0 . 1 M sodium citrate pH 5 . 6 , 0 . 5 M ammonium sulfate , and 1 . 0 M lithium sulfate . The largest crystals were formed 1 week after micro-seeding in 0 . 1 M sodium citrate , 0 . 4 M ammonium sulfate , and 0 . 7 M lithium sulfate pH 5 . 2 . Native TBCC crystals were soaked in mother liquor containing potassium hexacyanoplatinate , transferred to paratone-N oil , and then frozen in liquid nitrogen . TBCC diffraction data were collected from single crystals at the Stanford Synchrotron Radiation Laboratory ( SSRL ) . The best TBCC crystals diffracted at 2 . 0 Å resolution in a tetragonal ( P 43 ) space group . Phase information was determined using platinum-substituted crystals using the multi-wavelength anomalous dispersion ( MAD ) approach with data collected in 10° wedge increments . TBCC diffraction data were indexed using the program MOSFILM in a P 43 space group using the unit cell dimensions 70 . 03 , 70 . 03 , and 77 . 95 Å , and were scaled using the program SCALA ( Powell et al . , 2013 ) . Phase information was determined by locating platinum atom positions using the program RESOLVE in the Phenix program suite ( Terwilliger and Berendzen , 1999 ) . The initial locations for platinum atom positions were determined , refined to a 0 . 69 figure of merit ( FOM ) , and used to obtain initial TBCC density maps . TBCC density maps indicated two TBCC C-terminal domain molecules in the asymmetric unit; the TBCC N-terminal spectrin domain could not be identified in the density maps , suggesting it maybe disordered or underwent proteolysis during crystallization . A Matthew's coefficent calculation of the solvent content supports the idea that only the TBCC C-terminal domain is contained in the crystal rather than full length TBCC with a disordered N-terminus . A budding yeast TBCC C-terminal domain model was built starting at residue 100 and ending at residue 267 using the program COOT , and the resulting models were rigid-body refined using the Phenix program suite ( Emsley et al . , 2010; Adams et al . , 2010 ) . To generate an Arl2-TBCC interface model , an RP2-Arl3 ( PDB-ID 3BH7; Veltel et al . , 2008 ) model was used as homology templates , where Arl2 ( PDB-ID: 1KSJ; Hanzal-Bayer et al . , 2002b ) was aligned to Arl3 , and the TBCC C-terminal domain , determined in this study ( PDB-ID 5CYA ) , was aligned to RP2 using the Match-Maker structural analysis function in the program UCSF Chimera . The TBCC β-helix domain structure , homology models , and surface conservation images were generated using UCSF Chimera ( Pettersen et al . , 2004 ) . Yeast manipulation , media , and transformation were performed by standard methods ( Amberg et al . , 2005 ) . The Q73L substitution mutation was introduced into a CIN4 expression plasmid ( pJM0230 ) by site-directed mutagenesis , creating a cin4-Q73L expression plasmid ( pJM0231 ) . Q73L was introduced into CIN4 at the endogenous locus using methods similar to those described in Moore et al . ( 2009 ) . All mutations were confirmed by sequencing the complete open reading frame . Time=lapse images of cells expressing GFP-labeled microtubules ( plasmid pSK1050 , a gift from K Lee at the National Institutes of Health ) were collected on a Nikon Ti-E microscope equipped with a 1 . 45 NA 100× CFI Plan Apo objective , piezo electric stage ( Physik Instrumente , Auburn , MA ) , spinning disk confocal scanner unit ( CSU10; Yokogawa ) , 488 nm laser ( Agilent Technologies , Santa Clara , CA ) , and an EMCCD camera ( iXon Ultra 897; Andor Technology , Belfast , UK ) using NIS Elements software ( Nikon ) . Living cells from asynchronous cultures grown to early log phase were suspended in non-fluorescent medium , mounted on a slab of 2% agarose , and sealed beneath a coverslip with paraffin wax . Z series of 13 images separated by 400 nm were collected . The number of aMTs was determined in the first Z series of each acquisition . MT dynamics were analyzed by measuring aMT length at 4 or 5 s intervals for 10 min . This analysis was conducted in preanaphase cells , which typically exhibit one or two aMTs . Assembly and disassembly events were defined as continuous phases that produced a net change in aMT length of ≥0 . 5 µm and a coefficient of variation ≥0 . 85 . Pause events were defined as lasting at least four data points ( 12 s ) without significant change in aMT length . Catastrophes were defined as transitions from assembly or pause to disassembly . Catastrophe frequencies were determined for individual aMTs by dividing the number of catastrophe events by the total time spent in assembly and pause states . Rescues were defined as transitions from disassembly or pause to assembly . Rescue frequencies were determined for individual aMTs by dividing the number of rescue events by the total time for disassembly and pause states .	Cells contain a network of protein filaments called microtubules . These filaments are involved in many biological processes; for example , they help cells keep the right shape , and they help to transport proteins and other materials inside cells . Two proteins called α-tubulin and β-tubulin are the building blocks of microtubules . The filaments are very dynamic structures that can rapidly change length as individual tubulin units are either added or removed to the filament ends . Several proteins known as tubulin cofactors and an enzyme called Arl2 help to build a vast pool of tubulin units that are able attach to the microtubules . These units—called αβ-tubulin—are formed by α-tubulin and β-tubulin binding to each other , but it not clear exactly what roles the tubulin cofactors and Arl2 play in this process . Nithianantham et al . used a combination of microscopy and biochemical techniques to study how the tubulin cofactors and Arl2 are organised , and their role in the assembly of microtubules in yeast . The experiments show that Arl2 and two tubulin cofactors associate with each other to form a stable ‘complex’ that has a cage-like structure . A molecule of αβ-tubulin binds to the complex , followed by another cofactor called TBCC . This activates the enzyme activity of Arl2 , which releases the energy needed to alter the shape of the αβ-tubulin . Nithianantham et al . also found that yeast cells with a mutant form of Arl2 that lacked enzyme activity had problems forming microtubules . Together , these findings show that the tubulin cofactors and Arl2 form a complex that regulates the assembly and maintenance of αβ-tubulin . The next challenge is to understand how this regulation influences the way that microtubules grow and shrink inside cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Tubulin cofactors and Arl2 are cage-like chaperones that regulate the soluble αβ-tubulin pool for microtubule dynamics
Sleep loss can severely impair the ability to perform , yet the ability to recover from sleep loss is not well understood . Sleep regulatory processes are assumed to lie exclusively within the brain mainly due to the strong behavioral manifestations of sleep . Whole-body knockout of the circadian clock gene Bmal1 in mice affects several aspects of sleep , however , the cells/tissues responsible are unknown . We found that restoring Bmal1 expression in the brains of Bmal1-knockout mice did not rescue Bmal1-dependent sleep phenotypes . Surprisingly , most sleep-amount , but not sleep-timing , phenotypes could be reproduced or rescued by knocking out or restoring BMAL1 exclusively in skeletal muscle , respectively . We also found that overexpression of skeletal-muscle Bmal1 reduced the recovery response to sleep loss . Together , these findings demonstrate that Bmal1 expression in skeletal muscle is both necessary and sufficient to regulate total sleep amount and reveal that critical components of normal sleep regulation occur in muscle . The ability to recover from sleep loss is critical for preserving cognitive processes and executive functioning ( McCoy and Strecker , 2011; Simon et al . , 2015; Tucker et al . , 2010 ) . The mechanisms that are responsible for the recovery from sleep loss are not well understood . Genetic deletion of the circadian transcription factor Bmal1 ( brain and muscle ARNT-like factor; gene symbol Arntl ) in mice completely ablates circadian clock function ( Bunger et al . , 2000 ) and has effects on sleep that include: increased total sleep amount , increased non-rapid eye movement ( NREM ) sleep intensity and reduced ability to recover from sleep loss ( Laposky et al . , 2005 ) . Because Bmal1 whole-body deletion causes a broad range of physical abnormalities ( e . g . reduced locomotor activity , joint abnormalities , reduced lifespan; Bunger et al . , 2005; Kondratov et al . , 2006 ) , we sought to isolate sleep phenotypes from the potential effects of other phenotypes using tissue-specific Bmal1 rescue and knockout models . We first attempted to rescue electroencephalographic ( EEG ) sleep-phenotypes in Bmal1 knockout mice by restoring functional Bmal1 expression selectively in brain . To do this , we used a transgenic model that rescues Bmal1 expression in the brain of Bmal1 whole-body KO’s , thus restoring circadian behavior ( Scg2::tTa; tetO::Bmal1-HA; McDearmon et al . , 2006 ) . Surprisingly , transgenic Bmal1 expression in the brain ( i . e . , brain rescued ) did not restore NREM sleep-amount—one of the most prominent sleep changes in whole-body Bmal1 knockouts ( Figure 1A , B ) . Since circadian rhythms of locomotor activity are restored in this model ( McDearmon et al . , 2006 ) , our finding suggests that the sleep disturbances in Bmal1 knockout mice are not exclusively the result of disrupted circadian behavior . More importantly , the results show that Bmal1 expression in brain does not restore normal sleep amounts in Bmal1 knockout mice , suggesting that expression in other tissues may be important . 10 . 7554/eLife . 26557 . 003Figure 1 . Rescuing Bmal1 in skeletal muscle restores daily non-rapid eye movement ( NREM ) sleep amount . 24 hr electroencephalographic recordings were conducted in undisturbed mice listed in the legend . The 24 hr pattern of NREM and REM sleep are shown in A . Whole-body knockout of Bmal1 significantly increased NREM sleep when compared to WT controls ( B , ANOVA F ( 2 , 27 ) =11 . 3 , p=0 . 005; p<0 . 001 , posthoc Tukey’s test ) . Rescuing Bmal1in the brain of knockouts did not restore NREM sleep to WT levels ( B , p=0 . 001 vs . WT; Tukey’s test ) ; however , the effect of Bmal1 knockout was reversed when Bmal1 was rescued in the skeletal muscle ( C , ANOVA F ( 2 , 31 ) =9 . 9 , p<0 . 001; p=0 . 88 vs . WT , p<0 . 001 vs . KO , Tukey’s test ) . No differences were found in REM sleep or NREM slow-wave activity . Knockout animals are replotted in B and C to aid in comparison . KO mice were offspring from independent crosses of heterozygous Bmal1 KO’s . Representative electroencephalographic recordings are displayed in D . Grey boxes indicate the active/dark period . Bars and points represent mean ± s . e . m . * , p<0 . 05 . WT ( brain ) n = 11 , knockout n = 12 , muscle rescue n = 5 , brain rescue n = 4 , WT ( muscle ) n = 16 . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 00310 . 7554/eLife . 26557 . 004Figure 1—figure supplement 1 . Bmal1-HA is not detectable in the Brains of Acta1::Bmal1-HA mice . Fifty micrograms of proteins extracted from either an entire brain hemisphere or skeletal muscle ( gastrocnemius ) were subjected to western blotting with the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 004 To begin this investigation of other tissues , we chose mice harboring a transgene that restores Bmal1 specifically in skeletal muscle , but does not restore circadian behavior ( i . e . , muscle rescued; Acta1::Bmal1-HA ) . This model rescues several muscle-related phenotypes despite non-cycling BMAL1 levels ( McDearmon et al . , 2006 ) . To our surprise , although REM sleep was unaffected ( Figure 1B and C ) , we found that restoring Bmal1 in skeletal muscle completely restored NREM sleep amount to wild-type levels in otherwise Bmal1-deficient mice ( Figure 1C ) . We also assessed whether Bmal1 function in the skeletal muscle altered sleep intensity by measuring NREM slow wave activity ( SWA , 0 . 5–4 Hz; Borbély et al . , 1981; Dijk et al . , 1990 ) . In our hands , NREM SWA was not altered in Bmal1 knockout mice; similarly , NREM SWA was not significantly altered by rescuing BMAL1 in brain or muscle ( Figure 1B and C ) . These experiments demonstrate that restoring Bmal1 in the skeletal muscle of otherwise Bmal1-deficient mice is sufficient to restore normal NREM sleep amount , independently of Bmal1 expression in the brain . The diurnal rhythm in sleep amount , however , is not restored ( Figure 1A ) . To investigate the effect of Bmal1 function on recovery from sleep loss , we subjected the Bmal1 mouse lines to 6 hr of forced wakefulness and monitored sleep EEG’s during recovery ( Figure 2A ) . The amount and type of recovery sleep observed in response to forced wakefulness is commonly used to assess changes in sleep homeostasis ( i . e . , sleep drive ) —this recovery is typically characterized by increased sleep amount and increased SWA ( Ehlen et al . , 2013 ) . Whole-body KO of Bmal1 did not affect NREM recovery-sleep amount ( Figure 2A , B ) , but did significantly prevent increased NREM-SWA ( Figure 2C , D ) . Rescue of Bmal1 in either brain- or muscle-rescued mice reduced NREM recovery-sleep following forced wakefulness ( Figure 2A , B ) , a finding that indicates Bmal1 restoration in either tissue reduces sleep drive following forced wakefulness . Notably , NREM SWA in whole-body KO mice was rescued by restoring Bmal1 in skeletal muscle , but not brain ( Figure 2A , B ) . These findings demonstrate that restoring Bmal1 in the skeletal muscle of Bmal1-deficient mice is sufficient to restore normal SWA following sleep loss . 10 . 7554/eLife . 26557 . 005Figure 2 . Rescuing Bmal1 in skeletal muscle or brain reduces the amount of NREM sleep recovered after forced wakefulness . Continuous sleep recordings during 18 hr of recovery sleep were obtained from the mice in Figure 1 after 6 hr of forced wakefulness ( A; yellow double arrow = forced wakefulness ) . The total NREM recovery sleep during this 18 hr period in brain-rescued and muscle-rescued mice was reduced when compared to WT mice ( B , ANOVA brain rescued , F ( 2 , 28 ) =6 . 47 , p=0 . 005 , p=0 . 004 Tukey’s; ANOVA muscle rescued , F ( 2 , 30 ) =5 . 45 , p=0 . 01 Tukey’s ) . Values in A and B represent sleep time gained after forced wakefulness ( calculated using the corresponding interval during undisturbed sleep ) as a percentage of total sleep lost ( mean ± s . e . m . ) . The distribution of EEG power during NREM sleep for representative animals from each genotype is shown in ( panel C ) . Slow wave activity ( highlighted area ) represents power in the 0 . 5 to 4 Hz frequency band . NREM slow wave activity was reduced in knockout mice following forced wakefulness—compared to WT mice ( D , % change over corresponding baseline; p<0 . 01 , Tukey’s test ) . This reduction in slow wave activity was absent when Bmal1 was rescued in the skeletal muscle ( D ) . Knockout animals in B and D are replicated between graphs to aid in comparison . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 00510 . 7554/eLife . 26557 . 006Figure 2—figure supplement 1 . Rescuing Bmal1 in skeletal muscle or brain reduces the amount of NREM sleep recovered after forced wakefulness . Continuous sleep recordings during 18 hr of recovery sleep were obtained from the mice in Figure 1 after 6 hr of forced wakefulness ( squares , dotted line ) . NREM sleep amount is plotted as a percentage of total sleep for 2 hr bins across 24 h days . Each plot includes baseline sleep ( circles , solid line ) on the preceding day for comparison . ( yellow double arrow = forced wakefulness ) . Points represent mean ± s . e . m . p<0 . 05 Tukey’s . Statistics are not reported for the first 6 hr ( sleep deprivation period ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 006 Our results indicate that the rescue of Bmal1 in skeletal muscle is sufficient to restore both NREM sleep amount and the SWA recovery-responses to lost sleep . We next sought to determine if Bmal1 in skeletal muscle was necessary for the effects we observed on sleep processes by specifically deleting Bmal1 in skeletal muscle ( McCarthy et al . , 2012a ) . Mice lacking BMAL1 in their skeletal muscle had significantly increased baseline NREM sleep amount , a result similar to whole-body Bmal1 knockout mice ( Figure 3A–C ) . Moreover , when muscle-specific knockouts were examined after 6 hr of forced wakefulness , recovery sleep was nearly half that of WT mice ( Figure 3D ) . Forced wakefulness also increased SWA in these mice when compared to controls ( Figure 3D ) . This effect on SWA suggests that these mice have higher sleep intensity than WTs during recovery from sleep loss . Together , these data support a role for skeletal muscle and Bmal1 in regulating the ability to recover from sleep loss . Moreover , these data support the conclusion that Bmal1 expression in skeletal muscle is both necessary and sufficient for the regulation of normal NREM sleep amount . 10 . 7554/eLife . 26557 . 007Figure 3 . Knockout of Bmal1 in skeletal muscle increases NREM sleep amount and confers resistance to sleep loss . Selective knockout of Bmal1 in skeletal muscle significantly increased NREM sleep , but not REM sleep , when compared to the same animals prior to tamoxifen treatment ( A , B; t ( 10 ) =2 . 52 , p=0 . 036 ) . Treatment of floxed mutant control animals with tamoxifen did not significantly alter NREM sleep or REM sleep ( C ) . Following tamoxifen treatment , mice underwent 6 hr of sleep deprivation . Muscle knockout mice also had a significantly altered recovery response to this treatment . Significantly less NREM recovery sleep ( t ( 11 ) =2 . 44 , p=0 . 033 ) and increased NREM slow wave activity ( t ( 11 ) =2 . 2 , p=0 . 05 ) was observed in muscle knockout mice when compared to tamoxifen-treated floxed mutants ( D ) . n = 6 per group; p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 00710 . 7554/eLife . 26557 . 008Figure 3—figure supplement 1 . Slow wave activity in inducible muscle knockout mice and controls during undisturbed baseline sleep . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 008 That many of the sleep phenotypes caused by whole-body BMAL1 deficiency are either rescued by muscle-specific Bmal1 expression or recapitulated by muscle-specific Bmal1 deletion suggest that sleep phenotypes in the Bmal1 knockout mice are , in whole or in part , due to loss of BMAL1 in skeletal muscle . Thus , these results suggest an important role for skeletal muscle in sleep regulation , implying that Bmal1-dependent processes in skeletal muscle may be useful therapeutic targets for sleep disorders . In an effort to investigate the therapeutic potential of muscle Bmal1 , we examined sleep architecture in wild type mice harboring either the brain or muscle-specific ( Brager et al . , 2017 ) Bmal1 transgene . In these mice , transgene expression is in addition to endogenous Bmal1 expression . Neither baseline sleep nor SWA was significantly altered in brain overexpressed mice ( Figure 4B ) . Baseline sleep amount was not significantly altered in muscle- overexpressed mice , however , baseline SWA was significantly reduced ( Figure 4A ) , suggesting that overexpressing BMAL1 in the muscle renders mice resistant to sleep loss ( Dijk et al . , 1987 ) . Furthermore , cholinergic neurons of the basal forebrain , which are important for recovery from sleep loss ( Kalinchuk et al . , 2015 ) , are more active in Bmal1 muscle-overexpressed mice ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 26557 . 009Figure 4 . Overexpression of Bmal1 in skeletal muscle confers resistance to sleep loss . 24 hr electroencephalographic recordings were conducted in undisturbed mice overexpressing Bmal1 in skeletal muscle ( A ) or brain ( B ) . No differences in baseline sleep amount were found , however , NREM slow wave activity was significantly reduced in mice overexpressing Bmal1 in skeletal muscle ( A ) . Overexpression of Bmal1 in skeletal muscle also significantly decreased the NREM-recovery response to one ( 24 hr ) day of forced wakefulness by means of a slowly rotating wheel ( C , D ) . It is not unusual for mice to obtain brief amounts of sleep ( i . e . micro-sleep ) during such an extended regimen of forced wakefulness . This sleep obtained during forced wakefulness was also significantly lower in mice overexpressing Bmal1 ( A , ANOVA main effect of genotype F ( 1 , 96 ) =10 . 12 , p=0 . 002 , main effect of time F ( 11 , 96 ) =4 . 1 , p<0 . 001; B , NREM sleep amount t ( 8 ) =1 . 85 , p=0 . 047 , NREM recovery sleep t ( 8 ) =1 . 9 , p=0 . 039 ) . NREM slow wave activity ( a standard marker of sleep intensity ) was consistently lower in Bmal1 muscle-overexpressed mice throughout the 72 hr protocol ( D , repeated-measures ANOVA F ( 1 , 8 ) =6 . 57 , p=0 . 04 ) . In contrast , during forced wakefulness waking slow-wave activity was significantly increased in Bmal1 muscle-overexpressed mice ( D , repeated-measures ANOVA main effect of genotype , F ( 1 , 8 ) =5 . 7 , p=0 . 045 ) . Grey bars indicate forced wakefulness , black bars indicate darkness ( active period ) . Data presented as mean ± s . e . m , n = 16 muscle overexpression , n = 16 littermate controls ( muscle ) , brain overexpression n = 8 , littermate controls ( brain ) n = 11 . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 00910 . 7554/eLife . 26557 . 010Figure 4—figure supplement 1 . Fos-immunoreactivity ( IR ) in densely cholinergic areas is increased in Bmal1 muscle-overexpressed mice . Fos-IR in three brain areas known to be involved in sleep—wake regulation was assessed . Male mice were sacrificed at mid-day by CO2 inhalation in the presence ( SD ) or absence ( Bsln ) of 6 hr sleep deprivation . *p<0 . 05; wild-type n = 5; muscle overexpressed n = 6 . Basal forebrain , F1 , 16=26 . 6 , p<0 . 001 , univariate ANOVA; lateral habenula ( Hb ) , F1 , 16=75 . 6 , p<0 . 001; pedunculopontine tegmentum ( PPT ) , F1 , 16=31 . 0 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 01010 . 7554/eLife . 26557 . 011Figure 4—figure supplement 2 . Activity rhythms in Bmal1 muscle-overexpressed mice are modestly increased when compared to wildtype mice . No differences were found in the circadian rhythm of wheel-running activity between either genotype [Left panel] . Overall locomotor activity measured with infrared sensors revealed increased activity of muscle-overexpressed mice at several time points ( n = 7–8/genotype; Right panel ) . p<0 . 05 , repeated measures ANOVA [one-way ANOVA , post-hoc]; zeitgeber time 12 = activity onset; LD , light—dark cycle; DD , constant darkness . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 01110 . 7554/eLife . 26557 . 012Figure 4—figure supplement 3 . Bmal1 muscle-overexpressed mice do not have not altered levels of CLOCK:BMAL1-target genes in the brain or muscle . No significant differences in CLOCK:BMAL1-target gene expression between wildtype and muscle-overexpressed mice were found in the hypothalamus at either time point investigated ( ZT5 and ZT17 ) . As expected , Bmal1 expression was significantly elevated in the muscle of muscle-overexpressed mice ( 32 . 4-fold increase at ZT5 vs . WT; F1 , 22=15 . 3 , p=0 . 001 , two-way ANOVA; 5 . 1-fold increase at ZT 17 , p=0 . 003 ) . *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 26557 . 012 To further investigate recovery from sleep loss , muscle-overexpressed mice were subjected to 24 hr of forced wakefulness by placing mice in a slowly rotating wheel ( the 6 hr of forced wakefulness used previously is relatively mild ) . It is common for mice to exhibit some sleep during prolonged forced-wakefulness paradigms , however , Bmal1 muscle-overexpressed mice were awake more than WT littermates during these 24 hr of forced wakefulness ( Figure 4C ) . Similar to baseline , Bmal1 muscle-overexpressed mice also had less NREM recovery after forced wakefulness ( recovery = sleep gained in recovery/sleep lost during forced wakefulness; Figure 4C ) and reduced SWA throughout the 72 hr protocol compared to WT mice ( Figure 4D ) , despite sleeping less . In addition , waking SWA ( a measure of accumulating sleep pressure during extensive durations of sleep loss; Cajochen et al . , 2002 ) rose more rapidly and was consistently higher than in WT mice during 24 hr of forced wakefulness ( Figure 4D ) . Combined , these results demonstrate that overexpressing Bmal1 in the skeletal muscle renders mice less sensitive to the effects of sleep loss , portending muscle as a potential therapeutic target for sleep loss . How might peripheral tissues such as muscle influence sleep ? The rapidly emerging area of muscle-derived factors on systemic health provides a potential model for our findings . In particular , there are several examples of muscle-derived factors that alter brain processes . Notably , overexpression of PGC-1α ( peroxisome proliferator-activated receptor gamma coactivator 1α , gene symbol: Ppargc1a ) selectively in mouse skeletal muscle reduces the depressive phenotypes induced by stress by preventing plasma kynurenine from reaching the brain ( Agudelo et al . , 2014 ) . Furthermore , PGC-1α activation stimulates release of the muscle-derived peptide irisin into circulation ( Boström et al . , 2012 ) . Plasma irisin , in turn , induces BDNF expression in the hippocampus ( Wrann et al . , 2013 ) . Indeed , PGC-1α expression is rhythmic in skeletal muscle ( Liu et al . , 2007 ) raising the possibility that alterations in Bmal1 expresion/function may alter rhythmic PGC1α expression through a change in clock function . However , the studies presented here are not sufficient to determine if the sleep effects of loss/gain of Bmal1 function in skeletal muscle are via core clock or non-clock mechanisms . Furthermore , whole body deletions of other circadian factors such as Per1 and Per2 ( Shiromani et al . , 2004 ) , and Cry1 and Cry2 ( Wisor et al . , 2008 ) , do not have similar effects on sleep . Other potential contributors could be related to changes in muscle metabolism as Bmal1 metabolic phenotypes have been reported in both muscle mouse-lines used here ( Harfmann et al . , 2016; Brager et al . , 2017 ) . Recent studies have highlighted that sleep disruptions in humans are associated with peripheral circadian desynchrony ( Cedernaes et al . , 2015; Schroder and Esser , 2013 ) . The current study demonstrates that manipulating levels of the circadian transcription factor Bmal1 specifically in skeletal muscle alters sleep . Moreover , a majority of these effects of Bmal1 on sleep are not dependent on circadian timing in the brain—dependence on circadian timing in the skeletal muscle remains a possibility . Although it has been established that sleep is important for skeletal muscle function ( for a review , see Chase , 2013 ) , these investigations are the first to implicate molecular processes within skeletal muscle in signaling sleep regulatory mechanisms in the brain . Studies in our lab are currently underway to determine the nature of the pathway skeletal muscle uses to signal sleep regulatory mechanisms in the brain . All mice in the brain lines and muscle rescued/overexpressed lines were maintained on a 12-hr light:12-hr dark schedule throughout the study . Food and water were available ad libitum , and animals ( 10–12 w of age ) were individually housed for at least 2 weeks prior to experimental use . All protocols and procedures were approved by the Morehouse School of Medicine Institutional Animal Care and Use Committee . Bmal1 brain-rescued and brain overexpressed mice used the tetracycline transactivator ( tTA ) system , which requires two transgenes for expression of the target gene Bmal1 ( previously reported in McDearmon et al . , 2006; RRID:MGI:3714773 ) . The promoter sequence of the secretogranin II gene ( Scg2 ) , which is expressed exclusively in the brain , drives expression of the tetracycline transactivator ( tTA ) . The tTA protein binds to the tetracycline operator ( tetO ) sequence and drives expression of Bmal1 cDNA . The double transgenic mice were crossed onto a Bmal1 knockout background ( RRID:IMSR_JAX:009100 ) to create brain-rescued mice and were crossed onto a Bmal1 WT background to create brain overexpressed mice . Breeding was conducted in-house and genotypes were recorded as they became available from breeding . For rescue lines , animals from approximately 20–25 litters comprised the entire dataset . WT littermates obtained from these litters were kept separate for comparisons in each line . KO mice were offspring from independent crosses of heterozygous Bmal1 KO’s . In situ hybridization studies demonstrate that Scg2 mRNA is found throughout the brain with the higher expression in the hypothalamus and peak expression in the SCN . Moderate expression of Scg2 is detected in midbrain and hypothalamic nuclei of the ascending arousal system and in the sleep promoting ventrolateral preoptic area ( Lein et al . , 2007 ) . The mice used in the present study , constructed with a 9 . 8 kb promoter region of Scg2 , were characterized previously ( Hong et al . , 2007 ) . Briefly , Scg2::tTA mice express the tTA transcript broadly in the brain and is enriched in the SCN , when assessed by both an oligo to the tTA transcript and by crosses with tetO-promotor-linked reporter lines ( Hong et al . , 2007 ) . Furthermore , in situ hybridization in Scg2::tTA X tetO::Bmal1-HA and tetO::Bmal1-HA mice demonstrate that Bmal1 expression is under strict control of the tetracycline transactivator: Hemagglutinin ( HA ) -tag expression is only detectable in the double transgenic mouse . Both Bmal1 mRNA and protein are constitutively expressed in the brain of brain-rescued mice . ( McDearmon et al . , 2006 ) . Brain specificity has been demonstrated by western blot which shows an absence of HA-staining in both the muscle and liver of double transgenic mice ( McDearmon et al . , 2006 ) . Bmal1 muscle-rescued and muscle-overexpressed mice were generated with the use of a DNA construct consisting of human α-actin-1 promoter sequence positioned upstream of Bmal1 ( RRID:MGI:3714769; McDearmon et al . , 2006 ) . The transgenic mice were crossed onto a Bmal1 knockout background to create muscle-rescued mice , and also crossed onto a Bmal1 WT background to create brain-overexpressed mice . The skeletal muscle-specific Cre-recombinase mouse ( Acta1-cre/Esr1 , RRID:IMSR_JAX:025750 ) was generated in house ( McCarthy et al . , 2012a ) . Breeding with the floxed Bmal1 mouse ( Bmal1lox/lox , The Jackson Laboratory , RRID:IMSR_JAX:007668; Storch et al . , 2007 ) generated the inducible muscle knockout mouse . These offspring ( Bmal1lox/lox; Acta1-cre/Esr1* ) allow selective deletion of the bHLH domain of Bmal1 in skeletal muscle upon tamoxifen administration . The Acta1 promoter used for both mouse lines ( muscle rescued/overexpressed and muscle KO ) is a 2 . 2 kb sequence directly upstream from the human skeletal actin ( Acta1 gene ) translational start site . Proper developmental and tissue-specific expression has been verified previously using an Acta1::CAT mouse line . These findings included a demonstrated lack of expression in the brain ( Brennan and Hardeman , 1993 ) . Specific Cre-dependent excision of a loxP-flanked gene in mouse striated muscle fiber was also demonstrated in mice expressing Cre recombinase under the control of this same promoter ( Miniou et al . , 1999 ) . Transgene expression in the inducible muscle knockout mice ( Acta1-cre/Esr1* ) used here is constitutive and was not detectable in brain by western blot ( McCarthy et al . , 2012b ) . A lack transgene expression in the brain has also been demonstrated for the Acta1::Bmal1-HA line ( McDearmon et al . , 2006 ) . We verified this finding by western blotting an entire brain hemisphere or gastrocnemeous muscle in mice bred at our facility ( Figure 1—figure supplement 1 ) . HA-tag was detected in skeletal muscle , but not brain . EEG and EMG electrodes were implanted in anesthetized mice . A prefabricated head mount ( Pinnacle Technologies , KS ) was used to position three stainless-steel epidural screw electrodes . The first electrode ( frontal—located over the frontal cortex ) was placed 1 . 5 mm anterior to bregma and 1 . 5 mm lateral to the central suture , whereas the second two electrodes ( interparietal—located over the visual cortex and common reference ) were placed 2 . 5 mm posterior to bregma and 1 . 5 mm on either side of the central suture . The resulting two leads ( frontal-interparietal and interparietal-interparietal ) were referenced contralaterally . A fourth screw served as a ground . Electrical continuity between the screw electrode and head mount was aided by silver epoxy . EMG activity was monitored using stainless-steel Teflon-coated wires that were inserted bilaterally into the nuchal muscle . The head mount ( integrated 2 × 3 pin grid array ) was secured to the skull with dental acrylic . Mice were allowed to recover for at least 14 days before sleep recording . One week after surgery , mice were moved to the sleep-recording chamber and connected to a lightweight tether attached to a low-resistance commutator mounted over the cage ( Pinnacle Technologies ) . This enabled complete freedom of movement throughout the cage . Except for the recording tether , conditions in the recording chamber were identical to those in the home cage . Breeding was conducted in-house , and genotypes were recorded as they became available from breeding . Recordings from approximately 20–25 litters makup the dataset . All wildtype and KO littermates resulting from a litter , up to a maximum of 50% , were recorded with tissue-specific knockout/rescue mice . Mice were allowed a minimum of 7 additional days to acclimate to the tether and recording chamber . Recording of EEG and EMG waveforms began at zeitgeber time ( ZT ) 0 ( light onset ) . Data acquisition was performed on a personal computer running Sirenia Acquisition software ( Pinnacle Technologies ) , a software system designed specifically for polysomnographic recording in rodents . EEG signals were low-pass filtered with a 40 Hz cutoff and collected continuously at a sampling rate of 400 Hz . After collection , all waveforms were classified by a trained observer ( using both EEG leads and EMG ) as wake ( low-voltage , high-frequency EEG; high-amplitude EMG ) , NREM sleep ( high-voltage , mixed-frequency EEG; low- amplitude EMG ) or rapid eye movement ( REM ) sleep ( low-voltage EEG with a predominance of theta activity [6–10 Hz]; very low amplitude EMG ) . EEG epochs determined to have artifact ( interference caused by scratching , movement , eating , or drinking ) were excluded from analysis . Recordings where artifact comprised more than 5% of total recording time were excluded from analysis . Analysis of NREM delta power and NREM spectral distribution was accomplished by applying a fast Fourier transformation to raw EEG waveforms . Only epochs classified as NREM sleep were included in this analysis . Delta power was measured as spectral power in the 0 . 5 to 4 Hz frequency range and expressed as a percentage of total spectral power in the EEG signal ( 0 . 5–100 Hz ) during that time period . Homeostatic challenge: Six-hour forced wakefulness was conducted in all mouse lines ( Figure 2 ) . Following a 24 hr baseline recording , mice were sleep deprived during the first 6 hr of the light phase ( ZT 0–6 ) by gentle handling ( introduction of novel objects into the cage , tapping on the cage and when necessary delicate touching ) and allowed an18-hr recovery opportunity ( ZT 6–0 ) . Twenty-four-hour forced wakefulness: Following a 24 hr baseline recording , Bmal1 muscle-overexpressed lines and WT littermates were moved to a slowly rotating wheel ( nine inches in diameter; 1 rpm ) adjacent to the recording cage ( Figure 3 ) . Mice were confined to this wheel for 24 hr beginning at ZT 0 ( lights on ) during which time they had free access to food and water . Following sleep deprivation , animals were returned to the baseline recording cage and EEG acquisition was continued for a 24 hr recovery opportunity . Mice were sacrificed by CO2 inhalation at ZT 6 ( ZT0 represented lights-on under LD ) after 6 hr of forced wakefulness . Brains were immersion-fixed in 4% paraformaldehyde for 24 hr then sunk in 30% sucrose ( 24 hr at 4°C ) . Cryostat sections ( 40µm-thick ) were incubated with a rabbit polyclonal IgG antibody ( c-fos; Santa Cruz Biotechnology , Santa Cruz , CA; chicken polyclonal IgY antibody [ChAT]; Novus Biologicals , Littleton , CO ) and immunoreactivity was visualized using Vectastain Elite ABC kit with 3 , 3-diaminobenzidine tetrahydrochloride ( DAB ) as chromagen ( Vector Labs , Burlingame , CA ) . Sections were mounted with permount , and Fos expression was quantified using ImageJ ( National Institutes of Health , Bethesda , MD ) . Counts of immunostained nuclei were undertaken in the mid-to-posterior region of each brain site . Methods were described previously ( McCarthy et al . , 2012b ) . Briefly , brain ( hypothalamic ) and skeletal muscle ( gastrocnemius ) tissues were collected from mice sacrificed at ZT5 and ZT17 ( ZT12 was lights-off under a 12 hr: 12 hr light:dark cycle ) ; these are the mid-points of peak and trough Bmal1 gene expression in skeletal muscle ( McCarthy et al . , 2007 ) . Total RNA was extracted from frozen brain and skeletal muscle using Trizol ( Invitrogen , Carlsbad , CA ) and diluted to 0 . 1 mg/ml . Samples were converted from RNA to cDNA with Applied Biosystems RT-PCR kit reagents according to the manufacturer’s instruction . Real-time PCR assays were performed using the comparative amplification detection threshold of target gene expression ( CT ) method . mRNA levels detected with SYBR Green ( Bio-Rad; Hercules , CA ) were measured by determining the cycle number at which CT was reached . In each sample , CT was normalized to Gapdh expression ( ΔCT ) performed on the same plate . Normalized gene expression ( ΔΔCT ) for each gene with respect to genotype and time point was computed with Bio-Rad CFX Manager . Brain hemispheres were homogenized in a microfuge tube using a pellet pestle in 700 ul lysis buffer ( 20 mM Hepes pH 7 . 6 , 400 mM NaCl , 1 mM EDTA , 5 mM NaF , 0 . 3% Triton-X 100 , 5% glycerol , 1 mM DTT , 250 nM PMSF , and complete protease inhibitor mix ( Sigma ) , and tumbled at 4C overnight prior to quantification and loading on the gel . Muscle proteins were extracted similarly , except 300 µl to lysis buffer was used , and the lysis buffer contained 1% Triton X-100% and 10% glycerol . Protein concentrations were determined using a BCA assay kit ( Pierce ) , and separated on an Any-kDa minigel ( BioRad ) . Western blot was performed with anti-HA-HRP conjugated monoclonal antibody ( Roche ) and anti-Gapdh ( Santa Cruz ) . Sleep data were analyzed using one-way analysis of variance ( ANOVA ) , repeated-measures ANOVA or Student’s t . Significance was defined as p≤0 . 05 . Post hoc analysis was conducted using Tukey’s HSD method or student t-test where indicated . Tukey’s HSD method uses the studentized range statistic and maintains family-wise error-rate at 0 . 05 . An appropriate sample size of 5 was predicted with Type I error rate of 0 . 05 and Type II error rate of 0 . 2 . Standard deviation and mean difference were estimated as 46 . 4 and 100 min , respectively , based on the existing literature ( Laposky et al . , 2005 ) and previous studies in our lab . Sample sizes ( biological replicates ) for each experiment are indicated in the figure legends .	We spend nearly one third of our lives asleep . Sleep plays a critical role in human health and is regulated by multiple brain regions . Genes are some of the factors that control sleep . Recent studies have shown that mice in which a gene called Bmal1 had been completely removed , sleep more than mice that still have the gene . These Bmal1-deficient mice also respond differently to sleep loss . However , until now , it was not known which tissues and cells that carry active ( or ‘expressed’ ) Bmal1 are involved in regulating sleep . To find out if Bmal1 activity in the brain is sufficient to recover from sleep loss , Ehlen , Brager et al . compared genetically modified mice that either expressed Bmal1 only in the brain , or only in the muscle tissue that covers the skeleton . After the mice were kept awake for six hours , their sleep was monitored by measuring electrical signals on the surface of the skull . Contrary to what they expected , Ehlen et al . found that mice with Bmal1 expressed in the skeletal muscle were able to have a normal sleep pattern , while mice with Bmal1 expressed in the brain had an abnormal sleep pattern . Further experiments show that removing Bmal1 from the skeletal muscle of mice , but allowing the gene to be expressed in other tissues , produced sleeping patterns that were similar to those seen in mice that were completely missing the Bmal1 gene . These results indicate that Bmal1 in skeletal muscle is important to help regulate sleep , and that the signal for sleepiness does not only originate from the brain . This is the first study to show that skeletal muscle can regulate sleep . The next step will be to identify the specific signal the muscle uses to trigger the brain to sleep . Understanding the mechanisms that regulate sleep may help to develop new treatments for sleep disorders .	[ "Abstract", "Introduction", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2017	Bmal1 function in skeletal muscle regulates sleep
Motor coordination is supported by an array of highly organized heterogeneous modules in the cerebellum . How incoming sensorimotor information is channeled and communicated between these anatomical modules is still poorly understood . In this study , we used transgenic mice expressing GFP in specific subsets of Purkinje cells that allowed us to target a given set of cerebellar modules . Combining in vitro recordings and photostimulation , we identified stereotyped patterns of functional synaptic organization between the granule cell layer and its main targets , the Purkinje cells , Golgi cells and molecular layer interneurons . Each type of connection displayed position-specific patterns of granule cell synaptic inputs that do not strictly match with anatomical boundaries but connect distant cortical modules . Although these patterns can be adjusted by activity-dependent processes , they were found to be consistent and predictable between animals . Our results highlight the operational rules underlying communication between modules in the cerebellar cortex . Long-range connectivity between brain areas and the basic organization of microcircuits has been widely studied in many brain structures ( Shepherd and Grillner , 2010 ) . However , the synaptic connectivity at the mesoscale level , that is , the precise functional synaptic arrangement between cortical modules , has been described in few brain areas since module boundaries are often ill defined . In the rodent barrel cortex and visual cortex ( Callaway and Katz , 1993; Briggs and Callaway , 2001; Shepherd et al . , 2003 ) , anatomical modules contain a wide array of neuronal types that interconnect within complex neuronal circuits ( Harris and Mrsic-Flogel , 2013; Cossell et al . , 2015 ) and reveal inter-modular communication that might influence the output of individual modules ( e . g . see multi-whisker stimuli; Estebanez et al . , 2012 ) . However , whether neighboring modules strictly follow the same functional and cellular organization rules or display subtle differences in local architecture or neurochemical expression that determine module-specific function or processing rules is not established . In the cerebellar cortex , where anatomical modules are well defined and are associated with different features of motor control , the crystalline structure of the cortex initially led to the concept that different functions arose through distinct patterns of input/output connectivity as observed in the barrel cortex with a given vibrissa targeting one specific barrel ( Ito , 1984 ) . However , much anatomical and functional evidence suggests that cerebellar modules consist of an array of microcircuits , each of which has a specific identity ( Scott , 1963; Lange et al . , 1982; Brochu et al . , 1990; Dehnes et al . , 1998; Mateos et al . , 2000; Wadiche and Jahr , 2005; Sarna et al . , 2006; Shin et al . , 2009; Kim et al . , 2009; Xiao et al . , 2014; Zhou et al . , 2014 ) . Cerebellar modules are defined by the topographical organization of climbing fiber ( CF ) inputs to Purkinje cells ( PCs ) , originating in the inferior olive , and PC to cerebellar nuclei connections . Olivo-cortical and cortico-nuclear projections delimit parasagittal zones and subzones of the cerebellar cortex called ‘microzones’ having similar CF receptive fields ( Oscarsson , 1979; Voogd and Glickstein , 1998; Sugihara and Shinoda , 2004; Voogd and Ruigrok , 2004; Buisseret-Delmas and Angaut , 1993; De Zeeuw et al . , 2011; Voogd , 1967 ) . Thus , a microzone corresponds to the cortical element of a cerebellar module . PCs from a given microzone receive and integrate information from specific parts of the body and project to restricted areas of the cerebellar nuclei , which in turn send the cerebellar computation to the cerebral cortex , brainstem or spinal cord ( Voogd and Glickstein , 1998 ) . The complete olivo-cortico-nuclear loop defines the cerebellar module ( Figure 1A; Figure 1—figure supplement 1; for review see Ruigrok , 2011 ) . Mossy fiber ( MF ) inputs , the second major excitatory pathway of the cerebellum , partially overlap with the receptive fields of CFs when originating from the same body part ( Garwicz et al . , 1998; Pijpers et al . , 2006; Voogd et al . , 2003 ) . Therefore , it has been suggested that the cerebellar cortex is subdivided into individual microzones each controlling a given set of muscles ( Apps and Garwicz , 2005; Thach et al . , 1992 ) . Some boundaries between different microzones can be identified by a family of neurochemical markers , the zebrins , selectively expressed in PCs ( Figure 1—figure supplement 1 ) ( Apps and Hawkes , 2009; Brochu et al . , 1990 ) . The level of expression of some of these markers has been associated with specific synaptic properties and intracellular transduction pathways at the parallel fiber ( PF ) -PC or CF-PC synapses ( Wadiche and Jahr , 2005; Paukert et al . , 2010; Cerminara et al . , 2015; Hawkes , 2014 ) leading to the description of different intrinsic properties of PCs and different rules for plasticity induction between zebrin positive and negative bands . Combined with the regional differences in the cytoarchitecture of the cerebellar cortex ( for review see Cerminara et al . , 2015 ) , these findings demonstrate that the cerebellar cortex is composed of an ensemble of heterogeneous modules . Therefore , an appealing hypothesis would be that cell identity relies on its position in the cerebellar cortex and that specific communication rules between neighboring microzones might arise from this anatomical and neurochemical heterogeneity . Intermodular communication might involve granule cells ( GCs ) , which relay the MF input within the cerebellar cortex . GCs have long transverse axons , the PFs , that make glutamatergic synapses with hundreds of PCs across several microzones ( Harvey and Napper , 1991; Thach et al . , 1992 ) . PFs could thus define an associative pathway that enables MF inputs from one original microzone to communicate with a distant one . Paradoxically , several studies have shown that PCs are excited by a unique set of GCs localized in the same microzone as themselves ( Cohen and Yarom , 1998; Bower and Woolston , 1983; Brown and Bower , 2001; Isope and Barbour , 2002 ) . These results suggest that local GCs provide the major input to PCs and that the majority of GC-PC synapses are silent ( Ekerot and Jorntell , 2001; Isope and Barbour , 2002 ) . In contrast , other studies demonstrated that PCs can be driven by a restricted subset of GCs belonging to a different microzone ( Ekerot and Jörntell , 2003; Dean et al . , 2010; Jörntell and Ekerot , 2002; Ekerot and Jorntell , 2001 ) . Finally , recent studies have suggested that PCs can be excited by a broad range of GCs ( Walter et al . , 2009 ) and that feedforward inhibition through the MF-GC-molecular interneuron pathway might play a role in the restriction of receptive fields in PCs ( Cramer et al . , 2013; Santamaria et al . , 2007 ) . Finally , this wide range of functional connectivity might explain the regional differences in information processing ( Cramer et al . , 2013 ) underpinned by the molecular and cellular heterogeneity between individual microzones and by the specific topographical organization of MF and CF inputs in the cerebellar cortex . We therefore investigated the functional synaptic organization of the excitatory MF-GC-PC pathway . In order to understand how information is processed in individual and adjacent microzones , we systematically mapped the functional synaptic organization between GC and PCs , molecular interneurons ( MLIs ) and Golgi cells ( GoCs ) , targeting one set of identified cortical microzones in cerebellar modules of lobules III and IV of the anterior lobe . We describe here an activity-dependent stereotyped and predictable modular organization of GC inputs mediated by PFs , which pinpoints the operational rules of information processing in cerebellar modules . In order to address the hypothesis of specific functional synaptic communication between cerebellar microzones , we developed an experimental protocol that allowed us to target microzones in the medial parts of lobules III and IV , which receive inputs from the proximal hindlimb and forelimb . This region of the cortex is involved in the adaptive control of posture and locomotion; also , CF and MF inputs have already been well described in previous studies ( Ji and Hawkes , 1994; Voogd and Ruigrok , 2004; Sugihara and Shinoda , 2004; Garwicz et al . , 1998; Andersson and Oscarsson , 1978; Matsushita et al . , 1991; Matsushita and Tanami , 1987; Matsushita , 1988; Yaginuma and Matsushita , 1987 ) . Microzones are defined by their CF and cortico-nuclear projections ( Figure 1—figure supplement 1 ) ( Andersson and Oscarsson , 1978; Oscarsson , 1979; Voogd and Ruigrok , 2004 ) . Since CF projections share common boundaries with the zebrin II pattern of expression in PCs ( Apps and Hawkes , 2009; Sugihara and Shinoda , 2004 ) , experiments were performed in the EAAT4-GFP strain of mice , in which zebrin II bands are defined by the expression of eGFP in PCs ( Figure 1—figure supplement 1; Materials and methods ) ( Gincel et al . , 2007 ) . The use of two distinct specific antibodies against zebrin II/aldolase C ( Materials and methods ) confirmed that EAAT4-eGFP mice express GFP in zebrin II bands ( Figure 1—figure supplement 1 ) . By comparing EAAT4-eGFP/zebrin II expression and cortico-nuclear projections , we postulated that seven putative microzones extend to the P2+ band on both sides of the midline ( Figure 1—figure supplement 1 ) . So that the description of the cerebellar afferents could be completed , bilateral MF projections were reconstructed after the in vivo unilateral injection of AAV2/1-GFP viruses ( N = 4 animals ) or fluoro-ruby ( N = 2 animals ) into the external cuneate nucleus , which relays inputs from the forelimbs , or AAV2/1-GFP viruses into the lumbar spinal cord ( L3–L5; N = 1 animal ) , which conveys inputs from the hindlimbs ( Figure 1B , C; Figure 1—figure supplement 2; Materials and methods ) . Projections from these two precerebellar nuclei were mutually exclusive in lobules III and IV , identifying an alternation of two to three parasagittal bands ( mean width = 193 ± 24 µm , Figure 1C , D; n = 13 slices , N = 4 animals for cuneate projections; n = 4 slices , N = 1 animal for spinal projections ) in the mediolateral axis . Intensity plot profiles of GFP labeling in the GC layer ( Materials and methods ) allowed us to determine the boundaries of individual MF bands ( Figure 1D ) . Notably , when measuring the GFP level of expression , 22% of cuneate projections were found in the contralateral vermis , illustrating the importance of bilateral MF inputs . Interestingly , the comparison of the zebrin II pattern with MF inputs identified overlapping parasagittal bands ( Figure 1D ) ( Ji and Hawkes , 1994 ) with one shared boundary at the P2+/P2− transition . Also , MF inputs subdivided the P1− band into smaller regions . Finally , CF and MF ( Figure 1D ) inputs delimit 13 regions having a unique combination of inputs between P2+ipsi and P2+contra . These regions might therefore correspond to smaller processing units than microzones ( Ozden et al . , 2009; Tsutsumi et al . , 2015 ) . Using this anatomical description , we then targeted individual microzones in acute transverse cerebellar slices , and determined the spatial organization of GC excitatory inputs to PCs , GoCs and MLIs . 10 . 7554/eLife . 09862 . 003Figure 1 . Identification of anatomical cortical microzones by the zebrin II bands ( A ) Diagram of two cerebellar modules . One module is composed of a cortical microzone ( light red area ) , the target area of Purkinje cells ( PC , red ) in the cerebellar nuclei ( black and white cells ) and a group of olivary cells ( gray ) sending their climbing fibers ( CF , gray line ) to the cortical microzones and cerebellar nuclei . Mossy fibers ( MFs , green lines ) send sensorimotor information to the cerebellar nuclei and the microzones . The parallel fibers ( PFs ) , the axon of granule cells ( GCs ) , cross several microzones belonging to different modules . The red double arrow between the two modules illustrates intermodular communication . GoC: Golgi cell; MLI: molecular interneuron; filled triangles indicate excitatory synapses; empty triangles indicate inhibitory synapses . ( B ) Diagram illustrating AAV2/1-GFP or fluoro-ruby injection ( green pipettes ) sites in the external cuneate nucleus ( ECu ) and in segment L3–L5 of the spinal cord ( SC ) . The blue double arrowhead line indicates the localization of the coronal section shown in panel C . ( C ) Coronal section across lobule III of the cerebellar cortex showing GFP fluorescence in MF rosettes following viral injection in the external cuneate nucleus ( Ecu-MFs , green ) , aligned with anti-aldolase C/zebrin II ( ZII , red ) immunostaining . White dotted lines highlight positive zebrin II bands ( P1+ , P2− and b+ from the midline ) . ( D ) Upper panel , superimposed intensity plot profiles of the molecular layer ( red ) and granule cell layer ( green ) section shown in C illustrating bilateral MF projections . Upper labels , positive and negative zebrin bands . Black dotted line , midline . Lower panel , summary of MF inputs projection pattern in vermal lobule III from external cuneate ( 13 slices/4 animals ) and spinal cord ( 4 slices/1 animal ) compared to zebrin II bands ( two positive bands: P1+ , P2+ in red; and two negative bands: P1− and P2− ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 00310 . 7554/eLife . 09862 . 004Figure 1—figure supplement 1 . Zebrin bands as an accurate positioning tool . ( A ) Immunolabeling against aldolase C/zebrin II ( red ) , neurogranin and GlyT2 ( both in blue ) in the vermis of lobule IV from four different animals . Neurogranin and GlyT2 label Golgi cells ( GoCs ) . Note the stereotyped zebrin band pattern . ( B ) The EAAT4-GFP transgenic mice express GFP in zebrin II-positive bands . Top: immunolabeling for aldolase C/zebrin II ( red ) in a slice from an EAAT4-GFP mouse . Middle: GFP fluorescence in the same slice . Bottom: overlay . Scale bar , 100 μm . ( C ) Diagram illustrating the topographical organization of climbing fiber ( CF ) inputs to the cerebellar cortex ( green boxes refer to inferior olive subregions ) and the partial match with the zebrin band patterning ( red boxes refer to zebrin bands ) and the cortico-nuclear projection to cerebellar nuclei ( blue boxes refer to nuclear subregions ) . The olivo-cortico-nuclear loop defines the cerebellar module . Note that P1− is split onto two subregions . The array of CF projection in the cerebellar cortex and the cortico-nuclear projections define the microzone . MAOc: caudal medial accessory olive; Sub a: subnucleus a of the MAOc; Sub b lat . interm . : intermediate part of the lateral subnucleus b of the MAOc; Sub b lat . caudal: caudal part of the lateral subnucleus b of the MAOc; Sub a caudal: caudal part of subnucleus a of the MAOc . Ax and A zones refer to the zonal nomenclature of the olivo-cortical pathway ( Voogd , 1967 ) . FN/ICG: fastigial nucleus and interstitial cell group; FN/LVN: fastigial nucleus and lateral vestibular nucleus . Adapted from Sugihara , 2011 and Voogd and Ruigrok , 2004 . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 00410 . 7554/eLife . 09862 . 005Figure 1—figure supplement 2 . Mossy fiber projections from the external cuneate nucleus and spinal cord ( segment L3-L5 ) . ( A ) Section from the cerebellum and the brainstem after AAV2/1 injection of viruses into the external cuneate nucleus . Inset: mossy fiber rosettes in lobule III . Zebrin band boundaries are shown in red . Right panel: diagram illustrating the position of the external cuneate injections ( ECu ) . Bottom right: example of injection site in the external cuneate nucleus and infected neurons . ( B ) Section from the cerebellum after injection of AAV2/1 viruses into the spinal cord . Inset: mossy fiber rosettes in lobule III . Zebrin band boundaries are shown in red . Right: injection site and infected neurons in the spinal cord ( SC ) ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 005 We choose to specifically study monosynaptic excitatory inputs from GCs to other cell types of the cerebellar cortex , since PFs can cross many microzones and communicate information over a long distance ( Pichitpornchai et al . , 1994; Harvey and Napper , 1991 ) . Individual PCs ( n = 49 cells , N = 18 animals ) were whole-cell patch-clamped in P1+ , P1− and P2+ zebrin II bands of EAAT4-GFP mice , and RuBi-glutamate was systematically uncaged in the GC layer using point-scan laser photostimulation ( Figure 2A , B; width mapped = 664 µm centered on the recorded cell; Materials and methods ) on both sides of the midline while inhibitory transmission was blocked . As uncaging led to a compound synaptic current in PCs initiated by trains of action potentials in GCs ( Figure 2B , C; Figure 2—figure supplement 1; Materials and methods ) , the efficacy of GC patches in eliciting synaptic current in PCs was assessed by calculating the mean synaptic charge expressed as a Z-score ( Materials and methods; Figure 2—figure supplement 2; Figure 2C , D ) . This representation allowed us to compare GC input maps between PCs with different background noise . Since MFs from the external cuneate and the spinal cord project to the entire depth of the granule cell layer ( Figure 1C ) , glutamate uncaging sites at various depths of the GC layer but at the same mediolateral position were pooled and the evoked maximal synaptic input was considered . This maximal response was used as an indicator of the connectivity strength between the GCs and the PC along the mediolateral axis ( Figure 2B , D ) . Interestingly , patterns of GC inputs in PCs were found to be highly heterogeneous ( Figure 2D ) , with dense local inputs ( mean local charge = 10 . 3 ± 13 . 8 pC , n = 49 ) and silent sites , that is , no GC patch had a Z-score > 3 . 09 at any depth of the GC layer ( 38% of sites were silent; Z-score < 3 . 09; Materials and methods ) . Remotely connected GC patches separated by silent sites were always found and were also frequently observed on contralateral sites ( mean charge in distant patches = 6 . 1 ± 8 . 8 pC . Mean distance from the local peak = 215 ± 86 µm ) . In five cells out of 49 , no obvious local inputs were observed although distal inputs were present . Altogether , these findings demonstrate that while local GCs make a strong connection with PCs most of the time , exceptions may occur . Also , GCs located at several hundred micrometers from the recorded PC can make strong connections via their PFs , while neighboring patches can be silent , indicating that information originating in distant microzones can have a strong influence on a given PC . 10 . 7554/eLife . 09862 . 006Figure 2 . Granule cell input patterns to Purkinje cells reveal hotspots of connectivity . ( A ) Experimental design and simplified diagram of the cortical microcircuit . Purkinje cell ( PC ) synaptic inputs were recorded . Photorelease of RuBi-glutamate at multiple locations of the granule cell ( GC ) layer ( blue squares ) mimics GC ( black ) activation by mossy fibers ( MFs; green ) . GCs contact PCs , Golgi cells ( GoCs; orange ) and molecular interneurons ( MLIs; blue ) along the mediolateral axis . Inhibition is blocked and climbing fibers ( CFs ) are not activated . PF: parallel fibers . ( B ) Example of a PC ( red ) recorded in an EAAT4-GFP acute cerebellar slice and filled with biocytin . The recorded cell was reconstructed and located using both GFP expression ( not shown ) and aldolase C immunolabeling ( blue ) . The PC in this example is located in the P2+ zebrin band . Blue squares indicate uncaging sites . The size of the square is proportional to the synaptic charge of the evoked current . Mediolateral response is given by the strongest response at all depths of the GC layer ( i . e . maximal response in the white dotted box ) . Maximal responses along the mediolateral axis are reported in the gray area and define the connectivity pattern . The blue dashed box indicates the width of the photostimulation field . Please note the two scales: at the bottom is the distance from the recorded cell , while at the top is the distance to the midline . ( C ) Examples of evoked currents , from panel B . The blue bar indicates uncaging duration . The red area indicates the measured charge ( window = 200 ms ) . ( D ) The connectivity pattern was expressed as the Z-score of charge , as a function of the distance to the cerebellar midline . The significance threshold was defined at Z = 3 . 09 . Red areas are considered as functionally connected , while black areas indicate silent sites . Error bars illustrate the median from five mappings . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 00610 . 7554/eLife . 09862 . 007Figure 2—figure supplement 1 . Controls for photostimulation . ( A ) Normalized direct current extent in Purkinje cells ( PCs ) when the spot of illumination ( blue dots ) is moved through PC dendrites ( green ) ; illumination step: 20 μm . ( B ) Normalized direct current extent in granule cells ( GCs ) when the spot of illumination is moved through GC dendrites ( step: 20 μm ) . ( C ) Normalized number of spikes elicited in GCs when the spot of illumination is moved through GC dendrites ( step: 20 μm ) . ( D ) Normalized direct current response in GCs as a function of focal plane depth in slices . ( E ) Evaluation of independence between neighboring sites . Site 2 evoked a current in a recorded PC , but not sites 1 or 3 , although they are direct neighbors . All three sites are alternatively stimulated to test for a possible spread of glutamate between sites or desensitization of glutamate receptors on site 2 . ( F ) Distal silent sites might be due to a tilt in the slice . Slice angle was verified using Lugaro cell axons that parallel GC axons in Glyt2-GFP mice in order to rule out this hypothesis . ( G ) Left panel: a responding site was repetitively photostimulated to test the stability of recorded responses throughout mappings . Right panel: measured synaptic charge versus time . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 00710 . 7554/eLife . 09862 . 008Figure 2—figure supplement 2 . Z-score representation of the granule cell input map in one recorded cell . ( A ) Initial connectivity maps were built by determining the synaptic charge elicited by granule cells at a given site . ( B ) In each trial , light-evoked responses ( at 200 ms ) and noise ( at 800 ms ) were measured and a histogram of the noise was built for each cell . The mean and standard deviation of the noise were determined and used to calculate the Z-score at each site ( C ) using the following equation: Z-score = ( mean of synaptic charge at a given site − mean of noise distribution in the cell ) /standard deviation of the distribution of noise . A Z-score of 3 . 09 , corresponding to a significance level of 0 . 001 , was chosen to define significant and silent sites . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 008 The connectivity map of one single PC might not account for the full population of PCs in a given microzone . Also , a single group of GCs is unlikely to influence a microzone by targeting a single PC . Therefore , if microzones communicate together via the PFs , PCs belonging to a given microzone should share several groups of functionally connected GCs from a distant microzone . We tested this hypothesis by recording pairs of neighboring PCs having the same zebrin identity ( Figure 3 ) and determined their connectivity map . In the example shown in Figure 3A , B , strong GC inputs were observed from local GCs , in P2+ ipsilateral and in P2+ contralateral bands in both PCs , while no or few significant inputs originated in the P1− band . Although some strongly connected GC sites were specific to one PC of the pairs , we found a significant correlation between patterns of GC inputs for eight out of nine pairs of adjacent recorded PCs ( Figure 2C; r = 0 . 74 ± 0 . 14 , n = 9 pairs , N = 5 animals ) , indicating that in a given animal , PCs belonging to the same microzone and sharing similar MF inputs also displayed similar GC synaptic input maps . These findings suggest that neighboring PCs make powerful connections with a specific set of distant microzones while ignoring others . 10 . 7554/eLife . 09862 . 009Figure 3 . Neighboring Purkinje cells share similar granule cell input patterns . Two neighboring Purkinje cells ( PCs ) were simultaneously recorded and RuBi-glutamate was systematically uncaged . ( A ) Map of the recorded synaptic charge measured in PC1 ( white cell , top ) and PC2 ( white cell , bottom ) . Most of the responding and silent sites were observed at the same location or close by , although a few differences can be observed . ( B ) Corresponding mediolateral granule cell ( GC ) connectivity pattern to PC1 ( top panel ) and PC2 ( bottom panel ) expressed as a Z-score of the synaptic charge . PC positions are indicated in white . ( C ) Site by site correlation of GC connectivity patterns between neighboring PC pairs ( black dots , r = 0 . 74 ± 0 . 14; n = 8 pairs ) . Shuffled pairs showed no correlation ( gray dots , r = 0 . 02 ± 0 . 09 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 009 To investigate whether functional patterns of GC connectivity are specific to small groups of neighboring PCs in a given animal or are common features of the communication rules between cerebellar microzones , we set out to compare GC input patterns between groups of PCs belonging to different animals , but at the same location . PCs were grouped based on their position in the mediolateral axis between the P1+ and P2+ bands of lobule III/IV ( group width = 100 µm; shift between two groups = 10 µm; N = 49 PCs from 18 mice; Materials and methods ) . In each PC group , the median pattern was determined ( Materials and methods ) and correlated with the median pattern of all the other groups of PCs . A correlation matrix was then built using the Pearson coefficient ( r ) as a marker of correlation between GC input patterns ( Figure 3A; Figure 4A; Materials and methods ) . Strikingly , the correlation matrix revealed four clusters of PC groups displaying a highly correlated GC connectivity map ( Figure 4A; clusters 1–4 ) . Cluster identity was verified using a spectral co-clustering algorithm that simultaneously clusters rows and columns of a matrix without positional assumption ( Materials and methods ) . The four clusters found were all centered on the diagonal line indicating that correlated PCs are neighbors . They are identified on the correlation matrix by dashed blue squares ( Figure 4A ) . These results strongly argue that PCs from different animals but at the same specific location in the mediolateral axis ( i . e . the vermal part of lobule III/IV ) have selected similar GC sites . Notably , none of these clusters clearly matches with a set of anatomical boundaries as illustrated by the comparison with zebrin II bands and MF inputs from the external cuneate and the spinal cord ( green rectangles , Figure 4A ) indicating that boundaries of the functional organization between microzones differ from a pure anatomical description . 10 . 7554/eLife . 09862 . 010Figure 4 . Granule cell input patterns to Purkinje cells from different animals define stereotyped clusters . ( A ) Correlation matrix of groups of neighboring Purkinje cells ( PCs; N = 49 cells in 18 animals ) . Each group corresponds to the median granule cell ( GC ) input pattern of PCs located within a 100 µm window . Two consecutive groups are shifted by 10 µm . The x/y axes represent the center of each median pattern . All median patterns were compared to each other , and correlation was estimated using the Pearson coefficient . Four clusters of contiguous , correlated group of cells were identified using a co-clustering algorithm ( blue dotted boxes , numbered 1 to 4 from the midline to P2+ ) . Zebrin II bands , external cuneate mossy fiber input ( ECu MFs ) and spinal cord MF input ( SC MFs ) locations are indicated in green , for comparison . ( B ) Median GC input patterns to the four clusters of PCs identified in A . The PC cluster position is represented as a white bar . Black: non-connected areas , Z-score < 3 . 09; red: significant connections , Z-score > 3 . 09 . Error bars are in light gray . Local GC inputs ( dark red ) were usually observed , although weak for PCs belonging to cluster 2 ( see black cross ) . Distal hotspots mentioned in the main text are indicated with upwards arrows . Note the systematic hotspot in P2+ . Silent regions mentioned in the main text are indicated with downwards arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 010 We then determined the median pattern of GC inputs for each of the four identified cluster of PCs ( clusters 1 , 2 , 3 and 4 with n = 26 , n = 14 , n = 4 and n = 5 cells , respectively; Figure 4B ) . PCs defining cluster 1 displayed strong connections with local GCs and GCs located below ipsilateral and contralateral P2+ but did not display any inputs from a large part of contralateral P1− . The P2+/P2− boundary also clearly identifies limits for cluster 2 that is characterized by weak local inputs . Also , in this group of PCs , a narrow contralateral band of GCs located 200 µm from the midline was always strongly connected . Contrasting with these two clusters , PCs that extend between 200 and 270 µm from the midline ( cluster 3 ) were found to be broadly and strongly connected ipsilaterally by GCs spanning all zebrin positive and negative bands . Finally , PCs from the fourth cluster , located essentially in P2+ , were contacted by local GCs and GCs located in medial P1− and P2− bands . As for the four identified clusters of correlated PCs , no systematic link between GC hotspots and anatomical boundaries has been observed . However , it should be noted that for all PC groups , strong GC connections were always found in ipsilateral P2+ while a large area of the P1− band was silent . These findings showed that the highly heterogeneous patterns of GC connectivity are stereotyped and associated with four identified clusters of PCs , suggesting that a distinct functional organization linking distant regions might be superimposed on the anatomical microzonal framework . In the cerebellar cortex , GCs also contact local interneurons , MLIs and GoCs , which respectively perform feedforward inhibition onto local PCs and feedback inhibition onto GCs . We therefore tested whether GC connectivity patterns to GoCs and MLIs displayed similar spatial functional properties . GC inputs to GoCs were examined using GlyT2-EAAT4-GFP mice ( Figure 5A; Materials and methods ) . As opposed to PCs , patches of GCs connected to GoCs ( n = 19 cells ) were always found in the vicinity of the recorded cell . Within the mapped area , we found that 81% of GC synaptic inputs to GoCs were made by local GCs as assessed by the histogram of the median Z-score ( Figure 5A ) , for GoCs recorded in any of the four clusters of PCs identified ( Figure 5—figure supplement 1 ) . Since dendritic gap junctions might have shunted and filtered GC inputs located in distant apical GoC dendrites ( Vervaeke et al . , 2012 ) , potentially resulting in undetectable current at the soma , one subset of GoCs ( n = 6 ) was recorded with the gap junction blocker carbenoxolone ( 100 µM ) in the bath . No differences were observed in these mappings , and all cells were pooled . The pattern of GC inputs was centered on GoCs with a mean extension of 114 . 1 ± 70 . 1 µm , which falls within the average range of the reconstructed GoC axonal plexus ( 214 ± 30 µm , n = 10; inset Figure 5A ) . Neither the connectivity map nor the extension of the axonal plexus were limited by zebrin boundaries ( Figure 5—figure supplement 1 ) as opposed to GoC apical dendrites ( Sillitoe et al . , 2008 ) . Thus , unlike GC inputs to PCs , excitatory inputs to a given GoC appear to be mostly restricted to GCs that can be targeted by the local axon of the same GoC , confirming that they implement a local inhibitory feedback circuit in the GC layer ( Cesana et al . , 2013 ) . Occasionally , small distal inputs were observed . 10 . 7554/eLife . 09862 . 011Figure 5 . Spatial organization of granule cell ( GC ) input patterns to molecular interneurons and Golgi cells is distinct from GC input maps to Purkinje cells . ( A ) Median granule cell ( GC ) input pattern to Golgi cells ( GoCs; Z-score of charge ) for GoCs located at the same location as Purkinje cells ( PCs ) from cluster 1 . Black: Z-score <3 . 09 , orange: Z-score >3 . 09 . Error bars are in light gray . Inset: overlay of all 3D-reconstructed GoCs , showing the extension of the axonal plexus , that is , the maximal region in which GoCs could inhibit GCs ( same scale as the median pattern ) . ( B ) Median GC input pattern to molecular interneurons ( MLIs; Z-score of number of excitatory post-synaptic currents , EPSCs ) at the same position as PCs from cluster 1 . The upward arrow indicates a hotspot of GCs contacting MLIs located in the cluster 1 region , but not PCs . Inset: example of EPSCs recorded following photostimulation . Stars indicate detected EPSCs . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 01110 . 7554/eLife . 09862 . 012Figure 5—figure supplement 1 . Spatial organization of granule cell input patterns to Golgi cells recorded in the area of cluster 2 , 3 and 4 for Purkinje cells . Upper panel: histogram of the median Z-score for Golgi cells ( GoCs ) recorded in clusters 2 , 3 and 4 of Purkinje cells ( PCs ) . Lower panel: reconstruction of GoC recorded in P1− close to the P1+ boundary . Note that the axon crosses two zebrin band boundaries . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 012 We then compared GC connectivity to MLIs ( n = 7 cells , N = 4 animals ) for cells recorded in the area of cluster 1 for PCs ( Figure 5B ) . In MLIs , patterns of GC connectivity were measured using the number of events ( expressed as a Z-score; Materials and methods ) instead of the charge , as this measurement was less sensitive to background noise in this set of experiments . Similarly to GC-PC and GC-GoC connectivity , local GC inputs were identified . However , several patches of distant GC connections were also found in the histogram of the median Z-score of synaptic events ( Figure 5B; local peak , 80 µm from midline , two ipsilateral peaks , 195 µm and 320 µm from midline , and one contralateral peak , 135 µm from midline ) . Strikingly , some hotspots did not overlap perfectly with patches of GCs that elicit strong inputs in PCs from cluster 1 ( Figure 5B , see arrow and compare with PC from cluster 1 in Figure 4 ) , indicating that for PCs and MLIs located in the same area , different groups of GC inputs can be selected . These findings suggest that GCs can relay MF information to PCs through two distinct pathways , the MF-GC-PC and MF-GC-MLI-PC pathways . Since the GC connectivity maps to PCs , MLIs and GoCs strictly match neither with the topography of the CF inputs nor with the boundaries of the MF projections , these heterogeneous and stereotyped functional maps might arise from a combination of the topographical organization of MF and CF projections and an activity-dependent synapse selection through long-term plasticity ( long-term depression [LTD] and long-term potentiation [LTP] ) . In a set of PCs from cluster 1 , we therefore tested whether GC connectivity maps could be altered by a protocol known to induce plasticity , either LTP or LTD ( Coesmans et al . , 2004; Hartell , 1996 ) . After producing a first series of GC input maps ( Figure 6A ) , we applied an electrical stimulation ( 1 Hz stimulation/5 min ) to a large number of PFs in the molecular layer ( mean evoked current in PCs = 1285 ± 500 pA , n = 12 which correspond to around 130 non-silent PFs; Isope and Barbour , 2002 ) , and resumed the mapping procedure for at least 15 min . The initial averaged map was then compared site by site to the averaged updated map following stimulation ( Figure 6A-figure supplement 1 ) . After plasticity induction , 86 sites ( 22%; n = 8 cells ) displayed a significant modification in synaptic charge ( △Z-score > 3 . 09 or < −3 . 09; Figure 6A-figure supplement 1; Materials and methods ) , while no effect was observed in the remaining sites . As already observed , this protocol induced postsynaptic LTP ( green squares , Figure 6A; Coesmans et al . , 2004 ) or LTD ( blue squares , Figure 6A; Hartell , 1996 ) at the GC-PC synapse , indicating that we stimulated different PF beams converging on the recorded PC . All these changes were blocked by a combination of drugs that prevented the induction of plasticity ( Figure 6—figure supplement 1 ) . A negative correlation was observed ( slope 0 . 31 and r = 0 . 38 ) between the initial synaptic weight of the GC site ( high Z-score ) and the sign of the effect after induction , with stronger connections leading to LTD while weaker connections undergo LTP ( Figure 6—figure supplement 1 ) . Indeed , selecting strong connections ( Z-score > 6 . 5 ) and determining the averaged time course for all these sites led to a mean depressed charge of 27% after plasticity induction . Conversely , of the 56 sites ( 14% of total sites ) that were potentiated , 33 were silent in the initial map ( white ‘x’ in green squares in Figure 6A; Figure 6—figure supplement 1 ) . This suggests that previously undetectable synaptic connections were awakened and that the overall connectivity map can be modified by activity . Indeed , we compared the histograms of the median Z-score of charge of this set of PCs before and after plasticity induction ( Figure 6B ) . Although a few percent of the total number of PFs crossing the dendritic tree of the PCs had been stimulated , a new distant region , initially silent , became significantly connected to the PCs belonging to cluster 1 , demonstrating that the connectivity map is adjustable ( see * in the panel ‘Difference’ in Figure 6B ) . Therefore , functional microzones may communicate through the selection of GC-PC synapses in a specific set of PC clusters . 10 . 7554/eLife . 09862 . 013Figure 6 . Tunable maps of granule cell inputs to Purkinje cells . ( A ) Photostimulation grid superimposed on a reconstructed slice before ( top ) and after ( bottom ) the long-term synaptic plasticity induction protocol ( 300 stimulations at 1 Hz ) . The stimulation pipette was positioned in the molecular layer far from the recorded Purkinje cell ( PC ) . Zebrin II bands were confirmed by anti-aldolase C immunolabeling ( blue ) and are highlighted by a white dotted line . Recorded PCs are shown in red . Connected sites are in red , silent sites in black , depressed sites in blue , and potentiated sites in green . Awaken sites are indicated with a white cross . Traces illustrate examples of synaptic currents recorded in three sites showing different plasticity ( same color code ) . ( B ) Median granule cell ( GC ) input patterns to PCs were computed before ( top panel ) and after the plasticity induction protocol ( middle panel ) for a group of PCs from cluster 1 ( n = 7 ) . A difference between GC input patterns ( bottom panel ) identified a newly connected region . ( C ) Silent sites can be specifically awakened . Silent sites were identified in an initial map and RuBi-glutamate was uncaged repetitively ( one flash every 3 s for 5 min , blue bar ) at these sites . Right panel , time course of evoked responses before and after induction ( blue ticks indicates photostimulations ) . Individual responses ( n = 5 cells ) are indicated as circles . Average responses are indicated by squares . Non-detectable connections are in black , while significant connections are in red . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 01310 . 7554/eLife . 09862 . 014Figure 6—figure supplement 1 . Tunable maps of granule cell inputs to Purkinje cells . ( A ) Variation of granule cell ( GC ) synaptic charge ( in ΔZ-score ) per site plotted against the initial Z-score of synaptic charge ( n = 8 cells ) . Linear regression fit ( gray dotted line ) illustrates the correlation between the initial response and the post-induction response . ( B ) Top left: time course of the averaged Z-score of synaptic charge for silent GC sites ( Z-score < 3 . 09 ) before and after the induction protocol ( yellow arrow ) . Top right: time course of the averaged Z-score of synaptic charge for strongly connected sites ( Z-score > 6 . 5 ) . Bottom left: same experiment as in ( A ) , with antagonists in the bath ( n = 5 cells; see Experimental procedures ) . Bottom right: same experiment as in ( A ) , but with no stimulation ( n = 7 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09862 . 014 In order to assess whether silent sites can be specifically awakened at any position in the mediolateral axis , the induction protocol was reproduced by uncaging glutamate specifically on silent patches of GCs ( chosen at random distances from the recorded cell ) at 1 Hz for 5 min ( n = 5 cells; Figure 6C ) . In all cells tested , GC inputs became detectable after the induction protocol ( mean △charge from −0 . 4 ± 0 . 33 to −1 . 27 ± 1 pC ) . These findings demonstrate that the functional pattern of GC inputs to PCs results from the active tuning of synapses . We identified highly heterogeneous patterns of GC excitatory inputs to PCs and MLIs highlighting hotspots of connected GCs in distant microzones . Hotspots were frequently observed at several hundred micrometers from the recorded PC or MLI . At first glance , these findings appear to be in contradiction with previous studies ( Harvey and Napper , 1991; Walter et al . , 2009 ) . Walter et al . ( 2009 ) demonstrated that GC inputs are evenly distributed in the mediolateral axis although local connections were slightly stronger . However , as patterns of GC inputs were aligned on PC somas , positional information could not be considered in this study . Indeed , removing the positional information from our dataset reproduced these findings ( Figure 7—figure supplement 1 ) , indicating that the identification of the organization of GC input patterns is position specific . Therefore , whether we define a microzone by the topography of the CF inputs or the combination of CF and MF inputs ( Figure 1 and Figure 1—figure supplement 1B ) , our results strongly argue that PFs communicate excitatory MF information between distant microzones . Because inhibition was blocked in our experiments , we have to acknowledge that the GC input patterns of connectivity described here do not describe the net excitatory/inhibitory balance affecting PC dendrites . However , although no prediction about PC discharges can be made , the GC connectivity map onto MLIs observed in Figure 5 suggests that PCs can be potentially inhibited by GCs from specific locations ( Figure 7 ) . Indeed , several studies demonstrated that feedforward inhibition participates in controlling the firing rate of PCs in individual microzones ( Dizon and Khodakhah , 2011; Santamaria et al . , 2007 ) and prevents the activation of distant PCs . Nonetheless , our data provide the first description of a conserved and predictable functional synaptic organization between one cortical layer – the granule cell layer – and the output stage of the cerebellar cortex – the Purkinje cell layer . The preferred organization of GC inputs to PCs and MLIs ( in lobules III and IV of the anterior vermis ) among different animals suggests that microzones may have specific roles in the context of the behavioral tasks associated with this cerebellar region , notably during adaptive locomotion ( Apps and Garwicz , 2005; Dean et al . , 2010 ) . Our results are in agreement with in vivo studies demonstrating that PCs are activated by a restricted group of distant GCs ( Ekerot and Jörntell , 2001; 2003 ) , although in these studies , local GC inputs were never observed . There are several possible explanations for this discrepancy: ( 1 ) the different areas of the cerebellum studied ( i . e . the C3 zone rather than medial lobules III and IV ) might have specific operational rules . Indeed , other groups studying other regions of the cerebellar cortex have demonstrated that local GCs provide the majority of inputs for the underlying PCs ( Bower and Woolston , 1983; Cohen and Yarom , 1998; Isope and Barbour , 2002 ) , for example in Crus I and II . Furthermore , in a recent in vivo study , while restricted patches of PC activation were observed in Crus II following forelimb stimulation , beam-like PC excitation elicited by PFs from distant GC inputs were identified in Crus I ( Cramer et al . , 2013 ) . ( 2 ) In vivo , the receptive fields were defined by modulation of the PC firing rate . This parameter could have been misleading since a null net effect could still have masked correlated modifications of excitatory and inhibitory inputs . Notably , local GCs activate both PCs and MLIs in our experiments , suggesting that the resulting PC discharge is difficult to predict . In vivo , excitation or inhibition could be favored depending on the subset of GCs activated in a given context . Because similar patterns of GC inputs to PCs , GoCs and MLIs were found in the different animals tested , a major issue was to assess whether this organization was the result of a genetically encoded developmental program or if it could be modified by activity . We demonstrated that PF stimulation could induce either LTD at individual sites ( Hartell , 1996 ) or LTP at individual sites ( Coesmans et al . , 2004 ) indicating that activity-dependent processes can tune GC input maps . Such variability in the sign of the plasticity is due to the fact that the stimulation pipette sampled a limited set of PFs constituting a small proportion of the total number of PFs in the lobule , but belonging to many different GC patches . This explains why only 22% of GC patches were affected by plasticity . At the center of the stimulated PF beam , focal synaptic inputs to PC dendrites have likely evoked large calcium transients , similar to CF inputs ( Hartell , 1996 ) , and thus induced LTD at these specific GC-PCs synapses . Focally evoked calcium transients could have been amplified at GC-PC synapses with a high synaptic weight , as stronger depolarization would have been elicited . Although focal stimulation in the molecular layer may appear non-physiological , a recent study demonstrated in vivo that sensory inputs can lead to clustered activation of PFs ( Wilms and Häusser , 2015 ) . At the periphery of the stimulation beam , sparse activation of PFs might have induced moderate calcium transients leading to LTP ( Coesmans et al . , 2004 ) . Interestingly , new connections appeared systematically when we used a plasticity-inducing protocol at silent sites , demonstrating that quiescent synapses can be awakened following GC activation . These findings explain the appearance of a new peak of connected GCs in the median histogram of the Z-score of charges after plasticity induction . Indeed , plasticity occurred at almost every position in the mediolateral axis ( see difference in Figure 6B ) , but at silent sites a strong bias toward LTP was observed ( Figure 6—figure supplement 1 ) . Because of the high efficiency of the stimulation protocol to awaken silent synapses ( Figure 6B ) , a strong alteration of the GC connectivity map was likely to occur . Further experiments using specific MF stimulation will be required to assess whether such modifications in GC input maps are observed in physiological conditions . However , our results clearly demonstrate that the heterogeneity of the spatial map is an active process . These results are in agreement with previous in vivo studies ( Jörntell and Ekerot , 2002; Schonewille et al . , 2010; Gao et al . , 2012; Badura et al . , 2013 ) , suggesting that distant microzones are associated through PFs by LTP mechanisms . LTP could favor information transfer between a set of GCs from one microzone and a group of PCs from another microzone . Conversely , LTD , which is physiologically induced by the conjunction of PF and CF inputs to PCs , could limit the communication between these microzones . We have demonstrated here that PFs can link distant cortical regions that do not always match with known zebrin band markers or MF inputs . Notably , neither the PC clusters nor the GC spatial patterns of connections strictly match with the topography of the spinocerebellar or cuneate MF inputs carrying information from the hindlimb and forelimb , indicating that GC patches cannot simply be identified by the anatomical organization of a single source of MFs , and might combine MF information from different sources , as suggested in two recent studies ( Chabrol et al . , 2015; Huang et al . , 2013; but see also Bengtsson and Jorntell , 2009 ) . Discrepancies between anatomical and functional boundaries have also been shown using in vivo calcium imaging of CF signaling in PCs following sensory stimulation ( Tsutsumi et al . , 2015 ) . Since P2+ bands receive specific MF inputs from the hindlimb ( Jörntell et al . , 2000 ) , the functional organization observed in our experiments might illustrate the importance of connecting cortical microzones controlling similar muscles in both limbs . The term microzone was originally associated with both the description of the fine anatomical CF projection and the concept of unitary functional unit in the cerebellar cortex ( Oscarsson , 1979 ) . We described here clusters of neighboring PCs and MLIs having similar GC input connectivity maps that might represent individual functional units . However , they differ from the original anatomical microzone definition as they can span multiple zebrin bands . Therefore , a functional microzone likely computes information from different sources of MF inputs and frequently belongs to two adjacent zebrin bands as also proposed by ( Graham and Wylie , 2012 ) . Since zebrin bands also define an array of biochemical markers that can have an important role in synaptic transmission , signal integration and plasticity ( for example EAAT4 transporter , PLCbeta3 and 4 , IP3R1 and mGluR1a receptors follow a specific zebrin pattern of expression ) ( Cerminara et al . , 2015; Hawkes , 2014; Paukert et al . , 2010; Wadiche and Jahr , 2005 ) , a functional microzonal unit might then combine several toolboxes in order to process incoming information . In light of the stereotyped organization of GC input patterns to PCs and MLIs , and the local restriction of the GC inputs to GoCs , we postulate that the area between P2+ ipsi- and contralateral bands could be composed of 10–13 functional units with a mean extension of 70–100 µm , as already suggested by in vivo imaging studies ( Ozden et al . , 2009; Tsutsumi et al . , 2015; Schultz et al . , 2009 ) . In the posterior lobe , zebrin bands and cerebellar inputs have a different topographical organization , notably in the vestibular lobules ( Voogd , 2011 ) , but a modular system has also been described suggesting that the communication between microzones through PFs is a general mechanism involved in information processing in the cerebellum . However , whether conserved and stereotyped patterns across animals is the rule in the posterior lobe needs to be demonstrated . In Crus I/II , animal-specific processing linked to vibrissa tactile discrimination might lead to unique communication rules , as sensory discrimination relies on individual history . In conclusion , we demonstrated that cortical functional microzones of the cerebellar cortex communicate via the PFs and underlie an important level of coordination between cerebellar modules . These properties will favor the synchronization of PCs within specific modules and improve information readout by nuclear cells ( Person and Raman , 2012a; 2012b ) . MF projections from the cuneocerebellar and dorsal spinocerebellar tracts were labeled by in vivo injection of viruses or fluoro-ruby into the external cuneate nucleus ( N = 6 ) and the lumbar region of the spinal cord ( N = 1 ) , respectively . Male CD1 mice ( P28–P35 ) were injected with recombinant adeno-associated viral particles ( rAAV2/1 , 2 . 8×1012 GU/ml , N = 5 ) carrying cDNA for GFP expression under the CMV promoter , or with dextran tetramethylrhodamin ( 10 000 MW , fluoro-ruby , N = 2 ) . Mice were anesthetized with an intraperitoneal injection of a mixture of ketamine ( 100 mg/kg ) , medetomidine ( 1 mg/kg ) and acepromazine ( 3 mg/kg ) . For external cuneate targeting , stereotaxic injections were performed using the coordinates AP: −7 . 4 mm; Lat: 1 . 4 mm; DV: 3 . 3 mm from bregma . The virus or the dextran was loaded into a graduated pipette equipped with a piston for manual injections ( Wiretrol II , Drummond Scientific Company , Broomall , USA ) . By applying gentle pressure , final volumes of 1 . 2 µl and 0 . 75 µl were delivered into the spinal cord and the external/cuneate nucleus , respectively , at an approximate speed of 250 nl/min . The pipette was left in place for at least 10 min after injection for virus diffusion . Spinal injections were achieved by inserting the pipette between adjacent vertebrae in the lumbar region . After 2–11 weeks of recovery , injected mice were sacrificed by transcardiac perfusion of paraformaldehyde 4% and cerebellar slices ( 50 µm ) were prepared for subsequent immunohistochemistry and analysis . Zebrin bands were identified using a monoclonal ( 1/100 dilution; gift of Richard Hawkes , Calgary ) ( Brochu et al . , 1990 ) or polyclonal ( 1/50 dilution; gift from Izumi Sugihara , Japan ) ( Sugihara and Shinoda , 2004 ) antibody against aldolase C . Intensity plot profiles were generated after confocal imaging in individual slices by measuring the intensity of the MF signal in the granular layer and the intensity of aldolase C/zebrin II labeling in the molecular layer of lobule IV . Measurements were performed using ImageJ software . For group data , MF intensity plot profiles were aligned to the P1+ band ( indicating the midline ) . In four out of six animals injected in the external cuneate nucleus , the injection site extended into the entire lateral and dorso-ventral axes and 350–400 µm into the antero-posterior axis ( Figure 1—figure supplement 2 ) of the nucleus . Cortical projections in lobules III and IV were measured in 13 different slices in these four animals . Injection in the lumbar segments of the spinal cord ( L3–L5; 1 . 2 µl ) resulted in an infected area of ≈1 cm in length centered around L3–L5 ( see Figure 1—figure supplement 2 ) . Cortical projections were measured in four different cerebellar slices . Slices were prepared from P17–P90 male CD1 EAAT4-GFP or EAAT4-GlyT2-GFP mice . This strain was obtained by crossing EAAT4-GFP mice with a line expressing GlyT2-GFP ( gift from H . U . Zeilhofer ) ( Zeilhofer et al . , 2005 ) , in which some GoCs express GFP . Mice were anesthetized by inhalation of isoflurane , and then killed by decapitation . Transverse slices were prepared as previously described ( Valera et al . , 2012 ) . The cerebellum was dissected out and placed in cold artificial cerebrospinal fluid ( ACSF ) bubbled with carbogen ( 95% O2 , 5% CO2 ) , containing ( in mM ) : NaCl 120 , KCl 3 , NaHCO3 26 , NaH2PO4 1 . 25 , CaCl2 2 . 5 , MgCl2 2 , glucose 10 and minocycline 0 . 00005 ( Sigma-Aldrich , USA ) . Then 300 µm-thick transverse slices were prepared ( Microm HM 650V , Microm , Germany ) in potassium-based medium , containing ( in mM ) : K-gluconate 130 , KCl 14 . 6 , EGTA 2 , HEPES 20 , glucose 25 , minocycline 0 . 00005 and D-AP5 0 . 05 ( Sigma-Aldrich ) . After cutting , slices were soaked in a sucrose-based medium at 34°C , containing ( in mM ) : sucrose 230 , KCl 2 . 5 , NaHCO3 26 , NaH2PO4 1 . 25 , glucose 25 , CaCl2 0 . 8 , MgCl2 8 , and minocyclin 0 . 00005 ( Sigma-Aldrich ) and maintained in a water bath at 34°C in bubbled ACSF . Whole-cell patch-clamp recordings in voltage-clamp mode were obtained using a Multiclamp 700B amplifier ( Molecular Devices , USA ) and acquired with WinWCP 4 . 2 . x freeware ( John Dempster , SIPBS , University of Strathclyde , UK ) . Pipette resistance was 3–4 MΩ for PCs , 6–8 MΩ for GoCs and 10 MΩ for MLIs . Series resistance was monitored and compensated ( 80%–90% typically ) in all experiments , and cells were held at −60 mV . The internal pipette solution contained ( in mM ) : CsMeSO4 135 , NaCl 6 , HEPES 10 , MgATP 4 and Na2GTP 0 . 4 . pH was adjusted to 7 . 3 with KOH and osmolarity was set at 300 mOsm . Biocytin ( Sigma Aldrich ) and neurobiotin ( Vector Laboratories , USA ) were added ( 1 mg/ml each ) for cell reconstruction . Voltages were not corrected for the liquid junction potential , which was calculated to be 9 mV ( i . e . the membrane potential was 9 mV more hyperpolarized than reported ) . We accepted recordings for which the inward current at −60 mV did not exceed 1 nA for PCs and 250 pA for other cells . Synaptic currents in PCs were low-pass filtered at 2 kHz , then sampled at 20–50 kHz . All recorded cells were located in lobule III or IV . All experiments were performed at 34°C using the same bubbled ACSF . We blocked inhibitory transmission and NMDA , adenosine , CB1 , GABAB and mGluR1 receptors to limit the modulation of excitatory post-synaptic current ( EPSC ) amplitude by activity-dependent activation of these receptors . They were respectively blocked using ( in mM ) : picrotoxin 0 . 1 , strychnine 0 . 001 , D-AP5 0 . 05 ( Ascent Scientific , Abcam Inc ) , DPCPX 0 . 0005 , AM251 0 . 001 , CGP52432 0 . 001 and JNJ16259685 0 . 002 ( Tocris-Cookson , UK ) . An initial set of mappings was obtained during plasticity induction protocols . A glass pipette was positioned in the upper half of the molecular layer , 500 µm away from the recorded PC , and 300 stimulations were performed at 1 Hz in current-clamp mode . After a resting period of 5 min , photostimulation was resumed and the entire mapping was repeated at least four times . Perfusion contained picrotoxin , strychnine , CGP52432 and AM251 at the concentrations described above . For plasticity experiment in Figure 6C , the following internal solution was used ( in mM ) : K-gluconate 136 , KCl 4 , MgCl2 1 , HEPES 10 , Na2ATP 4 , NaGTP 0 . 4 , sucrose 16 , pH 7 . 3 , and inhibition was blocked with bicuculine ( 20 µM ) . Uncaging experiments were performed using bath-applied RuBi-glutamate ( 100 µM , Ascent Scientific ) , as described by Fino et al . , 2009 . We used the point-scan mode of a confocal microscope ( FV300 , Olympus , Japan ) for illumination , mounted with a diode-pumped solid-state blue laser ( 20 ms and 30 mW laser pulses at 473 nm [CrystaLaser , USA] , through a 20× objective [Olympus , Japan] ) . The beam of blue light ( 473 nm ) was controlled by a set of mirror galvanometers and focused using a low-aperture objective ( 20×/0 . 5 NA ) . This optical arrangement was designed to generate an almost cylindrical beam that could penetrate the slice , as assessed by the constant synaptic charge recorded when the focal plane of the photostimulation was increased to a depth of 100 µm in the slice ( Figure 2—figure supplement 1D ) . Mapping was carried out using a software-based point-scan stimulation ( PAPP , Fluoview 300 ) at identified positions , with a step of 41 . 5 µm every 3 . 5 s . The stability of the photostimulated current was assessed ( Figure 2—figure supplement 1G ) . In each experiment , the recorded cell was placed at the center of the field and the slice was positioned with the PC layer aligned with the X-axis of the grid . Grid extension was 332 µm on each side of the recorded cell above the GC layer . A range of 34–85 sites was sampled several times for each cell ( mean number of mappings/cell = 5 . 1 ± 2 . 3 ) . No sites were stimulated with frequencies higher than 0 . 008 Hz . RuBi-glutamate uncaging was used to study the functional excitatory synaptic connectivity of the cerebellar cortex . This compound was chosen because it incurs fewer non-specific effects than MNI-glutamate ( Fino et al . , 2009 ) during quantitative laser-scanning photostimulation ( Shepherd et al . , 2005 ) . Since these caged compounds partially block GABAergic currents , GC inhibitory transmission was blocked . The diffusion of uncaged RuBi-glutamate during laser stimulation was estimated by recording currents evoked in PC dendrites in the horizontal plane in 20 µm steps ( Figure 2—figure supplement 1A ) . Action potentials were blocked with tetrodotoxin ( TTX , 1 µM ) . Because of the thickness of the PC dendritic tree itself , direct stimulation of glutamate receptors yielded an upper limit for the size of the spot ( half-width = 33 . 0 ± 1 . 8 µm , n = 9 ) . Since the lateral extension of GC dendrites could increase the responsive area , we also estimated this extent by recording GCs both in loose cell-attached mode ( measuring evoked action potentials , the half-width of evoked action potentials was 45 . 5 ± 10 . 9 µm , n = 7 ) and in whole-cell mode ( measuring evoked currents , the half-width of direct stimulation was 59 . 6 ± 2 . 1 µm , n = 7 ) while uncaging glutamate in10 µm steps ( Figure 2—figure supplement 1B , C ) . The average maximum number of evoked action potentials at the center of the spot was 19 . 6 ± 6 . 6 ( n = 7 ) indicating that even GC-PC connections with a very low probability of release could evoke a current to the recorded PCs ( Isope and Barbour , 2002 ) . The distance between two photostimulated sites ( 41 . 5 µm ) was chosen to ensure that neighboring sites were almost independent , and that all parts of the granular layer was sampled ( Figure 2—figure supplement 1 ) . We estimated that each beam could illuminate several hundred GCs per 10 µm of slice thickness . A burst of action potentials was triggered in this population ( Figure 2—figure supplement 1C ) , where glutamate uncaging elicited currents of up to 10 pC . Since the synaptic weight of a single GC-PC connection is approximately 0 . 1 pC ( Isope and Barbour , 2002 ) , tens of active GC connections ( depending on the depth of the GC layer in the slice ) could be stimulated simultaneously , attesting to the efficacy of the photostimulation protocol . We estimated the impact of the focal plane on GC responses . GC inputs were recorded in PCs while the focal plane was changed in 10 µm steps . The average current was not significantly different between focal planes ( Figure 2—figure supplement 1D ) . Since the detection of distal connected GCs could be affected by the inclination of the slice , we used GFP-positive Lugaro cell axons , which run parallel to the PF , from EAAT4-GlyT2-GFP slices and determined granule cell layer slope ( Figure 2—figure supplement 1F ) . A total of 130 axons were traced in 13 slices after the experiment , and the mean deviation from the horizontal axis was 0 . 85 ± 0 . 82 deg , suggesting that the absence of distal responses was not due to a tilt in the slice . Measurements and analyses were performed using Python software written in-house and open source software , OpenElectrophy ( http://neuralensemble . org/OpenElectrophy/ ) ( Garcia and Fourcaud-Trocmé , 2009 ) . Data were stored in SQL databases . Synaptic charges evoked by photostimulation of granule cells were detected during a window of 0–200 ms after illuminationFigure 2—figure supplement 2 . The standard deviation of the background noise was detected as follows: peak amplitude was detected using a 200 ms window at the end of each trace , then the histogram of noise was built and its standard deviation was determined . The Z-score of synaptic charges for each site was calculated using the following equation: Z-scoresite = ( mean synaptic charge at one site − mean charge of the noise ) / ( standard deviation of the noise ) . A Z-score of 3 . 09 , corresponding to a significance level of 0 . 001 , was chosen to define significant and silent sites . This conservative choice of threshold was chosen in agreement with visual inspection of the traces , to ensure that any evoked current was not simply due to spontaneous activity . Although glutamate uncaging in the GC layer did not evoke direct currents in PCs , the direct stimulation of GoC somata or basolateral dendrites occurred at some sites . The slow direct current was filtered using a median filter applied to each individual trace . Evoked synaptic currents are much faster and were slightly affected . The MLI Z-score was determined using the number of individual EPSCs evoked by glutamate uncaging in a time window of 200 ms after uncaging compared to the number of individual EPSCs in a distant time window . After the recordings , slices were fixed in 4% paraformaldehyde and zebrin bands were identified using a monoclonal ( gift from Richard Hawkes , Calgary ) ( Brochu et al . , 1990 ) or polyclonal ( Izumi Sugihara , Japan ) ( Sugihara and Shinoda , 2004 ) antibody against aldolase C . Biocytin/neurobiotin-filled cells were visualized with a streptavidin Alexa Fluor conjugate ( Thermo Fisher Scientific , USA ) . In order to compared reconstructed cell positions between animals , the average width of zebrin bands in lobules III and IV was determined ( average P1+ width = 48 . 5 ± 13 . 6 µm , average P1− width = 275 . 6 ± 70 . 3 µm , average P2+ width = 64 . 6 ± 19 . 3 µm and average P2− width = 445 . 2 ± 60 . 5 µm , n = 89 ) and cells were positioned within these coordinates . A 3D reconstruction of GoCs was obtained using the free software Vaa3d ( Peng et al . , 2014 ) . In each mediolateral position of the GC layer , the site with the maximum Z-score value was used to build spatial patterns ( Figure 2B , C ) . For paired correlation and median Z-score histograms , spatial patterns were convolved , using a triangular kernel ( half-width = 18 µm ) . The correlation matrix was built as follows: the representative pattern of GC inputs at each position was given by the median Z-score of synaptic charges for a group of PCs in a range of 100 µm . A full set of median patterns was generated by shifting the averaging window by 10 µm . The Pearson correlation was calculated between each couple of median patterns , and the reported location in Figure 4A corresponds to the center of the median pattern . Biclustering was performed using spectral co-clustering from the sklearn Python library ( Abraham et al . , 2014 ) . In brief , the algorithm rearranges the rows and the columns of the correlation matrix to make biclusters contiguous without assumption . If the order of rows and columns is not changed , it indicates that the initial data order already presented a clustered organization . Means are reported with standard deviations , while error bars in figures represent SEMs or MADs ( median absolute deviation ) in cases where the mean or the median , respectively , was used . Unless otherwise stated , statistical tests used were the non-parametric Mann-Whitney U and Spearman rank order tests . Correlations were calculated using the Pearson coefficient .	The human brain is essentially organised into modules made from groups of connected neurons . A part of the brain at the back of the head , called the cerebellum , is organised into particularly well-defined modules and is important for fine-tuning and learning new movements . However , it is not clear if the pattern of connections between modules in the brain is broadly the same across different animals from the same species , or if it varies between individuals . This is partly because it has not been possible to identify the same cells across different individual animals , and thus measure whether the connections are different . Valera et al . used genetically engineered mice that produced a fluorescent marker in specific subsets of easily identifiable neurons in the cerebellum known as Purkinje cells . These groups of labeled Purkinje cells helped identify the boundaries of modules in the cerebellum . Valera et al . then stimulated another group of neurons , called the granule cells , that connect to Purkinje cells from several different modules . Measuring the resulting changes in the electrical activity of the Purkinje cells revealed the pattern of connections between these two types of neurons . These experiments showed that granule cells can connect Purkinje cells from distant modules and that neighbouring Purkinje cells often display similar connection patterns . Valera et al . also found that Purkinje cells from different animals but at the same location in the cerebellum had similar patterns of connections with granule cells . This suggests that , in identified modules , there is little variability in connection patterns between individuals . However , the patterns of connectivity could be altered by processes related to learning , indicating that they can be organized in different ways and still work . Further experiments then showed that different individuals display consistent patterns of connectivity between granule cells and other major cell types in the cerebellum . This finding suggests that while the basic modules in the cerebellum are similar across different individuals , the connections between modules can be shaped by experience . An important challenge for the future is now to understand how the connected modules combine the information they receive and drive the output of the cerebellum . This knowledge will help better describe how movements are controlled .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Stereotyped spatial patterns of functional synaptic connectivity in the cerebellar cortex
The tumor suppressor protein Scribble ( SCRIB ) plays an evolutionary conserved role in cell polarity . Despite being central for its function , the molecular basis of SCRIB recruitment and stabilization at the cell cortex is poorly understood . Here we show that SCRIB binds directly to the CH1 domain of β spectrins , a molecular scaffold that contributes to the cortical actin cytoskeleton and connects it to the plasma membrane . We have identified a short evolutionary conserved peptide motif named SADH motif ( SCRIB ABLIMs DMTN Homology ) which is necessary and sufficient to mediate protein interaction with β spectrins . The SADH domains contribute to SCRIB dynamics at the cell cortex and SCRIB polarity function . Furthermore , mutations in SCRIB SADH domains associated with spina bifida and cancer impact the stability of SCRIB at the plasma membrane , suggesting that SADH domain alterations may participate in human pathology . The protein SCRIB has been implicated in a staggering array of cellular processes including polarity , migration , proliferation , differentiation , apoptosis , stemcell maintenance , and vesicle trafficking ( Humbert et al . , 2008 ) . SCRIB is a membrane associated protein localizing at cell junctions ( Navarro et al . , 2005; Qin et al . , 2005 ) . Alteration of SCRIB localization at the membrane often mimics SCRIB loss-of-function phenotype ( Zhan et al . , 2008; Cordenonsi et al . , 2011; Elsum and Humbert , 2013 ) and SCRIB aberrant accumulation in the cytosol strongly correlates with poor survival in human cancers ( Nakagawa et al . , 2004; Navarro et al . , 2005; Gardiol et al . , 2006; Kamei et al . , 2007; Ouyang et al . , 2010; Pearson et al . , 2011; Feigin et al . , 2014 ) . Despite the crucial importance of SCRIB subcellular localization , the molecular basis of SCRIB recruitment and stabilization to the cell cortex is not fully understood . SCRIB has a tripartite domain organization that consists of an N-terminal region composed of leucine-rich repeats ( LRR ) , PDZ domains and a C-terminal region with no identified protein domain . While the LRR region is crucial for membrane targeting ( Legouis et al . , 2003; Zarbalis et al . , 2004; Zeitler et al . , 2004 ) , mutations affecting the PDZ and C-terminal regions have also been reported to affect SCRIB subcellular localization ( Robinson et al . , 2011; Lei et al . , 2013 ) . The LRR and PDZ domains are highly conserved between Drosophila and human ( 60% of homology ) . The C-terminal region displays a poor conservation between the two species ( 13% ) ( Figure 1—figure supplement 1 ) and its function remains unclear . This is surprising considering that about 40% ( 5/12 ) of all the pathological germline mutations identified in the mouse and human Scrib/SCRIB genes lies within the C-terminal domain of the protein ( Murdoch et al . , 2003; Zarbalis et al . , 2004; Wansleeben et al . , 2010; Stottmann et al . , 2011; Lei et al . , 2013 ) . Here we show that the C-terminal part of SCRIB contains three spectrin binding motifs which are crucial for SCRIB cortical dynamics and polarity function . We have previously reported that over-expression of the C-terminal part of SCRIB impedes the polarized orientation of the centrosome during astrocyte migration ( Osmani et al . , 2006 ) . The same SCRIB fragment consisting of the last 407 amino acids was used as a bait to screen exhaustively a human fetal brain cDNA library in a two-hybrid screen . Nine independent clones encoding β2-spectrin ( SPTBN1 gene ) and seven independent clones encoding β3-spectrin ( SPTBN2 gene ) were recovered ( Figure 1—figure supplement 2 ) . The overlapping sequences of these clones map to the calponin homology one ( CH1 ) domain of spectrins . GST-SCRIB proteins including the C-terminal fragment ( 1223–1630aa ) or the C1 fragment ( 1223–1424aa ) bound endogenous β2 spectrin from cell extracts ( Figure 1A , B ) . In contrast , the C2 fragment ( 1425–1630 ) did not interact with spectrin ( Figure 1B ) . These interactions were also confirmed by co-immunoprecipitation in HEK cells ( Figure 1C ) . 10 . 7554/eLife . 04726 . 003Figure 1 . Dissection of the SCRIB β-spectrin interaction . ( A ) Schematic representation of SCRIB and different C-terminal constructs used in this study . ( B ) GST-C , GST-C1 and GST-C2 pull-down on astrocytes cell extract . Samples were analyzed by ponceau staining and immunoblotting using the indicated antibody . ( C ) Immunoprecipitation was performed with anti Flag antibody using HEK293 cell lysates co-expressing Flag CH1 spectrin domain with indicated GFP-SCRIB constructs . Samples were analyzed by immunoblotting using the indicated antibodies . ( D ) Schematic representation of the C1 SCRIB constructs with internal deletions of putative spectrin binding motif sequences . GST-C1 , GST-Δ1 , GST-Δ2 , GST-Δ1Δ2 pull-down assay on HEK293 cell lysates expressing GFP-CH1 spectrin domain . Samples were analyzed by immunoblotting using anti GFP . ( E ) GST and GST-SCRIB repeat 1 , 2 and 3 pull down assay on HEK293 cell lysates expressing GFP-CH1 spectrin domain . Samples were analyzed by ponceau staining and immunoblotting using the indicated antibody . ( F ) Immunofluorescence images of 16HBE monolayers fixed and stained for SCRIB , SPTBN1 ( spectrin β2 ) and SPTBN2 ( spectrin β3 ) . ( G ) HeLa cells were singly transfected with indicated GFP-SCRIB constructs ( upper panel ) or in combination with spectrin RFP-CH1 domain ( bottom panels ) . Images are representative of at least three independent experiments . Bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 00310 . 7554/eLife . 04726 . 004Figure 1—figure supplement 1 . SCRIB C-terminal sequence alignment . Sequence alignment of SCRIB C-terminal domain of indicated different species . Boxes indicate the emplacement of the SADH motifs . Note that no sequences fitting the SADH consensus can be detected in Drosophila Scribble C-terminal domain sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 00410 . 7554/eLife . 04726 . 005Figure 1—figure supplement 2 . Schematic representation of SCRIB , β2-spectrin and β3 spectrin proteins . ( A ) The SCRIB C-terminal sequence encompassing the 1223–1630 region was used as a bait in the two hybrid screen . ( B ) Minimal β2-spectrin and β3 spectrin sequences recovered in the two hybrid screen are indicated ( SID ) as well as the exact start and stop of each individual selected clones . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 005 Insertion of the alternative exon 36 brings an additional 25aa sequence within the SCRIB C2 protein fragment and forms a long SCRIB isoform ( Figure 1A ) . In contrast to the short GFP-C2 fragment ( C2 − exon36 ) domain , the long GFP-C2 isoform ( C2 + exon36 ) strongly interacted with spectrin Flag-CH1 ( 27–167aa ) , suggesting that the exon 36 encodes a putative β-spectrin binding sequence ( Figure 1C ) . Sequence analysis of the C1 fragment revealed the presence of two 25aa sequences displaying strong similarity with exon 36 sequence . Single deletion of each repeated motif impaired GST-C1 binding to the GFP-tagged CH1 domain of spectrin . The deletion of both repeats totally abolished spectrin binding ( Figure 1D ) . GST fusion protein including the 25aa peptide corresponding to the exon 36 as well as the two other similar repeat motifs identified in SCRIB C1 , but not GST alone , interacted robustly with GFP tagged CH1 spectrin domain ( Figure 1E ) . We then performed co-localization experiments . In bronchial epithelial cell line ( 16HBE ) SCRIB and both β-spectrins colocalized to adherens junctions ( Figure 1F ) . In contrast , in HeLa cells which do not express endogenous cadherin ( Lock and Stow , 2005 ) , GFP-SCRIB was mainly cytosolic . In these cells , exogenously expressed RFP-CH1 domain of β2 spectrin localises into filamentous structures and induces the relocalisation of GFP-SCRIB to these structures ( Figure 1G ) . In these conditions the GFP-SCRIB lacking putative spectrin binding domains remained cytosolic ( Figure 1G ) . Together our results indicate that SCRIB interacts directly with β-spectrins via three spectrin binding motifs with one of them included in an alternative exon . The β2 spectrin N-terminal actin binding region is composed of a tandem calponin homology domain designated CH1-CH2 ( Figure 2A ) . GST coupled C1 SCRIB fragment interacted with GFP tagged β spectrin CH1 but neither with CH2 nor with CH1-CH2 tandem domains expressed in HEK cells ( Figure 2B ) . In agreement with this observation RFP-CH1 but not RFP-CH1-CH2 β2 spectrin constructs strongly colocalises with F-actin and GFP-C1 SCRIB fragment in HeLa cells . Furthermore , C1 SCRIB fragment interacted with actin-associated CH1 domain ( Figure 2D ) suggesting that SCRIB as a preferential affinity with actin-associated β2 spectrin . Structural studies have shown that the CH1–CH2 tandem domains can switch between an open conformation where the CH1 domain binds to F-actin robustly ( Way et al . , 1992 ) and a closed conformation in which the CH1 and CH2 domains are closely apposed and display a weak F-actin affinity ( Sjöblom et al . , 2008; Galkin et al . , 2010 ) . 10 . 7554/eLife . 04726 . 006Figure 2 . SCRIB spectrin binding motifs bind to CH1 domains of the β spectrin family . ( A ) Schematic representation of full length human β2 spectrin protein ( left ) and different β2 spectrin deletion constructs used in the study ( right ) . ( B ) GST pull down using a GST-SCRIB C1 resin from HEK293 cell lysates expressing indicated GFP-tagged β2 spectrin CH domains . Samples were analyzed by ponceau staining and immunoblotting using anti GFP . ( C ) Fluorescence images showing HeLa cells expressing GFP-SCRIB C1 construct together with RFP-NCH1CH2 or SID β2 spectrin domains and stained with phalloidin . ( D ) GST-SCCRIB C1 pull down from HEK293 cell lysates expressing indicated GFP tagged β2 spectrin CH domains . Samples were analyzed by ponceau staining and immunoblotting using anti GFP . ( E ) Phylogenetic analysis of the Calponin Homology ( CH ) domains type1 , 2 and 3 used in this study . ( F ) GST pull down from HEK293 cell lysates expressing the indicated CH domains using a GST-SCRIB SADH3 resin . Pull down was done in low or high stringency conditions ( LSB: Low Salt Buffer , HSB: High Salt Buffer ) . Ponceau staining indicates the relative amount of GST tagged proteins bound to the resin . Samples were analyzed by ponceau staining and immunoblotting using anti GFP . Bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 006 Calponin homology domains are widespread in the human genome ( 157 domains referenced in SMART database ) and are classified in three classes CH1 , CH2 and CH3 . We could not detect any interaction with the β2 spectrin CH2 domain ( Figure 2B , F ) or the CH3 domain of Vav3 or IQGAP proteins ( Figure 2E , F ) . However , we found that at physiological salt stringency a SCRIB GST-SADH3 bound to the CH1 domains of spectrins β , β2 , β3 , β5 and α-actinin 2 ( Figure 2E , F ) . At high salt stringency ( 500 mM NaCl ) , only the spectrin β , β2 and β3 displayed binding with the GST-SADH3 motif ( Figure 2F ) . These results indicate that SCRIB has a preferential affinity with the CH1 domain from the close homologues spectrin β , β2 and β3 . Phylogenetic analysis revealed that the spectrin binding repeats are conserved in all vertebrate SCRIB sequences tested ( primate , rodent , bird , amphibian and fish ) but are not present in Drosophila or Caenorhabditis elegans scribble/LET-413 ( Figure 3—figure supplement 1A ) . Alignment of all spectrin binding repeats revealed a conserved [+]-X-X-Y-[+]-X-ϕ-A-A-ϕ-P sequence ( Figure 3A ) . We examined the consequences of alanine or glycine substitutions in SCRIB repeat 3 . Mutations of both positively charged residues did not noticeably affect the binding to spectrin CH1 domain . However , mutations of the tyrosine , the alanine doublet or the proline of the repeat 3 severely weakened or completely abolished GST-repeat 3 binding to spectrin CH1 domain ( Figure 3B ) . 10 . 7554/eLife . 04726 . 007Figure 3 . The SADH motif ( SCRIB , ABLIM , dematin homology motif ) . ( A ) Alignment of vertebrate SCRIB spectrin binding repeats 1 ( pink ) , 2 ( yellow ) and 3 ( green ) . Conserved residues are shown in blue . [+] indicates positively charged amino acids and Φ represents hydrophobic residues . ( B ) Alanine or glycine substitutions ( in red ) performed on SCRIB repeat 3 conserved residues . WT and mutants GST-repeat 3 pull down assay on HEK293 cell lysates expressing spectrin GFP-CH1 domain . Samples were analyzed by ponceau staining and immunoblotting using the indicated antibody . ( C ) Schematic representation and sequence of DMTN and ABLIM1 , 2 and 3 proteins displaying strong homology with the SCRIB putative spectrin binding motif Y-[KR]-X-[FL]-A-A-[ILV]-P ( boxed in pink ) . Position of the motif in the protein is indicated in brackets in aa . MyoX , MPP7 and Afadin sequences with noncanonical motifs consensus are shown on the bottom . VHP ( Villin Hedapiece domain ) . ( D ) Pull down assay with GST resin fused to 25aa motif from indicated proteins performed in HEK293 cell lysates expressing spectrin GFP-CH1 domain . Samples were analyzed by ponceau staining and immunoblotting using the indicated antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 00710 . 7554/eLife . 04726 . 008Figure 3—figure supplement 1 . SADH domains in vertebrates . ( A ) Schematic representation of SCRIB gene with spectrin binding motifs ( up ) . Table displaying the distribution of SADH domains in SCRIB vertebrate protein sequences . SADH motifs could not be identified in drosophila and C . elegans Scribble/LET-413 ortholog genes . ( B ) Schematic representation of SADH motifs in DMTN and in ABLIM123 genes ( up ) and conservation of SADH motif sequence in vertebrates . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 008 A search in the human protein database ( Swiss-prot ) for the simplified Y-[KR]-X-[FL]-A-A-[ILV]-P motif sequence identified four other proteins displaying a perfect fit with the SCRIB motif consensus: DMTN ( dematin/Band 4 . 9 ) and the three member of the ABLIM family of protein ( Actin Binding LIM domain containing proteins ) ( Figure 3C ) . We found that the DMTN and the ABLIMs motifs retained GFP tagged spectrin CH1 domain from cell extract ( Figure 3D ) . Interestingly DMTN is a constituent of the spectrin-actin junctional complex and was known to bind directly to spectrin ( Koshino et al . , 2012 ) . Phylogenetic analyses of the three ABLIMs showed that the spectrin CH1 binding motif was conserved in all vertebrate ABLIMs but could not be detected in invertebrate ABLIMs proteins ( Figure 3—figure supplement 1B ) . We also observed that the noncanonical motifs identified in MyoX , MPP and Afadin did not bind to spectrin CH1 domain ( Figure 3D ) . Altogether those results indicate that the consensus sequence Y-[KR]-X-[FL]-A-A-[ILV]-P is a spectrin binding motif specific of the vertebrate sub-phylum and present in at least five different genes in the human genome . We named this sequence the SADH motif ( SCRIB ABLIMs DMTN Homology ) . No homology between the SADH motif sequences and the previously characterized 10 kDa spectrin-actin-binding domain of the band 4 . 1 family of proteins could be detected ( Gimm et al . , 2002 ) . We then investigated the effect of SADH motif mutations ( P > A mutant described in Figure 2B ) on SCRIB cortical localization . We determined the Cortical Localisation Index ( CLI , Figure 4—figure supplement 1 ) , defined as the ratio between the mean GFP fluorescence intensity in the cortical region and the mean fluorescence intensity in the cytoplasm of the cell . A CLI superior to one indicates that GFP is enriched at the cell cortex while a CLI equal to one shows that the GFP is equally distributed between the cytoplasm and the cortex . GFP alone had a CLI around 1 ( 0 . 9 ) whereas a control GFP-CAAX construct which essentially localized at the cell membrane had a CLI of 1 . 75 . Short and long WT SCRIB ( −/+ exon36 ) constructs showed a strong cortical accumulation ( CLI = 1 . 8 and 2 respectively ) . In contrast the P305L SCRIB mutant that carries a mutation in the N-terminal LRR region which impedes SCRIB plasma membrane localization ( Legouis et al . , 2003 ) was mainly cytosolic ( CLI = 1 ) . SCRIB SADH1+2 mutant showed a statistically weaker cortical recruitment ( CLI = 1 . 4 ) than the WT protein ( Figure 4A ) suggesting that SADH mutations may influence SCRIB recruitment and/or stabilization at the membrane . We assessed SCRIB WT and SADH mutant protein dynamics at the membrane by monitoring FRAP ( Fluorescence Recovery After Photobleaching ) at cell–cell contact in confluent transiently transfected 16HBE cells . A membrane associated GFP-CAAX showed a fast recovery rate ( t1/2 = 10 s ) and the vast majority of the protein was available for exchange ( mobile fraction of 97% ) ( Figure 4B , C , D ) . In contrast GFP-WT SCRIB exchanged less rapidly ( t1/2 = 26 s ) and had a smaller mobile fraction ( 60% ) . SCRIB SADH12 mutant displayed a significantly faster exchange ( t1/2 = 18 s ) than the WT SCRIB protein with 82% of the protein pool available for exchange ( Figure 4B , C , D ) . These results indicate that the SADH motifs are important for SCRIB stabilization at the cell cortex . 10 . 7554/eLife . 04726 . 009Figure 4 . SADH motif influence SCRIB membrane stability and is required for SCRIB polarity function . ( A ) 16HBE cells were transiently nucleofected with the indicated GFP constructs and analyzed by live confocal microscopy to calculate their cortical localization index . ( n = 50 for each conditions ) . ( B ) FRAP experiment on 16HBE cells nucleofected with the indicated GFP constructs . High magnification images of adherens junction before and at the indicated time points after photobleaching . ( C ) Quantitative analysis of FRAP from experiments similar to those shown in B ( n = 30 for each conditions ) . ( D ) The mobile fraction and t1/2 of recovery for each protein were calculated from the recovery curves in C . ( E ) Centrosome reorientation in migrating astrocytes expressing the indicated DNA constructs ( up ) . Results shown are means SEM of 3–4 independent experiments , with a total of at least of 150 cells . ( F ) Representative images of astrocytes expressing the indicated constructs and stained with anti-pericentrin ( centrosome , red ) and Hoechst ( nucleus , blue ) . ( G ) Primary astrocytes were nucleofected with control ( Ctrl ) or β2 spectrin ( si SPTBN1 ) siRNA and incubated for 72 hr . Protein levels were analyzed by western blots using anti-SPTBN1 antibody and anti-actin . ( H ) Centrosome reorientation assay in migrating astrocytes nucleofected with control or β2 spectrin ( SPTBN1 ) siRNA and microinjected with GFP SCRIB Cter construct . Results shown are means ± SEM of three independent experiments , with a total of at least 100 cells . *p < 0 . 001; p < 0 . 01; *p < 0 . 05 . Bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 00910 . 7554/eLife . 04726 . 010Figure 4—figure supplement 1 . Cortical localization of SCRIB SADH mutant . ( A ) Representative images of 16HBE cells nucleofected with the indicated GFP constructs acquired live and used for the CLI analysis . ( B ) Schematic representation of the process used for CLI calculation . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 01010 . 7554/eLife . 04726 . 011Figure 4—figure supplement 2 . SADH domains are implicated in SCRIB-mediated control of Cdc42 localization . ( A ) Cherry-Cdc42 recruitment at the leading edge of migrating astrocytes cells expressing the indicated constructs . Results as shown are means ± SEM of three independent experiments , with a total of at least 200 cells . Representative images are shown on the right . White dotted lines show scratch positions . The scale bar represents 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 01110 . 7554/eLife . 04726 . 012Figure 4—figure supplement 3 . SCRIB SADH domain are required for tight junction assembly . ( A ) WB analysis of SCRIB and actin expression in Caco-2 cells , 3 days after nucleofection with control siRNA ( Ctrl ) or siRNAs directed against SCRIB ORF ( ORF1 and ORF2 ) or 3′ UTR ( UTR1 and 2 ) . ( B ) Percentage of Caco-2 cells nucleofected with control or SCRIB UTR1 siRNA displaying intact of ZO-1 labeling at the cell–cell contacts . ( C ) Representative example of Caco-2 cells displaying intact or short and disconnected areas of ZO-1 labeling at the cell–cell contacts in the indicated conditions . ( D ) Caco-2 cells nucleofected with indicated siRNA or construct were scored 1 hr after calcium switch for their tight junction integrity using ZO-1 staining . Results shown are means ± SEM of three independent experiments , with a total of at least 100 cells . The scale bar represents 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 012 SCRIB is involved in both polarization and orientation of migrating cells in in vitro scratch ‘wound-healing’ assay ( Qin et al . , 2005; Osmani et al . , 2006; Dow et al . , 2007 ) . Expression of GFP-SCRIB C-terminal constructs and to a lesser extent of GFP SADH3 perturbed centrosome reorientation in migrating astrocytes ( Figure 4E , F ) ( Osmani et al . , 2006 ) . Mutations of SADH domains strongly reduced the ability to disrupt centrosome reorientation indicating that the SADH domains are involved in SCRIB polarity fonction . β2 spectrin depletion moderately but significantly perturbed scratch-induced centrosome reorientation ( Figure 4G ) . This effect , likely underestimated because of the incomplete knock down or the compensatory role of other spectrins , indicates a role of spectrin in the control of centrosome positioning . Moreover , spectrin β2 depletion impaired SCRIB Cter ability to disrupt centrosome reorientation ( Figure 4G ) . Together , these results suggest that SCRIB interaction with spectrin contributes to SCRIB polarity function . SCRIB is required for the recruitment and the activation of Cdc42 at the cell front edge leading to the centrosome reorientation ( Osmani et al . , 2006; Dow et al . , 2007 ) . GFP SCRIB Cter overexpression but not GFP alone significantly reduced Cherry-Cdc42 membrane recruitment at the leading edge ( Figure 4—figure supplement 2 ) while SCRIB SADH12 mutant Cter construct had significantly weaker effect , suggesting that SCRIB SADH domains are implicated in the SCRIB-mediated recruitment of Cdc42 at the leading edge of migrating cells . In addition to its role in polarity , SCRIB has been previously implicated in tight junction ( TJ ) assembly in intestinal epithelium ( Qin et al . , 2005; Ivanov et al . , 2010 ) . After calcium switch , SCRIB-depleted Caco-2 cell monolayers showed short and disconnected areas of ZO-1 labeling at the cell–cell contacts , indicative of a significant delay in TJ reassembly ( Figure 4—figure supplement 3A , B , C ) . This tight junction phenotype could be partially rescued by the expression of a siRNA resistant WT SCRIB but not by SCRIB SADH12 mutant , suggesting that SCRIB interaction with spectrin plays a role in tight junction assembly ( Figure 4—figure supplement 3D ) . Altogether these results strongly suggest that SCRIB SADH motifs control SCRIB dynamics at the cell cortex and are important for polarity and tight junction assembly . 61 missense mutations in the SCRIB gene coding sequence have been identified so far ( COSMIC , [Forbes et al . , 2015] ) . Almost 10% of these mutations ( 6/61 ) fall within the SCRIB SADH motifs ( Figure 5A ) , which account for less than 5% of SCRIB sequence ( 75/1657aa ) . One of the mutations ( R1322W ) identified in a lung cancer patient directly impacts the core SADH2 consensus motif . In contrast to GST-SADH2 WT sequence , GST-SADH2 R1322W mutant did not bind to the spectrin GFP-CH1 domain pointing to the R1322W mutation as a spectrin binding loss of function mutation ( Figure 5B ) . SCRIB mutations have also been described in congenital diseases ( Robinson et al . , 2011; Lei et al . , 2013 ) . In particular , six SCRIB mutations have been recently described in spina bifida ( Lei et al . , 2013 ) and two of those six fall within the SADH2 motif sequence ( Figure 5C ) . The A1315T mutation did not affect the ability of a GST-SADH2 to bind to GFP spectrin . In contrast the P1332L mutation noticeably increased GST-SADH2 affinity for GFP spectrin in vitro ( Figure 5C , D , E ) . These mutations did not impact significantly SCRIB overall recruitment to the cellular cortex ( Figure 5—figure supplement 1 ) . However , FRAP experiments showed that GFP R1322W and GFP P1332L SCRIB mutants exchanged more rapidly than the GFP WT SCRIB at the plasma membrane ( Figure 5F ) . Surprisingly , P1332L mutation also increased SCRIB exchange at the plasma membrane , suggesting that , in the context of the SCRIB full length molecule , the P1332L mutation prevents rather than increases spectrin binding . We cannot however exclude the possibility that this mutation also affects other yet-unidentified function of the SADH domain . Altogether these observations indicate that mutations of the SADH motifs may participate in human pathology by impacting the stability of SCRIB at the cell cortex . 10 . 7554/eLife . 04726 . 013Figure 5 . SADH motifs mutations in human pathology . ( A ) Table of identified somatic SCRIB SADH motifs mutations in human cancers . Data sourced from COSMIC ( http://cancer . sanger . ac . uk/cosmic ) . ( B ) GST-SCRIB SADH2 WT and SADH2 R1322W pull down assay on HEK293 cell lysates expressing spectrin GFP-CH1 domain . Ponceau staining indicates the relative amount of GST tagged proteins bound to the resin . Samples were analyzed by immunoblotting using anti GFP . ( C ) Table of identified germinal SCRIB SADH motifs mutations in spina bifida . ( D ) GST-SCRIB SADH2 WT , A1315T and P1332L pull down assay on HEK293 cell lysates expressing spectrin GFP-CH1 domain . Ponceau staining indicates the relative amount of GST tagged proteins bound to the resin . Samples were analyzed by immunoblotting using anti GFP . ( E ) GST-SCRIB SADH2 WT and P1332L pull down assays similar to D performed in indicated salt stringency . ( F ) Quantitative analysis of FRAP experiments in 16HBE adherens junction expressing the indicated GFP constructs ( n = 30 for each conditions ) . The mobile fraction and t1/2 of recovery for R1322W and P1332L proteins were calculated from the recovery curves in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 01310 . 7554/eLife . 04726 . 014Figure 5—figure supplement 1 . ( A ) 16HBE cells were transiently nucleofected with the indicated GFP constructs and analyzed by live confocal microscopy to calculate their cortical localization index . ( n = 50 for each conditions ) . ( B ) Quantitative analysis of FRAP experiment on 16HBE cells nucleofected with the indicated GFP constructs ( n = 30 for each conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04726 . 014 Dow et al have shown that a transgene encoding the human gene SCRIB was able to partially rescue the Drosophila scribble mutant ( 23% survival rescue to adult stage ) , arguing for an indisputable conserved role of scribble/SCRIB during evolution ( Dow et al . , 2003 ) . Nevertheless it is clear that some phenotypic differences exist between Drosophila and mouse Scrib mutant models like the severity of apico-basal epithelium polarity phenotype , and that the scribble/SCRIB genes have undergone some intra-phylum specific adaptation ( Bilder et al . , 2000; Murdoch et al . , 2003 ) . No phylum specific SCRIB interacting partner has been characterized so far that could account for the differences observed between invertebrates and vertebrates SCRIB function . The vertebrate-specific spectrin binding motifs SADH located inside SCRIB divergent C-terminal region appears to be a good candidate to bear such role . Yeast two-hybrid screening was performed by Hybrigenics Services , SAS , Paris , France ( http://www . hybrigenics-services . com ) . The coding sequence for human protein SCRIB ( aa 1224–1630 ) ( GenBank accession number gi: 18141296 ) was PCR-amplified and cloned into pB27 as a C-terminal fusion to LexA ( N-LexA-SCRIB-C ) and into pB66 as a C-terminal fusion to Gal4 DNA-binding domain ( N-Gal4-SCRIB-C ) . The constructs were checked by sequencing and used as a bait to screen a random-primed Human Fetal Brain cDNA library . A confidence score ( PBS , for Predicted Biological Score ) was attributed to each interaction as previously described ( Formstecher et al . , 2005 ) . The following reagents were used in this study: Anti SCRIB C-20 ( Goat , Santa Cruz Biotechnology , Dallas , TX ) , Anti-SPTBN1 and E-cadherin ( Mouse , BD Transduction Laboratories , San Jose , CA ) , Anti-SPTBN2 A301-117A ( Rabbit , Bethyl , Montgomery , TX ) , Anti-GFP HRP ab6663 ( Abcam , Cambridge , MA ) , Anti-FLAG HRP clone M2 ( Sigma , Saint Louis , MO ) , Anti-pericentrin PRB-432C ( Rabbit , Covance , Princeton , NJ ) , Anti-Actin AC-40 ( Sigma ) . SiRNA sequences: Non-Targeting control siRNA ( luciferase ) ( UAAGGCUAUGAAGAGAUAC ) , SPTBN1 ( UGAUGGCAAAGAGUACCUCTT ) , SCRIB-ORF1 ( GCACUGAGGAGGAGGACAATT ) , ORF2 ( GAACCUCUCUGAGCUGAUCTT ) , UTR1 ( GUUCUGGCCUGUGACUAACTT ) and UTR2 ( GGUUUAAGGAGAAUAAAGUTT ) were ordered at Eurofins ( France ) . The SCRIB domains were amplified by PCR from a human SCRIB template provided by Jean Paul Borg and cloned into the NotI-EcoRI sites mammalian expression vectors CB6-N-GFP and pEGFP or the Escherichia coli expression vector pMW172-GST . SADH motifs were obtained by annealed oligo cloning and cloned likewise in the above-mentioned vectors . The SCRIB internal deletions and point mutagenesis were generated using PCR-based site-directed mutagenesis . The spectrin sequence corresponding to the 27–167aa prey clone encompassing the β2-spectrin CH1 domain was cloned in CB6-N-GFP , CB6-N-RFP or CB6-N-Flag . For sequence homology search the Y-[KR]-X-[FL]-A-A-[ILV]-P motif was blasted on the Pattern Search program at http://www . expasy . org/ ( Sigrist et al . , 2010 ) against the Homo sapiens Swiss-prot database . Sequence alignments and phylogeny of calponin homology domains were done on the mobile website ( Neron et al . , 2009 ) and using the seaview program ( Gouy et al . , 2010 ) . Statistical analysis was performed using GraphPad Prism 5 . 0 . SCRIB somatic mutations in cancers were obtained at www . sanger . ac . uk ( Forbes et al . , 2015 ) . All proteins were expressed in E . coli BL21 ( DE3 ) Rosetta strain . Bacterial cell pellets were lysed 1 hr at RT in 150 mM NaCl , 50 mM Tris pH8 and 25% sucrose supplemented with 5000 units of lyzozyme ( Sigma ) . Cleared supernatants were mixed for 90 min with glutathione-Sepharose 4B beads ( GE Healthcare ) . The resulting resins were washed three times with PBS containing 200 mM NaCl and 0 . 1% Triton ( buffer A ) . HEK 293 cells were transiently transfected using the phosphate calcium method . Cell lysates were prepared by scraping cells in lysis buffer 50 mM Tris pH7 . 5 , triton 2% , NP40 1% , 200 mM NaCl with Complete protease inhibitor tablet ( Roche , Indianapolis , IN ) and centrifuged for 10 min at 13 , 000 rpm 4°C to pellet cell debris . Soluble detergent extracts were either incubated with GST-SCRIB resins or Anti FLAG coupled protein G-Sepharose ( GE healthcare ) for 2 hr at 4°C prior to washing three times with buffer A and processed for western blot analysis . For the calponin homology domain binding experiment ( Figure 2B ) the high stringency binding was done in lysis buffer containing 500 mM NaCl . 16HBE cells were maintained in DMEM/F12 medium ( Invitrogen ) , Caco-2 and HEK cells in DMEM supplemented with 10% FBS ( Invitrogen ) and penicillin ( 100 U/ml ) -streptomycin ( 100 μg/ml; Invitrogen ) at 37°C in 5% CO2 . For DNA and siRNA transfection 5 × 106 16HBE or Caco-2 cells were nucleofected with DNA ( 5 µg ) using Lonza kitT ( program A-23 and B-024 respectively ) and nucleofector device , according to manufacturer's protocol . Primary rat astrocytes were prepared as described previously ( Etienne-Manneville , 2006 ) . For immunofluorescence , cells were fixed in 4% paraformaldehyde , permeabilized with 0 . 1% Triton X-100 in PBS , and blocked in PBS 10% Serum for 1 hr before incubation with antibodies . For scratch-induced assays , primary astrocytes were seeded on poly-L-ornithine-coated coverslips and were grown in serum to confluence . The medium was changed 16 hr before scratching . Individual wounds ( approximately 300 mm wide ) were made with a microinjection needle and front row migrating astrocytes were immediately micro-injected with the indicated GFP tagged constructs . Centrosome reorientation was determined as described previously ( Etienne-Manneville and Hall , 2001 ) . Briefly , 8 hr after the wound , centrosomes located in front of the nucleus of GFP positive front row cells , within the quadrant facing the wound were scored as correctly oriented . In these assays , a score of 25% ( astrocytes ) represents the absolute minimum corresponding to random centrosome positioning . Calcium Switch was performed as described previously ( Ivanov et al . , 2010 ) . Briefly , Caco-2 cells were incubated for 1 hr in the low-calcium medium and supplemented with 2 mmol/l EGTA before being returned to normal cell culture media ( calcium repletion ) for indicated times at 37°C . Fixed cells were imaged on a microscope ( DM6000 B; Leica ) using an HCX Plan Apochromat 40×/1 . 25 NA oil confocal scanning or HCX Plan Apochromat 63×/1 . 40 NA oil confocal scanning objective ( Leica ) . Microscopes were equipped with a camera ( DFC350FX; Leica ) , and images were collected with LAS software ( Leica ) . 16HBE cells grown on MatTek ( P35G-1 . 5-14-C ) Petri dishes were analyzed 72 hr after nucleofection . Live cell imaging was performed on a spinning disk confocal microscope Zeiss Axiovert 200 with UltraView ERS ( Perkin–Elmer ) , at 37°C with a Plan-Apochromat X63/1 . 4 objective . To calclulate the CLI we used the following equation: CLI = ( i/[i + I] ) / ( a/[a + A] ) where ( i ) represents the pixel intensity contained in a 0 . 625 µm thick cortical band encompassing the cell edge , ( I ) the cytoplasmic pixel intensity , ( a ) the cortical region area and ( A ) the cytoplasmic area ( Figure 3—figure supplement 1 ) . The CLI was calculated for each cell on three different Z plan and then averaged . FRAP experiment was performed using the FRAP module of confocal Volocity software ( Perkin–Elmer ) . A 9 µm2 square region of interest to be bleached was defined for the FRAP and maximum laser power at 488 nm for one iteration was used to bleach signals . After bleaching , images were taken within the same focal plane at regular intervals ( between 3 and 10 s ) to monitor fluorescence recovery . After background subtraction the recovery of the GFP signal was measured using ImageJ and fitted using the Prism software and the equation Y ( t ) = ( Ymax − Ymin ) ( 1 − e2kt ) − Ymin ( Weisswange et al . , 2009 ) , where Y ( t ) is the intensity of fluorescence at time t , Ymax and Ymin are respectively the maximum and minimum intensities of fluorescence post-bleaching and k is the rate constant of recovery . Mobile fraction was determined as Mf = ( Ymax − Y0 ) / ( 1 − Y0 ) . All data are presented as the mean ± s . e . m . One-way ANOVA analysis of the variance was followed by the Tukey's multiple comparison post-hoc test . A p value of <0 . 05 was considered as statistically significant .	Proteins found in cells often have more than one role . Scribble is one such multi-tasking protein that is found in a diverse range of species , including fruit flies and humans . Although Scribble commonly helps to ensure that the components of a cell are in their correct locations , its exact roles vary between species . To perform its role well , Scribble itself must localize to the cell cortex—the inside surface of the cell membrane—at the regions where cells connect to one another . How this localization occurs is not fully understood; and defects in the human form of Scribble have been linked to diseases including spina bifida and cancer . Much of the Scribble protein is very similar across different species , but the fruit fly and human version of the protein have large differences in their ‘C-terminal region’ that makes up one end of each protein . Boëda and Etienne-Manneville now show that in humans and other animals with backbones—but not in fruit flies—the C-terminal region of Scribble contains three repeats of a sequence called the SADH motif . These motifs can bind to proteins called beta spectrins , which connect the cell's outer membrane to the scaffolding-like structure inside the cell that provides support . Mutations that alter the SADH motif interfere with Scribble's ability to bind to the scaffolding , and alters Scribble localization at cell–cell contacts or the cell cortex . Boëda and Etienne-Manneville also found that some mutations linked to spina bifida and cancer affect the SADH motif , suggesting that this motif has a wider role in disease . While the abnormal localization of Scribble inside cells is frequently observed in particularly difficult to survive cancers , the molecular mechanism that causes Scribble to fail to localize to the cell periphery is still poorly understood . Boëda and Etienne-Manneville's work establishes the beta spectrin family of proteins as regulators that stabilize Scribble at the cell cortex and suggests that Scribble-associated diseases might depend on the integrity of the spectrin network .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "short", "report", "biochemistry", "and", "chemical", "biology" ]	2015	Spectrin binding motifs regulate Scribble cortical dynamics and polarity function
In the course of global climate change , Central Europe is experiencing more frequent and prolonged periods of drought . The drought years 2018 and 2019 affected European beeches ( Fagus sylvatica L . ) differently: even in the same stand , drought-damaged trees neighboured healthy trees , suggesting that the genotype rather than the environment was responsible for this conspicuous pattern . We used this natural experiment to study the genomic basis of drought resistance with Pool-GWAS . Contrasting the extreme phenotypes identified 106 significantly associated single-nucleotide polymorphisms ( SNPs ) throughout the genome . Most annotated genes with associated SNPs ( >70% ) were previously implicated in the drought reaction of plants . Non-synonymous substitutions led either to a functional amino acid exchange or premature termination . An SNP assay with 70 loci allowed predicting drought phenotype in 98 . 6% of a validation sample of 92 trees . Drought resistance in European beech is a moderately polygenic trait that should respond well to natural selection , selective management , and breeding . Climate change comes in many different facets , amongst which are prolonged drought periods ( Christensen et al . , 2007 ) . The Central European droughts in the years 2018 and 2019 caused severe water stress in many forest tree species , leading to the die-off of many trees ( Schuldt et al . , 2020 ) . Among the suffering tree species was European beech , Fagus sylvatica L . As one of the most common deciduous tree species in Central Europe , F . sylvatica is of great ecological importance: beech forests are a habitat for more than 6000 different animal and plant species ( Brunet et al . , 2010; Dorow et al . , 2010 ) . The forestry use of beech in 2017 generated a turnover of more than €1 billion in Germany alone ( Thünen_Institute , 2020 ) , without taking the economic and societal value of the ecosystem services of woods into account ( Elsasser et al . , 2016 ) . However , the drought years 2018 and 2019 severely impacted the beech trees in Germany ( Paar and Dammann , 2019 ) . Official reports on drought damage in beech recorded 62% of trees with rolled leaves and 20–30% of small leaves , mainly in the crown , resulting in 7% of badly damaged or dead trees . As shown before ( Bressem , 2008 ) , mainly medium- to old-aged trees were affected by drought stress ( >60 years ) . Under favourable conditions , beech is a competitive and shade-tolerant tree species , dominating mixed stands ( Pretzsch et al . , 2013 ) . High genetic diversity within populations supports adaptation to local conditions ( Kreyling et al . , 2012 ) . Significant differences between local populations in tolerance to various stress factors such as early frost ( Czajkowski and Bolte , 2006 ) , drought ( Cocozza et al . , 2016; Harter et al . , 2015 ) , or air pollution ( Müller-Stark , 1985 ) are known . The distribution of F . sylvatica is mainly limited by water availability as the tree does not tolerate particularly wet or dry conditions ( Sutmöller et al . , 2008 ) . Therefore , it is quite conceivable that the species could suffer even more under the predicted future climatic conditions than today ( Sutmöller et al . , 2008 ) . Despite the widespread , severe drought damage , a pattern observed in all beech forests was very noticeable ( personal observations ) . Using crown deterioration as a significant indicator for drought damage ( Choat et al . , 2018 ) , not all trees in a beech stand were equally damaged or healthy . The damage occurred rather in a mosaic-like pattern instead . Even though the extent of drought damage varied among sites , apparently completely healthy trees immediately neighboured severely damaged ones and vice versa . This observation gave rise to the hypothesis that not the local environmental conditions might be decisive for the observed drought damage , but rather the genetic make-up of the individual trees . We decided to draw on this natural ‘experimental set-up’ to infer the genomic basis underlying the drought susceptibility in F . sylvatica . We identified more than 200 neighbouring pairs of trees with extreme phenotypes and used a Pool-GWAS approach ( Bastide et al . , 2013 ) to infer associated single-nucleotide polymorphism ( SNP ) loci by contrasting allele frequencies with replicated pools of drought-susceptible and -resistant individuals . In addition , we individually resequenced a subset of 51 pairs of susceptible and resistant trees . If the observed pattern indeed has a genetic basis , identifying the associated loci would enable the genomic prediction of drought resistance ( Stocks et al . , 2019 ) . Constructing an SNP assay from the most highly phenotype-associated SNPs , we validated 70 identified loci by predicting the drought phenotype of an additional set of beech trees from their genotype at these loci using linear discriminant analysis ( LDA ) and a new Machine Learning approach ( Horenko , 2020 ) . These accurate genomic prediction tools , for example , the choice of drought-resistant seed-producing trees and selective logging , could help accelerate and monitor natural selection and thus harness beech forests against climate change ( Waldvogel et al . , 2020 ) . Damaged and healthy beech tree pairs were sampled from woods in the lowland Rhein-Main plain , the adjacent low mountain ranges of Odenwald and Taunus , and mountain ranges from Central and Northern Hessen ( Figure 1A ) . When summarising the climatic conditions from 1950 to 2019 for the sampling sites in a principal component analysis ( PCA ) , the sites were divided into two groups by axis 1 , a temperature gradient . The Taunus mountain sites grouped with those from the northern part of Hessen , while the Rhein-Main plain clustered with the Odenwald sites ( Figure 1B ) . This grouping was also used to construct the GWAS pools ( see below ) . Comparing the climate from the 1950s , when most of the trees sampled were already in place , with the decade from 2010 to 2019 , showed that all local conditions changed substantially and similarly in the direction and extent of warmer and drier conditions ( Figure 1B ) . The steepest temperature increase occurred in the 1980s , while precipitation patterns mainly changed in the last decade ( Figure 1—figure supplement 1 ) . A wide range of parameters , potentially relevant as selection pressures , changed drastically during this period: the mean January daily minimum temperature at the sampling sites increased by 1 . 49°C from −2 . 64°C ( s . d . 1 . 68°C ) in the 1950s to −1 . 15°C ( s . d . 2 . 50°C ) during the last decade . The mean August daily maximum temperatures increased even more by 2 . 37°C from 22 . 06°C ( s . d . 1 . 95°C ) to 24 . 43°C ( s . d . 2 . 35°C ) . Simultaneously , mean annual precipitation decreased by 40 . 5 mm or 5 . 5% from 741 . 2 mm ( s . d . 85 . 8 mm ) to 700 . 7 mm ( s . d . 70 . 9 mm ) . Most of the precipitation loss ( 84% ) occurred during the main growth period between April and September , with a decrease of 33 . 9 mm from 410 . 4 mm ( s . d . 36 . 1 mm ) to 376 . 5 mm ( s . d . 25 . 6 mm ) . Mean monthly evaporation potential , available from 1991 onwards , showed that , compared to the beginning of the 1990s , the main growth period of beech from April to September became increasingly drier , with up to 30 mm more evaporation per month . The drought dynamics suggested that the years 2018 and 2019 were not outliers , but rather part of a long-term , accelerating trend ( Figure 1C ) , following the overall global pattern ( Büntgen et al . , 2021; Trenberth et al . , 2014 ) . There was a strong negative correlation ( r = 0 . 695 ) between the drought strength during the main growth period ( Apr– Sept ) and a proxy for ( green ) leaf cover ( leaf area index [LAI] ) for the sampled plots in the years 2015–2019 ( Figure 1—figure supplement 2 ) . This observation suggested that leaf loss and dried leaves are good indicators for drought stress . The mean distance between paired trees was 5 . 1 m ( s . d . 3 . 4 m , Figure 2—figure supplement 1 ) . Phenotypic measurements generally confirmed the study design and selection of trees: healthy and damaged trees within each tree pair did not differ significantly in trunk circumference , tree height , canopy closure , and competition index ( Figure 2A–D , Supplementary file 1B ) . Hence , these parameters were not considered in further analyses . As expected , and confirming the assignment of damage status , the quantity of dried leaves and leaf loss differed substantially between damaged and healthy ones ( Figure 2E , F , Supplementary file 1B ) . A sample of photographs contrasting damaged and healthy paired trees can be found in Figure 2—figure supplement 2 . For a subsample of 300 out of the 402 sampled beech trees , we generated four DNA pools from two climatically distinct regions ( North and South Hessen , Figure 1B ) , contrasting trees that were either healthy or highly drought damaged , respectively ( Supplementary file 1A ) . The ‘South’ pools consisted of 100 individuals each , whereas the ‘North’ pools contained 50 individuals each . We created ~50 GB 150 bp-paired end reads with insert size 250–300 bp on an Illumina HiSeq 4000 system per pool . More than 96% of the reads mapped against the repeat-masked chromosome-level beech reference genome ( accession no . PRJNA450822 ) . After filtering the alignment for quality and a coverage between 15× and 70× , and removing indels , allele frequencies for 9 . 6 million SNPs were scored . All 100 individuals from the North population were additionally individually resequenced to ~20× coverage each ( for more details , see Materials and methods ) . This data was used to ( i ) determine individual variability in allele frequencies and ( ii ) validate the information content of the candidate SNP set . Using all individually resequenced individuals , we inferred the extent of genome-wide linkage disequilibrium ( LD ) . The plot of LD r² against the distance from the focal SNP showed that LD fell to r² ~ 0 . 3 within <120 bp , which means that genome positions such a distance apart are on average effectively unlinked ( Figure 3—figure supplement 1A ) . The PCA on SNP variation of the individually resequenced trees from the North population explained 12 . 3% of accumulated variation on the first two axes ( Figure 3—figure supplement 1B ) . Trees from the same sampling site ( within the North population ) did not tend to cluster together ( Figure 3—figure supplement 1B ) . FST estimates among pools for non-overlapping 1 kb windows were virtually identical among healthy/damaged pools within region as compared to between regions ( Figure 3—figure supplement 2 ) . Trees within a phenotypic class were genomically not more similar than between classes ( Figure 3—figure supplement 3 , ANOSIM R = −0 . 008 , p=0 . 76 , 9999 permutations ) . Pool-GWAS analysis identified 106 SNPs significantly associated with the drought damage status using a Cochran–Mantel–Haenszel test on the two pairs of damaged and healthy pools after false discovery rate correction and a cutoff at 1 × 10−2 ( Figure 3A , Figure 3—figure supplement 4 ) . Some of the 106 SNPs were in close physical proximity ( <120 bp ) and thus probably linked . Taking this into account , 80 independent genomic regions were associated with the drought damage status . None of the significantly differentiated SNP loci was mutually fixed; the observed allele frequency differences between healthy and damaged trees at associated loci ranged between 0 . 12 and 0 . 51 ( Figure 3B ) . Of the 106 significant SNPs , 24 were found in 20 protein coding genes ( Table 1 ) . Forty-nine genes were the closest genes to the remaining 82 SNPs . For 61 of these genes , the best BLAST hit was with a tree , mainly from the Fagales genera Quercus and Castanea ( Table 1 , Supplementary file 1C ) . Among the 24 SNPs in genes , we observed 13 non-synonymous changes . In 11 of these changes , the alternate allele was associated with the damaged phenotype and only in two cases with the healthy phenotype . Three of the non-synonymous substitutions resulted in a stop codon . Of the remaining 10 , 8 exchanges caused a major change in amino acid characteristics and thus probably in protein folding or function ( Table 1 ) . One gene , a PB1 domain-containing protein tyrosine kinase , contained four non-synonymous changes , suggesting that the allele version associated with the damaged phenotype lost its function ( Table 1 ) . From the 20 genes with significant SNPs , functional information could be obtained from the UniProt database for 14 ( Supplementary file 1C ) . Of these , 10 genes were associated in previous studies with either environmental stress response ( two ) or specifically with drought stress response ( eight; Supplementary file 1C ) . Of the 49 predicted genes closest to the remaining significant SNPs ( Table 1 ) , 16 could be reliably annotated ( Supplementary file 1B ) . Twelve had been directly related to drought in previous studies , while three were previously associated with other environmental stress responses ( Supplementary file 1C ) . We furthermore set out to determine how many SNPs were needed to successfully predict the drought susceptibility of individual trees , that is , to develop a genotyping assay . All Pool-GWAS SNPs in addition to the top 20 individual resequencing SNPs were used to create an SNP combination to reach a genotyping success threshold of min 90% . After excluding loci due to technical reasons and filtering for genotyping success , 70 loci proved to be suitable for reliable genotyping with an SNP assay . We genotyped only individuals sampled in 2019 that were not used to identify the SNPs in the first place plus paired individuals sampled in August 2020 . On average , each of the 95 individuals was successfully genotyped at 67 . 7 loci ( 96 . 7% ) . We coded the genotypes as 0 for homozygous reference allele , 1 for heterozygous , and 2 for the homozygous alternate allele , thus assuming a linear effect relationship . Figure 3B shows the genotypogram for the tested individuals . Linear discriminant analysis ( LDA ) correctly predicted the observed phenotype from the genotype in 91 of 92 cases ( 98 . 9% ) . Prediction success decreased to 65% when successively removing loci from the analysis ( Figure 3—figure supplement 5 ) . Nevertheless , ordering the individuals according to the LDA score of axis 1 revealed no clear genotype pattern that distinguished healthy from damaged trees ( Figure 3C ) . Observed heterozygosity at loci used in the SNP assay of individuals in the upper half of predictive values for a healthy phenotype was not significantly different from heterozygosity of the lower half ( Figure 3—figure supplement 6 ) . Ordering the loci according to their squared loadings showed that loci’s contribution to the genomic prediction differed substantially ( Figure 3 ) . As expected , the histogram of LDA scores showed two peaks , corresponding to the two phenotypes ( Figure 3—figure supplement 7 ) . To validate the results of the LDA prediction and circumvent potential overfitting due to the small sample size , we also applied a non-parametric Machine Learning algorithm for feature selection and clustering that was especially designed for small sample sizes ( Gerber et al . , 2020; Horenko , 2020 ) . The method identified the 20 most-significant SNPs allowing to make an almost 85% correct classification that distinguished healthy from damaged trees ( Supplementary file 1G ) . Over the last two decades , increasing drought periods caused severe damage to European forests ( Schuldt et al . , 2020; Etzold et al . , 2019; Pretzsch et al . , 2013 ) . Conifers seem to suffer the most , but deciduous trees were also strongly affected ( Schuldt et al . , 2020 ) . Weather data from our study area from 1950 onwards suggested that the climatic conditions for beech trees in the area investigated changed dramatically during this period . Roughly estimating the tree age from their trunk circumference ( Bošeľa et al . , 2014 ) , more than a third of the trees were already in place at the beginning of this period . About 60% were recruited prior to the acceleration of temperature change from the 1980s onwards . As a result , trees in the mountainous regions of the study area today experience climatic conditions comparable to those experienced by lowland trees in the 1950s , which in turn now experience a climate that used to be typical for regions much further South . Given the documented propensity of beech for local adaptation ( Gárate‐Escamilla et al . , 2019; Pluess et al . , 2016; Aranda et al . , 2015 ) , including drought ( Bolte et al . , 2016 ) , it is therefore conceivable that current conditions exceed the reaction norm of some previously locally well-adapted genotypes with detrimental consequences for their fitness . If the trend of an increasingly drier vegetation period persists , this will likely affect an even larger proportion of the currently growing beeches . Evolutionary genomics will be indispensable to predict and manage the impact of global change on biodiversity ( Waldvogel et al . , 2020 ) . As already shown for other partially managed ( tree ) species ( Stocks et al . , 2019 ) , in particular pool-GWAS approaches ( Endler et al . , 2016 ) have proven to be useful in guiding conservation management . Our strictly pairwise sampling design avoided many pitfalls of GWAS studies , arising , for example , from cryptic population structure and shared ancestry ( Hoban et al . , 2016; Wellenreuther and Hansson , 2016 ) . Despite presented evidence from this and other studies ( Schuldt et al . , 2020 ) that the observed crown damages in large parts of Central Europe used for phenotyping here are directly or indirectly due to the severe drought years 2018 and 2019 , we must acknowledge that we have no direct physiological proof that the trees surveyed here indeed suffered from drought stress . In addition , the observed diagnostic symptoms are not specific to drought stress . Nevertheless , an unknown independent stressor would have needed to accidentally co-occur spatially and temporally with the drought . The phenotypical drought response of individual trees may also be influenced by microspatial variation ( Carrière et al . , 2020 ) . In the present study , however , the mean distance between sampled paired trees of about 5 m assured that their roots systems largely overlapped . Thus , environmental variation in soil quality , rooting depth , water availability , or other factors should have been minimal . Please note that any phenotypical misclassification due to such microspatial differences would have rather dissimulated the genotypic differences found in GWAS than enhanced them artificially . As expected from previous studies ( Rajendra et al . , 2014 ) , we found no population structure among the sampling sites . Applying relatively strict significance thresholds , we found systematic genomic differences between the healthy and damaged trees . In all cases , these differences were quantitative and not categorical , that is , we found allele frequency changes but no fixed SNPs between phenotypes . Significant SNPs were mostly not clustered – we found on average 1 . 4 selected SNPs in a particular genomic region . These findings were in line with the observed very short average LD in F . sylvatica , indicating that polymorphisms associated with the two phenotypes were likely old-standing genetic variation ( Harris and Nielsen , 2013 ) . Moreover , such SNPs are mostly detached from the background in which they arose and they are therefore often the actual causal variants or at least in very close proximity ( Schaid et al . , 2018 ) . This observation is underlined by the high proportion of non-synonymous significant SNPs within genes , which in most cases caused substitution to an amino acid with different properties or even premature termination . Such deviant variants with likely substantial functional or conformational changes in the resulting proteins may be selectively neutral or nearly neutral under ancestral benign conditions , but may become selectively relevant under changing conditions ( Paaby and Rockman , 2014 ) . Interestingly , most of the allelic variants associated with a healthy phenotype were also the variants in the reference genome . This might be due to the choice of the F . sylvatica individual from which the reference genome was gained ( Mishra et al . , 2018 ) . This more than 300-year-old individual is standing at a particularly dry site on a rocky outcrop on the rim of a scarp where precipitation swiftly runs off . Trees at such sites were likely selected for drought tolerance . Even though the area sampled for this study was limited relative to the species distribution range , it comprised its core area . In addition , the climatic variation covered by the sampling sites for this study is representative for large parts of the species range ( Baumbach et al . , 2019 ) . The relatively limited population structure over large parts of the species range ( Magri et al . , 2006 ) , together with the propensity for long-range gene flow ( Belmonte et al . , 2008 ) , suggested that the genomic variation responsible for drought tolerance identified here is widely distributed ( Lander et al . , 2021 ) . Nevertheless , an assessment of the geographic distribution of the drought-related genomic variants over the entire distribution range would yield general insight into the species-wide architecture of this important trait . None of the genes found here was involved in a transcriptomic study on drought response in beech saplings ( Müller et al . , 2017 ) . However , most of the reliably annotated genes with or close to SNP loci significantly associated with drought phenotypes had putative homologs in other plant species previously shown to be involved in drought or different environmental stress response ( for citations , see Supplementary file 1C , D ) . For the remaining , not annotated genes , it remained unclear whether they had really never been associated with drought before , or whether we were just unable to make this link due to the lack of ( ecological ) annotation and standardised reporting ( Waldvogel et al . , 2021 ) . The involvement of in total 67 genes , together with the relatively flat effect size distribution , suggested that drought resistance in F . sylvatica is a moderately polygenic trait , which should respond well to artificial breeding attempts and natural selection . However , given the relatively strict threshold criteria , it is likely that more yet undetected loci contribute to the respective phenotypes . The low LD in beech predicts that an adaptation to drought will not compromise genome-wide genetic diversity and thus adaptation potential to other stressors . We achieved a high level of accuracy using genomic data to predict the drought phenotype from individuals not used to identify drought-associated SNP loci . However , due to the small sample size , LDA might have resulted in overfitting ( Hawkins , 2004 ) . We therefore also used a non-parametric Machine Learning algorithm that has been shown to produce more robust results , especially for small sample sizes ( Horenko , 2020 ) . Both analyses confirmed that we mainly identified alleles widespread throughout the sampled range and not locally specific . Besides , we confirmed a considerable level of genetic variation in the sampled regions . The observation that trees with the highest predictive values showed no loss of heterozygosity indicated that there is still adaptive potential for drought adaptation in the species ( Gienapp et al . , 2017 ) . With the SNP assay , we therefore created a tool that can ( i ) support the choice of seed trees for reforestations , ( ii ) provide decision guidance for selective logging , and ( iii ) monitor whether natural selection on this quantitative trait is already acting in the species . The current study can also serve as a starting point for molecular and physiological research on how the identified loci or variants may , alone or in concert , confer resilience or tolerance to a range of drought stress symptoms . In August/early September 2019 , we sampled leaf tissue of 402 F . sylvatica trees from 32 locations in Hessen/Germany ( set 1 , Figure 1 ) , of which 300 were used for the ( pool ) GWAS analysis . 43 , plus additional 53 trees which were sampled in August 2020 , additional 52 trees from four sites were sampled ( set 2 , Figure 1 ) which made up the confirmation set . The coordinates and characteristics of each site can be found in Supplementary file 1A . The sampling was performed in a strictly pairwise design . The pairs consisted of one tree with heavy damage of the crown ( lost or rolled up , dried leaves ) and one with an unaffected crown , respectively . This categorisation into least and most damaged trees was taken compared to the other trees in the respective forest patch . The pairs were a priori chosen such that the two trees were ( i ) mutually the closest neighbours with contrasting damage status ( i . e . no other tree in the direct sight line ) , ( ii ) free from apparent mechanical damage , fungal infestations , or other signs of illness , similar ( iii ) in tree height , ( iv ) trunk circumference , ( v ) light availability , and ( vi ) canopy closure . In addition , each pair was situated at least 30 m from the closest forest edge . For each tree of the chosen pairs , we recorded the exact position , distance to the pair member and the estimated tree height ( in 1 m increments ) , measured the trunk circumference at 150 cm height above the ground ( in 10 cm increments ) , and estimated the leaf loss of the crown and the proportion of dried leaves ( in 5% increments ) . We also recorded the estimated distance ( in 1 m resolution ) and the specific identity of the two closest neighbour trees for each pair member and calculated a competition index C as follows: C = S1/D1 +S2/D2 , where S1and S2 are the trunk diameter at 150 cm and D1and D2 the distances of the nearest and second nearest neighbour tree of the same size or larger than the focal tree . Photographs from the crown and the trunk were taken from the trees sampled in 2019 . From each tree , we sampled 5–10 fully developed leaves from low branches . The leaves sampled from each tree were placed in paper bags . After returning from the field , they were dried at 50°C for 30–90 min and then kept on salt until they could be stored at −80°C . Monthly daily mean minimum and maximum temperature values and precipitation data were obtained for the 1 × 1 km grid cells harbouring the sampling sites for the period between 1950 and 2019 . Data on the accumulated potential evapotranspiration during the growth season was obtained for the same grid cells . The data is publicly available from https://opendata . dwd . de/climate_environment/CDC/grids_germany/monthly/ . LAI data for the above grid cells was obtained from Copernicus ( http://www . copernicus . eu ) for the period 2014–2019 , considering only the month of August . To see whether drought conditions influenced leaf coverage of the woods at the sampling sites , we calculated the relative annual deviation of LAI from the 2014 value . We correlated it to the relative deviation of the cumulated potential evatransporation over the growth season from 2014 . The year 2014 was used as a baseline because of the significant drought increase since then ( Büntgen et al . , 2021 ) . Please note that the absolute level of LAI depends on the wood coverage , vegetation density , and species composition of each plot . Changes in LAI are thus not exclusively due to drought damage in beech . DNA was extracted from 12 . 5 mm² of a single leaf from each tree following the NucleoMag Plant Kit ( Macherey Nagel , Düren , Germany ) protocol . We setup four DNA pools for Pool-GWAS by pooling equal amounts of DNA from each individual: damaged individuals from the Southern part ( dSouth ) , healthy individuals from the South ( hSouth ) , damaged North ( dNorth ) , and healthy North ( hNorth ) . The Southern pools consisted of 100 individuals each , the Northern pools of 50 individuals each . The pools were sent to Novogene ( Cambridge , UK ) for library construction and 150 bp paired end sequencing with 350 bp insert size with 25 Gb data for the Northern and 38 Gb data for the Southern samples . The 100 individuals used to construct the Northern pools were also individually resequenced . The exact composition of the genomic pools can be found in Supplementary file 1A . All sequence information can be found on the European Nucleotide Archive ( ENA ) under project accession number PRJEB24056 . We used an improved version of the recently published reference genome for the European beech ( Mishra et al . , 2018 ) . Contiguity was improved to chromosome level using Hi-C reads with the help of the allhic software after excluding the probable organelle backbones from the earlier assembly that was generated from the Illumina-corrected PacBio reads using Canu assembler ( Mishra et al . , 2021 ) accession no . PRJNA450822 . Reads of pools and individual resequencing were trimmed using the wrapper tool autotrim v0 . 6 . 1 ( Waldvogel et al . , 2018 ) that integrates trimmomatic ( Bolger et al . , 2014 ) for trimming and fastQC ( Andrews , 2010 ) for quality control . The trimmed reads were then mapped on the latest chromosome-level build of the F . sylvatica genome using the BWA mem algorithm v . 0 . 7 . 17 ( Li and Durbin , 2009 ) . Low-quality reads were subsequently filtered and SNPs were initially called using samtools v . 1 . 10 ( Li et al . , 2009 ) . A PCA was conducted on unlinked SNPs using the R package Factoextra v . 1 . 0 . 7 ( Kassambara and Mundt , 2017 ) . The PoPoolation pipeline 2_2012 ( Kofler et al . , 2011a; Kofler et al . , 2011b ) was used to call SNPs and remove indels from the four pools . Allele frequencies for all SNPs with a coverage between 15× and 100× with a minimum allele count of 3 were estimated with the R library PoolSeq v . 0 . 35 ( Taus et al . , 2017 ) . The statistical test to detect significant allele frequency differences among damaged and healthy trees was the Cochran–Mantel–Haenszel test . With this test , a 2 × 2 table was created for each variable position and region with two phenotypes ( healthy and damaged ) . The read counts of each allele for each phenotype were treated as the dependent variables . We controlled for false discovery rate using the Benjamini–Hochberg correction in R package p . adjust . For the individual resequencing data , we followed the GATK-pipeline 4 . 1 . 3 . 0 ( DePristo et al . , 2011 ) . In short , Picard tools v . 2 . 20 . 8 was used to mark duplicates . GVCF files were created with HaplotypeCaller and genotyped with GEnotypeGVCFs . Since we did not have a standard SNP set , we hard-filtered SNPs with VariantFiltration QD < 2 . 0 , MQ < 50 . 0 , MQRankSum < 12 . 5 , ReadPosRankSum < 8 . 0 , FS > 80 . 0 , SOR > 4 . 0 , and QUAL < 10 . 0 . This conservative SNP set was used for base recalibration before running the HaplotypeCaller pipeline a second round . Finally , the genotyped vcf-files were filtered using vcftools with --maf 0 . 03 --max-missing 0 . 9 --minQ 25 --min-meanDP 10 --max-meanDP 50 --minDP 10 --maxDP 50 . The detailed pipeline can be found in Supplementary file 1G . To conduct the GWAS association on the above-generated SNP set with phenotypes being either damaged or healthy and to generate a PCA on the SNP positions of the individually resequenced trees , we used PLINK 1 . 9 ( Purcell et al . , 2007 ) . The detailed workflow can be found in Supplementary file 1G . We calculated a non-parametric ANOSIM on an inter-individual Euclidean distance matrix based on the first 10 principal components to infer whether the trees within phenotype groups are overall genetically more similar than within groups ( 9999 permutations; Hammer et al . , 2001 ) . The expected length of segregating haplotypes in a species depends on the recombination rate and their age . The former can be approximated by an estimate of LD . To determine LD decay based on individually resequenced data , we used the software LDkit v 1 . 0 . 0 ( Tang et al . , 2020 ) , in 1 kb and 100 kb windows . We inferred whether significantly differentiated SNPs within genes lead to a ( non- ) synonymous amino acid substitution using tbg-tools v0 . 2 ( https://github . com/Croxa/tbg-tools; Schoennenbeck et al . , 2021 ) . The protein sequences of the identified genes were used in a blastp search against all non-redundant GenBank CDS translations , PDB , SwissProt , PIR , PRF to infer potential gene functions . For each search , only the single BLAST- top hit was considered . For the design of SNPtype assays , we used the web-based D3 assay design tool ( Fluidigm Corp . ) . We aimed in first preference for the most significant SNPs of each genomic region identified by Pool-GWAS ( 80 loci ) . If this was technically impossible and the region harboured more than a single significant SNP , we opted for the second most significant SNP and so forth . This resulted finally in 76 suitable loci . The remaining 20 loci were recruited from the 20 most significant SNPs of the PLINK analysis that were not scored in the Pool-GWAS . For validation of drought susceptibility-associated SNPs , we conducted SNP genotyping on 96 . 96 Dynamic Arrays ( Fluidigm ) with integrated fluidic circuits ( Wang et al . , 2009 ) ( N = 96 ) to validate the effectiveness of the identified SNPs in discriminating healthy from damaged trees . Prior to genotyping PCR , DNA extracts were normalised to approximately 5–10 ng/µl . They underwent a pre-amplification PCR ( Specific Target Amplification [STA] ) according to the manufacturer's protocol to enrich target loci . PCR products were diluted 1:10 with DNA suspension buffer ( TEKnova , PN T0221 ) before further use . Genotyping was performed according to the manufacturer's recommendations . Four additional PCR cycles were added to accommodate for samples of lower quality or including inhibitors ( von Thaden et al . , 2020 ) . Fluorescent data were measured using the EP1 ( Fluidigm ) and analysed with the SNP Genotyping Analysis Software version 4 . 1 . 2 ( Fluidigm ) . The automated scoring of the scatter plots was checked visually and , if applicable , manually corrected . To predict drought susceptibility from genotype data , we used an LDA on 92 genotypes scored with the Fluidigm assay at 70 loci . Genotypes homozygous for the reference allele were scored as 0 , heterozygous as 1 and homozygous alternate alleles as 2 . We used the LDA option implemented in PAST v . 4 . 05 . ( Hammer et al . , 2001 ) . We also used a non-parametric entropy-based Scalable Probabilistic Analysis framework ( eSPA ) . This method allows simultaneous solution of feature selection and clustering problems , meaning that does not rely on a particular choice of user-defined parameters and has been shown to produce more robust results , especially for small sample sizes ( Gerber2020 , Horenko2020 ) . Following the suggestion of the user manual , eSPA was run 100 times with independent cross-validations of the area under the curve ( AUC ) on the validation data .	Climate change is having a serious impact on many ecosystems . In the summer of 2018 and 2019 , around two thirds of European beech trees were damaged or killed by extreme drought . It is critical to keep these beech woods healthy , as they are central to the survival of over 6 , 000 other species of animals and plants . The level of damage caused by the drought varied between forests . However , not all the trees in each forest responded in the same way , with severely damaged trees often sitting next to fully healthy ones . This suggests that the genetic make-up of each tree determines how well it can adapt to drought rather than its local environment . To investigate this further , Pfenninger et al . studied the genome of over 400 European beech trees from the Hesse region in Germany . The samples came from pairs of neighbouring trees that had responded differently to the droughts . The analysis found more than 80 parts of the genome that differed between healthy and damaged trees . Pfenninger et al . then used this information to create a genetic test which can quickly and inexpensively predict how well an individual beech tree might survive in a drought . Applying this test to another 92 trees revealed that it can reliably detect which ones were healthy and which ones were damaged . Beech forests are typically managed by private owners , agencies or breeders that could use this genetic test to select and reproduce trees that are better adapted to drought . The goal now is to develop the test so that it can be used more widely to manage European beech trees and potentially other species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "genetics", "and", "genomics" ]	2021	Genomic basis for drought resistance in European beech forests threatened by climate change
The Drosophila embryo transiently exhibits a double-segment periodicity , defined by the expression of seven 'pair-rule' genes , each in a pattern of seven stripes . At gastrulation , interactions between the pair-rule genes lead to frequency doubling and the patterning of 14 parasegment boundaries . In contrast to earlier stages of Drosophila anteroposterior patterning , this transition is not well understood . By carefully analysing the spatiotemporal dynamics of pair-rule gene expression , we demonstrate that frequency-doubling is precipitated by multiple coordinated changes to the network of regulatory interactions between the pair-rule genes . We identify the broadly expressed but temporally patterned transcription factor , Odd-paired ( Opa/Zic ) , as the cause of these changes , and show that the patterning of the even-numbered parasegment boundaries relies on Opa-dependent regulatory interactions . Our findings indicate that the pair-rule gene regulatory network has a temporally modulated topology , permitting the pair-rule genes to play stage-specific patterning roles . Segmentation is a developmental process that subdivides an animal body axis into similar , repeating units ( Hannibal and Patel , 2013 ) . Segmentation of the main body axis underlies the body plans of arthropods , annelids and vertebrates ( Telford et al . , 2008; Balavoine , 2014; Graham et al . , 2014 ) . In arthropods , segmentation first involves setting up polarised boundaries early in development to define 'parasegments' ( Martinez-Arias and Lawrence , 1985 ) . Parasegment boundaries are maintained by an elaborate and strongly-conserved signalling network of 'segment-polarity' genes ( Ingham , 1988; Perrimon , 1994; DiNardo et al . , 1994; Sanson , 2001; Janssen and Budd , 2013 ) . In all arthropods yet studied , the segmental stripes of segment-polarity genes are initially patterned by a group of transcription factors known as the 'pair-rule' genes ( Green and Akam , 2013; Peel et al . , 2005; Damen et al . , 2005 ) . The pair-rule genes were originally identified in a screen for mutations affecting the segmental pattern of the Drosophila melanogaster larval cuticle ( Nüsslein-Volhard and Wieschaus , 1980 ) . They appeared to be required for the patterning of alternate segment boundaries ( hence 'pair-rule' ) and were subsequently found to be expressed in stripes of double-segment periodicity ( Hafen et al . , 1984; Akam , 1987 ) . Early models of Drosophila segmentation suggested that the blastoderm might be progressively patterned into finer-scale units by some reaction-diffusion mechanism that exhibited iterative frequency-doubling ( reviewed in Jaeger , 2009 ) . The discovery of a double-segment unit of organisation seemed to support these ideas , and pair-rule patterning was therefore thought to be an adaptation to the syncytial environment of the early Drosophila embryo , which allows diffusion of gene products between neighbouring nuclei . However , the transcripts of pair-rule genes are apically localised during cellularisation of the blastoderm , and thus pair-rule patterning occurs in an effectively cellular environment ( Edgar et al . , 1987; Davis and Ish-Horowicz , 1991 ) . Furthermore , double-segment periodicity of pair-rule gene expression is also found in some sequentially segmenting ( 'short germ' ) insects ( Patel et al . , 1994 ) , indicating that pair-rule patterning predates the evolution of simultaneous ( 'long germ' ) segmentation ( Figure 1 ) . 10 . 7554/eLife . 18215 . 003Figure 1 . The evolution of pair-rule patterning predates the evolution of long germ segmentation . ( A ) Single segment periodicity is ancestral in arthropod segmentation , being found in spiders , millipedes , crustaceans and some insects ( Davis et al . , 2005; Pueyo et al . , 2008 ) . 'Pair-rule' patterning , involving an initial double segment periodicity of pair-rule gene expression , appears to have evolved independently at least twice . It is found in insects and certain centipedes ( Davis et al . , 2001; Chipman et al . , 2004 ) . ( B ) Long germ segmentation is likely to have evolved independently multiple times within holometabolous insects , from an ancestral short germ state ( Liu and Kaufman , 2005 ) . Light blue boxes for the Lepidoptera and Hymenoptera indicate that short germ segmentation is relatively uncommon in these clades . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 003 The next set of models for pair-rule patterning were motivated by genetic dissection of the early regulation of the segment-polarity gene engrailed ( en ) . It was found that odd-numbered en stripes – and thus the anterior boundaries of odd-numbered parasegments ( hereafter 'odd-numbered parasegment boundaries' ) – require the pair-rule gene paired ( prd ) , but not another pair-rule gene fushi tarazu ( ftz ) , while the opposite was true for the even-numbered en stripes and their associated ( 'even-numbered' ) parasegment boundaries ( DiNardo and O'Farrell , 1987 ) . Differential patterning of alternate segment-polarity stripes , combined with the observation that the different pair-rule genes are expressed with different relative phasings along the anterior-posterior ( AP ) axis , led to models where static , partially overlapping domains of pair-rule gene expression form a combinatorial regulatory code that patterns the blastoderm with single-cell resolution ( DiNardo and O'Farrell , 1987; Gergen and Butler , 1988; Weir et al . , 1988; Coulter et al . , 1990; Morrissey et al . , 1991 ) . However , pair-rule gene expression domains are not static . One reason for this is that their upstream regulators , the gap genes , are themselves dynamically expressed , exhibiting expression domains that shift anteriorly over time ( Jaeger et al . , 2004; El-Sherif and Levine , 2016 ) . Another major reason is that , in addition to directing the initial expression of the segment-polarity genes , pair-rule genes also cross-regulate one another . Pair-rule proteins and transcripts turn over extremely rapidly ( Edgar et al . , 1986; Nasiadka and Krause , 1999 ) , and therefore regulatory feedback between the different pair-rule genes mediates dynamic pattern changes throughout the period that they are expressed . Most strikingly , many of the pair-rule genes undergo a transition from double-segment periodicity to single-segment periodicity at the end of cellularisation . The significance of this frequency-doubling is not totally clear . In some cases , the late , segmental stripes are crucial for proper segmentation ( Cadigan et al . , 1994b ) , but in others they appear to be dispensable ( Coulter et al . , 1990; Fujioka et al . , 1995 ) , or their function ( if any ) is not known ( Klingler and Gergen , 1993; Jaynes and Fujioka , 2004 ) . More recent models of pair-rule patterning recognise that the pair-rule genes form a complex gene regulatory network that mediates dynamic patterns of expression ( Edgar et al . , 1989; Sánchez and Thieffry , 2003; Jaynes and Fujioka , 2004 ) . However , whereas other stages of Drosophila segmentation have been extensively studied from a dynamical systems perspective ( reviewed in Jaeger , 2009; Grimm et al . , 2010; Jaeger , 2011 ) , we do not yet have a good systems-level understanding of the pair-rule gene network ( Jaeger , 2009 ) . This appears to be a missed opportunity: not only do the pair-rule genes exhibit fascinating transcriptional regulation , but their interactions are potentially very informative for comparative studies with short germ arthropods . These include the beetle Tribolium castaneum , in which the pair-rule genes form a segmentation oscillator ( Sarrazin et al . , 2012; Choe et al . , 2006 ) . To better understand exactly how pair-rule patterning works in Drosophila , we carried out a careful analysis of pair-rule gene regulation during cellularisation and gastrulation , drawing on both the genetic literature and a newly generated dataset of double-fluorescent in situs . Surprisingly , we found that the majority of regulatory interactions between pair-rule genes are not constant , but undergo dramatic changes just before the onset of gastrulation . These regulatory changes mediate the frequency-doubling phenomena observed in the embryo at this time . We then realised that all the regulatory interactions specific to the late pair-rule gene regulatory network seem to require the non-canonical pair-rule gene odd-paired ( opa ) . opa was identified through the original Drosophila segmentation screen as being required for the patterning of the even-numbered parasegment boundaries ( Jürgens et al . , 1984 ) . However , rather than being expressed periodically like the rest of the pair-rule genes , opa is expressed ubiquitously throughout the trunk region ( Benedyk et al . , 1994 ) . The reported appearance of Opa protein temporally correlates with the time we see regulatory changes in the embryo , indicating that it may be directly responsible for these changes . We propose that Opa provides a source of temporal information that acts combinatorially with the spatial information provided by the periodically expressed pair-rule genes . Pair-rule patterning thus appears to be a two-stage process that relies on the interplay of spatial and temporal signals to permit a common set of patterning genes to carry out stage-specific regulatory functions . We carried out double fluorescent in situ hybridisation on fixed wild-type Drosophila embryos for all pairwise combinations of the pair-rule genes hairy , even-skipped ( eve ) , runt , fushi tarazu ( ftz ) , odd-skipped ( odd ) , paired ( prd ) and sloppy-paired ( slp ) . Because the expression patterns of these genes develop dynamically but exhibit little embryo-to-embryo variability ( Surkova et al . , 2008; Little et al . , 2013; Dubuis et al . , 2013 ) , we were able to order images of individual embryos by inferred developmental age . This allowed us to produce pseudo time-series that illustrate how pair-rule gene expression patterns change relative to one another during early development ( Figure 2 ) . 10 . 7554/eLife . 18215 . 004Figure 2 . Representative double fluorescent in situ hybridisation data for three combinations of pair-rule genes . This figure shows a small subset of our wild-type dataset . Each column represents a different pairwise combination of in situ probes , while each row shows similarly-staged embryos of increasing developmental age . All panels show a lateral view , anterior left , dorsal top . Individual channels are shown in grayscale below each double-channel image . For ease of comparison , the signal from each gene is shown in a different colour in the double-channel images . Time classes are arbitrary , meant only to illustrate the progressive stages of pattern maturation between early cellularisation ( t1 ) and late gastrulation ( t6 ) . Note that the developing pattern of odd expression in the head provides a distinctive and reliable indicator of embryo age . Scale bar = 100 μm . The complete dataset is available from the Dryad Digital Repository ( Clark and Akam , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 004 The expression profile of each individual pair-rule gene has been carefully described previously ( Hafen et al . , 1984; Ingham and Pinchin , 1985; Macdonald et al . , 1986; Kilchherr et al . , 1986; Gergen and Butler , 1988; Coulter et al . , 1990; Grossniklaus et al . , 1992 ) , and high-quality relative expression data are available for pair-rule proteins ( Pisarev et al . , 2009 ) . In addition , expression atlases facilitate the comparison of staged , averaged expression profiles of many different blastoderm patterning genes at once ( Fowlkes et al . , 2008 ) . However , because the pair-rule genes are expressed extremely dynamically and in very precise patterns , useful extra information can be gleaned by directly examining relative expression patterns in individual embryos . In particular , we have found these data invaluable for understanding exactly how stripe phasings change over time , and for interrogating regulatory hypotheses . In addition , we have characterised pair-rule gene expression up until early germband extension , whereas blastoderm expression atlases stop at the end of cellularisation . Our entire wild-type dataset ( 32 gene combinations , >600 individual embryos ) is available from the Dryad Digital Repository ( Clark and Akam , 2016 ) . We hope it proves useful to the Drosophila community . We classify the striped expression of the pair-rule genes into three temporal phases ( Figure 3A ) . Phase 1 ( equivalent to phase 1 of Schroeder et al . , 2011; timepoint 1 in Figure 2 ) corresponds to early cellularisation , before the blastoderm nuclei elongate . Phase 2 ( spanning phases 2 and 3 of Schroeder et al . , 2011; timepoints 2–4 in Figure 2 ) corresponds to mid cellularisation , during which the plasma membrane progressively invaginates between the elongated nuclei . Phase 3 ( starting at phase 4 of Schroeder et al . , 2011 but continuing beyond it; timepoints 5–6 in Figure 2 ) corresponds to late cellularisation and gastrulation . Our classification is a functional one , based on the times at which different classes of pair-rule gene regulatory elements ( Figure 3B ) have been found to be active in the embryo . 10 . 7554/eLife . 18215 . 005Figure 3 . Three phases of pair-rule gene expression , usually mediated by different classes of regulatory element . ( A ) Representative expression patterns of each of the seven pair-rule genes at phase 1 ( early cellularisation ) , phase 2 ( mid cellularisation ) and phase 3 ( gastrulation ) . Pair-rule genes are classified as 'primary' or 'secondary' based on their regulation and expression during phase 1 ( see text ) . All panels show a lateral view , anterior left , dorsal top . Note that the cephalic furrow may obscure certain anterior stripes during phase 3 . ( B ) Illustrative diagrams of the different kinds of regulatory elements mediating pair-rule gene expression . 'Stripe-specific' elements are regulated by gap genes and give rise to either one or two stripes each . 'Zebra' elements are regulated by pair-rule genes and give rise to seven stripes . 'Late' expression patterns may be generated by a single-element generating segmental stripes , or by a combination of two elements each generating a distinct pair-rule pattern . The coloured outlines around the panels in ( A ) correspond to the colours of the different classes of regulatory elements in ( B ) , and indicate how each phase of expression of a given pair-rule gene is thought to be regulated . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 005 During phase 1 , expression of specific stripes is established through compact enhancer elements mediating gap gene inputs ( Howard et al . , 1988; Goto et al . , 1989; Harding et al . , 1989; Pankratz and Jäckle , 1990 ) . hairy , eve and runt all possess a full set of these 'stripe-specific' elements , together driving expression in all seven stripes , while ftz lacks an element for stripe 4 , and odd lacks elements for stripes 2 , 4 and 7 ( Schroeder et al . , 2011 ) . These five genes are together classified as the 'primary' pair-rule genes , because in all cases the majority of their initial stripe pattern is established de novo by non-periodic regulatory inputs . The regulation of various stripe-specific elements by gap proteins has been studied extensively ( for example Small et al . , 1992 , 1996 ) . Phase 2 is dominated by the expression of so-called 'zebra' ( or '7-stripe' ) elements ( Hiromi et al . , 1985; Dearolf et al . , 1989; Butler et al . , 1992 ) . These elements , which tend to be relatively large ( Gutjahr et al . , 1994; Klingler et al . , 1996; Schroeder et al . , 2011 ) , are regulated by pair-rule gene inputs and thus produce periodic output patterns . The stripes produced from these elements overlap with the stripes generated by stripe-specific elements , and often the two sets of stripes appear to be at least partially redundant . For example , ftz and odd lack a full complement of stripe-specific elements ( see above ) , while the stripe-specific elements of runt are dispensable for segmentation ( Butler et al . , 1992 ) . Neither hairy nor eve appears to possess a zebra element , and thus their expression during phase 2 is driven entirely by their stripe-specific elements . ( Note that the 'late' ( or 'autoregulatory' ) element of eve ( Goto et al . , 1989; Harding et al . , 1989 ) does generate a periodic pattern and has therefore been considered to be analogous to the zebra elements of other pair-rule genes . However , because it is not expressed until phase 3 ( Schroeder et al . , 2011 ) , we do not classify it as such . ) In addition to the five primary pair-rule genes , there are two other pair-rule genes , prd and slp , that turn on after regular periodic patterns of the other genes have been established . These genes possess only a single , anterior stripe-specific element , and their trunk stripes are generated by a zebra element alone ( Schroeder et al . , 2011 ) . Because ( ignoring the head stripes ) these genes are regulated only by other pair-rule genes , and not by gap genes , they are termed as the 'secondary' pair-rule genes . The third , 'late' phase of expression is the least understood . Around the time of gastrulation , most of the pair-rule genes undergo a transition from double-segmental stripes to single-segmental stripes . For prd , this happens by splitting of its early , broad pair-rule stripes . In contrast , odd , runt and slp show intercalation of 'secondary' stripes between their 'primary' 7-stripe patterns . Secondary stripes of eve also appear at gastrulation , but these 'minor' stripes ( Macdonald et al . , 1986 ) are extremely weak ( usually undetectable in our fluorescent in situs ) , and not comparable to the rapidly developing segmental expression of prd , odd , runt and slp . Expression of hairy and ftz remains double segmental . In some cases , discrete enhancer elements have been found that mediate just the secondary stripes ( Klingler et al . , 1996 ) , while in other cases , all 14 segmental stripes are likely to be regulated coordinately ( Fujioka et al . , 1995 ) . In certain cases , non-additive interactions between enhancers play a role in generating the segmental pattern ( Prazak et al . , 2010; Gutjahr et al . , 1994 ) . The functional significance of the late patterns is not always clear , since they are usually not reflected in pair-rule gene mutant cuticle phenotypes ( Kilchherr et al . , 1986; Coulter et al . , 1990 ) . In the remainder of this paper , we investigate the nature and causes of the pattern transitions that occur between the end of phase 2 and the beginning of phase 3 . A detailed analysis of the timing and dynamics of pair-rule gene expression during phase 2 will be covered elsewhere . As noted above , four of the seven pair-rule genes undergo a transition from double-segment periodicity to regular single-segment periodicity at the end of cellularisation ( Figure 3 ) . These striking pattern changes could be caused simply by feedback interactions within the pair-rule and segment-polarity gene networks . Alternatively , they could be precipitated by some extrinsic temporal signal ( or signals ) . Comparing between genes , we find that the pattern changes develop almost simultaneously ( Figure 4; Figure 4—figure supplement 1 ) , although there are slight differences in the times at which the first signs of frequency-doubling become detectable . ( The prd trunk stripes split just before the odd secondary stripes start to appear , while the secondary stripes of slp and runt appear just after ) . These events appear to be spatiotemporally modulated: they show a short but noticeable AP time lag , and also a DV pattern – frequency-doubling occurs first mid-laterally , and generally does not extend across the dorsal midline . In addition , the secondary stripes of slp are not expressed in the mesoderm , while the ventral expression of odd secondary stripes is only weak . 10 . 7554/eLife . 18215 . 006Figure 4 . Frequency-doubling of pair-rule gene expression patterns is almost simultaneous and coincides with the first expression of the segment-polarity genes . Each row shows the expression of a particular pair-rule gene or segment-polarity gene , while each column represents a particular developmental timepoint . Late phase 2 and early phase 3 both correspond to late Bownes stage 5; gastrulation is Bownes stage 6 , and early germband extension is Bownes stage 7 ( Bownes , 1975; Campos-Ortega and Hartenstein , 1985 ) . All panels show a lateral view , anterior left , dorsal top . GBE = germband extension . The figure represents about 20 min of development at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 00610 . 7554/eLife . 18215 . 007Figure 4—figure supplement 1 . Relative expression of pair-rule genes during frequency-doubling . Each row shows the relative expression of two pair-rule genes , while each column represents a particular developmental timepoint . Late phase 2 and early phase 3 both correspond to late Bownes stage 5; gastrulation is Bownes stage 6 , and early germband extension is Bownes stage 7 ( Bownes , 1975; Campos-Ortega and Hartenstein , 1985 ) . All panels show lateral or ventrolateral views , anterior left , dorsal top . Single-channel images are shown in greyscale below each double-channel image ( the channel listed first in the row label is always on the left ) . Each gene is shown as a different colour in the double-channel images . GBE = germband extension . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 00710 . 7554/eLife . 18215 . 008Figure 4—figure supplement 2 . Relative expression of segment-polarity genes and pair-rule genes during frequency-doubling . Each row shows the relative expression of a particular pair-rule gene and segment-polarity gene combination , while each column represents a particular developmental timepoint . Late phase 2 and early phase 3 both correspond to late Bownes stage 5; gastrulation is Bownes stage 6 , and early germband extension is Bownes stage 7 ( Bownes , 1975; Campos-Ortega and Hartenstein , 1985 ) . All panels show a lateral view , anterior left , dorsal top . Single-channel images are shown in greyscale below each double-channel image ( the channel listed first in the row label is always on the left ) . Each segment-polarity gene is shown in a different colour , while pair-rule gene expression is shown in red . GBE = germband extension . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 008 We also investigated the timing of the frequency-doubling events relative to the appearance of expression of the segment-polarity genes en , gooseberry ( gsb ) and wingless ( wg ) ( Figure 4; Figure 4—figure supplement 2 ) . We find that the spatiotemporal pattern of segment-polarity gene activation coincides closely with that of pair-rule frequency-doubling – starting at the beginning of phase 3 , and rapidly progressing over the course of gastrulation . Only around 20 min separate a late stage 5 embryo ( with double-segment periodicity of pair-rule gene expression and no segment-polarity gene expression ) from a late stage 7 embryo ( with regular segmental expression of both pair-rule genes and segment-polarity genes ) ( Campos-Ortega and Hartenstein , 1985 ) . We can make three conclusions from the timing of these events . First , segment-polarity gene expression cannot be precipitating the frequency-doubling of pair-rule gene expression , because frequency-doubling occurs before segment-polarity proteins would have had time to be synthesised . Second , the late , segmental patterns of pair-rule gene expression do not play a role in regulating the initial expression of segment-polarity genes , because they are not reflected at the protein level until after segmental patterns of segment-polarity gene transcripts are observed . Third , the synchrony of pair-rule gene frequency-doubling and segment-polarity gene activation is consistent with co-regulation of these events by a single temporal signal . It is clear that a dramatic change overtakes pair-rule gene expression at gastrulation . For a given gene , an altered pattern of transcriptional output could result from an altered spatial pattern of regulatory inputs , or , alternatively , altered regulatory logic . Pair-rule proteins provide most of the spatial regulatory input for pair-rule gene expression at both phase 2 and phase 3 . Therefore , the fact that the distributions of pair-rule proteins are very similar at the end of phase 2 and the beginning of phase 3 ( Pisarev et al . , 2009 ) suggests that it must be the 'input-output functions' of pair-rule gene transcription that change to bring about the new expression patterns . For example , consider the relative expression patterns of prd and odd ( Figure 5 ) . There is abundant experimental evidence that the splitting of the prd stripes is caused by direct repression by Odd protein . The primary stripes of odd lie within the broad prd stripes , and the secondary interstripes that form within the prd stripes at gastrulation correspond precisely to those cells that express odd ( Figure 5D ) . Furthermore , the prd stripes do not split in odd mutant embryos ( Baumgartner and Noll , 1990; Saulier-Le Dréan et al . , 1998 ) , and prd expression is largely repressed by ectopically expressed Odd protein ( Saulier-Le Dréan et al . , 1998; Goldstein et al . , 2005 ) . 10 . 7554/eLife . 18215 . 009Figure 5 . Odd does not repress prd transcription until phase 3 . Relative expression of prd and odd is shown in a late phase 2 embryo ( just prior to frequency doubling ) and an early phase 3 embryo ( showing the first signs of frequency doubling ) . ( A , B ) Whole embryos , lateral view , anterior left , dorsal top . Individual channels are shown to the right of each double-channel image , in the same vertical order as the panel label . ( C , D ) Blow-ups of expression in stripes 2–6; asterisks mark the location of odd primary stripes . Thresholded images ( C’ , D’ ) highlight regions of overlapping expression ( yellow pixels ) . Considerable overlap between prd and odd expression is observed at phase 2 but not at phase 3 . Note that the prd expression pattern is the combined result of initially broad stripes of medium intensity , and intense two-cell wide 'P' stripes overlapping the posterior of each of the broad stripes ( arrowheads in C’’’ , D’’’ ) . The two sets of stripes are mediated by separate stretches of DNA ( Gutjahr et al . , 1994 ) , and must be regulated differently , since the 'P' stripes remain insensitive to ectopic Odd even during phase 3 ( Saulier-Le Dréan et al . , 1998; Goldstein et al . , 2005 ) . Scale bars = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 009 However , prior to prd stripe splitting , prd and odd are co-expressed in the same cells , with no sign that prd is sensitive to repression by Odd ( Figure 5C ) . Because prd expression begins at a time when Odd protein is already present ( Pisarev et al . , 2009 ) , this co-expression cannot be explained by protein synthesis delays . We therefore infer that Odd only becomes a repressor of prd at gastrulation , consistent with previous observations that aspects of Odd regulatory activity are temporally restricted ( Saulier-Le Dréan et al . , 1998 ) . This apparent temporal switch in the regulatory function of Odd is not unique . We have carefully examined pair-rule gene stripe phasings just before and just after the double-segment to single-segment transition , and find that these patterns do indeed indicate significant changes to the control logic of multiple pair-rule genes . The results of this analysis are presented in Appendix 1 . In summary , a number of regulatory interactions seem to disappear at the beginning of phase 3: repression of odd by Hairy , repression of odd by Eve , and repression of slp by Runt . These regulatory interactions are replaced by a number of new interactions: repression of prd by Odd , repression of odd by Runt , repression of runt by Eve and repression of slp by Ftz . At the same time that these regulatory changes are observed , new elements for eve and runt turn on and various segment-polarity genes start to be expressed . The outcome of all these regulatory changes is a coordinated transition to single-segment periodicity . We have schematised this transition in Figure 6 . Our diagrams are in broad agreement with the interpretation of Jaynes and Fujioka ( Jaynes and Fujioka , 2004 ) , although we characterise the process in greater temporal detail and distinguish between transcript and protein distributions at each timepoint . 10 . 7554/eLife . 18215 . 010Figure 6 . Schematic diagram of the transition to single-segment periodicity . Schematic diagram showing segmentation gene expression at late phase 2 ( A ) , early phase 3 ( B ) and late phase 3 ( C ) . The horizontal axis represents an idealised portion of the AP axis ( ~12 nuclei across ) . The grey vertical lines in ( A , B ) demarcate a double parasegment repeat ( ~8 nuclei across ) , while black lines in ( C ) indicate future parasegment boundaries . The patterns of protein expression ( intense colours ) and transcript expression ( paler colours ) of the pair-rule genes are shown at each timepoint . Those of the segment-polarity genes en and wg are additionally shown at the later timepoints . Transcript distributions were inferred from our double in situ data , while pair-rule protein distributions were inferred mainly from triple antibody stains in the FlyEx database ( Pisarev et al . , 2009 ) . Additional protein expression information for late phase 3 ( equivalent to the onset of germband extension ) was gathered from published descriptions ( Frasch et al . , 1987; DiNardo et al . , 1985; van den Heuvel et al . , 1989; Gutjahr et al . , 1993; Lawrence and Johnston , 1989; Carroll et al . , 1988 ) . Fading expression of Eve and Runt is represented by lighter red and green sections in ( B ) . The transient 'minor' stripes of Eve are represented by faint red in ( C ) . Note that this diagram does not capture the graded nature of pair-rule protein distributions during cellularisation . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 010 Having identified the regulatory changes detailed above , we wanted to know how they are made to happen in the embryo . Because they all occur within a very short time window ( Figure 4 ) , they could potentially all be co-regulated by a single temporal signal that would instruct a regulatory switch . We reasoned that if this hypothetical signal were absent , the regulatory changes would not happen . This would result in a mutant phenotype in which frequency-doubling events do not occur , and segment-polarity expression is delayed . We then realised that this hypothetical phenotype was consistent with descriptions of segmentation gene expression in mutants of the non-canonical pair-rule gene , odd-paired ( opa ) ( Benedyk et al . , 1994 ) . This gene is required for the splitting of the prd stripes and the appearance of the secondary stripes of odd and slp ( Baumgartner and Noll , 1990; Benedyk et al . , 1994; Swantek and Gergen , 2004 ) . It is also required for the late expression of runt ( Klingler and Gergen , 1993 ) , and for the timely expression of en and wg ( Benedyk et al . , 1994 ) . The opa locus was originally isolated on account of its cuticle phenotype , in which odd-numbered segments ( corresponding to even-numbered parasegments ) are lost ( Jürgens et al . , 1984 ) . For many years afterwards , opa was assumed to be expressed in a periodic pattern of double-segment periodicity similar to the other seven pair-rule genes ( for example , see Coulter and Wieschaus , 1988; Ingham and Baker , 1988; Weir et al . , 1988; Baumgartner and Noll , 1990; Lacalli , 1990 ) . When opa , which codes for a zinc finger transcription factor , was finally cloned , it was found – surprisingly – to be expressed uniformly throughout the trunk ( Benedyk et al . , 1994 ) . Presumed to be therefore uninstructive for spatial patterning , it has received little attention in the context of segmentation since . However , we realised that Opa could still be playing an important role in spatial patterning . By providing temporal information that would act combinatorially with the spatial information carried by the canonical pair-rule genes , Opa might permit individual pair-rule genes to carry out different patterning roles at different points in time . We examined opa expression relative to other segmentation genes , and found an interesting correlation with the spatiotemporal pattern of segmentation ( Figure 7 ) . As previously reported ( Benedyk et al . , 1994 ) , the earliest expression of opa is in a band at the anterior of the trunk , which we find corresponds quite closely with the head stripe of prd ( data not shown ) . Expression in the rest of the trunk quickly follows , and persists until germband extension , at which point expression becomes segmentally modulated ( Figure 7I ) . 10 . 7554/eLife . 18215 . 011Figure 7 . Spatiotemporal expression of opa relative to odd . Expression of opa relative to odd from early cellularisation until mid germband extension . ( A ) phase 1 , lateral view; ( B ) early phase 2; ( C–E ) late phase 2; ( F ) early phase 3; ( G , H ) gastrulation; ( I ) early germband extension . Anterior left; ( A , B , C , F , I ) lateral views; ( D ) dorsal view; ( E ) ventral view; ( G ) ventrolateral view; ( H ) dorsolateral view . Single-channel images are shown in greyscale below each double-channel image ( opa on the left , odd on the right ) . Arrowheads in ( C–E ) point to the new appearance of odd stripe 7 , which abuts the posterior boundary of the opa domain . Note that odd stripe 7 is incomplete both dorsally ( D ) and ventrally ( E ) . By gastrulation , opa expression has posteriorly expanded to cover odd stripe 7 ( G , H ) . opa expression becomes segmentally modulated during germband extension ( I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 01110 . 7554/eLife . 18215 . 012Figure 7—figure supplement 1 . The cellular localisation of opa transcripts changes over the course of segmentation . Relative expression of opa and ftz is shown in embryos at phase 1 , phase 2 and phase 3 . ( A–C ) Whole embryos , lateral view , anterior left , dorsal top . Single-channel images are shown in greyscale below each double-channel image ( opa on the left , ftz on the right ) . ( D–F ) Blown-up regions from each of the embryos in ( A–C ) . Panels with superscripts show individual channels from the double-channel images in ( D–-F ) . opa transcript is largely nuclear during phase 1 , and largely cytoplasmic during phase 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 012 opa begins to be transcribed throughout the trunk during phase 1 , before regular patterns of pair-rule gene expression emerge ( Figure 7A ) . The sharp posterior border of the opa domain at first lies just anterior to odd stripe 7 ( Figure 7B–E ) , but gradually shifts posteriorly over the course of gastrulation to encompass it ( Figure 7F–H ) . Notably , odd stripe 7 is the last of the primary pair-rule gene stripes to appear , and segmentation of this posterior region of the embryo appears to be significantly delayed relative to the rest of the trunk ( Kuhn et al . , 2000 ) . The timing of opa transcription has been shown to rely on nuclear / cytoplasmic ratio ( Lu et al . , 2009 ) , and begins relatively early during cellularisation . However , it takes a while for the opa expression domain to reach full intensity . Unlike the periodically expressed pair-rule genes , which have compact transcription units ( all <3 . 5 kb , FlyBase ) consistent with rapid protein synthesis , the opa transcription unit is large ( ~17 kb , FlyBase ) , owing mainly to a large intron . Accordingly , during most of cellularisation , we observe a punctate distribution of opa , suggestive of nascent transcripts located within nuclei ( Figure 7—figure supplement 1 ) . Unfortunately , the available polyclonal antibody against Opa ( Benedyk et al . , 1994 ) did not work well in our hands , so we have not been able to determine precisely what time Opa protein first appears in blastoderm nuclei . However , Opa protein levels have been reported to peak at late cellularisation and into gastrulation ( Benedyk et al . , 1994 ) , corresponding to the time at which we observe regulatory changes in the embryo , and consistent with our hypothesised role of Opa as a temporal signal . If our hypothesised role for Opa is correct , patterning of the pair-rule genes should progress normally in opa mutant embryos up until the beginning of phase 3 , but not undergo the dramatic pattern changes observed at this time in wild-type . Instead , we would expect that the double-segmental stripes would persist unaltered , at least while the activators of phase 2 expression remain present . The pair-rule gene expression patterns that have been described previously in opa mutant embryos ( see above ) seem consistent with this prediction; however , we wanted to characterise the opa mutant phenotype in more detail to be sure . Throughout cellularisation , we find that pair-rule gene expression is relatively normal in opa mutant embryos ( Figure 8; Figure 8—figure supplement 1 ) , consistent with our hypothesis that Opa function is largely absent from wild-type embryos during these stages . During late phase 2 , we observe only minor quantitative changes to the pair-rule stripes: the odd primary stripes seem wider than normal , the prd primary stripes seem more intense than normal , and the slp primary stripes – which normally appear at the very end of phase 2 – are weakened and delayed . 10 . 7554/eLife . 18215 . 013Figure 8 . Pair-rule gene expression is perturbed from gastrulation onwards in opa mutant embryos . Pair-rule gene expression in wild-type and opa mutant embryos at late cellularisation , late gastrulation , and early germband extension . During cellularisation , pair-rule gene expression in opa mutant embryos is very similar to wild-type . Expression from gastrulation onwards is severely abnormal; in particular , note that single-segment patterns do not emerge . All panels show a lateral view , anterior left , dorsal top . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 01310 . 7554/eLife . 18215 . 014Figure 8—figure supplement 1 . Pair-rule gene expression in opa mutant embryos at cellularisation . Relative expression patterns of pair-rule genes in wild-type and opa mutant embryos at late cellularisation . All images are double in situs for odd and one other pair-rule gene . Individual channels are shown to the right of each double-channel image ( odd on the left , other pair-rule genes on the right ) . All panels show a lateral view , anterior left , dorsal top . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 01410 . 7554/eLife . 18215 . 015Figure 8—figure supplement 2 . Pair-rule gene expression in opa mutant embryos at gastrulation . Relative expression patterns of pair-rule genes in wild-type and opa mutant embryos at gastrulation . All images are double in situs for odd and one other pair-rule gene . Individual channels are shown to the right of each double-channel image ( odd on the left , other pair-rule genes on the right ) . All panels show a lateral view , anterior left , dorsal top . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 01510 . 7554/eLife . 18215 . 016Figure 8—figure supplement 3 . Pair-rule gene expression in opa mutant embryos at early germband extension . Relative expression patterns of pair-rule genes in wild-type and opa mutant embryos at early germband extension . All images are double in situs for odd and one other pair-rule gene . Individual channels are shown to the right of each double-channel image ( odd on the left , other pair-rule genes on the right ) . All panels show a lateral view , anterior left , dorsal top . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 01610 . 7554/eLife . 18215 . 017Figure 8—figure supplement 4 . The transition to single-segment periodicity does not occur in opa mutant embryos . Comparison of early phase 3 segmentation gene expression in wild-type and opa mutant embryos . The horizontal axis represents an idealised portion of the AP axis ( ~12 nuclei across ) . The grey vertical lines demarcate a double parasegment repeat ( ~8 nuclei across ) , of an odd- followed by an even-numbered parasegment ( see Figure 6 ) . The pattern of protein ( intense colour ) and transcript expression ( paler colour ) of the pair-rule genes , and the segment-polarity genes en and wg , are shown for each genotype . Wild-type patterns are the same as in Figure 6B . Transcript distributions for opa mutant embryos were inferred from our double in situ data , while protein distributions were extrapolated from transcript data . Fading expression of Eve and Runt is represented by lighter sections at the posterior of the stripes . In opa mutant embryos , expression of eve and runt fades prematurely , while the expression of odd , prd and slp remains double segmental . Only the even-numbered stripes of wg emerge , with en expression delayed until mid-germband extension ( Benedyk et al . , 1994; Figure 10 ) . Stronger expression in the posterior of the Eve stripes in opa mutants is inferred from the observation that the eve stripes remain broad at a time when they would have already narrowed in wild-type ( compare panels A and F in Appendix 2—figure 2 , or see Figure 8—figure supplement 5 ) . For simplicity , the low-level or residual expression of eve and runt observed in opa mutant embryos is not included in the schematic . See text for further details . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 01710 . 7554/eLife . 18215 . 018Figure 8—figure supplement 5 . Opa activates the eve “'late' element . eve and odd expression in wild-type and opa mutant embryos at various timepoints spanning mid-phase 2 ( mid-cellularisation , top row ) to late phase 3 ( onset of germband extension , bottom row ) . In opa mutant embryos , eve stripes are initially expressed normally ( row 1 ) , but fail to narrow and refine at the end of cellularisation ( row 3 ) , and largely fade away at gastrulation ( row 4 ) . Residual eve expression persists in some stripes into germband extension ( bottom row ) in opa mutant embryos , particularly in ventral regions . Individual channels are shown to the right of the double-channel images . All panels show a lateral view , anterior left , dorsal top . Embryo morphology and the pattern of odd expression in the head were used for staging . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 01810 . 7554/eLife . 18215 . 019Figure 8—figure supplement 6 . 'Late' eve expression is observed in cells that do not express prd . eve and prd expression in wild-type embryos during phase 3 . During early phase 3 ( left ) , eve is strongly expressed in stripes ~2 cells wide . These stripes only partially overlap with the 'P' stripes of prd expression ( asterisks ) , meaning that the eve 'late' element is active in many cells that have never expressed prd . eve expression is largely lost from non-prd expressing cells by the end of gastrulation ( late phase 3 , right ) , indicating that Prd protein may nevertheless be required for the maintenance of eve late element expression . Individual channels are shown below each double-channel image . All panels show a lateral view , anterior left , dorsal top . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 019 In contrast , pair-rule gene expression becomes dramatically different from wild-type at gastrulation ( Figure 8; Figure 8—figure supplement 2 ) . Most notably , the transition from double-segment to single-segment periodicity is not observed for any pair-rule gene – for example , the secondary stripes of odd and slp do not appear , and the prd stripes do not split . In addition , the primary stripes of ftz and odd remain broad , similar to their expression during phase 2 , rather than narrowing from the posterior as in wild-type . Not all the pair-rule genes remain expressed in pair-rule stripes . Except for stripes 6 and 7 , the runt primary stripes are lost , replaced by fairly ubiquitous weak expression which nevertheless retains a double-segmental modulation . eve expression – which has not to our knowledge been previously characterised in opa mutant embryos – fades from stripes 3–7 , with no sign of the sharpened 'late' expression normally activated in the anteriors of the early stripes ( Figure 8—figure supplement 5 ) . hairy expression fades much as it does in wild-type , except that there is reduced separation between certain pairs of stripes . The expression patterns seen at gastrulation persist largely unaltered into germband extension ( Figure 8; Figure 8—figure supplement 3 ) , with the exception that the slp stripes expand anteriorly , overlapping the domains of odd expression . The persistence of the intense prd stripes ( which overlap those of ftz , odd and slp , and remain strongly expressed throughout germband extension ) is especially notable given that prd expression fades from wild-type embryos soon after gastrulation . In summary , in opa mutant embryos odd , prd and slp remain expressed in pair-rule patterns after gastrulation , while expression of eve and runt is largely lost ( schematised in Figure 8—figure supplement 4 ) . The aberrant expression patterns of odd , prd and slp appear to directly reflect an absence of the regulatory changes normally observed in wild-type at phase 3 . For example , the altered prd pattern is consistent with Odd failing to repress prd , indicating that Odd only acts as a repressor of prd in combination with Opa . Similarly , the expression pattern of slp is consistent with continued repression from Runt ( a phase 2 interaction ) and an absence of repression from Ftz ( a phase 3 interaction ) , indicating that Runt only represses slp in the absence of Opa , while the opposite is true for Ftz . In Appendix 2 , we demonstrate how an Opa-dependent switch from repression of odd by Eve ( phase 2 ) to repression of odd by Runt ( phase 3 ) is important for the precise positioning of the anterior borders of the odd primary stripes , in addition to being necessary for the emergence of the odd secondary stripes . The loss of eve and runt expression in opa mutant embryos indicates first that the activators that drive expression of eve and runt during phase 2 do not persist in the embryo after the end of cellularisation , and second that the expression of these genes during phase 3 is activated by the new appearance of Opa . The inference of different activators at phase 2 and phase 3 is not too surprising for eve , which has phase 2 expression driven by stripe-specific elements and phase 3 expression driven by a separate 'late' element ( see below ) . Indeed , expression of stripe-specific elements is known to fade away at gastrulation , as seen for endogenous expression of hairy ( Ingham et al . , 1985; Figure 8 ) , for stripe-specific reporter elements of eve ( Bothma et al . , 2014 ) , or for transgenic embryos lacking eve late element expression ( Fujioka et al . , 1995 ) . However , a single stretch of DNA drives runt primary stripe expression at both phase 2 and phase 3 ( Klingler et al . , 1996 ) . This suggests that the organisation and regulatory logic of this element may be complex , as it is evidently activated by different factors at different times . Opa is also likely to contribute to the activation of the slp primary stripes , explaining why they are initially weaker than normal in opa mutant embryos . ( However , in this case Opa must act semi-redundantly with other activators , in contrast to its effects on eve and runt . ) A resulting delay in the appearance of Slp protein in opa mutant embryos could account for the broadened stripes of ftz and odd , which normally narrow during phase 3 in response to repression from Slp at the posterior . Alternatively , these regulatory functions of Slp could themselves be directly Opa-dependent . Our discovery that Opa was required for late eve expression ( Figure 8—figure supplement 5 ) was surprising , because the enhancer element responsible for this expression has been studied in detail ( Goto et al . , 1989; Harding et al . , 1989; Jiang et al . , 1991; Fujioka et al . , 1996; Sackerson et al . , 1999 ) , and Opa has not previously been implicated in its regulation . The eve 'late' element is sometimes referred to as the eve 'autoregulatory' element , because expression from it is lost in eve mutant embryos ( Harding et al . , 1989; Jiang et al . , 1991 ) . However , the observed 'autoregulation' appears to be indirect ( Goto et al . , 1989; Manoukian and Krause , 1992; Fujioka et al . , 1995; Sackerson et al . , 1999 ) . Instead of being directly activated by Eve , the element mediates regulatory inputs from repressors such as Runt and Slp , which are ectopically expressed in eve mutant embryos ( Vavra and Carroll , 1989; Klingler and Gergen , 1993; Riechmann et al . , 1997; Jaynes and Fujioka , 2004 ) . The element is thought to be directly activated by Prd , and functional Prd-binding sites have been demonstrated within it ( Fujioka et al . , 1996 ) . However , while Prd protein appears at roughly the right time to activate the eve late element ( Pisarev et al . , 2009 ) , activation by Prd cannot explain all the expression generated from this element , because during early phase 3 it drives expression in many cells that do not express prd ( Figure 8—figure supplement 6 ) . Instead , it seems that the eve late element is directly activated by Opa . The lack of late eve expression in opa mutant embryos cannot be explained by the ectopic expression of repressive inputs , since none of runt , odd or slp are ectopically expressed in the domains where eve late element expression would normally be seen ( Figure 8; Figure 8—figure supplement 4 ) . Furthermore , the total loss of eve expression in certain stripes despite the presence of appropriately positioned prd expression indicates that Prd alone is not sufficient to drive strong eve expression . Activation of en expression by Prd also requires the presence of Opa ( Benedyk et al . , 1994 ) , suggesting that cooperative interactions between Prd and Opa might be common . Not all the Opa-dependent expression pattern changes we identified through our analysis of opa mutant embryos happen at exactly the same time in wild-type embryos . Specifically , the splitting of the prd stripes and the appearance of the slp primary stripes occur a few minutes earlier than the other changes , such as the appearance of the secondary stripes of odd and slp , and the late expression of eve . If we assume that Opa concentration increases in the embryo over time as more protein is synthesised , these timing discrepancies could be explained by the former events being driven a lower level of Opa activity than required for the latter events . In order to investigate this hypothesis , we examined pair-rule gene expression in mutants for a 'weak' allele of opa ( opa5 , also known as opa13D92 ) which we presume to represent an opa hypomorph . Whereas mutants for the null allele we investigated ( opa8 , also known as opa11P32 ) develop cuticles with complete pairwise fusion of adjacent denticle belts , mutants for opa5 develop less severe patterning defects where denticle belts remain separate or only partially fuse ( Baumgartner et al . , 1994 ) . Figure 9 compares expression patterns in opa hypomorphic embryos to both the wild-type and null situations . At cellularisation , expression patterns are similar for all three genotypes ( data not shown ) . At gastrulation , expression patterns in the hypomorphic embryos tend to resemble those in the null embryos . However , there are two significant differences , corresponding to the two Opa-dependent patterning events that occur first in wild-type embryos . First , the slp primary stripes are expressed more strongly in the hypomorphic embryos than in the null embryos ( although their appearance is still slightly delayed ) , and second , the prd stripes in the hypomorphic embryos show weak expression in the centre of the stripes ( arrowheads in Figure 9 ) , a situation intermediate between the wild-type situation of full splitting , and the null situation of completely uniform stripes . Later , during germband extension , expression patterns in the hypomorphic embryos diverge further from the null situation , with multiple genes exhibiting evidence of Opa-dependent regulation ( arrowheads in Figure 9 ) . For example , the prd stripes fully split , some evidence of odd and slp secondary stripes can be seen , and strong runt expression is reinitiated . 10 . 7554/eLife . 18215 . 020Figure 9 . Opa-dependent expression pattern changes are delayed in opa hypomorphic embryos . Expression of selected pair-rule genes compared between embryos wild-type , hypomorphic ( opa5 ) , or null mutant ( opa8 ) for opa . Arrowheads mark evidence of Opa-dependent regulatory interactions in opa5 embryos ( see text for details ) . All panels show a lateral view , anterior left , dorsal top . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 020 Together , this evidence suggests that different regulatory targets of Opa respond with differential sensitivity . As the level of Opa increases over time , 'sensitive' targets would show expression changes as soon as a low threshold of Opa activity was reached , whereas other targets would respond later , when a higher threshold was reached . In wild-type embryos , the low threshold events occur slightly earlier than the high threshold events , at late phase 2 rather than early phase 3 . In opa hypomorphic embryos , in which the rate of increase in Opa activity would be slower , these events happen later but still in the same temporal sequence , with the low threshold events occurring at gastrulation , and the high threshold events not detected until germband extension . In opa null embryos , both classes of events of course do not happen at all . Explaining the aetiology of the opa pair-rule phenotype requires understanding why the loss of Opa activity results in the mispatterning of parasegment boundaries by segment-polarity genes . In wild-type embryos , en and wg are initially regulated cell-autonomously by pair-rule proteins ( for example , see Ingham and Baker , 1988; Weir et al . , 1988; Manoukian and Krause , 1993; Mullen and DiNardo , 1995 ) . During germband extension , they become dependent on intercellular signalling for their continued expression , with the Wingless and Hedgehog signalling pathways forming a positive feedback loop that maintains each parasegment boundary ( DiNardo et al . , 1988 , 1994; Perrimon , 1994; von Dassow et al . , 2000 ) . Expression of en and wg has previously been characterised in opa mutant embryos , demonstrating that the even-numbered parasegment boundaries fail to establish properly ( Benedyk et al . , 1994; Ingham , 1986; DiNardo and O'Farrell , 1987; see also Figure 10—figure supplement 1 ) . To summarise , although their initial appearance is somewhat delayed , the even-numbered wg stripes ( which normally contribute to the odd-numbered parasegment boundaries ) and some of the even-numbered en stripes ( which normally contribute to the even-numbered parasegment boundaries ) become established in their normal locations by the beginning of the germband extension . Later on in germband extension , odd-numbered en stripes become established adjacent to the even-numbered wg stripes , leading to the formation of the odd-numbered parasegment boundaries . In contrast , the odd-numbered wg stripes never appear , the even-numbered en stripes eventually fade away , and the even-numbered parasegment boundaries are not established . Our characterisation of pair-rule gene expression in opa mutant embryos enables us to make sense of these patterns . First , Opa appears to regulate slp and wg in a very similar way ( Figure 10—figure supplement 1 ) . The even-numbered wg stripes overlap with the primary stripes of slp and show the same expression delays in opa mutant embryos , while the odd-numbered wg stripes and the slp secondary stripes , which would normally be activated at the same time and in the same places , both fail to appear . Second , the activation of en by Prd seems to strictly require Opa activity , whereas the activation of en by Ftz does not . Therefore , while the odd-numbered ( Prd-activated ) en stripes are initially absent in opa mutant embryos , some of the even-numbered ( Ftz-activated ) stripes do appear , although this is compromised by ectopic expression of Odd ( Benedyk et al . , 1994 , Appendix 2 ) . Third , the capacity for a partially specified parasegment boundary to later recover depends on the presence of an appropriate segmental pattern of pair-rule gene expression , despite these patterns arising too late to regulate the initial expression of segment-polarity genes at gastrulation . For example , the Slp stripes play an important segment-polarity role during germband extension , defining the posterior half of each parasegment . They repress en expression and are also necessary for the maintenance of wg expression ( Cadigan et al . , 1994b ) . In the case of the odd-numbered parasegment boundaries , slp and wg are properly patterned in opa mutant embryos , but the en stripes are absent ( Figure 10—figure supplement 1 ) . However , repressors of en such as Odd and Slp are not ectopically expressed in their place . Therefore , the odd-numbered en stripes are able to be later induced in their normal positions , presumably in response to Wg signalling coming from the Slp primary stripes , and therefore properly patterned boundaries eventually emerge . However , in the case of the even-numbered parasegment boundaries , while the en stripes are usually present in opa mutant embryos , both the slp stripes and the wg stripes are not ( Figure 10—figure supplement 1 ) . The absence of the Slp secondary stripes means that the cells anterior to the even-numbered En stripes are not competent to express wg . Hedgehog signalling from the even-numbered En stripes is therefore unable to induce the odd-numbered wg stripes , and consequently the boundaries do not recover . Based on regulatory interactions analysed in Appendix 1 and Appendix 2 , we present an updated model for how the even-numbered parasegment boundaries are specified in wild-type embryos ( Figure 10 ) . We propose that the spatial information directly responsible for patterning these boundaries derives from overlapping domains of Runt and Ftz activity ( Appendix 1—figure 3G , H ) . Ftz and Runt combinatorially specify distinct expression domains of slp , en and odd , by way of late acting , Opa-dependent regulatory interactions . As described above , the loss of these interactions in opa mutant embryos results in mispatterning of slp and odd ( Figure 8—figure supplement 4 ) , which later has significant repercussions for segment-polarity gene expression . 10 . 7554/eLife . 18215 . 021Figure 10 . Model for the Opa-dependent patterning of the even-numbered parasegment boundaries . ( A ) Schematic showing the phasing of odd , slp and en relative to Runt and Ftz protein at phase 3 . The horizontal axis represents part of a typical double-segment pattern repeat along the AP axis of the embryo ( ~4 nuclei across , centred on an even-numbered parasegment boundary ) . ( B ) Inferred regulatory interactions governing the expression of odd , slp and en at phase 3 . Regular arrows represent activatory interactions; hammerhead arrows represent repressive interactions . Solid arrows represent interactions that are currently in operation; pale dashed arrows represent those that are not . Red arrows represent interactions that depend on the presence of Opa protein . Overlapping domains of Runt and Ftz expression ( A ) subdivide this region of the AP axis into three sections ( black dashed lines ) . Opa-dependent repression restricts odd expression to the posterior section , resulting in offset anterior boundaries of Ftz and Odd activity ( Appendix 2—figure 1; Appendix 2—figure 1—figure supplement 2 ) . slp expression is restricted to the anterior section by the combination of Opa-dependent repression from Ftz and Opa-dependent de-repression from Runt ( Appendix 1—figure 2—figure supplement 1 ) . en is restricted to the central section by the combination of activation from Ftz ( likely partially dependent on Opa ) , and repression by Odd . Later , mutual repression between odd , slp and en will maintain these distinct cell states . The even-numbered parasegment boundaries will form between the en and slp domains . Note that , in this model , Eve has no direct role in patterning these boundaries . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 02110 . 7554/eLife . 18215 . 022Figure 10—figure supplement 1 . Segment-polarity gene expression in opa mutant embryos . Development of segment-polarity gene expression in wild-type and opa mutant embryos . Arrowheads marks segment-polarity stripes that normally contribute to odd-numbered parasegment boundaries ( even-numbered wg stripes , and odd-numbered en stripes , respectively ) . Asterisks mark segment-polarity stripes that normally contribute to even-numbered parasegment boundaries ( odd-numbered wg stripes and even-numbered en stripes , respectively ) . ( Note that wg stripes are traditionally numbered from 0 ) . In opa mutant embryos , odd-numbered wg stripes never emerge , while even-numbered en stripes do emerge , but are not maintained . In contrast , even-numbered wg stripes emerge fairly normally , while odd-numbered en stripes are delayed initially , but later recover . All panels show a lateral view , anterior left , dorsal top . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 18215 . 022 We have found that many regulatory interactions between the pair-rule genes are not constant over the course of Drosophila segmentation , but instead undergo coordinated changes at the end of cellularisation . We are not the first to notice that certain regulatory interactions do not apply to all stages of pair-rule gene expression ( Baumgartner and Noll , 1990; Manoukian and Krause , 1992; Manoukian and Krause , 1993; Fujioka et al . , 1995; Saulier-Le Dréan et al . , 1998 ) . However , cataloguing and analysing these changes for the whole pair-rule system led us to the realisation that they are almost simultaneous and mediate the transition from double-segment to single-segment periodicity . We propose that the product of the non-canonical pair-rule gene opa acts as a temporal signal that mediates these changes , and simultaneously activates the expression of segment-polarity genes . Analysis of pair-rule gene expression patterns in opa mutant embryos indicates that the phase-specific regulatory interactions we inferred from wild-type embryos appear to be modulated by Opa , and thus explained by the onset of Opa regulatory activity at gastrulation . We argue that the pair-rule system should not be thought of as a static gene regulatory network , but rather two temporally and topologically distinct networks , each with its own dynamical behaviour and consequent developmental patterning role . Pair-rule patterning can therefore be thought of as a two-stage process . In the absence of Opa , the early network patterns the template for the odd-numbered parasegment boundaries . Then , when Opa turns on , Opa-dependent regulatory interactions lead to the patterning of the even-numbered parasegment boundaries . Each stage of patterning uses the same source of positional information ( the primary stripes of the pair-rule genes ) , but uses different sets of regulatory logic to exploit this information in different ways . Opa thus plays a crucial timing role in segmentation , orchestrating the transition from pair-rule to segmental patterning . Notably , the role of Opa in activating the initial stages of segment-polarity gene expression demonstrates that segment-polarity gene expression is not simply induced by the emergence of an appropriate pattern of pair-rule proteins , as in textbook models of hierarchical gene regulation . The necessity for an additional signal had been surmised previously , based on the delayed appearance of odd-numbered en stripes in cells already expressing Eve and Prd ( Manoukian and Krause , 1993 ) . Because correct segmentation depends upon the initial expression of segment-polarity genes being precisely positioned , it is imperative that a regular pair-rule pattern is present before the segment-polarity genes first turn on . Therefore , explicit temporal control of segment-polarity gene activation by Opa makes good sense from a patterning perspective . There are likely to be a number of analogous regulatory signals that provide extrinsic temporal information to the Drosophila segmentation cascade . For example , a ubiquitously expressed maternal protein , Tramtrack , represses pair-rule gene expression during early embryogenesis , effectively delaying pair-rule gene expression until appropriate patterns of gap gene expression have been established ( Harrison and Travers , 1990; Read et al . , 1992; Brown and Wu , 1993 ) . opa is the Drosophila ortholog of zinc finger of the cerebellum ( zic ) ( Aruga et al . , 1994 ) . zic genes encode zinc finger transcription factors closely related to Gli proteins that have many important developmental roles . In the Drosophila embryo , in addition to its role in segmentation , Opa is also involved in the formation of visceral mesoderm ( Cimbora and Sakonju , 1995; Schaub and Frasch , 2013 ) . Opa is later highly expressed in the larval and adult brain ( FlyAtlas – Chintapalli et al . , 2007 ) and is likely to be involved in neuronal differentiation ( Eroglu et al . , 2014 ) . It is also involved in the regulation of adult head development ( Lee et al . , 2007 ) . The neuronal function is likely to reflect an ancestral role of Zic , as involvement of Zic genes in nervous system development and neuronal differentiation is pervasive throughout metazoans ( Layden et al . , 2010 ) . Lineage-specific duplications have resulted in five zic genes in most vertebrate taxa , and seven in teleosts ( Aruga et al . , 2006; Merzdorf , 2007 ) . While partial redundancy between these paralogs complicates the interpretation of mutant phenotypes , it is clear that in vertebrates Zic proteins play crucial roles in early embryonic patterning , neurogenesis , left-right asymmetry , neural crest formation , somite development , and cell proliferation ( reviewed in Merzdorf , 2007; Houtmeyers et al . , 2013 ) . Zic proteins have been shown to act both as classical DNA-binding transcription factors , and as cofactors that modulate the regulatory activity of other transcription factors via protein-protein interactions ( reviewed in Ali et al . , 2012; Winata et al . , 2015 ) . They show context-dependent activity and can both activate and repress transcription ( Yang et al . , 2000; Salero et al . , 2001 ) . In particular , they appear to be directly involved in the modulation and interpretation of Wnt and Hedgehog signalling ( Murgan et al . , 2015; Pourebrahim et al . , 2011; Fujimi et al . , 2012; Koyabu et al . , 2001; Chan et al . , 2011; Quinn et al . , 2012 ) . Finally , they may play a direct role in chromatin regulation ( Luo et al . , 2015 ) . The roles that Opa plays in the Drosophila segmentation network appear to be consistent with the mechanisms of Zic regulatory activity that have been characterised in vertebrates . Opa appears to transcriptionally activate a number of enhancers , including those driving late expression of eve , runt and slp . In the case of the slp enhancer , this has been verified experimentally ( Sen et al . , 2010 ) . In other cases , the role of Opa is likely to be restricted to modulating the effect of other regulatory inputs , such as mediating the repressive effect of Odd on prd expression , or the activatory effect of Prd on en expression . It will be interesting to investigate the enhancers mediating late pair-rule gene expression and early segment polarity gene expression , and to determine how Opa interacts with them to bring about these varied effects . Our data seem consistent with Opa being 'the' temporal signal that precipitates the 7 stripe to 14 stripe transition . However , it remains possible that Opa acts in conjunction with some other , as yet unidentified , temporally patterned factor , or has activity that is overridden during cellularisation by some maternal or zygotic factor that disappears at gastrulation . Indeed , combinatorial interactions with DV factors do seem likely to be playing a role in restricting the effects of Opa: despite the opa expression domain encircling the embryo , many Opa-dependent patterning events do not extend into the mesoderm or across the dorsal midline . Identification of these factors should yield insights into cross-talk between the AP and DV patterning systems of the Drosophila blastoderm . The activity of Opa has previously been suggested to be concentration-dependent ( Swantek and Gergen , 2004 ) . By comparing pair-rule gene expression in embryos with varying levels of Opa activity , we found evidence that different enhancers show different sensitivity to the concentration of Opa in a nucleus , explaining why different Opa-dependent regulatory events happen at slightly different times in wild-type embryos . One of the earliest responses to Opa regulatory activity is the appearance of the slp primary stripes . However , we note that while Opa may contribute to their timely activation , these stripes still emerge in opa null mutant embryos . This is not surprising , as the slp locus has been shown to possess multiple partially redundant regulatory elements driving spatially and temporally overlapping expression patterns ( Fujioka and Jaynes , 2012 ) . From our own observations , we have found multiple cases where mutation of a particular gene causes the slp primary stripes to be reduced in intensity , but not abolished ( data not shown ) , suggesting that regulatory control of these expression domains is redundant at the trans level as well as at the cis level . Partially redundant enhancers that drive similar patterns , but are not necessarily subject to the same regulatory logic , appear to be very common for developmental transcription factors ( Cannavò et al . , 2016; Hong et al . , 2008; Perry et al . , 2011; Staller et al . , 2015; Wunderlich et al . , 2015 ) . By carefully analysing pair-rule gene expression patterns in the light of the experimental literature ( Appendix 1 ) , we have clarified our understanding of the regulatory logic responsible for generating them . In particular , we propose significantly revised models for the patterning of odd , slp and runt . Because the structure of a regulatory network determines its dynamics , and its structure is determined by the control logic of its individual components , these subtleties are not merely developmental genetic stamp-collecting . Our reappraisal of the pair-rule gene network allows us to re-evaluate some long-held views about Drosphila blastoderm patterning . First , pair-rule gene interactions are combinatorially regulated by an extrinsic source of temporal information , something not allowed for by textbook models of the Drosophila segmentation cascade . We have characterised the role of Opa during the 7 stripe to 14 stripe transition , but there may well be other such signals acting earlier or later . Indeed , context-dependent transcription factor activity appears to be very common ( Stampfel et al . , 2015 ) . Second , our updated model of the pair-rule network is in many ways simpler than previously thought . While we do introduce the complication of an Opa-dependent network topology , this effectively streamlines the sub-networks that operate early ( phase 2 ) and late ( phase 3 ) . At any one time , each pair-rule gene is only regulated by two or three other pair-rule genes . We do not see strong evidence for combinatorial interactions between these inputs ( cf . DiNardo and O'Farrell , 1987; Baumgartner and Noll , 1990; Swantek and Gergen , 2004 ) . Instead , pair-rule gene regulatory logic seems invariably to consist of permissive activation by a broadly expressed factor ( or factors ) that is overridden by precisely positioned repressors ( Edgar et al . , 1986; Weir et al . , 1988 ) . This kind of regulation appears to typify other complex patterning systems , such as the vertebrate neural tube ( Briscoe and Small , 2015 ) . Finally , pair-rule gene cross-regulation has traditionally been thought of as a mechanism to stabilise and refine stripe boundaries ( e . g . Edgar et al . , 1989; Schroeder et al . , 2011 ) . Consistent with this function , as well as with the observed digitisation of gene expression observed at gastrulation ( Baumgartner and Noll , 1990; Pisarev et al . , 2009 ) , we find that the late network contains a number of mutually repressive interactions ( Eve/Runt , Eve/Slp , Ftz/Slp , Odd/Runt , Odd/Slp , and perhaps Odd/Prd ) . However , the early network does not appear to utilise these switch-like interactions , but is instead characterised by unidirectional repression ( e . g . of ftz and odd by Eve and Hairy , or of runt by Odd ) . Interestingly , pair-rule gene expression during cellularisation has been observed to be unexpectedly dynamic ( Keränen et al . , 2006; Surkova et al . , 2008 ) , something that is notable given the oscillatory expression of pair-rule gene orthologs in short germ arthropods ( Sarrazin et al . , 2012; El-Sherif et al . , 2012; Brena and Akam , 2013 ) . We have shown that for the pair-rule genes , the transition to single-segment periodicity is mediated by substantial re-wiring of regulatory interactions . In addition , we have shown that this re-wiring is controlled by the same signal , Opa , that activates segment-polarity gene expression . We propose that Opa's effective role is to usher in a 'segment-polarity phase' of expression , in which both canonical segment-polarity factors , and erstwhile pair-rule factors , work together to define cell states . This hypothesis is consistent with the spatial patterns and regulatory logic of late pair-rule gene expression: most pair-rule genes become expressed in narrow segmental stripes , and partake in switch-like regulatory interactions consistent with segment-polarity roles . Furthermore , regulatory feedback from segment-polarity genes suggests the pair-rule genes become integrated into the segment-polarity network: for example , En protein is involved in patterning the late expression of eve , odd , runt and slp ( Harding et al . , 1986; Mullen and DiNardo , 1995; Klingler and Gergen , 1993; Fujioka et al . , 2012 ) . However , the hypothesis that pair-rule factors perform segment-polarity roles is at odds with that fact that their mutants generally do not exhibit segment-polarity defects . We argue that this discrepancy can be resolved by accounting for partial redundancy with paralogous factors . For example , slp has a closely linked paralog , slp2 , expressed almost identically , ( Grossniklaus et al . , 1992 ) , and simultaneous disruption of both genes is required in order to reveal that the Slp stripes are a critical component of the segment-polarity network ( Cadigan et al . , 1994a; 1994b ) . prd and odd also have paralogs , expressed in persistent segmental stripes coincident with their respective phase 3 expression patterns ( Baumgartner et al . , 1987; Hart et al . , 1996 ) . The prd paralog , gsb , gives a segment-polarity phenotype if mutated , but Prd and Gsb are able to substitute for each other if expressed under the control of the other gene's regulatory region ( Li and Noll , 1993 , 1994; Xue and Noll , 1996 ) , indicating that the same protein can fulfil both pair-rule and segment-polarity functions . Moreover , we have found that a deficiency removing both odd and its closely linked paralogs , sob and drm , gives a cuticle phenotype that shows segment-polarity defects corresponding to the locations of odd secondary stripes , in addition to the pair-rule defects characteristic of odd mutants ( data not shown ) . We envisage that ancestrally , the orthologs of prd/gsb and odd/sob/drm would have sequentially fulfilled both pair-rule and segment-polarity functions , employing different regulatory logic in each case . Later , these roles would have been divided between different paralogs , leaving the transient segmental patterns of prd and odd as evolutionary relics . Consistent with this hypothesis , the roles of prd and gsb seem to be fulfilled by a single co-ortholog , pairberry1 , in grasshoppers , with a second gene , pairberry2 , expressed redundantly ( Davis et al . , 2001 ) . Therefore , of the four pair-rule factors expressed in segmental patterns after gastrulation ( Runt , Odd , Prd , Slp ) , at least three appear to have segment-polarity functions , although they may perform these roles only transiently before handing over the job to their paralogs . ( No function has as yet been assigned to late Runt expression . ) Because Hairy expression fades away after phase 2 , that leaves only the functions of the late , double-segmental expression patterns of Eve and Ftz to be accounted for . Both these factors partake in the segment-polarity network by repressing slp and wg ( Fujioka et al . , 2002; Swantek and Gergen , 2004; Copeland et al . , 1996 ) . However , unlike canonical segment-polarity factors , their expression fades during germband extension . Functional equivalence with each other explains why , from a patterning perspective , they need not be expressed in every segment . Functional redundancy with En ( Fujioka et al . , 2012 ) explains why they need not be persistently expressed ( indeed , En is the factor responsible for switching off late eve expression [Harding et al . , 1986] ) . Given that eve shows a phase of single-segment periodicity in many pair-rule insects ( Patel et al . , 1994; Binner and Sander , 1997; Rosenberg et al . , 2014; Mito et al . , 2007 ) , ( although not in Bombyx mori [Nakao , 2010] ) , it will be interesting to investigate whether a loss of regular segmental eve expression in the lineage leading to Drosophila is associated with changes to the roles of Ftz ( and/or its cofactor , Ftz-F1 ) in segment patterning ( Heffer et al . , 2013 , 2011 ) . In light of our data , it will be interesting to characterise the role of Opa in other arthropod model organisms . The best studied short germ insect is the beetle Tribolium castaneum , which also exhibits pair-rule patterning . An RNAi screen of pair-rule gene orthologs reported no segmentation phenotype for opa knock-down , and concluded that opa does not function as a pair-rule gene in Tribolium ( Choe et al . , 2006 ) . However , the authors also state that opa knock-down caused high levels of lethality and most embryos did not complete development , indicating that this conclusion may be premature . In contrast to this study , iBeetle-Base ( Dönitz et al . , 2015 ) reports a segmentation phenotype for opa knock-down ( TC number: TC010234; iBeetle number: iB_04791 ) . The affected cuticles show a reduced number of segments including the loss of the mesothorax ( T2 ) . This could indicate a pair-rule phenotype in which the even-numbered parasegment boundaries are lost , similar to the situation in Drosophila . If true , this suggests that at least some aspects of the role of Opa are conserved between long germ and short germ segmentation . Wild-type embryos were Oregon-R . The pair-rule gene mutations used were opa5 ( Bloomington stock no . 5334 ) , opa8 ( Bloomington stock no . 5335 ) and ftz11 ( gift of Bénédicte Sanson ) . These mutations were balanced over TM6C Sb Tb twi::lacZ ( Bloomington stock no . 7251 ) to allow homozygous mutant embryos to be easily distinguished . Two to four hours old embryos were collected on apple juice agar plates at 25°C , fixed in 4% paraformaldehyde ( PFA ) for 20 min according to standard procedures , and stored at -20°C in methanol until required . Digoxigenin- ( DIG ) and fluorescein ( FITC ) -labelled riboprobes were generated using full-length pair-rule gene cDNAs from the Drosophila gene collection ( Stapleton et al . , 2002 ) and either DIG or fluorescein RNA labelling mix ( Roche , Basel , Switzerland ) . The clones used were RE40955 ( hairy ) ; MIP30861 ( eve ) ; GH02614 ( runt ) ; IP01266 ( ftz ) ; GH22686 ( prd ) ; GH04704 ( slp ) ; LD30441 ( opa ) ; LD16125 ( en ) ; FI07617 ( gsb ) ; RE02607 ( wg ) . Embryos were post-fixed in 4% PFA then washed in PBT ( PBS with 0 . 1% Tween-20 ) prior to hybridisation . Hybridisation was performed at 56°C overnight in hybridisation buffer ( 50% formamide , 5x SSC , 5x Denhardt’s solution , 100 μ g/ml yeast tRNA , 2 . 5% w/v dextran sulfate , 0 . 1% Tween-20 ) , with at least 1 hr of prehybridisation before introducing the probes . Embryos were simultaneously hybridised with one DIG probe and one FITC probe to different segmentation genes . Embryos from mutant crosses were additionally hybridised with a DIG probe to lacZ . Post-hybridisation washes were carried out as in Lauter et al . , 2011 . Embryos were then incubated in peroxidase-conjugated anti-FITC and alkaline phosphatase ( AP ) -conjugated anti-DIG antibodies ( Roche , Basel , Switzerland ) diluted 1:4000 . Tyramide biotin amplification ( TSA biotin kit , Perkin Elmer , Waltham , MA ) followed by incubation in streptavidin Alexa Fluor 488 conjugate ( ThermoFisher Scientific , Waltham , MA ) was used to visualise the peroxidase signal . A Fast Red reaction ( Fast Red tablets , Kem-En-Tec Diagnostics , Taastrup , Denmark ) was subsequently used to visualise the AP signal . Embryos were mounted in ProLong Diamond Antifade Mountant ( ThermoFisher Scientific ) before imaging . Embryos were imaged on a Leica SP5 Upright confocal microscope , using a 20x objective . For each pairwise combination of probes , a slide of ~100 embryos was visually examined , and around 20 images taken for further analysis . Occasional embryos with severe patterning abnormalities were discounted from analysis . Minor brightness and contrast adjustments were carried out using Fiji ( Schindelin et al . , 2012 , 2012 ) . Thresholded images were produced using the 'Make Binary' option in Fiji . Our full wild-type dataset of over 600 double channel confocal images is available from the Dryad Digital Repository ( Clark and Akam , 2016 ) .	The basic body plan of an animal is set up in the early embryo , where key developmental genes are expressed in specific patterns across the organism . These patterns emerge from the way in which the proteins encoded by these genes act to regulate each other’s expression . The fruit fly Drosophila is often used as a simple model for studying how regulatory interactions between genes lead to the formation of complex developmental patterns . One example is segmentation , the process by which the trunk ( main body ) region of the Drosophila embryo is subdivided into 14 segments . A group of transcription factor proteins that are encoded by the so-called “pair-rule” genes play a crucial role in producing the final pattern . Early in development , the pair-rule genes are expressed in patterns of seven stripes , dividing the embryo into double segment units . As the embryo develops , these patterns change to form more precise patterns of 14 stripes , corresponding to single segments . Although many of the regulatory interactions between the pair-rule genes were known , how the system works as a whole to produce this change in expression patterns was not well understood . Clark and Akam have now examined how the patterns of pair-rule gene expression change over time inside fly embryos . This revealed that a “rewiring” of the network of pair-rule genes occurs at the end of the seven stripe stage to produce fourteen stripes . Further investigation revealed that a transcription factor encoded by the gene odd-paired causes this rewiring , and that the timing of the expression of the Odd-paired protein determines when the rewiring happens . Further studies could now investigate whether Odd-paired’s role as a timer extends to species where segments emerge sequentially , instead of the simultaneous formation of segments seen in Drosophila . Future challenges will be to find out how Odd-paired interacts with other pair-rule transcription factors , and whether there are other timing factors that help to coordinate embryonic patterning .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "developmental", "biology", "computational", "and", "systems", "biology" ]	2016	Odd-paired controls frequency doubling in Drosophila segmentation by altering the pair-rule gene regulatory network
Our experiments address two long-standing models for the function of the Drosophila brain circadian network: a dual oscillator model , which emphasizes the primacy of PDF-containing neurons , and a cell-autonomous model for circadian phase adjustment . We identify five different circadian ( E ) neurons that are a major source of rhythmicity and locomotor activity . Brief firing of PDF cells at different times of day generates a phase response curve ( PRC ) , which mimics a light-mediated PRC and requires PDF receptor expression in the five E neurons . Firing also resembles light by causing TIM degradation in downstream neurons . Unlike light however , firing-mediated phase-shifting is CRY-independent and exploits the E3 ligase component CUL-3 in the early night to degrade TIM . Our results suggest that PDF neurons integrate light information and then modulate the phase of E cell oscillations and behavioral rhythms . The results also explain how fly brain rhythms persist in constant darkness and without CRY . Animals use endogenous circadian pacemakers to control their physiology and behavior with roughly 24-hr periodicity ( Bass and Takahashi , 2010; Thut et al . , 2012 ) . Intracellular timekeeping mechanisms include transcriptional feedback loops , which involve many key genes in Drosophila . They include period ( per ) , timeless ( tim ) , clock ( clk ) , cycle ( cyc ) , and doubletime ( dbt ) . Coordination of their encoded protein activities ( period , PER; timeless , TIM; clock , CLK; cycle , CYC: and doubletime , DBT ) contributes to the intracellular cycling of the molecular pacemaker , which is similar between flies and mammals ( Dubruille and Emery , 2008; Menet and Rosbash , 2011 ) . PER and TIM concentrations increase during the day , and they eventually negatively regulate their own transcription . Biochemical oscillations in the head occur in part through a direct interaction of PER and TIM with the positive transcription factor CLK: CYC ( Dubruille and Emery , 2008; Menet and Rosbash , 2011 ) . They also require the photoreceptor cryptochrome ( CRY ) as well as a cycling light:dark ( LD ) environment , that is , RNA and protein oscillations damp rapidly in constant darkness ( Stanewsky et al . , 1998 ) . The central brain is probably different as its molecular and behavioral rhythms persist in constant darkness and is CRY-independent ( Stanewsky et al . , 1998 ) . Nonetheless , CRY is expressed within many of the 75 pairs of central brain circadian neurons and is necessary for an important biological feature of circadian rhythms , namely , light-mediated phase adjustment or phase-shifting ( Emery et al . , 1998 ) . There is good evidence in favor of a cell-autonomous view of Drosophila phase-shifting: light penetrates the thin insect cuticle ( Fogle et al . , 2011 ) and causes a CRY conformational change within circadian neurons ( Ozturk et al . , 2011 ) . The altered CRY molecules recruit the E3 ligase JETLAG ( JET ) to TIM ( Koh et al . , 2006; Peschel et al . , 2009 ) . Premature TIM degradation then causes phase advances in the late night , whereas inappropriate TIM degradation during the TIM accumulating phase in the early night causes phase delays . Two groups of central brain circadian neurons appear particularly important for behavioral rhythms . The 4 PDF-expressing small ventrolateral neurons ( sLNvs ) dictate morning activity as well as the rhythmicity in constant darkness ( Renn et al . , 1999; Blanchardon et al . , 2001; Stoleru et al . , 2004 ) . This latter feature , free running locomotor activity rhythms , has caused the s-LNvs to be considered the major fly pacemaker neurons . A less well-defined set of cells directs evening activity . These neurons ( E cells ) also dictate circadian behavior in constant light and probably include the 6 LNds and the PDF-negative 5th s-LNv ( Picot et al . , 2007 ) . We identify here five E neurons as a major source of circadian and behavioral rhythmicity as well as locomotor activity , of which the 2 CRY+ E neurons are most important . The phase of these cells is shifted by the M cell firing and requires PDF as well as the PDF receptor within these five E cells . Moreover , brief firing of M cells at different times of day generates a phase response curve ( PRC ) , which resembles a proper light-mediated PRC . Brief M cell firing also resembles light by causing rapid TIM degradation within downstream circadian neurons , but firing-mediated phase-shifting is CRY-independent and exploits the E3 ligase components CUL-3 to degrade TIM , at least in the early night-delay zone . Our findings suggest that E cells are very important for timekeeping under light–dark conditions whereas an important function of PDF cells is for light-mediated phase adjustment , that is , to integrate light information and appropriately fire . This degrades TIM within E cells , which modulates the phase of E cell oscillations and behavioral rhythms . To address further the roles of M and E cells , we made use of a recently described DvPdf-GAL4 driver , which expresses strongly and specifically in the M cells and subsets of the E cells . To verify this expression pattern , we crossed DvPdf-GAL4 with UAS-mCD8::GFP . Exactly as reported ( Bahn et al . , 2009 ) , GFP is expressed in the PDF-positive LNv cells and in several E cells in each hemisphere ( Figure 1A ) ; the latter consist of the single PDF-negative 5th-sLNv ( Figure 1C ) and 4 LNds ( subsequently referred to as the 5 Dv-E cells ) . Cell identification of these circadian neurons was confirmed by co-staining with anti-PER antibodies ( Figure 1A ) . 10 . 7554/eLife . 02780 . 003Figure 1 . Characterization of the five E cells labeled by DvPdf-GAL4 , Pdf-GAL80 . ( A ) The brain expression pattern of DvPdf-GAL4; UAS-mCD8::GFP flies . Immunostaining with anti-PER only ( red , left panel ) and with anti-GFP ( green ) as well as anti-PER ( right panel ) . GFP and PER co-localize in the PDF positive cells , LNds and the 5th-sLNv . The scale bar = 50 μm . A maximum intensity projection of confocal image stacks containing the cell bodies regions is shown . ( B ) Rescue of PER expression with DvPdf-GAL4 in a per0 background restores both morning and evening anticipation peaks . per0 flies have no morning and evening anticipation peaks ( left panel ) , whereas per0; DvPdf-GAL4; UAS-per flies show normal morning and evening peaks ( right panel ) . Black and gray arrows point to morning and evening anticipation peaks , respectively . White and black bars indicate activity events/30 min bin during the day and night of the LD cycle . Error bars represent standard error of the mean ( SEM ) . n = 15–16 for each group . ( C ) GFP immunostaining of DvPdf-GAL4 , Pdf-GAL80; UAS-mCD8::GFP flies . GFP ( red ) is expressed in 3–4 LNds and the 5th-sLNv ( middle panel ) , and PDF ( green ) is expressed in the large and small LNvs ( left panel ) . The white arrow shows additional tracts from Dv-E cells that project to the accessory medulla ( aMe ) . The scale bar = 50 μm . ( D ) Averaged group activity profiles from UAS-DBTs expressed with Pdf-GAL4 , Clk4 . 1m-GAL4 and DvPdf-GAL4 , Pdf-GAL80 flies . The black arrow shows the advanced evening anticipation peak . Averaged over 2 days of LD data . White and black bars indicate activity events/30 min bin during the day and night of the LD cycle , respectively . Error bars represent standard error of the mean ( SEM ) . n = 15–16 for each group . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 003 Rescuing PER expression with the DvPdf-GAL4 driver restored both morning and evening anticipation to the arrhythmic per01 strain ( Figure 1B ) , indicating an important role of these cells in circadian behavior . The five E cells labeled with DvPdf-GAL4 , Pdf-gal80 ( henceforth called the Dv-E driver and the Dv-E cells ) project to the dorsal brain and to the accessory medulla ( aMe ) as previously described ( Figure 1C , arrows ) ( Bahn et al . , 2009 ) ; the latter is based on a projection from the E cells , which contacts the M cells and is similar to that recently described with anti-ITP and anti-CRY antibodies ( Yoshii et al . , 2008; Johard et al . , 2009 ) . To compare the contribution of different clock neurons to circadian timing under standard LD conditions , we overexpressed the period-altering kinase DBTS ( Muskus et al . , 2007 ) under the control of three different circadian drivers . Although M cells ( PDF neurons ) are considered master pacemakers , neither expression of DBTS in M cells nor in its direct downstream target neurons-DN1ps ( Zhang et al . , 2010a , 2010b; Seluzicki et al . , 2014 ) changes the locomotor activity pattern or phase ( Figure 1D ) . In contrast , the evening peak phase is dramatic advanced when accelerating the endogenous clock in the 5 Dv-E cells ( Figure 1D ) , indicating that the E cells independently determine the major activity phase in LD ( Stoleru et al . , 2007 ) . Because differential neuron expression profiling suggests that the 5 Dv-E cells use excitatory neurotransmitter machinery to communicate with other neurons ( unpublished data ) ( Johard et al . , 2009 ) , we blocked neurotransmitter release from 5 Dv-E cells by expressing the tetanus toxin light chain ( TNT ) ( Sweeney et al . , 1995 ) . Compared to control flies , that is , flies that express inactive tetanus toxin UAS-Tet , Dv-E GAL4/UAS-TNT flies have a severe circadian locomotor activity deficiency . It includes a significant attenuation of morning anticipation as well as evening anticipation ( Figure 2A , arrows ) , a reduced major activity peak in DD and high levels of arrhythmicity ( Figure 2A; Table 1 ) . The mean locomotor activity of Dv-E GAL4/UAS-TNT flies compared to control flies was 44% in LD and 45% in DD . 10 . 7554/eLife . 02780 . 004Figure 2 . The five Dv-E cells are essential for circadian activity . ( A ) Group activity profiles during LD ( top ) and DD ( bottom ) cycles from DvPdf-GAL4 , Pdf-GAL80/UAS-Tet and DvPdf-GAL4 , Pdf-GAL80/UAS-TNT flies averaged over 3 days of LD or DD data . White/black bars , LD cycle; Grey/black bars , DD cycle . Error bars represent standard error of the mean ( SEM ) and n = 16 for each group . Arrows indicate morning anticipation ( black ) and evening anticipation ( light gray ) , and dashed arrows indicate attenuated morning anticipation ( black ) and evening anticipation ( light gray ) . The activity ( with the standard error of the mean ) , DD rhythmicity and period are also shown . ( B ) Averaged group activity profiles during LD cycles from UAS-Kir/+; TubGAL80ts/+ ( left panel ) , DvPdf-GAL4 , Pdf-GAL80/+ ( middle panel ) and DvPdf-GAL4 , Pdf-GAL80/UAS-Kir; TubGAL80ts/+ ( right panel ) flies at 30°C . The left and middle panels are the two parental control strains and the right panel is the activity-inhibited strain . At the high temperature the control flies show a typical increased morning peak and delayed evening peak . In the right panel , the GAL80ts become inactivated at 30°C allowing KIR expression in the E cells . The dashed arrows indicate attenuated morning and evening peaks . White/black bars indicate activity events in day/night as above . n = 24 for DvPdf-GAL4 , Pdf-GAL80/UAS-Kir; TubGAL80ts/+flies and n = 20 for parental groups . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 00410 . 7554/eLife . 02780 . 005Table 1 . Circadian behavior parameters of different genotypes under constant darkness ( DD ) DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 005GenotypeNRhythmic NPercent rhythmicPeriod ( hr ) ± S . D . Power ±S . D . Pdf-GAL4/+151493 . 324 . 3 ± 0 . 193 ± 15UAS-DBTs/+161610023 . 7 ± 0 . 3125 ± 19Clk4 . 1m-GAL4/+161487 . 524 . 2 ± 0 . 186 ± 15DvPdf-GAL4/+322990 . 623 . 6 ± 0 . 2143 ± 22DvPdf-GAL4 , Pdf-GAL80/+302686 . 724 . 5 ± 0 . 579 ± 14DvPdf-GAL4/+;Pdf-GAL80/+161487 . 524 . 2 ± 0 . 3113 ± 26Pdf-GAL4/+;UAS-DBTs/+161168 . 719 . 7 ± 1 . 274 ± 16Clk4 . 1m-GAL4/UAS-DBTs15128023 . 1 ± 0 . 4104 ± 23DvPdf-GAL4 , Pdf-GAL80/+;UAS-DBTs/+161487 . 523 . 5 ± 0 . 169 ± 7DvPdf-GAL4 , Pdf-GAL80/UAS-TNT1642524 . 1 ± 0 . 535 ± 6DvPdf-GAL4 , Pdf-GAL80/UAS-Tet161487 . 524 . 3 ± 0 . 2134 ± 32UAS-Kir/+;Tub-GAL80ts/+ ( 30°C ) 20168025 . 7 ± 1105 ± 16DvPdf-GAL4 , Pdf-GAL80/UAS-Kir;Tub-GAL80ts/+ ( 30°C ) 24520 . 826 . 5 ± 2 . 227 ± 5DvPdf-GAL4 , Pdf-GAL80/+ ( 30°C ) 20157524 . 1 ± 0 . 371 ± 13UAS-perS/+;Pdf-GSG/+ ( RU- ) 151510023 . 3 ± 0 . 2121 ± 12UAS-perS/+;Pdf-GSG/+ ( RU+ ) 161062 . 520 . 5 ± 1 . 473 ± 15UAS-perS/UAS-Kir;Pdf-GSG/+ ( RU− ) 141392 . 923 ± 0 . 394 ± 13UAS-perS/UAS-Kir;Pdf-GSG/+ ( RU+ ) 16318 . 722 . 8 ± 1 . 224 ± 6UAS-perS/+;Pdf-GSG/UAS-PDF RNAi ( RU− ) 151493 . 323 . 3 ± 0 . 288 ± 15UAS-perS/+;Pdf-GSG/UAS-PDF RNAi ( RU+ ) 16212 . 522 . 3 ± 0 . 819 ± 3UAS-perS/+;Pdf-GSG/+ ( RU+ to RU− ) 60549023 . 2 ± 0 . 194 ± 13UAS-perS/UAS-Kir;Pdf-GSG/+ ( RU+ to RU− ) 64416423 . 6 ± 0 . 656 ± 11UAS-perS/+;Pdf-GSG/UAS-PDF RNAi ( RU+ to RU− ) 623556 . 523 . 8 ± 1 . 248 ± 7UAS-dTrpA1/+ ( 21°C ) 161610024 . 4 ± 0 . 4117 ± 8DvPdf-GAL4/UAS-dTrpA1 ( 21°C ) 323196 . 824 . 2 ± 0 . 393 ± 6DvPdf-GAL4 , Pdf-GAL80/UAS-dTrpA1 ( 21°C ) 16127523 . 6 ± 0 . 572 ± 8Pdf-GAL4/UAS-dTrpA1 ( 21°C ) 323196 . 924 . 2 ± 0 . 3127 ± 15UAS-dTrpA1/+;TH-GAL4/+ ( 21°C ) 161610023 . 6 ± 0 . 4166 ± 19perS;UAS-dTrpA1/+ ( 21°C ) 161610020 . 3 ± 0 . 2131 ± 12perS;Pdf-GAL4/UAS-dTrpA1 ( 21°C ) 323196 . 919 . 7 ± 0 . 6102 ± 10perS;DvPdf-GAL4/UAS-dTrpA1 ( 21°C ) 323196 . 920 . 1 ± 0 . 297 ± 13perS;DvPdf-GAL4 , Pdf-GAL80/UAS-dTrpA1 ( 21°C ) 161487 . 521 . 1 ± 0 . 7125 ± 19pdfr−;UAS-PDFR16637 . 521 . 9 ± 1 . 234 ± 7pdfr−;DvPdf-GAL4 , Pdf-GAL80/UAS-PDFR181372 . 224 . 8 ± 0 . 477 ± 10pdfr−;DvPdf-GAL4/UAS-PDFR;Cry-GAL80/+16743 . 822 . 1 ± 0 . 737 ± 8pdfr−;DvPdf-GAL4 , UAS-dTrpA1/UAS-PDFR ( 21°C ) 30248023 . 7 ± 0 . 969 ± 9pdfr−;DvPdf-GAL4 , UAS-dTrpA1/UAS-PDFR;Pdf-GAL80/+ ( 21°C ) 24187523 . 8 ± 0 . 765 ± 13pdfr−;DvPdf-GAL4 , UAS-dTrpA1/UAS-PDFR;Cry-GAL80/+ ( 21°C ) 241145 . 821 . 5 ± 133 ± 5pdfr−;DvPdf-GAL4 , UAS-dTrpA1/+ ( 21°C ) 321340 . 621 . 2 ± 1 . 129 ± 8DvPdf-GAL4 , UAS-dTrpA1/+;cry01 ( 21°C ) 322578 . 123 . 2 ± 0 . 764 ± 14DvPdf-GAL4 , UAS-dTrpA1/+ ( 21°C ) 30248024 . 1 ± 0 . 371 ± 13DvPdf-GAL4 , UAS-dTrpA1/+;UAS-Cul-3 RNAi #1/+ ( 21°C ) 312580 . 624 . 6 ± 0 . 385 ± 8DvPdf-GAL4 , UAS-dTrpA1/UAS-Cul-3 RNAi #2 ( 21°C ) 322784 . 424 . 2 ± 0 . 473 ± 9DvPdf-GAL4 , UAS-dTrpA1/+;UAS-Cul-3 RNAi #3/+ ( 21°C ) 322578 . 124 . 1 ± 0 . 382 ± 19UAS-Cul-3 RNAi #1/+ ( 21°C ) 161610024 . 1 ± 0 . 2131 ± 11UAS-Cul-3 RNAi #2/+ ( 21°C ) 161593 . 823 . 5 ± 0 . 2103 ± 21UAS-Cul-3 RNAi #3/+ ( 21°C ) 161593 . 823 . 7 ± 0 . 195 ± 13 Use of a different reagent , expression of the potassium channel Kir , to suppress neuronal activity in the 5 Dv-E cells also causes a significant reduction in both morning and evening anticipation as well as high percentage of DD arrhythmicity ( Figure 2B , right , arrows and Table 1 ) compared to the two parental strains ( Figure 2B , left and middle ) . The use of Tub-GAL80ts and high temperature was to prevent Kir expression during development . The unusual level of nocturnal activity at high temperature ( previously reported , Majercak et al . , 1999 ) does not obscure identification of the morning and evening anticipation peaks ( Figure 2B ) . The M cells ( s-LNvs ) impact wild-type E cell molecular oscillations ( Stoleru et al . , 2005 ) as well as control DD rhythmicity . To identify relevant M cell signals , we repeated the previous strategy ( Stoleru et al . , 2005 ) but altered circadian period with M cell overexpression of UAS-perS; perS is a missense mutation that causes a short period phenotype ( Li and Rosbash , 2013 ) . This was combined with simultaneous M cell expression of RNAi constructs to screen for the candidate signal molecule . Since long-term expression of RNAi constructs can have developmental effects , we used the Pdf-geneswitch system and fed flies with drug to activate GeneSwitch protein for 5 days , that is , 2 LD days and 3 DD days ( Depetris-Chauvin et al . , 2011 ) . This protocol should temporally express PERS , which indeed shortened the circadian period to 20 . 5 hr ( Figure 3A; Table 1 ) . Since sibling flies on normal food maintained a normal 23 . 5 hr period and returning the flies to normal food also reversed the short period to 23 . 5 hr , short-term drug feeding is sufficient to induce the transgene and accelerate circadian period to generate an approximate 14 hr phase advance at the end of the protocol ( Figure 3A , B ) . 10 . 7554/eLife . 02780 . 006Figure 3 . M cells use the PDF peptide and neuronal activity to adjust period in DD . ( A ) Accelerated M cells send PDF as a resetting signal to cause a daily advanced activity phase . Double plotted averaged actograms of representative individual flies from each genotype are shown . Experimental group were transferred to RU486 food for 2 days in LD and the maintained for 3 days in DD before being returned back to food containing vehicle . Control group were maintained in vehicle food . Flies expressing Kir and PDF RNAi in M cells gradually lose rhythmicity during the 3 drug feeding DD days . In the actograms , white background represents day , gray background represents darkness . Red lines indicate the DD phase of the experimental group , and the blue lines indicate the DD phase of control group . The genotype of each group was labeled above the panels . ( B ) The quantification of phase change for the genotypes described in ( A ) . n = 60-64 for each group . ‘*’ means p<0 . 001 as determined by one way analysis of variance ( ANOVA ) , Tukey post hoc test , and the error bars indicate SEM . ( C ) Co-expression of PDF RNAi in accelerates M cells prevented period shortening . The histogram shows the period of five genotypes: UAS-SGG; Pdf-GAL4/+ , UAS-SGG; Pdf-GAL4/UAS-TNT , UAS-SGG; Pdf-GAL4/UAS-Tet , UAS-SGG; Pdf-GAL4/UAS-Trh RNAi , UAS-SGG; Pdf-GAL4/UAS-PDF RNAi . Tetanus toxin light chain ( TNT ) was used to block neurotransmitter releasing , and Tet is an inactive form of TNT , Trh RNAi was used as control of PDF RNAi since PDF cells do not express tryptophan hydroxylase ( Trh ) , PDF RNAi is as in Figure 2B n = 13–15 for each group . ‘’ represents p<0 . 001 as determined by one way analysis of variance ( ANOVA ) , Tukey post hoc test , and the error bar indicates SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 006 Flies co-expressing PDF RNAi along with UAS-perS showed no short period phenotype ( Figure 3A , B ) . A series of control RNAi and tetanus toxin light chain ( TNT ) constructs suggests that the effect is specific to this single neuropeptide . For example , PDF cells may not express a classical neurotransmitter that causes E cells to follow accelerated PDF neurons ( Figure 3C ) . Temporal co-expression of the inwardly rectifying potassium channel ( Kir ) with UAS-perS similarly inhibited the short period phenotype , indicating that PDF cell firing is also required for maintaining PDF-mediated communication from M cells ( Figure 3A , B ) . A recent paper also concluded that circadian period is determined by multiple independent oscillators , M cells and subset of E cells , which are coordinated by PDF signaling ( Yao and Shafer , 2014 ) . The data taken together indicate that PDF-cell firing and PDF generate the M cell resetting signal . The importance of M cell neuronal activity suggested that firing might be more generally relevant to phase resetting , despite the prevailing cell-autonomous model . To address this possibility , we artificially activated circadian neurons with the thermo-activated dTrpA1 cation channel under control of the DvPdf-GAL4 driver and assayed the phase response . We used the well-established anchored PRC protocol ( APRC; Kaushik et al . , 2007 ) with firing induced by a 2 hr temperature shift of flies from 21°C to 30°C during the night of the last LD cycle; conditions were returned to 21°C and constant darkness for the rest of the experiment . We focused initially on a time in the early evening ( ZT15 ) when M cell activity should be low ( Cao and Nitabach , 2008; Cao et al . , 2013 ) and more importantly when exposure to light ( or even a 37°C heat pulse [Kaushik et al . , 2007] ) elicits a maximal phase-shift in wild-type flies . This protocol caused a phase delay of approximately 3–4 hr ( Figure 4A ) , similar to the delay caused by a 2 hr saturating ( 2000 lux ) light pulse at ZT15 ( Bao et al . , 2001 ) . Addition of Pdf-GAL80 to DvPdf-GAL4 , that is , use of the Dv-E driver , had a dramatically reduced phase-shift , which indicates that PDF morning cell firing is necessary for the DvPdf-GAL4 firing-induced phase-shift . Firing was next restricted to PDF neurons , that is , a 2-hr temperature shift of Pdf-GAL4/UAS-dTrpA1 flies at ZT15 . This protocol elicited a similar 3-hr phase delay ( Figure 4A ) , indicating that PDF morning cell firing is not only necessary but also sufficient for a quasi-normal firing-mediated phase-shift at ZT15 . 10 . 7554/eLife . 02780 . 007Figure 4 . Activation of M cells is necessary and sufficient to trigger a phase-shift without light . ( A ) Only the drivers that express dTrpA1 in the PDF+ morning cells can cause a phase-shift in constant darkness . All of the GAL4>UAS-dTrpA1 flies were first entrained during 3 LD days to synchronize their endogenous clock and the transferred to 30°C at ZT15 for 2 hr during the last LD night and then returned to 21°C for the following days in DD . GAL4 lines that exhibit substantial phase delays label the PDF-positive pacemaker neurons . n = 16–32 for each group . Genotype of each group was labeled above the histogram . DA = dopaminergic neurons . M and E cells were described in main text . The letters ‘a’ and ‘b’ indicate significantly different groups ( p<0 . 001 ) , by one way analysis of variance ( ANOVA ) , Tukey post hoc test . The error bars indicate SEM . ( B ) Firing of PDF positive morning cells induces quasi-normal phase-shifts compared to light-induced phase-shifts . Pdf-GAL4/UAS-dTrpA1 flies exhibit a phase delay at ZT15 and a phase advance at ZT21 after a 2 hr 30°C pulse or a 2 hr light pulse . n = 32 for each group . The error bars indicate SEM . Note that phase-shift values are not very different from a standard 10 min light pulse . ( C ) The magnitude of a neuronal firing-induced phase-shift is larger in perS flies . perS; pdf-GAL4/UAS-dTrpA1 , perS; DvPdf-GAL4/UAS-dTrpA1 and perS; UAS-dTrpA1 flies were exposed to a 2 hr 30°C pulse at different circadian times . n = 16-32 for each group . The error bars indicate SEM . ( D ) The detailed phase shift panel of the 30°C pulse at ZT15 and ZT18 of perS; DvPdf-GAL4/UAS-dTrpA1 flies . Red box indicates the time of the 2 hr pulse of 30°C . The dashed line represents the 30°C pulsed group and the solid line the control group . Red arrows indicate the direction of phase change . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 007 Firing of morning pacemakers also causes proper phase advances at ZT21 , and control flies with TrpA1 expression in E cells but not in M cells show no phase advance ( Figure 4B and data not shown ) . This indicates that PDF cell firing more broadly mimics light-induced phase-shifting . It is also notable that firing of many other neuronal subsets , including the strongly activity-promoting dopaminergic neuronal group with TH-GAL4 , had little or no phase-shifting effect ( Figure 4A and data not shown ) . These negative results confirm that a 2 hr 30°C pulse alone is unable to cause a substantial phase-shift . To extend further this relationship , we assayed firing-induced phase-shifting in a perS genetic background . This is because previous studies had shown that the perS mutation not only shortens circadian period to 19 hr but also dramatically alters light-induced phase-shifting , from a low amplitude type 1 PRC to a high amplitude type 0 PRC ( Bao et al . , 2001 ) . Similar to light , neuronal firing induced an exaggerated phase change in a perS background compared to wild type in the advance zone ( maximum 6–7 hr; Figure 4C ) . The phase-shifts of parental control strains without dTrpA1 expression were statistically indistinguishable from unheated groups , further indicating that the 2 hr 30°C pulse has no phase-shifting effect ( Kaushik et al . , 2007 ) . Addition of Pdf-GAL80 to the perS; DvPdf-GAL4; UAS-dTrpA1 background dramatically reduced both advance and delay phase-shifts ( data not shown ) , confirming that morning cell activity makes the major contribution to the generation of this enhanced phase-shift . We note that the ability of PDF to contribute to phase-shifting has been previously demonstrated in other insects ( Petri and Stengl , 1997 ) . Given the important role of the Dv-E cells to circadian timing and to locomotor activity in DD as well as LD ( Figures 1 and 2 ) , we suspected that they are important downstream targets of PDF and therefore important sites of PDFR expression . Like pdf01 mutant flies , a mutant in PDFR ( pdfr5304 flies ) exhibits a reduced morning peak and a phase advanced evening peak in LD as well as a short period and a high percentage of arrhythmic flies in DD ( Hyun et al . , 2005; Lear et al . , 2005 ) ( Figure 5A , left ) . Although a large number of circadian cells including M and E cells normally express PDFR ( Im et al . , 2011 ) , we could restrict PDFR expression to only the 5 Dv-E cells of pdfr5304 flies with the DvPdf-GAL4 , Pdf-GAL80 driver expressing PDFR in the pdfr5304 background . Essentially all major circadian deficiencies were rescued ( Figure 5A , middle compared to 5A left and Table 1 ) . Because CRY and PDFR are co-expressed in the same groups of circadian cells ( Im et al . , 2011 ) , we inhibited PDFR expression in CRY-positive Dv-E cells with Cry-GAL80 . This almost completely eliminated the rescue of the morning , evening peaks and rhythmicity on the pdfr5304; DvPdf-GAL4 , Pdf-GAL80/UAS-PDFR flies ( Figure 5A , right compared to Figure 5A , middle and Table 1 ) , suggesting that it is the CRY-positive subset of Dv-E cells that are the key targets of PDF and that makes the major contribution to circadian behavior in LD as well as DD . 10 . 7554/eLife . 02780 . 008Figure 5 . Restoring E cell PDFR rescues period , rhythmicity and firing-induced phase shifts . ( A ) pdfr mutant flies show no morning peak and an advanced evening peak as expected ( left panel ) . Rescue of PDFR expression with the Dv-E cell driver in this mutant background restores both morning and evening anticipation peaks , that is , these flies show an intact morning peak and more normal onset of evening peak activity ( middle panel ) . Inhibition of rescue in CRY-positive E cells with Cry-GAL80 prevents the restoration of morning and evening activity peaks . n = 16–18 for each group . Genotypes are shown under each panel . The error bar indicates SEM . ( B ) CRY staining pattern in DvPdf-GAL4 , Pdf-GAL80; UAS-mCD8::GFP brains . GFP ( green ) is expressed in 4 LNds ( left and middle panel ) , and CRY signal ( magenta ) is visible in the large PDF cells ( below and to the left of the arrow ) as well as 3 LNds ( left and right panel ) . The arrow shows the one CRY-positive DV-E cell . Magnified images of the upper dash line boxed area are shown in lower solid line box . ( C ) Adding a copy of Cry-GAL80 eliminates GFP staining of the CRY-positive Dv-E cells as well as their branches , which are adjacent to the PDF cell dorsal projections . Brains from DvPdf-GAL4/+; UAS-mCD8::GFP/+ ( left panel ) and DvPdf-GAL4/+; Cry-GAL80/UAS-mCD8::GFP ( right panel ) were stained with a GFP antibody ( red ) . The upper arrow shows the typical projection from DvPdf -labeled CRY positive LNd neuron and the lower arrow points to likely branches from the 5th-sLNv . With Cry-GAL80 , both of these projections are absent . Note that a copy of Cry-GAL80 can block GAL4 activity in CRY positive Dv-E cells but not in sLNv , the dorsal projections from sLNvs are still visible in DvPdf-GAL4/+; Cry-GAL80/UAS-mCD8::GFP brains ( right panel ) . ( D ) Flies were exposed to a 30°C pulse for 2 hr at ZT15 ( dark gray bars ) or ZT21 ( light gray bars ) . The expression pattern of UAS-dTrpA1 and UAS-PDFR is shown above the histogram . Genotypes shown below each histogram are ( from left to right ) : DvPdf-GAL4/UAS-dTrpA1 ( n = 32 ) , pdfr−; DvPdf-GAL4/UAS-dTrpA1 ( n = 32 ) , pdfr−; DvPdf-GAL4 , UAS-dTrpA1/UAS-PDFR; Pdf-GAL80 ( n = 24 ) , pdfr−; DvPdf-GAL4 , UAS-dTrpA1/UAS-PDFR; Cry-GAL80 ( n = 24 ) and pdfr−; DvPdf-GAL4 , UAS-dTrpA1/UAS-PDFR ( n = 30 ) . ‘’ represents p<0 . 001 as determined by the student's t test and indicates a significant phase change . The error bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 008 To localize these CRY-positive Dv-E cells , we stained DvPdf-GAL4 , Pdf-GAL80>UAS-mCD8:GFP brains with an anti-CRY antibody and identified a single LNd ( Figure 5B , arrow ) as well as the 5th small LNv ( [Yoshii et al . , 2008] and data not shown ) . To confirmed that there is only 1 CRY-positive LNd , adding a copy of Cry-GAL80 to DvPdf-GAL4>UAS-mCD8:GFP flies eliminated GFP expression in that single LNd and the 5th-sLNv ( Figure 5C; arrow points to a fiber from CRY+ LNd to the accessory medulla [Yoshii et al . , 2008] ) but not in M cells ( Figure 5C ) . Based on the behavior ( Figure 5A ) , these 2 CRY+ Dv-E cells play a more important circadian and activity role . We next assayed the importance of PDFR expression in Dv-E cells for firing-induced phase-shifting . Although flies without PDF are too arrhythmic to assay phase-shifting ( data not shown ) , the strain without the cognate receptor ( pdfr5304 flies ) is for unknown reasons a bit more rhythmic ( 30–60% rhythmicity ) ( Lear et al . , 2009; Im et al . , 2011 ) , sufficient to support a phase-shifting assay . We therefore assayed firing-induced phase-shifting in rhythmic PDFR mutant groups with and without PDFR rescue in the 5 Dv-E cells . The magnitude of the phase-shift was severely inhibited in pdfr5304 flies , consistent with the importance of PDF signaling for communication ( Figure 5D ) , and expressing PDFR and dTrpA1 with the DvPdf driver rescued both phase delays and advances . Although the same experiment with the Dv-E cell driver ( DvPdf-GAL4; Pdf-GAL80 ) failed to induce a phase-shift , the negative result probably reflects the requirement for dTrpA1 expression within PDF cells , which is inhibited by the presence of the Pdf-GAL80 transgene . Also because rescue of PDFR expression in PDF cell and CRY negative Dv-E cells with DvPdf-GAL4; Cry-GAL80 driver is not sufficient to rescue the major circadian deficiency ( Figure 5A; Table 1; Lear et al . , 2009 ) and phase-shifting of the PDFR mutant strain ( Figure 5D ) , we favor the interpretation that PDFR expression in the 5 Dv-E cells , especially in the 2 CRY+ Dv-E cells , is sufficient to rescue PDF cell firing-induced phase-shifting like its rescue of circadian behavior ( Figure 5C ) . The data taken together suggest that firing of morning cells releases PDF and that a subsequent association with its cognate receptor on Dv-E cells is important for both phase delays and advances . Because rapid CRY-dependent TIM degradation is a key event in light or heat-mediated phase-shifting ( Suri et al . , 1998; Yang et al . , 1998 ) , PDF cell firing might also trigger TIM degradation . There was indeed a dramatic and rapid reduction of TIM but not PER staining intensity in circadian neurons after M cell firing at ZT15 when TIM and PER are cytoplasmic ( Figure 6A , B ) ; the staining intensity of control flies without dTrpA1 expression was unaffected by the temperature shift ( Figure 6D ) . M cell firing also caused nuclear TIM degradation in all circadian neurons at ZT21 ( Figure 6E ) , consistent with the firing-mediated phase advances observed at this time ( Figure 4B ) . 10 . 7554/eLife . 02780 . 009Figure 6 . TIM but not PER in downstream circadian neurons responds to M cell firing . ( A ) TIM and PER staining in central pacemakers of fly brains after firing of M cells at ZT15 . PER or TIM staining intensity is measured in pdf-GAL4/UAS-dTrpA1 flies after a 2 hr 30°C pulse at ZT15 . Brains were co-staining with anti-PDF ( green ) to visualize PDF+ cells . Asterisk indicates LNds ( top ) and s-LNvs cells ( bottom ) . Note that staining in l-LNvs ( higher than asterisk ) does not change very much with firing . ( B ) Quantification of PER and TIM staining intensity in each group of clock neurons with or without a 30°C pulse ( standard error of the mean [±SEM] ) . 5 brains and 10 hemispheres were quantified in each group . Scale bar = 20 μm . ( C ) TIM and PER staining in dorsal region ( asterisk ) of flies as described in ( A ) . ( D ) TIM levels are not changed in wild-type fly brains after 2 hr at 30°C at ZT15 . TIM levels in central ( left panel ) and dorsal circadian neurons ( right panel ) are shown in representative brains . ( E ) TIM levels are significantly decreased in Pdf-GAL4/UAS-dTrpA1 brains ( left panel ) but not in WT control brains ( right panel ) after 2 hr 30°C pulse at ZT21 . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 009 To complement the in vivo firing , we assayed the effect of in vitro firing on TIM degradation using the ATP-gated P2X2 cation channel ( Lima and Miesenbock , 2005; Yao et al . , 2012 ) . Brains were dissected from Pdf-GAL4; UAS-p2x2; cry01 flies at ZT21 and incubated with 2 . 5 mM ATP for 2 hr . The inclusion of homozygous cry01 ( null for CRY expression ) in the genetic background was so that the brains would be light-blind , to minimize the effect of light on the TIM degradation assay . TIM levels were dramatically reduced after ATP application , and there was no significant effect in control cry01 flies ( Figure 7A ) . The results mimic the similar effects of in vivo dTrpA1 activation , confirming that PDF neuron firing causes TIM degradation . 10 . 7554/eLife . 02780 . 010Figure 7 . CRY is not required for M cell firing-induced TIM degradation and Phase Shifts . ( A ) TIM staining intensity strongly decreases in response to M cell firing caused by 2 hr incubation with 2 . 5 mM ATP ( Pdf-GAL4/UAS-p2x2; cry01 , left panel ) at ZT21 . TIM levels in the control cry01 group were not affected by the 2 . 5 mM ATP incubation ( right panel ) . Asterisks indicate DNs , LNds and LNvs ( top to bottom ) . ( B ) Firing-induced phase-shifting behavioral data in a cry null mutant strain . This strain shows a normal response to firing at ZT15 but an exaggerated response at ZT21 . n = 32 for each group . ‘’ represents p<0 . 001 as determined by the student's t test for normally distributed data . The error bars indicate SEM . ( C ) TIM is degraded in DNs , LNds and LNvs ( asterisks , top and bottom , respectively ) after firing at ZT21 even without CRY . An anti-PDF antibody ( green ) was used to visualize LNvs , and TIM was visualized with an anti-TIM antibody ( red ) . TIM levels were markedly decreased after a ZT21 2 hr 30°C pulse of DvPdf-GAL4/UAS-dTrpA1; cry01 flies . TIM was not decreased in control UAS-cry; cry01 flies . The control strains are in the left four panels and the firing panels in the right two panels . Odd panels show staining with PDF and anti-TIM , whereas the even panels show staining only with anti-TIM . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 010 The canonical cell-autonomous model for TIM degradation and phase-shifting requires functional CRY ( Emery et al . , 2000 ) . However , the in vitro P2X2 results suggest that firing-induced phase-shifting is CRY-independent . We therefore repeated the in vivo dTrpA1 experiments in a cry01 background . Light pulses failed to change the phase of these DvPdf-GAL4 , UAS-dTrpA1; cry01 flies as expected from the role of CRY in photoreception ( data not shown ) , but they still responded to M cell firing . Although ZT21 phase advances showed a dramatic and enigmatic 10-hr advance in this background ( Figure 7B ) , ZT15 phase delays in the cry01 background were only marginally different from wild-type . This suggests that delays may be more simply firing-dependent than advances , which may respond to a more complicated combination of light and firing . We also examined TIM levels in these cry01 brains . Like the results in a wild-type strain , the pacemaker firing paradigm dramatically reduced TIM levels in these circadian neurons at ZT21 , especially in the LNds; this further confirms that the neuronal firing-phase resetting pathway does not require CRY ( Figure 7C ) . Our data taken together indicate that neural firing causes TIM degradation and phase resetting , which are controlled by a mechanism at least partially independent of CRY-mediated protein degradation . The E3 ubiquitin ligase component Cullin-3 ( CUL-3 ) was recently shown to regulate TIM degradation and promote TIM/PER oscillations in DD ( Grima et al . , 2012 ) . Moreover , the effect of CUL-3 is particularly striking when TIM is cytoplasmic in the early night . This suggests that CUL-3 might be important for PDF mediated , CRY-independent TIM degradation and phase-shifting , especially in the early night-delay zone . Indeed , co-expression of CUL-3 RNAi with dTrpA1 under DvPdf-GAL4 control did not significantly affect DD rhythmicity and period at 21°C ( Table 1; Grima et al . , 2012 ) . However , these flies had reduced firing-mediated phase delays at ZT15 ( Figure 8A ) . TIM staining revealed that the co-expression also substantially reduced TIM degradation in DvPdf-GAL4 labeled M and E cells ( Figure 8B , C ) . In contrast , TIM degradation within the DNs was potent and indistinguishable from control strains , presumably because the DvPdf-GAL4 driver and UAS constructs are not expressed in the DNs ( Figure 8B ) . This suggests that the effects of the CUL-3 RNAi constructs are principally cell-autonomous and that dTrpA1-mediated M cell-firing remains potent despite co-expression of the RNAi constructs . The results taken together indicate that M cell firing in the early night activates a PDFR-CUL-3 pathway to reduce cytoplasmic TIM accumulation . The results also imply that TIM levels within the DvPdf-GAL4-labeled pacemakers make a substantial contribution to the magnitude of the phase delay . We therefore suggest that E cells are major pacemakers under more natural L-D conditions and that M cells serve principally to integrate light information and phase adjust the E cells through firing and PDF release ( Figure 9 ) . 10 . 7554/eLife . 02780 . 011Figure 8 . CUL-3 is involved in the delay zone phase shift response to PDF cell activation . ( A ) CUL-3 RNAi lines and background control lines were co-expressed with dTrpA1 under control of DvPdf-GAL4 . Their phase-shifts were measured after 2 hr 30°C pulse at ZT15 or ZT21 . Genotypes were shown below each group . n = 30–32 for each group . ‘’ means p<0 . 001 as determined by one way analysis of variance ( ANOVA ) , Tukey post hoc test , and the error bars indicate SEM . ( B ) TIM levels were much less affected by neuron firing in DvPdf positive neurons expressing CUL-3 RNAi . Flies were transferred to 30°C and pulsed at ZT15 for 2 hr . Their brains were then dissected and immunostained with antibodies against TIM ( red ) and PDF ( not shown ) . Different groups of circadian neurons were imaged and classified by their positions relative to PDF staining . Individual representative brains are shown . The experiment was repeated three times with qualitatively identical results in all three groups: LNds , s-LNvs and DNs . ( C ) TIM levels in different clock neuron groups for the genotypes described in ( A ) were quantified and normalized to the values of the control ( same genotype without 30°C pulse , set to 100% ) . 10–12 hemispheres were examined in each group . *p<0 . 001 compared to the controls as determined by the student's t test . The error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 01110 . 7554/eLife . 02780 . 012Figure 9 . A model depicting how PDF and PDF-positive morning cells respond to light cues and control the pace of downstream E cells . Light activates morning cells directly or only indirectly through the visual system or from other clock neurons and induces PDF secretion . PDF then acts on downstream evening cells to promote cytoplasmic TIM degradation through a CRY-independent and CUL-3 dependent pathway , which causes phase or period adjustment . DOI: http://dx . doi . org/10 . 7554/eLife . 02780 . 012 Diurnal behavior in Drosophila is currently best explained by a dual-oscillator model , which emphasizes the M cells as master pacemakers and the E cells as secondary slave oscillators . Here we provide a modified configuration with a more prominent role for 5 Dv-E cells ( Figure 9 ) . This view is based on several new and unanticipated results . First , the 5 Dv-E cells are not only responsible for the evening anticipation peak and its timing , but they also contribute to the morning peak , free running rhythms and total activity . Otherwise put , these cells appear to play a key role in all aspects of circadian rhythms and locomotor activity . Second , E cell resetting can occur in a non cell-autonomous manner , through M cell neuron firing , PDF release and E cell PDFR activation . Third , TIM level changes within the 5 Dv-E cells in response to M cell firing likely make a major contribution to the resulting phase changes , at least in the early night when phase delays occur , and therefore make a major contribution to circadian timekeeping under more normal LD conditions . The classical cell-autonomous model of Drosophila phase-shifting has long contrasted with the well-established view of phase-shifting in the mammalian SCN . In this system , firing from optic track activates SCN NMDA receptors , which is necessary and sufficient for light-mediated phase shifts . Importantly , the contribution of PDF cell firing and PDF to Drosophila phase-shifting brings this system closer to that of mammals . Because many aspects of Vasoactive intestinal polypeptide ( VIP ) function in mammals resemble those of PDF in flies ( Vosko et al . , 2007 ) , it will be interesting to know if VIP cell firing can mimic light and phase-shift mammalian rhythms . An important contribution of the E cell TIM level change to phase delays is consistent with previous work showing that a full light pulse at ZT15 induces rapid and potent TIM degradation in all 6 LNds as well as reduced TIM degradation in response to a light pulse of decreased intensity ( Tang et al . , 2010 ) . This conclusion is also consistent with more recent data from Emery et al . ( Lamba et al . , 2014 ) . Our data here also point to a more important role of the LNds rather than the M cells: the LNds show more TIM degradation in response to firing and a bigger response to the CUL-3 knockdown , which correlates with the decrease in phase delay caused by the knockdown ( Figure 8 ) . This previous work also indicates that there is no observed TIM degradation in the PDF-positive M cells after a light pulse ( Tang et al . , 2010 ) . This is consistent with older work showing that cry rescue in non M cell circadian neurons ( with Tim-GAL4/Pdf-GAL80 ) provides a robust light mediated PRC ( Stoleru et al . , 2007 ) , suggesting that canonical CRY-JET-mediated TIM degradation within M cells is unnecessary to convey light information to the brain clock . Our data extend this picture to the more dorsal circadian DNs by showing that TIM degradation within this circadian subpopulation is probably also not essential , at least to phase delays ( Figure 8B , C ) ( Lamba et al . , 2014 ) . Although these data do not exclude a more inclusive view , our results indicate that acute PDF cell activation causes TIM degradation within the 5 Dv-E cells and that this degradation correlates with the magnitude of the phase delay . We note however that PDF cell activation may also function in other ways , for example by chronically affecting PER levels ( Li et al . , 2014 ) or by activating other downstream neurons ( Seluzicki et al . , 2014 ) . CUL-3 is the first ubiquitin ligase component other than JET to connect with phase-shifting . Importantly , CUL-3 has been shown to participate in light-independent TIM degradation within the cytoplasm ( Grima et al . , 2012 ) . This is where TIM is localized during its accumulating phase in the early night ( see Figure 6A ) , which perfectly matches the preferential role of CUL-3 in promoting delay zone phase-shifts . However , the CUL-3 RNAi effects are incomplete . The new non cell-autonomous pathway described here may therefore cooperate with the more traditional cell-autonomous light-CRY-JET pathway to effect phase shifting . As the latter is essential for light-mediated phase-shifting , another non-exclusive possibility is that the two pathways are in series rather than in parallel , that is , that the light-CRY-JET pathway is upstream of neuronal firing and that the RNAi effects reflect incomplete knockdowns . This notion fits with recent studies suggesting that the light-CRY interaction promotes neuronal firing ( Fogle et al . , 2011 ) . At least 2 and probably 3 LNds are CRY-negative by antibody staining criteria even after many days of extended incubation in DD ( Yoshii et al . , 2008 ) . As all 6 LNds respond similarly to a full light pulse and to neuronal firing , this indicates that TIM degradation within the CRY-negative LNds is non cell-autonomous and probably requires communication from M cells and/or from other neurons like their CRY-positive LNd neighbors ( Lamba et al . , 2014 ) . If the CRY-negative LNds are also PDFR-negative ( Im et al . , 2011 ) , other neuropeptides , neurotransmitters or perhaps even gap junctions may participate in the transfer of light or firing information upstream of TIM degradation . Surprisingly , output from the 5 Dv-E cells controls a substantial fraction of evening activity as well as overall locomotor activity ( Figure 2 ) . These conclusions are also based on the phenotypes of Kir as well as dTrpA1 expression under Dv-E GAL4 control ( Figure 2B and data not shown ) . As Dv-E GAL4/UAS-TNT flies have significantly impaired activity in DD as well as in LD ( Figure 2A ) , the residual locomotor activity may be clock-independent . However , these flies still manifest an evening anticipation peak under LD conditions ( Figure 2A ) , suggesting incomplete suppression by TNT or a role of other circadian activity neurons . As the circadian activity-promoting subset of circadian neurons expresses CRY and PDFR and the Dv-E GAL4 driver only labels 1 CRY positive LNd ( Figure 2 ) , a simple explanation is that one or more of the other 2 CRY+ LNds also promotes locomotor activity . The Dv-E cells also contribute to morning activity . This is because rescue of PDFR expression only in the Dv-E cells is sufficient to rescue the morning anticipation peak ( Figure 5C ) . In addition , Dv-E cell output is necessary for a robust morning peak ( Figure 2A , B ) . Yet previous results connect the M cells to morning activity ( Choi et al . , 2012 ) . These cells also communicate via PDF with the DN1p dorsal neurons , which also impact the morning anticipation peak ( Zhang et al . , 2010a , 2010b ) . Although all of these considerations suggest that redundant circuits downstream of the M cells underlie the morning anticipation peak , only output from the Dv-E cells has been shown to be necessary for morning activity , and there is evidence that even the M cells are not necessary for the morning activity peak ( Sheeba et al . , 2010 ) . These PDF+ M cells have been considered the master pacemakers based on their critical role in keeping time in constant darkness . However , their firing-resetting properties , the identification of the Dv-E cells as major activity neurons even in DD and the likely importance of TIM degradation within these cells for phase adjustment suggest a different reason for the strong molecular rhythms of the PDF neurons in DD and their contribution to behavioral rhythmicity under these conditions . These features likely reflect the important role of the PDF neurons as light-sensitive cells , directly through CRY and indirectly as post-synaptic targets of other light-sensitive cells such as photoreceptor cells within the eye and dorsal brain ( Malpel et al . , 2002; Yoshii et al . , 2008 and unpublished data ) . We therefore suggest that a major function of the PDF neurons is to integrate light information and use it to phase-adjust the Dv-E cells , which are important pacemakers under LD conditions ( Figure 9 ) . Phase is a more important parameter than period under these more ‘natural’ conditions , and expression of period-altering mutants under Dv-E control appropriately alters the phase of the major evening peak but fails to do so under PDF control ( Figure 1D and data not shown ) . Lastly , our data offer a simple mechanistic explanation for a long-standing difference between peripheral and central circadian oscillators in Drosophila . The amplitude of peripheral oscillators is dependent on light as they damp rapidly after transfer of flies to constant darkness . Fly peripheral oscillators are also dependent on the photoreceptor CRY ( Stanewsky et al . , 1998; Malpel et al . , 2002 ) . In contrast , the circadian brain network and resultant rhythmic behavior persist in constant darkness and in the absence of CRY ( Emery et al . , 1998 , 2000 ) . Our work here can explain this difference in a simple way: neuronal firing replaces the function of light and CRY and even acts at the same step in the circadian cycle , namely , to promote TIM degradation within circadian neurons . We suggest that this step is intrinsically weak and that TIM degradation is normally potentiated every day , by light and/or by firing . This ensures the maintenance of robust rhythms of brain and behavior without compromising the light- and firing-sensitivity needed for phase adjustment . DvPdf-GAL4 was provided by Dr . JH Park; per0;UAS-per was from Dr . Francois Rouyer; UAS-dTrpA1 was from Dr . Paul Garrity; pdfr5304; UAS-PDFR were from Dr . Paul Taghert; pdf-GSG was from Dr . Fernanda Ceriani ; the cry01 mutant was from Dr . JC . Hall; UAS-TNT and UAS-Tet were from Dr . Hubert Amrein; UAS-DBTs was from Dr . Jeffrey Price; UAS-SGG; Pdf-GAL4 , Pdf-GAL80 , Cry-GAL80 were described by Stoleru et al . ( 2004 ) ; UAS-p2x2 was from Dr . Orie Shafer; UAS-CUL-3 RNAi 1 ( VDRC 25875 ) , UAS-CUL-3 RNAi 2 ( VDRC 109415 ) , UAS-PDF RNAi ( VDRC 4382 ) , UAS-Trh RNAi ( VDRC 105414 ) were from VDRC . UAS-CUL-3 RNAi 3 ( BL 36684 ) was from Bloomington stock center . All of the other GAL4 and UAS lines were obtained from the Bloomington stock center . Flies were reared on standard cornmeal/agar medium supplemented with yeast . The adult flies were entrained in 12:12 light–dark ( LD ) cycles at 25°C . The flies carrying GAL4 and UAS-dTrpA1 were kept at 21°C to inhibit dTrpA1 activity . Locomotor activity of individual male flies ( aged 3–7 days ) was measured with Trikinetics Activity Monitors ( Waltham , MA ) for at least 4 days under 12:12 LD conditions followed by at least 7 days of constant darkness . The period and rhythmicity analysis was performed with a signal-processing toolbox implemented in MATLAB ( MathWorks , Natick , MA ) as described by Stoleru et al . ( 2004 ) . Group activity was also generated and analyzed with MATLAB . For neuronal firing- induced phase-shift experiments , flies were entrained in LD for 4 days at 21°C , transferred to 30°C for 2 hr at ZT15 or ZT21 in the last night of the LD cycle and were put back to 21°C for the following DD days . For RU486 experiments , flies were fed with normal food for first 2 LD days and then transferred to tubes containing 200 μg/ml RU486 food ( mifepristone , Sigma , USA ) for 5 days ( 2 LD days and 3 DD days ) . At CT0 of DD4 , the flies were put back to normal food . The phase difference was calculated by comparing the phase of RU486 feeding group with the vehicle feeding group . All statistical analysis was conducted using IBM SPSS software . The Wilks–Shapiro test was used to determine normality of data . Normally distributed data were analyzed with 2-tailed , unpaired Student's t tests or one way analysis of variance ( ANOVA ) followed by a Tukey–Kramer HSD Test as the post hoc test . Data were presented as mean behavioral responses , and error bars represent the standard error of the mean ( SEM ) . Differences between groups were considered significant if the probability of error was less than 0 . 05 ( p<0 . 05 ) . All of the experimental and control flies were in a cry01 background to avoid light-induced TIM degradation during brain dissection . Adult flies were maintained longer in LD , for at least 5 days of entrainment , because of the cry01 background . Flies were collect and dissected in PBS at ZT21 when TIM levels are high . Fresh brains were incubated with AHL medium containing 2 . 5 mM ATP or vehicle for 2 hr and then fixed using the standard brain immunocytochemistry procedure . Immunostaining was performed as described ( Tang et al . , 2010 ) . Fly heads were removed and fixed in PBS with 4% paraformaldehyde and 0 . 008% Triton X-100 for 45–50 min at 4°C . Fixed heads were washed in PBS with 0 . 5% Triton X-100 and dissected in PBS . The brains were blocked in 10% goat serum ( Jackson Immunoresearch , West Grove , PA ) and subsequently incubated with primary antibodies at 4°C overnight or longer . For TIM/PER/CRY and PDF co-staining , a rat anti-TIM ( 1:200 ) , rabbit anti-PER ( 1:1000 ) , rabbit anti-CRY ( 1:1000 ) and mouse anti-PDF ( 1:10 ) antibody ( Developmental Studies Hybridoma Bank , University of Iowa , Iowa city , IA ) were used as primary antibodies . For GFP staining , a mouse anti-GFP antibody ( Invitrogen ) was used at a 1:100 dilution . After washing with 0 . 5% PBST three times , the brains were incubated with Alexa Fluor 633 conjugated anti-rabbit ( PER ) and Alexa Fluor 488 conjugated anti-mouse ( PDF ) ( Molecular Probes , Carlsbad , CA ) at 1:500 dilution . The brains were washed three more times before being mounted in Vectashield Mounting Medium ( Vector Laboratories , Burlingame , CA ) and viewed sequentially in 1 . 1 μm sections on a Leica confocal microscope . To compare the fluorescence signals from different conditions , the laser intensity and other settings were set at the same level during each experiment . Fluorescence signals were quantified by ImageJ as described .	Most animals have daily rhythms of activity: some are awake during the day and asleep at night , whilst others are more active at night , or during the twilight hours around dawn and dusk . These cycles of activity are driven by an internal body clock , which is reset in response to external cues , like light and temperature , and which keeps the animal in sync with the day–night cycle . The fruit fly Drosophila has daily—or circadian—rhythms of behavior , which are controlled by a network of genes that are switched ‘on’ or ‘off’ at different times in every 24-hr period . These circadian genes encode various proteins , including PERIOD and TIMELESS . The levels of these two proteins increase during the day and into the night , until they reach a point at which they cause their own genes to be switched off . PERIOD and TIMELESS are then destroyed each morning , and the cycle begins anew . Most of these same proteins perform similar functions in mammals . In the fly brain , two groups of neurons express these key proteins and control the timings of activity or movement . One group , called M cells , regulates activity in the morning and also produces a small molecule called PDF . Another group , called E cells , controls evening activity , but is less well-defined . Since M cells can maintain circadian rhythms even in total darkness , these cells were also considered key ‘pacemaker neurons’ . However , Guo et al . now challenge this view by identifying five E cells that are the major source of circadian activity . Blocking the release of signaling molecules from these neurons severely disrupted movement in both the morning and the evening . The E cells are also critical for timekeeping under a normal day–night cycle . Guo et al . found that the rhythm of the E cells is reset when the M cell neurons fire , which causes a release of PDF molecules . Further , PDF molecules reset the E cells by causing the degradation of the TIMELESS protein—which is similar to the effect of light , although light cause TIMELESS to be degraded via a different biochemical pathway . Guo et al . suggest that under normal light–dark conditions the E cells are important for driving the flies' activity as well as for overall timekeeping . The M cells , instead , appear to function primarily to integrate information about light and reset the E cell clock . Challenges moving forward will include understanding other ways in which the firing of neurons can affect timekeeping , as well as looking if there any differences between the five E cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	PDF neuron firing phase-shifts key circadian activity neurons in Drosophila
Continuous adaptation allows survival in an ever-changing world . Adjustments in the synaptic coupling strength between neurons are essential for this capability , setting us apart from simpler , hard-wired organisms . How these changes can be mathematically described at the phenomenological level , as so-called ‘plasticity rules’ , is essential both for understanding biological information processing and for developing cognitively performant artificial systems . We suggest an automated approach for discovering biophysically plausible plasticity rules based on the definition of task families , associated performance measures and biophysical constraints . By evolving compact symbolic expressions , we ensure the discovered plasticity rules are amenable to intuitive understanding , fundamental for successful communication and human-guided generalization . We successfully apply our approach to typical learning scenarios and discover previously unknown mechanisms for learning efficiently from rewards , recover efficient gradient-descent methods for learning from target signals , and uncover various functionally equivalent STDP-like rules with tuned homeostatic mechanisms . How do we learn ? Whether we are memorizing the way to the lecture hall at a conference or mastering a new sport , somehow our central nervous system is able to retain the relevant information over extended periods of time , sometimes with ease , other times only after intense practice . This acquisition of new memories and skills manifests at various levels of the system , with changes of the interaction strength between neurons being a key ingredient . Uncovering the mechanisms behind this synaptic plasticity is a key challenge in understanding brain function . Most studies approach this monumental task by searching for phenomenological models described by symbolic expressions that map local biophysical quantities to changes of the connection strength between cells ( Figure 1A , B ) . Approaches to deciphering synaptic plasticity can be broadly categorized into bottom-up and top-down . Bottom-up approaches typically rely on experimental data ( e . g . , Artola et al . , 1990; Dudek and Bear , 1993; Bi and Poo , 1998; Ngezahayo et al . , 2000 ) to derive dynamic equations for synaptic parameters that lead to functional emergent macroscopic behavior if appropriately embedded in networks ( e . g . , Gütig et al . , 2003; Izhikevich , 2007; Clopath et al . , 2010 ) . Top-down approaches proceed in the opposite direction: from a high-level description of network function , for example , in terms of an objective function ( e . g . , Toyoizumi et al . , 2005; Deneve , 2008; Kappel et al . , 2015; Kutschireiter et al . , 2017; Sacramento et al . , 2018; Göltz et al . , 2019 ) , dynamic equations for synaptic changes are derived and biophysically plausible implementations suggested . Evidently , this demarcation is not strict , as most approaches seek some balance between experimental evidence , functional considerations and model complexity . However , the relative weighting of each of these aspects is usually not made explicit in the communication of scientific results , making it difficult to track by other researchers . Furthermore , the selection of specific tasks to illustrate the effect of a suggested learning rule is usually made only after the rule was derived based on other considerations . Hence , this typically does not consider competing alternative solutions , as an exhaustive comparison would require significant additional investment of human resources . A related problem is that researchers , in a reasonable effort to use resources efficiently , tend to focus on promising parts of the search space around known solutions , leaving large parts of the search space unexplored ( Radi and Poli , 2003 ) . Automated procedures , in contrast , can perform a significantly less biased search . We suggest an automated approach to discover learning rules in spiking neuronal networks that explicitly addresses these issues . Automated procedures interpret the search for biological plasticity mechanisms as an optimization problem ( Bengio et al . , 1992 ) , an idea typically referred to as meta-learning or learning-to-learn . These approaches make the emphasis of particular aspects that guide this search explicit and place the researcher at the very end of the process , supporting much larger search spaces and the generation of a diverse set of hypotheses . Furthermore , they have the potential to discover domain-specific solutions that are more efficient than general-purpose algorithms . Early experiments focusing on learning in artificial neural networks ( ANNs ) made use of gradient descent or genetic algorithms to optimize parameterized learning rules ( Bengio et al . , 1990; Bengio et al . , 1992; Bengio et al . , 1993 ) or genetic programming to evolve less constrained learning rules ( Bengio et al . , 1994; Radi and Poli , 2003 ) , rediscovering mechanisms resembling the backpropagation of errors ( Linnainmaa , 1970; Ivakhnenko , 1971; Rumelhart et al . , 1985 ) . Recent experiments demonstrate how optimization methods can design optimization algorithms for recurrent ANNs ( Andrychowicz et al . , 2016 ) , evolve machine learning algorithms from scratch ( Real et al . , 2020 ) , and optimize parametrized learning rules in neuronal networks to achieve a desired function ( Confavreux et al . , 2020 ) . We extend these meta-learning ideas to discover free-form , yet interpretable plasticity rules for spiking neuronal networks . The discrete nature of spike-based neuronal interactions endows these networks with rich dynamical and functional properties ( e . g . , Dold et al . , 2019; Jordan et al . , 2019; Keup et al . , 2020 ) . In addition , with the advent of non-von Neumann computing systems based on spiking neuronal networks with online learning capabilities ( Moradi et al . , 2017; Davies et al . , 2018; Billaudelle et al . , 2019 ) , efficient learning algorithms for spiking systems become increasingly relevant for non-conventional computing . Here , we employ genetic programming ( Figure 1C , D; Koza , 2010 ) as a search algorithm for two main reasons . First , genetic programming can operate on analytically tractable mathematical expressions describing synaptic weight changes that are interpretable . Second , an evolutionary search does not need to compute gradients in the search space , thereby circumventing the need to estimate a gradient in non-differentiable systems . We successfully apply our approach , which we refer to as ‘evolving-to-learn’ ( E2L ) , to three different learning paradigms for spiking neuronal networks: reward-driven , error-driven , and correlation-driven learning . For the reward-driven task , our approach discovers new plasticity rules with efficient reward baselines that perform competively and even outperform previously suggested methods . The analytic form of the resulting expressions suggests experimental approaches that would allow us to distinguish between them . In the error-driven learning scenario , the evolutionary search discovers a variety of solutions which , with appropriate analysis of the corresponding expressions , can be shown to effectively implement stochastic gradient descent . Finally , in the correlation-driven task , our method generates a variety of STDP kernels and associated homeostatic mechanisms that lead to similar network-level behavior . This sheds new light onto the observed variability of synaptic plasticity and thus suggests a reevaluation of the reported variety in experimentally measured STDP curves with respect to their possible functional equivalence . Our results demonstrate the significant potential of automated procedures in the search for plasticity rules in spiking neuronal networks , analogous to the transition from hand-designed to learned features that lies at the heart of modern machine learning . We introduce the following recipe to search for biophysically plausible plasticity rules in spiking neuronal networks . First , we determine a task family of interest and an associated experimental setup which includes specification of the network architecture , for example , neuron types and connectivity , as well as stimulation protocols or training data sets . Crucially , this step involves defining a fitness function to guide the evolutionary search towards promising regions of the search space . It assigns high fitness to those individuals , that is , learning rules , that solve the task well and low fitness to others . The fitness function may additionally contain constraints implied by experimental data or arising from computational considerations . We determine each individual’s fitness on various examples from the given task family , for example , different input spike train realizations , to discover plasticity rules that generalize well ( Chalmers , 1991; Soltoggio et al . , 2018 ) . Finally , we specify the neuronal variables available to the plasticity rule , such as low-pass-filtered traces of pre- and postsynaptic spiking activity or neuromodulator concentrations . This choice is guided by biophysical considerations , for example , which quantities are locally available at a synapse , as well as by the task family , for example , whether reward or error signals are provided by the environment . We write the plasticity rule in the general form Δ⁢w=η⁢f⁢ ( … ) , where η is a fixed learning rate , and employ an evolutionary search to discover functions f that lead to high fitness . We propose to use genetic programming ( GP ) as an evolutionary algorithm to discover plasticity rules in spiking neuronal networks . GP applies mutations and selection pressure to an initially random population of computer programs to artificially evolve algorithms with desired behaviors ( e . g . , Koza , 1992 ) . Here , we consider the evolution of mathematical expressions . We employ a specific form of GP , Cartesian genetic programming ( CGP; e . g . , Miller and Thomson , 2000; Miller , 2011 ) , that uses an indexed graph representation of programs . The genotype of an individual is a two-dimensional Cartesian graph ( Figure 2A , top ) . Over the course of an evolutionary run , this graph has a fixed number of input , output , and internal nodes . The operation of each internal node is fully described by a single function gene and a fixed number of input genes . A function table maps function genes to mathematical operations ( Figure 2A , bottom ) , while input genes determine from where this node receives data . A given genotype is decoded into a corresponding computational graph ( the phenotype , Figure 2B ) which defines a function f . During the evolutionary run , mutations of the genotype alter connectivity and node operations , which can modify the encoded function ( Figure 2C ) . The indirect encoding of the computational graph via the genotype supports variable-length phenotypes , since some internal nodes may not be used to compute the output ( Figure 2B ) . The size of the genotype , in contrast , is fixed , thereby restricting the maximal size of the computational graph and ensuring compact , interpretable mathematical expressions . Furthermore , the separation into genotype and phenotype allows the buildup of ‘silent mutations’ , that is , mutations in the genotype that do not alter the phenotype . These silent mutations can lead to more efficient search as they can accumulate and in combination lead to an increase in fitness once affecting the phenotype ( Miller and Thomson , 2000 ) . A μ+λ evolution strategy ( Beyer and Schwefel , 2002 ) drives evolution by creating the next generation of individuals from the current one via tournament selection , mutation and selection of the fittest individuals ( see section Evolutionary algorithm ) . Prior to starting the search , we choose the mathematical operations that can appear in the plasticity rule and other ( hyper ) parameters of the Cartesian graph and evolutionary algorithm . For simplicity , we consider a restricted set of mathematical operations and additionally make use of nodes with constant output . After the search has completed , for example , by reaching a target fitness value or a maximal number of generations , we analyze the discovered set of solutions . In the following , we describe the results of three experiments following the recipe outlined above . We consider a simple reinforcement learning task for spiking neurons . The experiment can be succinctly described as follows: N inputs project to a single readout modeled by a leaky integrator neuron with exponential postsynaptic currents and stochastic spike generation ( for details see section Reward-driven learning task ) . We generate a finite number M of frozen-Poisson-noise patterns of duration T and assign each of these randomly to one of two classes . The output neuron should learn to classify each of these spatio-temporal input patterns into the corresponding class using a spike/no-spike code ( Figure 3A , B ) . The fitness ℱ⁢ ( f ) of an individual encoding the function f is measured by the mean reward per trial averaged over a certain number of experiments nexp , each consisting of n classification trials ( 1 ) ℱ⁢ ( f ) :=1nexp⁢∑k=1nexpRk⁢ ( f ) , where Rk⁢ ( f ) :=1n⁢∑i=1nRk , i⁢ ( f ) is the mean reward per trial obtained in experiment k . The reward Rk , i is the reward obtained in the i th trial of experiment k . It is one if the classification is correct and -1 otherwise . In the following , we drop the subscripts from Rk , i where its meaning is clear from context . Each experiment contains different realizations of frozen-noise input spike trains with associated class labels . Previous work on reward-driven learning ( Williams , 1992 ) has demonstrated that in policy-gradient-based approaches ( e . g . , Sutton and Barto , 2018 ) , subtracting a so called ‘reward baseline’ from the received reward can improve the convergence properties by adjusting the magnitude of weight updates . However , choosing a good reward baseline is notoriously difficult ( Williams , 1988; Dayan , 1991; Weaver and Tao , 2001 ) . For example , in a model for reinforcement learning in spiking neurons , Vasilaki et al . , 2009 suggest the expected positive reward as a suitable baseline . Here , we consider plasticity rules which , besides immediate rewards , have access to expected rewards . These expectations are obtained as moving averages over a number of consecutive trials ( here: 100 trials , i . e . , 50 s ) during one experiment and can either be estimated jointly ( R¯∈[-1 , 1] ) or separately for positive ( R¯+∈[0 , 1] ) and negative ( R¯-∈[-1 , 0] ) rewards , with R¯=R¯++R¯- ( for details , see section Reward-driven learning task ) . In the former case , the plasticity rule takes the general form ( 2 ) Δ⁢wj=η⁢f⁢ ( R , Ejr⁢ ( T ) , R¯ ) , while for seperately estimated positive and negative rewards it takes the form ( 3 ) Δ⁢wj=η⁢f⁢ ( R , Ejr⁢ ( T ) , R¯+ , R¯- ) . In both cases , η is a fixed learning rate and Ejr⁢ ( t ) is an eligibility trace that contains contributions from the spiking activity of pre- and post-synaptic neurons which is updated over the course of a single trial ( for details see section Reward-driven learning task ) . The eligibility trace arises as a natural consequence of policy-gradient methods aiming to maximize the expected reward ( Williams , 1992 ) and is a common ingredient of reward-modulated plasticity rules for spiking neurons ( Vasilaki et al . , 2009; Frémaux and Gerstner , 2015 ) . It is a low-pass filter of the product of two terms: the first is positive if the neuron was more active than expected by synaptic input; this can happen because the neuronal output is stochastic , to promote exploration . The second is a low-pass filter of presynaptic activity . A simple plasticity rule derived from maximizing the expected rewards would , for example , change weights according to the product of the received reward and the eligibility trace: Δ⁢wj=R⁢Ejr . If by chance a neuron is more active than expected , and the agent receives a reward , all weights of active afferents are increased , making it more likely for the neuron to fire in the future given identical input . Reward and eligibility trace values at the end of each trial ( t=T ) are used to determine synaptic weight changes . In the following , we drop the time argument of Ejr for simplicity . Using CGP with three ( R , Ejr , R¯ ) , or four inputs ( R , Ejr , R¯+ , R¯- ) , respectively , we search for plasticity rules that maximize the fitness ℱ⁢ ( f ) ( Equation 1 ) . None of the evolutionary runs with access to the expected reward ( R¯ ) make use of it as a reward baseline ( see Appendix section Full evolution data for different CGP hyperparameter choices for full data ) . Some of them discover high-performing rules identical to that suggested by Urbanczik and Senn , 2009: Δ⁢wj=η⁢ ( R-1 ) ⁢Ejr ( LR0 , ℱ=216 . 2 , Figure 3C , D ) . This rule uses a fixed base line , the maximal reward ( Rmax=1 ) , rather than the expected reward . Some runs discover a more sophisticated variant of this rule with a term that decreases the effective learning rate for negative rewards as the agent improves , that is , when the expected reward increases: Δ⁢wj=η⁢ ( 1+R⁢R¯ ) ⁢ ( R-1 ) ⁢Ejr ( LR1 , ℱ=234 . 2 , Figure 3C , D; see also Appendix section Causal and homeostatic terms over trials ) . Using this effective learning-rate , this rule achieve higher fitness than the vanilla formulation at the expense of requiring the agent to keep track of the expected reward . Using the expected reward as a baseline , for example , Δ⁢wj=η⁢ ( R-R¯ ) ⁢Ejr , is unlikely to yield high-performing solutions: an agent may get stuck in weight configurations in which it continuously receives negative rewards , yet , as it is expecting negative rewards , does not significantly change its weights . This intuition is supported by our observation that none of the high-performing plasticity rules discovered by our evolutionary search make use of such a baseline , in contrast to previous studies ( e . g . , Frémaux and Gerstner , 2015 ) . Subtracting the maximal reward , in contrast , can be interpreted as an optimistic baseline ( cf . Sutton and Barto , 2018 , Ch2 . 5 ) , which biases learning to move away from weight configurations that result in negative rewards , while maintaining weight configurations that lead to positive rewards . However , a detrimental effect of such an optimistic baseline is that learning is sparse , as it only occurs upon receiving negative rewards , an assumption at odds with behavioral evidence . In contrast , evolutionary runs with access to separate estimates of the negative and positive rewards discover plasticity rules with efficient baselines , resulting in increased fitness ( see Appendix section Full evolution data for different CGP hyperparameter choices for the full data ) . In the following , we discuss four such high-performing plasticity rules with at least 10% higher fitness than LR0 ( Figure 3D ) . We first consider the rule ( LR2 , ℱ=242 . 0 , Figure 3D ) ( 4 ) Δ⁢wj=η⁢[R- ( R¯+-R¯- ) ]⁢Ejr=η⁢ ( R-R¯abs ) ⁢Ejr , where we introduced the expected absolute reward R¯abs:=R¯+-R¯-=\|R¯+\|+\|R¯-\| , with R¯abs∈[0 , 1] . Note the difference to the expected reward R¯=R¯++R¯- . Since the absolute magnitude of positive and negative rewards is identical in the considered task , R¯abs increases in each trial , starting at zero and slowly converging to one with a time constant of 50 s . Instead of keeping track of the expected reward , the agent can thus simply optimistically increase its baseline with each trial . Behind this lies the , equally optimistic , expectation that the agent improves its performance over trials . Starting out as R⁢Ejr and converging to ( R-1 ) ⁢Ejr this rule combines efficient learning from both positive and negative rewards initially , with improved convergence towards successful weight configuration during late learning by a reward-dependent modulation of the effective learning rate ( see also Appendix section Causal and homeostatic terms over trials ) . Note that such a strategy may lead to issues with un- or re-learning . This may be overcome by the agent resetting the expected absolute reward R¯abs upon encountering a new task , similar to a ‘novelty’ signal . Furthermore , our algorithm discovers a variation of this rule ( LR3 , ℱ=256 . 0 , Figure 3D ) , which replaces η with η/ ( 1+R¯+ ) to decrease the magnitude of weight changes in regions of the weight space associated with high performance . This can improve convergence properties . We next consider the rule ( LR4 , ℱ=247 . 2 , Figure 3D ) : ( 5 ) Δ⁢wj=η⁢[ ( R-1 ) ⁢Ejr+ ( R-1 ) ⁢ ( R+2⁢R¯+ ) ] . This rule has the familiar form of LR0 and LR1 , with an additional homeostatic term . Due to the prefactors R-1 , this rule only changes weights on trials with negative reward . Initially , the expected reward R¯+ is close to zero and the homeostatic term results in potentiation of all synapses , causing more and more neurons to spike . In particular , if initial weights are chosen poorly , this can make learning more robust . As the agent improves and the expected positive rewards increases , the homeostatic term becomes negative ( see also Appendix section Causal and homeostatic terms over trials ) . In this regime , it can be interpreted as pruning all weights until only those are left that do not lead to negative rewards . This term can hence be interpreted as an adapting action baseline ( Sutton and Barto , 2018 ) . Finally , we consider the rule ( LR5 , ℱ=254 . 8 , Figure 3D ) : ( 6 ) Δ⁢wj=η⁢{ 2⁢[R- ( R¯+-R⁢R¯- ) ]⁢Ejr-[R- ( R¯+-R⁢R¯- ) ]⁢R⁢R¯-} . To analyze this seemingly complex rule , we consider the expression for trials with positive and trials with negative reward separately:R=1:Δwj+=η{2 ( 1−R¯abs ) Ejr− ( 1−R¯abs ) R¯−} , R=−1:Δwj−=η{2 ( −1−R¯ ) Ejr− ( 1+R¯ ) R¯−} . Both expressions contain a ‘causal’ term depending on pre- and postsynaptic activity via the eligibility trace , as well as , and a ‘homeostatic’ term . Aside from the constant scaling factor , the causal term of Δ⁢wj+ is identical to LR2 ( Equation 4 ) , that is , it only causes weight changes early during learning , and converges to zero for later times . Similarly , the causal term of Δ⁢wj- is initially identical to that of LR2 ( Equation 4 ) , decreasing weights for connections contributing to wrong decisions . However it increases in magnitude as the agent improves and the expected reward increases . The homeostatic term of Δ⁢wj+ is potentiating , similarly to LR4 ( Equation 5 ) : it encourages spiking by increasing all synaptic weights during early learning and decreases over time . The homeostatic term for negative rewards is also potentiating , but does not vanish for long times unless the agent is performing perfectly ( R¯-→0 ) . Over time , this plasticity rule hence reacts less and less to positive rewards , while increasing weight changes for negative rewards . The reward-modulated potentiating homeostatic mechanisms can prevent synaptic weights from vanishing and thus encourage exploration for challenging scenarios in which the agent mainly receives negative rewards . In conclusion , by making use of the separately estimated expected negative and positive rewards in precise combinations with the eligibility trace and the instantaneous reward , our evolving-to-learn approach discovered a variety of reward-based plasticity rules , many of them outperforming previously known solutions ( e . g . , Urbanczik and Senn , 2009 ) . The evolution of closed-form expressions allowed us to analyze the computational principles that allow these newly discovered rules to achieve high fitness . This analysis suggests new mechanisms for efficient learning , for example from ‘novelty’ and via reward-modulated homeostatic mechanisms . Each of these new hypotheses for reward-driven plasticity rules makes specific predictions about behavioral and neuronal signatures that potentially allow us to distinguish between them . For example LR2 , LR3 , and LR5 suggest that agents initially learn both from positive and negative rewards , while later they mainly learn from negative rewards . In particular the initial learning from positive rewards distinguishes these hypotheses from LR0 , LR1 , and LR4 , and previous work ( Urbanczik and Senn , 2009 ) . As LR2 does not make use of the , separately estimated , expected rewards , it is potentially employed in settings in which such estimates are difficult to obtain . Furthermore , LR4 and LR5 suggest that precisely regulated homeostatic mechanisms play a crucial role besides weight changes due to pre- and post-synaptic activity traces . During early learning , both rules implement potentiating homeostatic mechanisms triggered by negative rewards , likely to explore many possible weight configurations which may support successful behavior . During late learning , LR4 suggests that homeostatic changes become depressing , thus pruning unnecessary or even harmful connections . In contrast , they remain positive for LR5 , potentially avoiding catastrophic dissociation between inputs and outputs for challenging tasks . Besides experimental data from the behavioral and neuronal level , different artificial reward-learning scenarios could further further select for strengths or against weaknesses of the discovered rules . Furthermore , additional considerations , for example achieving small variance in weight updates ( Williams , 1986; Dayan , 1991 ) , may lead to particular rules being favored over others . We thus believe that our new insights into reinforcement learning are merely a forerunner of future experimental and theoretical work enabled by our approach . We next consider a supervised learning task in which a neuron receives information about how far its output is from the desired behavior , instead of just a scalar reward signal as in the previous task . The widespread success of this approach in machine learning highlights the efficacy of learning from errors compared to correlation- or reward-driven learning ( Goodfellow et al . , 2016 ) . It has therefore often been hypothesized that evolution has installed similar capabilities in biological nervous systems ( see , e . g . Marblestone et al . , 2016; Whittington and Bogacz , 2019 ) . Urbanczik and Senn , 2014 introduced an efficient , flexible , and biophysically plausible implementation of error-driven learning via multi-compartment neurons . For simplicity , we consider an equivalent formulation of this learning principle in terms of two point neurons modeled as leaky integrator neurons with exponential postsynaptic currents and stochastic spike generation . One neuron mimics the somatic compartment and provides a teaching signal to the other neuron acting as the dendritic compartment . Here , the difference between the teacher and student membrane potentials drives learning: ( 7 ) Δ⁢wj⁢ ( t ) =η⁢[v⁢ ( t ) -u⁢ ( t ) ]⁢s¯j⁢ ( t ) , where v is the teacher potential , u the student membrane potential , and η a fixed learning rate . s¯j⁢ ( t ) = ( κsj ) ⁢ ( t ) represents the the presynaptic spike train sj filtered by the synaptic kernel κ . Equation 7 can be interpreted as stochastic gradient descent on an appropriate cost function ( Urbanczik and Senn , 2014 ) and can be extended to support credit assignment in hierarchical neuronal networks ( Sacramento et al . , 2018 ) . For simplicity , we assume the student has direct access to the teacher’s membrane potential , but in principle one could also employ proxies such as firing rates as suggested in Pfister et al . , 2010; Urbanczik and Senn , 2014 . We consider a one-dimensional regression task in which we feed random Poisson spike trains into the two neurons ( Figure 4A ) . The teacher maintains fixed input weights while the student’s weights should be adapted over the course of learning such that its membrane potential follows the teacher’s ( Figure 4B , C ) . The fitness ℱ⁢ ( f ) of an individual encoding the function f is measured by the root mean-squared error between the teacher and student membrane potential over the simulation duration T , excluding the initial 10% , averaged over nexp experiments: ( 8 ) ℱ⁢ ( f ) :=1nexp⁢∑k=1nexp∫0 . 1⁢TTd⁢t⁢[vk⁢ ( t ) -uk⁢ ( t ) ]2 . Each experiment consists of different input spike trains and different teacher weights . The general form of the plasticity rule for this error-driven learning task is given by: ( 9 ) Δ⁢wj=η⁢f⁢ ( v , u , s¯j ) . Using CGP with three inputs ( v , u , s¯j ) , we search for plasticity rules that maximize the fitness ℱ⁢ ( f ) . Starting from low fitness , about half of the evolutionary runs evolve efficient plasticity rules ( Figure 4D ) closely matching the performance of the stochastic gradient descent rule of Urbanczik and Senn , 2014 . While two runs evolve exactly Equation 7 , other solutions with comparable fitness are discovered , such as ( 10 ) Δ⁢wj=η⁢ ( v-u ) ⁢s¯j⁢2⁢u-1v⁢ , and ( 11 ) Δ⁢wj=η⁢s¯j⁢ ( v+u ) ⁢v⁢ ( v-u ) -s¯jv2 . At first sight , these rules may appear more complex , but for the considered parameter regime under the assumptions v≈u;v , u≫1 , we can write them as ( see Appendix section Error-driven learning – simplification of the discovered rules ) : ( 12 ) Δ⁢wj=η⁢c1⁢ ( v-u ) ⁢s¯j+c2 , with a multiplicative constant c1=𝒪⁢ ( 1 ) and a negligible additive constant c2 . Elementary manipulations of the expressions found by CGP thus demonstrate the similarity of these superficially different rules to Equation 7 . Consequently , they can be interpreted as approximations of gradient descent . The constants generally fall into two categories: fine-tuning of the learning rate for the set of task samples encountered during evolution ( c1 ) , which could be responsible for the slight increase in performance , and factors that have negligible influence and would most likely be pruned over longer evolutionary timescales ( c2 ) . This pruning could be accelerated , for example , by imposing a penalty on the model complexity in the fitness function , thus preferentially selecting simpler solutions . In conclusion , the evolutionary search rediscovers variations of a human-designed efficient gradient-descent-based learning rule for the considered error-driven learning task . Due to the compact , interpretable representation of the plasticity rules we are able to analyze the set of high-performing solutions and thereby identify phenomenologically identical rules despite their superficial differences . We now consider a task in which neurons do not receive any feedback from the environment about their performance but instead only have access to correlations between pre- and postsynaptic activity . Specifically , we consider a scenario in which an output neuron should discover a repeating frozen-noise pattern interrupted by random background spikes using a combination of spike-timing-dependent plasticity and homeostatic mechanisms . Our experimental setup is briefly described as follows: N inputs project to a single output neuron ( Figure 5A ) . The activity of all inputs is determined by a Poisson process with a fixed rate . A frozen-noise activity pattern of duration Tpatternms is generated once and replayed every Tinter⁢ms ( Figure 5B ) while inputs are randomly spiking in between . We define the fitness ℱ⁢ ( f ) of an individual encoding the function f by the minimal average signal-to-noise ratio ( SNR ) across nexp experiments: ( 13 ) ℱ⁢ ( f ) :=mink⁡{SNRk , k∈[1 , nexp]} . The signal-to-noise ratio SNRk , following Masquelier , 2018 , is defined as the difference between the maximal free membrane potential during pattern presentation averaged over multiple presentations ( ⟨uk , i , max⟩ ) and the mean of the free membrane potential in between pattern presentations ( ⟨uk , inter⟩ ) divided by its variance ( Var⁢ ( vk , inter ) ) : ( 14 ) SNRk:=⟨uk , i , max⟩-⟨uk , inter⟩Std⁢ ( uk , inter ) . The free membrane potential is obtained in a separate simulation with frozen weights by disabling the spiking mechanism for the output neuron . This removes measurement noise in the signal-to-noise ratio arising from spiking and subsequent membrane-potential reset . Each experiment consists of different realizations of a frozen-noise pattern and background spiking . We evolve learning rules of the following general form , which includes a dependence on the current synaptic weight in line with previously suggested STDP rules ( Gütig et al . , 2003 ) : ( 15 ) ΔwjSTDP=η{fdep ( wj , Ejc ) Δtj<0ffac ( wj , Ejc ) Δtj≥0 . Here , Ejc:=e-\|Δ⁢tj\|/τ represents an eligibility trace that depends on the relative timing of post- and presynaptic spiking ( Δ⁢tj=tpost-tpre , j ) and is represented locally in each synapse ( e . g . , Morrison et al . , 2008 ) . η represents a fixed learning rate . The synaptic weight is bound such that wj∈[0 , 1] . We additionally consider weight-dependent homeostatic mechanisms triggered by pre- and postsynaptic spikes , respectively . These are implemented by additional functions of the general form: ( 16 ) Δwjhom=η{fprehom ( wj ) uponpresynapticspikefposthom ( wj ) uponpostsynapticspike Weight changes are determined jointly by Equation 15 and Equation 16 as Δ⁢wj=Δ⁢wjSTDP+Δ⁢whom . Using CGP , we search for functions fdep , ffac , fprehom , and fposthom that maximize the fitness ℱ⁢ ( fdep , ffac ) ( Equation 13 ) . As a baseline we consider a rule described by Masquelier , 2018 ( Figure 5C ) . It is a simple additive spike-timing-dependent plasticity ( STDP ) rule that replaces the depression branch of traditional STDP variants with a postsynaptically triggered constant homeostatic term whom<0 ( Kempter et al . , 1999 ) . The synaptic weight of the projection from input j changes according to ( Figure 5G ) : ( 17 ) ΔwjSTDP=η{0Δtj<0 ( anticausalinteraction ) EjcΔtj≥0 ( causalinteraction ) , with homeostatic mechanisms: ( 18 ) Δwjhom=η{0uponpresynapticspikewhomuponpostsynapticspike . To illustrate the result of synaptic plasticity following Equation 17 and Equation 18 , we consider the evolution of the membrane potential of an output neuron over the course of learning ( Figure 5C ) . While the target neuron spikes randomly at the beginning of learning , its membrane potential finally stays subthreshold in between pattern presentations and crosses the threshold reliably upon pattern presentation . After 2000 generations , half of the runs of the evolutionary algorithm discover high-fitness solutions ( Figure 5D ) . These plasticity rules lead to synaptic weight configurations which cause the neuron to reliably detect the frozen-noise pattern . From these well-performing learning rules , we pick two representative examples ( Figure 5D , E ) to analyze in detail . Learning rule 1 ( LR1 , Figure 5D ) is defined by the following equations: ( 19 ) ΔwjSTDP=η{− ( wj−1 ) EjcΔtj<0EjcΔtj≥0 , Δwjhom=η{wjuponpresyn . spike−wjuponpostsyn . spike . Learning rule 2 ( LR2 , Figure 5E ) is defined by the following equations: ( 20 ) ΔwjSTDP=η{−Ejc/wjΔtj<0 ( wjEjc ) wjΔtj≥0 , Δwjhom=η{wjuponpresyn . spike−1uponpostsyn . spike . The form of these discovered learning rules and associated homeostatic mechanisms suggests that they use distinct strategies to detect the repeated spatio-temporal pattern . LR1 causes potentiation for small time differences , regardless of whether they are causal or anticausal ( note that - ( wj-1 ) ≥0 since wj∈[0 , 1] ) . In the Hebbian spirit , this learning rule favors correlation between presynaptic and postsynaptic firing . Additionally , it potentiates synaptic weights upon presynaptic spikes , and depresses them for each postsynaptic spike . In contrast , LR2 implements a similar strategy as the learning rule of Masquelier , 2018: it potentiates synapses only for small , positive ( causal ) time differences . Additionally , however , it pronouncedly punishes anticausal interactions . Similarly to LR1 , its homeostatic component potentiates synaptic weights upon presynaptic spikes , and depresses them for each postsynaptic spike . Note how both rules reproduce important components of experimentally established STDP traces ( e . g . , Caporale and Dan , 2008 ) . Despite their differences both in the form of the STDP kernel as well as the associated homeostatic mechanisms , both rules lead to high fitness , that is , comparable system-level behavior . Unlike the classical perception of homeostatic mechanisms as merely maintaining an ideal working point of neurons ( Davis and Bezprozvanny , 2001 ) , in both discovered plasticity rules these components support the computational goal of detecting the repeated pattern . By potentiating large weights more strongly than small weights , the pre-synaptically triggered homeostatic mechanisms support the divergence of synaptic weights into strong weights , related to the repeated pattern , and weak ones , providing background input . This observation suggests that homeostatic mechanisms and STDP work hand in hand to achieve desired functional outcomes , similar to homeostatic terms in stabilized Hebbian rules ( Oja , 1982; Miller and MacKay , 1994 ) . Experimental approaches hence need to take both factors into account and variations in observed STDP curves should be reconsidered from a point of functional equivalence when paired with data on homeostatic changes . In conclusion , for the correlation-driven task , the evolutionary search discovered a wide variety of plasticity rules with associated homeostatic mechanisms supporting successful task learning , thus enabling new perspectives for learning in biological substrates . Uncovering the mechanisms of learning via synaptic plasticity is a critical step toward understanding brain ( dys ) function and building truly intelligent , adaptive machines . We introduce a novel approach to discover biophysically plausible plasticity rules in spiking neuronal networks . Our meta-learning framework uses genetic programming to search for plasticity rules by optimizing a fitness function specific to the respective task family . Our evolving-to-learn approach discovers high-performing solutions for various learning paradigms , reward-driven , error-driven , and correlation-driven learning , yielding new insights into biological learning principles . Moreover , our results from the reward-driven and correlation-driven task families demonstrate that homeostatic terms and their precise interation with plasticity play an important role in shaping network function , highlighting the importance of considering both mechanisms jointly . The experiments considered here were mainly chosen due to their simplicity and prior knowledge about corresponding plasticity rules that provided us with a high-performance reference for comparison . Additionally , in each experiment , we restricted ourselves to a constrained set of possible inputs to the plasticity rule . Here , we chose quantities which have been previously shown to be linked to synaptic plasticity in various learning paradigms , such as reward , low-pass filtered spike trains , and correlations between pre- and postsynaptic activities . This prior knowledge avoids requiring the evolutionary algorithm to rediscover these quantities but limits the search space , thus potentially excluding other efficient solutions . A key point of E2L is the compact representation of the plasticity rules . We restrict the complexity of the expressions by three considerations . First , we assume that effective descriptions of weight changes can be found that are not unique to each individual synapse . This is a common assumption in computational neuroscience and based on the observation that nature must have found a parsimonious encoding of brain structure , as not every connection in the brain can be specified in the DNA of the organism ( Zador , 2019 ) ; rather , genes encode general principles by which the neuronal networks and subnetworks are organized and reorganized ( Risi and Stanley , 2010 ) . Our approach aims at discovering such general principles for synaptic plasticity . Second , physical considerations restrict the information available to the plasticity rule to local quantities , such as pre- and post-synaptic activity traces or specific signals delivered via neuromodulators ( e . g . , Cox and Witten , 2019; Miconi et al . , 2020 ) . Third , we limit the maximal size of the expressions to keep the resulting learning rules interpretable and avoid overfitting . We explicitly want to avoid constructing an opaque system that has high task performance but does not allow us to understand how the network structure is shaped over the course of learning . Since we obtain analytically tractable expressions for the plasticity rule , we can analyze them with conventional methods , in contrast to approaches representing plasticity rules with ANNs ( e . g . , Risi and Stanley , 2010; Orchard and Wang , 2016; Bohnstingl et al . , 2019 ) , for which it is challenging to fully understand their macroscopic computation . This analysis generates intuitive understanding , facilitating communication and human-guided generalization from a set of solutions to different network architectures or task domains . In the search for plasticity rules suitable for physical implementations in biological systems , these insights are crucial as the identified plasticity mechanisms can serve as building blocks for learning rules that generalize to the actual challenges faced by biological agents . Rather than merely applying the discovered rules to different learning problems , researchers may use the analytic expressions and prior knowledge to distill general learning principles – such as the computational role of homeostasis emerging from the present work – and combine them in new ways to extrapolate beyond the task families considered in the evolutionary search . Therefore , our evolving-to-learn approach is a new addition to the toolset of the computational neuroscientist in which human intuition is paired with efficient search algorithms . Moreover , simple expressions highlight the key interactions between the local variables giving rise to plasticity , thus providing hints about the underlying biophysical processes and potentially suggesting new experimental approaches . From a different perspective , while the learning rules found in the experiments described above were all evolved from random expressions , one can also view the presented framework as a hypothesis-testing machine . Starting from a known plasticity rule , our framework would allow researchers to address questions like: assuming the learning rule would additionally have access to variable x , could this be incorporated into the weight updates such that learning would improve ? The automated procedure makes answering such questions much more efficient than a human-guided manual search . Additionally , the framework is suitable to find robust biophysically plausible approximations for complex learning rules containing quantities that might be non-local , difficult to compute , and/or hard to implement in physical substrates . In particular , multi-objective optimization is suitable to evolve a known , complex rule into simpler versions while maintaining high task performance . Similarly , one could search for modifications of general rules that are purposefully tuned to quickly learn within a specific task family , outperforming more general solutions . In each of these cases , prior knowledge about effective learning algorithms provides a starting point from which the evolutionary search can discover powerful extensions . The automated search can discover plasticity rules for a given problem that exploit implicit assumptions in the task . It therefore highlights underconstrained searches , be this due to scarcity of biological data , the simplicity of chosen tasks or the omission of critical features in the task design . For instance , without asserting equal average spike rates of background and pattern neurons in the correlation-driven task , one could discover plasticity rules that exploit the rate difference rather than the spatio-temporal structure of the input . Evolved Plastic Artificial Neural Networks ( EPANNs; e . g . , Soltoggio et al . , 2018 ) and in particular adaptive HyperNEAT ( Risi and Stanley , 2010 ) , represent an alternative approach to designing plastic neural networks . In contrast to our method , however , these approaches include the network architecture itself into the evolutionary search , alongside synaptic plasticity rules . While this can lead to high-performance solutions due to a synergy between network architecture and plasticity , this interplay has an important drawback , as in general it is difficult to tease apart the contribution of plasticity from that of network structure to high task performance ( cf . Gaier and Ha , 2019 ) . In addition , the distributed , implicit representation of plasticity rules in HyperNEAT can be difficult to interpret , which hinders a deeper understanding of the learning mechanisms . In machine-learning-oriented applications , this lack of credit assignment is less of an issue . For research into plasticity rules employed by biological systems , however , it presents a significant obstacle . Future work needs to address a general issue of any optimization method: how can we systematically counter overfitting to reveal general solutions ? A simple approach would increase the number of sample tasks during a single fitness evaluation . However , computational costs increase linearly in the number of samples . Another technique penalizes the complexity of the resulting expressions , for example , proportional to the size of the computational graph . Besides avoiding overfitting , such a penalty would automatically remove ‘null terms’ in the plasticity rules , that is , trivial subexpressions which have no influence on the expressions’ output . Since it is a priori unclear how this complexity penalty should be weighted against the original fitness measures , one should consider multi-objective optimization algorithms ( e . g . , Deb , 2001 ) . Another issue to be addressed in future work is the choice of the learning rate . Currently , this value is not part of the optimization process and all tasks assume a fixed learning rate . The analysis of the reward- and error-driven learning rules revealed that the evolutionary algorithm tried to optimize the learning rate using the variables it had access to , partly generating complex terms that that amount to a variable scaling of the learning rate . The algorithm may benefit from the inclusion of additional constants which it could , for example , use for an unmitigated , permanent scaling of the learning rate . However , the dimensionality of the search space scales exponentially in the number of operators and constants , and the feasibility of such an approach needs to be carefully evaluated . One possibility to mitigate this combinatorial explosion is to combine the evolutionary search with gradient-based optimization methods that can fine-tune constants in the expressions ( Topchy and Punch , 2001; Izzo et al . , 2017 ) . Additionally , future work may involve less preprocessed data as inputs while considering more diverse mathematical operators . In the correlation-driven task , one could for example provide the raw times of pre- and postsynaptic spiking to the graph instead of the exponential of their difference , leaving more freedom for the evolutionary search to discover creative solutions . We expect particularly interesting applications of our framework to involve more complex tasks that are challenging for contemporary algorithms , such as life-long learning , which needs to tackle the issue of catastrophic forgetting ( French , 1999 ) or learning in recurrent spiking neuronal networks . In order to yield insights into information processing in the nervous system , the design of the network architecture should be guided by known anatomical features , while the considered task families should fall within the realm of ecologically relevant problems . The evolutionary search for plasticity rules requires a large number of simulations , as each candidate solution needs to be evaluated on a sufficiently large number of samples from the task family to encourage generalization ( e . g . , Chalmers , 1991; Bengio et al . , 1992 ) . Due to silent mutations in CGP , that is , modifications of the genotype that do not alter the phenotype , we use caching methods to significantly reduce computational cost as only new solutions need to be evaluated . However , even employing such methods , the number of required simulations remains large , in the order of 103-104 per evolutionary run . For the experiments considered here , the computational costs are rather low , requiring 24-48 node hours for a few parallel runs of the evolutionary algorithms , easily within reach of a modern workstation . The total time increases linearly with the duration of a single simulation . When considering more complex tasks which would require larger networks and hence longer simulations , one possibility to limit computational costs would be to evolve scalable plasticity rules in simplified versions of the tasks and architectures . Such rules , quickly evolved , may then be applied to individual instances of the original complex tasks , mimicking the idea of ‘evolutionary hurdles’ that avoid wasting computational power on low-quality solutions ( So et al . , 2019; Real et al . , 2020 ) . A proof of concept for such an approach is the delta rule: originally in used in small-scale tasks , it has demonstrated incredible scaling potential in the context of error backpropagation . Similar observations indeed hold for evolved optimizers ( Metz et al . , 2020 ) . Neuromorphic systems – dedicated hardware specifically designed to emulate neuronal networks – provide an attractive way to speed up the evolutionary search . To serve as suitable substrates for the approach presented here , these systems should be able to emulate spiking neuronal networks in an accelerated fashion with respect to real time and provide on-chip plasticity with a flexible specification of plasticity mechanisms ( e . g . , Davies et al . , 2018; Billaudelle et al . , 2019; Mayr et al . , 2019 ) . We view the presented methods as a machinery for generating , testing , and extending hypotheses on learning in spiking neuronal networks driven by problem instances and prior knowledge and constrained by experimental evidence . We believe this approach holds significant promise to accelerate progress toward deep insights into information processing in physical systems , both biological and biologically inspired , with immanent potential for the development of powerful artificial learning machines . We use a μ+λ evolution strategy ( Beyer and Schwefel , 2002 ) to evolve a population of individuals towards high fitness . In each generation , λ new offsprings are created from μ parents via tournament selection ( e . g . , Miller and Goldberg , 1995 ) and subsequent mutation . From these μ+λ , individuals the best μ individuals are selected as parents for the next generation ( Alg . 4 . 1 ) . In this work , we use a tournament size of one and a fixed mutation probability pmutate for each gene in an offspring individual . Since in CGP crossover of individuals can lead to significant disruption of the search process due to major changes in the computational graphs ( Miller , 1999 ) , we avoid it here . In other words , new offspring are only modified by mutations . We use neutral search ( Miller and Thomson , 2000 ) , in which an offspring is preferred over a parent with equal fitness , to allow the accumulation of silent mutations that can jointly lead to an increase in fitness . As it is computationally infeasible to exhaustively evaluate an individual on all possible tasks from a task family , we evaluate individuals only on a limited number of sample tasks and aggregate the results into a scalar fitness , either by choosing the minimal result or averaging . We manually select the number of sample tasks to balance computational costs and sampling noise for each task . In each generation , we use the same initial conditions to allow a meaningful comparison of results across generations . If an expression is encountered that cannot be meaningfully evaluated , such as division by zero , the corresponding individual is assigned a fitness of -∞ . HAL-CGP ( Schmidt and Jordan , 2020 , https://github . com/Happy-Algorithms-League/hal-cgp , Jordan , 2021b ) is an extensible pure Python library implementing Cartesian genetic programming to represent , mutate and evaluate populations of individuals encoding symbolic expressions targeting applications with computationally expensive fitness evaluations . It supports the translation from a CGP genotype , a two-dimensional Cartesian graph , into the corresponding phenotype , a computational graph implementing a particular mathematical expression . These computational graphs can be exported as pure Python functions , NumPy-compatible functions ( van der Walt et al . , 2011 ) , SymPy expressions ( Meurer et al . , 2017 ) or PyTorch modules ( Paszke et al . , 2019 ) . Users define the structure of the two-dimensional graph from which the computational graph is generated . This includes the number of inputs , columns , rows , and outputs , as well as the computational primitives , that is , mathematical operators and constants , that compose the mathematical expressions . Due to the modular design of the library , users can easily implement new operators to be used as primitives . It supports advanced algorithmic features , such as shuffling the genotype of an individual without modifying its phenotype to introduce additional drift over plateus in the search space and hence lead to better exploration ( Goldman and Punch , 2014 ) . The library implements a μ+λ evolution strategy to evolve individuals ( see section Evolutionary algorithm ) . Users need to specify hyperparameters for the evolutionary algorithm , such as the size of parent and offspring populations and the maximal number of generations . To avoid reevaluating phenotypes that have been previously evaluated , the library provides a mechanism for caching results on disk . Exploiting the wide availability of multi-core architectures , the library can parallelize the evaluation of all individuals in a single generation via separate processes . Spiking neuronal network simulations are based on the 2 . 16 . 0 release of the NEST simulator ( Gewaltig and Diesmann , 2007 , https://github . com/nest/nest-simulator; Eppler , 2021 commit 3c6f0f3 ) . NEST is an open-source simulator for spiking neuronal networks with a focus on large networks with simple neuron models . The computationally intensive propagation of network dynamics is implemented in C++ while the network model can be specified using a Python API ( PyNEST; Eppler et al . , 2008; Zaytsev and Morrison , 2014 ) . NEST profits from modern multi-core and multi-node systems by combining local parallelization with OpenMP threads and inter-node communication via the Message Passing Interface ( MPI ) ( Jordan et al . , 2018 ) . The standard distribution offers a variety of established neuron and plastic synapse models , including variants of spike-timing-dependent plasticity , reward-modulated plasticity and structural plasticity . New models can be implemented via a domain-specific language ( Plotnikov et al . , 2016 ) or custom C++ code . For the purpose of this study , we implemented a reward-driven ( Urbanczik and Senn , 2009 ) and an error-driven learning rule ( Equation 7; Urbanczik and Senn , 2014 ) , as well as a homeostatic STDP rule ( Equation 17; Masquelier , 2018 ) via custom C++ code . Due to the specific implementation of spike delivery in NEST , we introduce a constant in the STDP rule that is added at each potentiation call instead of using a separate depression term . To support arbitrary mathematical expressions in the error-driven ( Equation 9 ) and correlation-driven synapse models ( Equation 15 ) , we additionally implemented variants in which the weight update can be specified via SymPy compatible strings ( Meurer et al . , 2017 ) that are parsed by SymEngine ( https://github . com/symengine/symengine; SymEngine Contributors , 2021 ) a C++ library for symbolic computation . All custom synapse models and necessary kernel patches are available as NEST modules in the repository accompanying this study ( https://github . com/Happy-Algorithms-League/e2l-cgp-snn ( copy archived at swh:1:rev:2f370ba6ec46a46cf959afcc6c1c1051394cd02a ) , Jordan , 2021a ) . Experiments were performed on JUWELS ( Jülich Wizard for European Leadership Science ) , an HPC system at the Jülich Research Centre , Jülich , Germany , with 12 Petaflop peak performance . The system contains 2271 general-purpose compute nodes , each equipped with two Intel Xeon Platinum 8168 processors ( 2×24 cores ) and 12×8 GB main memory . Compute nodes are connected via an EDR-Infiniband fat-tree network and run CentOS 7 . Additional experiments were performed on the multicore partition of Piz Daint , an HPC system at the Swiss National Supercomputing Centre , Lugano , Switzerland with 1 . 731 Petaflops peak performance . The system contains 1813 general-purpose compute nodes , each equipped with two Intel Xeon E5-2695 v4 processors ( 2×18 cores ) and 64 GB main memory . Compute nodes are connected via Cray Aries routing and communications ASIC with Dragonfly network topology and run Cray Linux Environment ( CLE ) . Each experiment employed a single compute node . We consider a reinforcement learning task for spiking neurons inspired by Urbanczik and Senn , 2009 . Spiking activity of the output neuron is generated by an inhomogeneous Poisson process with instantaneous rate ϕ determined by its membrane potential u ( Pfister et al . , 2006; Urbanczik and Senn , 2009 ) : ( 21 ) ϕ⁢ ( u ) :=ρ⁢eu-uthΔ⁢u . Here , ρ is the firing rate at threshold , uth the threshold potential , and Δ⁢u a parameter governing the noise amplitude . In contrast to Urbanczik and Senn , 2009 , we consider an instantaneous reset of the membrane potential after a spike instead of an hyperpolarization kernel . The output neuron receives spike trains from sources randomly drawn from an input population of size N with randomly initialized weights ( winitial∼𝒩⁢ ( 0 , σw ) ) . Before each pattern presentation , the output neurons membrane potential and synaptic currents are reset . The eligibility trace in every synapse is updated in continuous time according to the following differential equation ( Urbanczik and Senn , 2009; Frémaux and Gerstner , 2015 ) : ( 22 ) τM⁢E˙jr=-Ejr+1Δ⁢u⁢[∑s∈yδ⁢ ( t-s ) -ϕ⁢ ( u⁢ ( t ) ) ]⁢s¯j⁢ ( t ) , where τM governs the time scale of the eligibility trace and has a similar role as the decay parameter γ in policy-gradient methods ( Sutton and Barto , 2018 ) , Δ⁢u is a parameter of the postsynaptic cell governing its noise amplitude , y represents the postsynaptic spike train , and s¯j⁢ ( t ) = ( κsj ) ⁢ ( t ) the presynaptic spike train sj filtered by the synaptic kernel κ . The learning rate η was manually tuned to obtain high performance with the one suggested by Urbanczik and Senn , 2009 . Expected positive and negative rewards in trial i are separately calculated as moving averages over previous trials ( Vasilaki et al . , 2009 ) : ( 23 ) R¯i+⁣/-= ( 1-1mr ) ⁢R¯i-1+⁣/-+1mr⁢[Ri-1]+/- , where mr determines the number of relevant previous trials and [x]+:=max⁢ ( 0 , x ) , [x]-:=min⁢ ( 0 , x ) . Note that R¯+∈[0 , 1] and R¯-∈[-1 , 0] , since R∈{-1 , 1} . We obtain the average reward as a sum of these separate estimates R¯=R¯++R¯-;R¯∈[-1 , 1] , while the expected absolute reward is determined by their difference R¯abs=R¯+-R¯-;R¯abs∈[0 , 1] . We consider an error-driven learning task for spiking neurons inspired by Urbanczik and Senn , 2014 . N Poisson inputs with constant rates ( ri∼𝒰⁢[rmin , rmax] , i∈[1 , N] ) project to a teacher neuron and , with the same connectivity pattern , to a student neuron . As in section Reward-driven learning task , spiking activity of the output neuron is generated by an inhomogeneous Poisson process . In contrast to section Reward-driven learning task , the membrane potential is not reset after spike emission . Fixed synaptic weights from the inputs to the teacher are uniformly sampled from the interval [wmin , wmax] , while weights to the student are all initialized to a fixed value w0 . In each trial we randomly shift all teacher weights by a global value wshift to avoid a bias in the error signal that may arise if the teacher membrane potential is initially always larger or always smaller than the student membrane potential . Target potentials are read out from the teacher every δ⁢t and provided instantaneously to the student . The learning rate η was chosen via grid search on a single example task for high performance with Equation 7 . Similar to Urbanczik and Senn , 2014 , we low-pass filter weight updates with an exponential kernel with time constant τI before applying them . We consider a correlation-driven learning task for spiking neurons similar to Masquelier , 2018: a spiking neuron , modeled as a leaky integrate-and-fire neuron with delta-shaped post-synaptic currents , receives stochastic spike trains from N inputs via plastic synapses . To construct the input spike trains , we first create a frozen-noise pattern by drawing random spikes 𝒮ipattern∈[0 , Tpattern] , i∈[0 , N-1] from a Poisson process with rate ν . Neurons that fire at least once in this pattern are in the following called ‘pattern neurons’ , the remaining are called ‘background neurons’ . We alternate this frozen-noise pattern with random spike trains of length Tinter generated by a Poisson process with rate ν ( Figure 5B ) . To balance the average rates of pattern neurons and background neurons , we reduce the spike rate of pattern neurons in between patterns by a factor α . Background neurons have an average rate of νinter=ν⁢TinterTinter+Tpattern . We assume that pattern neurons spike only once during the pattern . Thus , they have an average rate of rate of ν=α⁢νinter+1Tinter+Tpattern=α⁢νinter+νpattern . Plugging in the previous expression for νinter and solving for α yields α=1-νpatternνinter . We choose the same learning rate as Masquelier , 2018 . Due to the particular implementation of STDP-like rules in NEST ( Morrison et al . , 2007 ) , we do not need to evolve multiple functions describing correlation-induced and homeostatic changes separately , but can evolve only one function for each branch of the STDP window . Terms in these functions which do not vanish for Ejc→0 are effectively implementing pre-synaptically triggered ( in the acausal branch ) and post-synaptically triggered ( in the causal branch ) homeostatic mechanisms .	Our brains are incredibly adaptive . Every day we form memories , acquire new knowledge or refine existing skills . This stands in contrast to our current computers , which typically can only perform pre-programmed actions . Our own ability to adapt is the result of a process called synaptic plasticity , in which the strength of the connections between neurons can change . To better understand brain function and build adaptive machines , researchers in neuroscience and artificial intelligence ( AI ) are modeling the underlying mechanisms . So far , most work towards this goal was guided by human intuition – that is , by the strategies scientists think are most likely to succeed . Despite the tremendous progress , this approach has two drawbacks . First , human time is limited and expensive . And second , researchers have a natural – and reasonable – tendency to incrementally improve upon existing models , rather than starting from scratch . Jordan , Schmidt et al . have now developed a new approach based on ‘evolutionary algorithms’ . These computer programs search for solutions to problems by mimicking the process of biological evolution , such as the concept of survival of the fittest . The approach exploits the increasing availability of cheap but powerful computers . Compared to its predecessors ( or indeed human brains ) , it also uses search strategies that are less biased by previous models . The evolutionary algorithms were presented with three typical learning scenarios . In the first , the computer had to spot a repeating pattern in a continuous stream of input without receiving feedback on how well it was doing . In the second scenario , the computer received virtual rewards whenever it behaved in the desired manner – an example of reinforcement learning . Finally , in the third ‘supervised learning’ scenario , the computer was told exactly how much its behavior deviated from the desired behavior . For each of these scenarios , the evolutionary algorithms were able to discover mechanisms of synaptic plasticity to solve the new task successfully . Using evolutionary algorithms to study how computers ‘learn’ will provide new insights into how brains function in health and disease . It could also pave the way for developing intelligent machines that can better adapt to the needs of their users .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2021	Evolving interpretable plasticity for spiking networks
Synaptic currents display a large degree of heterogeneity of their temporal characteristics , but the functional role of such heterogeneities remains unknown . We investigated in rat cerebellar slices synaptic currents in Unipolar Brush Cells ( UBCs ) , which generate intrinsic mossy fibers relaying vestibular inputs to the cerebellar cortex . We show that UBCs respond to sinusoidal modulations of their sensory input with heterogeneous amplitudes and phase shifts . Experiments and modeling indicate that this variability results both from the kinetics of synaptic glutamate transients and from the diversity of postsynaptic receptors . While phase inversion is produced by an mGluR2-activated outward conductance in OFF-UBCs , the phase delay of ON UBCs is caused by a late rebound current resulting from AMPAR recovery from desensitization . Granular layer network modeling indicates that phase dispersion of UBC responses generates diverse phase coding in the granule cell population , allowing climbing-fiber-driven Purkinje cell learning at arbitrary phases of the vestibular input . Sensory stimuli are encoded by populations of neurons with a diversity of spatio-temporal selectivity properties , which allow sensory systems to obtain accurate representations of such stimuli over the whole range of ecological spatial and temporal time scales . In the vestibular system , primary vestibular neurons primarily encode the rotational velocity ( semicircular canals ) and linear acceleration ( gravitational or inertial , in the otholitic organs ) of the head ( Arenz et al . , 2008; Fernandez and Goldberg , 1971; Goldberg , 2000; Goldberg and Fernandez , 1971a , 1971b; Green and Angelaki , 2010; Sadeghi et al . , 2007 ) . These vestibular inputs are in turn used to adapt body posture , generate compensatory and orienting eye movements , and modify vasomotor autonomic functions . The vestibulo-cerebellum , which is innervated by primary vestibular afferents in the vermal part ( Barmack et al . , 1993; Chabrol et al . , 2015; Gerrits et al . , 1989; Korte and Mugnaini , 1979 ) and by secondary vestibular afferents in all its subsections ( Barmack et al . , 1992; Magras and Voogd , 1985; Thunnissen et al . , 1989 ) , is well known to perform the sensory-motor integration required to orient the body in space and to generate accurate eye movements . Computational models have suggested that , in order to perform such functions , vestibular inputs should be pre-processed by filters with a wide diversity of temporal scales so that learning can take place at synapses onto Purkinje cells ( Dean et al . , 2010; Fujita , 1982 ) . A similar idea was proposed for suppression of the self-generated sensory signal in the electric fish ( Kennedy et al . , 2014; Roberts and Bell , 2000 ) . Despite the central importance of this filtering process for vestibulo-cerebellar function , it remains unclearat which stage and through which underlying mechanisms this pre-processing takes place . In vivo electrophysiological recordings during passive head rotation indicate that primary vestibular mossy fibers activity displays a high level of stereotypy with mostly in-phase firing during sinusoidal velocity modulations ( Arenz et al . , 2008 ) . Filtering may nevertheless occur upstream of granule cells , as the available in vivo recordings of granule cells show a large diversity of phase shifts in response to head rotations ( Barmack and Yakhnitsa , 2008 ) . This suggests that pre-processing occurs in the granular layer , as recently suggested in the electric fish electrosensory lobe ( Kennedy et al . , 2014 ) . Extrinsic vestibular extrinsic mossy fibers ( eMFs ) project to granule cells but also to another cell type , the unipolar brush cells ( UBCs ) . Each UBC receives a single mossy fiber input , and projects onto granule cells by forming several intrinsic mossy fiber terminals ( iMFs ) . The single mossy fiber contacting the dendritic brush of UBCs forms a giant synapse of unique morphology ( Mugnaini et al . , 1994 ) at which various types of glutamate receptors are expressed ( Jaarsma et al . , 1995 ) . Entrapment of glutamate in the synaptic cleft can lead to responses on slow time scales ( Kinney et al . , 1997; van Dorp and De Zeeuw , 2014 ) . UBCs have been divided into different subtypes by a number of studies ( Borges-Merjane and Trussell , 2015; Dino et al . , 1999; Mugnaini et al . , 1997; Nunzi et al . , 2002 ) . In particular , Borges-Merjane and Trussell ( 2015 ) showed recently that dorsal cochlear nucleus UBCs can be divided into ON UBCs , which have an excitatory response to their mossy fiber inputs mediated by both AMPA and mGluR1 receptors , and into OFF UBCs , the response of which is inhibitory and is produced by mGluR2 receptor activation . These studies still leave unanswered the question whether and how UBC pre-processing generates the diversity of tuning properties of granule cells . To answer this question , we first set out to characterize the input/output relationship of UBCs in the presence of sinusoidal MF stimulation . Our experiments show that UBCs have a wide diversity of tuning properties , due to the diversity of their synaptic responses . We then investigated the functional consequences of this diversity on the computational properties of the vestibulocerebellar circuit , using a network model . We demonstrate that UBCs , through the generation of diverse response profiles , greatly enhance the ability of Purkinje cells to learn arbitrary input/output mappings . In vivo , afferent vestibular inputs fire steadily at mean rates of 10–50 Hz at rest ( Arenz et al . , 2008; Barmack and Yakhnitsa , 2008; Dickman et al . , 1991; Fernandez and Goldberg , 1971; Goldberg , 2000; Goldberg and Fernandez , 1971a , 1971b; Green and Angelaki , 2010; Loe et al . , 1973; Sadeghi et al . , 2007; Tomko et al . , 1981 ) . We first examined the impact of a continuous MF input on the discharge rate of UBCs in acute brain slices using a typical firing rate of 26 Hz . GABAergic and glycinergic components were blocked during the experiments by perfusion of SR95531 ( 2 μM ) , CGP 55 , 845 ( 0 . 5 mM ) , and strychnine ( 1 μM ) . This simple protocol evoked three types of behavior . In a first group of cells ( OFF-UBCs , n= 25 , Figure 1A ) , the glutamatergic MF input evoked a hyperpolarization ( −9 . 5 ± 4 . 2 mV at steady-state ) . In 5 cells the hyperpolarization was preceded by a transient period of action potential firing ( 4 ± 2 spikes at 26 . 3 ± 15 . 2 Hz , duration 0 . 12 ± 0 . 06 s ) . Out of 25 OFF-UBCs , 6 fired spontaneously ( 12 . 8 ± 8 . 1 Hz ) at rest . MF stimulation silenced 5 of these UBCs and decreased firing rate by 28% in the last one . 10 . 7554/eLife . 15872 . 003Figure 1 . OFF- and ON- responses to MF stimulations in two classes of UBCs . ( A ) Current-clamp response of a OFF-UBC to a prolonged stimulation of its afferent MF at 26 Hz showing the OFF hyperpolarizing behavior . Steady stimulation is followed by a 1 Hz sinusoidal modulation of the MF stimulation rate . ( B ) Enlargement of the selected region in ( A ) showing the phase of UBC spiking relative to the MF stimulation modulation . ( C ) Fit of the OFF-UBC instantaneous firing rate vs phase relationship , obtained with the ten stimulation cycles shown in ( A ) , using an exponentiated cosine function ( see Materials and methods ) Here and in the next figures the MF stimulation frequency is depicted in black and peaks at 90° . The blue arrow represents the UBC phase shift relative to the MF stimulation . ( D ) Polar plot of the phase shift and firing frequency modulation of OFF-UBCs ( n = 20 ) obtained with 1 Hz sinusoidal modulations of the MF stimulation rate . The cell in ( A ) is plotted in red . ( E ) Distribution of the concentration factor k1/2 yielded by the exponentiated cosine fit of OFF-UBC responses in ( C ) ( n = 20 ) . ( F–J ) Same as ( A–E ) for ON-UBCs . ( K ) Typical UBC current-clamp responses to the current injections displayed on top traces , where a sinusoidal current modulation of constant amplitude is superimposed on a variable holding current . ( L ) Phase shift of the UBC responses to sinusoidal current injections shown in ( K ) as a function of the injected holding current . ( M ) k1/2 values of the UBC responses to sinusoidal current injections shown in ( K ) as a function of the injected holding current . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 00310 . 7554/eLife . 15872 . 004Figure 1—source data 1 . Numerical data corresponding to panels D , E , I , J , L , M of Figure 1 . Data in the L/M tab are raw data for each cell of the experimental set . Other numbers are averages across cells . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 004 In a second group of cells ( ON-UBCs , n= 22 , Figure 1F ) MF synaptic inputs depolarized the neuron , as expected from a glutamatergic fiber . At steady-state ON-UBCs were depolarized by 10 . 4 ± 5 . 9 mV from a resting membrane potential of −60 . 7 ± 3 . 7 mV and fired at 11 . 7 ± 9 . 2 Hz with an interspike interval CV2 of 0 . 435 ± 0 . 239 ( see Experimental Procedures ) . Overall , ON-UBCs covered a range of frequencies ( 1 – 29 . 9 Hz ) similar to that found in vivo under anesthesia ( Barmack and Yakhnitsa , 2008; Ruigrok et al . , 2011; Simpson et al . , 2005 ) . Only 1 of 22 ON UBCs fired at rest ( 1 . 7 Hz ) in the cell-attached configuration . Finally , a group of complex UBCs ( n = 10 ) displayed delayed or gradual depolarization which did not reach steady-state at the end of the MF stimulation protocol ( Figure 1—source data 1A , E ) . Similarly to OFF-UBCs , 8 of these 10 UBCs first hyperpolarized and then resumed firing at 10 . 9 ± 10 . 5 Hz with a delay from stimulation onset of 1 . 2 ± 1 . 3 s ( from 0 . 44 to 4 . 16 s ) . The slow inward current underlying the depolarization ( Figure 1—source data 1B ) was blocked by the group I mGluR antagonist JNJ16259685 ( Figure 1—source data 1C ) . In some cells the slow depolarization persisted after MF stimulation was stopped , leading to several seconds ( min 2 s , max >10 s ) of firing ( 5 . 6 ± 3 . 7 Hz , min 3 . 3; max 10 . 9 ) . These complex UBCs behave as the very low frequency integrators predicted by vestibular psychophysics ( Green and Angelaki , 2010 ) . They could nonetheless not be studied here within the timescale of our stimulation protocols and were discarded from further experiments . OFF-UBCs and ON-UBCs were further exposed to 1 Hz modulations of their MF input rate ( Figures 1A and F ) ( see Experimental Procedures ) , as seen for vestibular MFs during sinusoidal head rotations in vivo ( Arenz et al . , 2008; Fernandez and Goldberg , 1971 ) . OFF- and ON-UBCs responded to MF rate modulation with phase-locked firing ( Figures 1B and G ) , as assessed by fitting the phase distribution of firing rates with a circular normal function , f ( θ ) =rmin+ ( rmax−rmin ) ( ek2cos ( θ−ϕ ) −e−k2 ) / ( ek2−e−k2 ) , ( see Experimental Procedures ) , where rmax ( rmin ) is the maximal ( minimal ) firing rate , ϕ is the preferred phase shift and k is a measure of concentration of the spike phases ( n = 47 ) ( Figures 1C and H ) . A striking feature of the UBC responses was the broad diversity of their phase shift ( Figures 1D and I ) relative to the peak of the MF stimulation ( Figures 1B and G , blue arrow ) . The phase shift of ON-UBCs was broadly distributed ( SD of 105°; Figure 1I ) and covered essentially all possible phase shifts , with a sum of phase preferences weighted by the firing modulation strength of 140° . OFF-UBCs fired within a narrower range of phases ( 274 ± 28°; Figure 1D ) and had a significantly higher phase lag than ON-UBCs ( p<0 . 001; Figure 1J ) . During 1 Hz modulation 4 OFF-UBCs were silenced at all phases . The firing modulation amplitude of OFF and ON cells was similar ( 12 ± 8 and 12 ± 23 Hz; respectively p=0 . 353 ) but the dispersion of action potentials along the modulation cycle differed between these two populations . We obtained the spike phase concentration factor k from the fit to the phase histogram . OFF-UBCs fired in short bouts ( Figure 1E; k = 2 . 2 ± 1 . 2; 16 out of 18 cells with k>1 ) . In contrast most ON-UBCs displayed a near-sinusoidal behavior ( Figure 1J; k = 0 . 9 ± 1 . 5; 15 out of 22 cells with k<1; p=0 . 0006 ) , with only 7 cells displaying somewhat clustered responses ( k>1; mean 2 . 2 ) . Thus , OFF and ON UBCs respond to time-modulated MF inputs with different phase-modulated firing behaviors and non-linearity . UBCs are endowed with a complex set of active membrane conductances which may influence their phase response ( Afshari et al . , 2004; Diana et al . , 2007; Locatelli et al . , 2013; Russo et al . , 2007 ) . To quantify the effects of these intrinsic conductances we recorded the response of UBCs to sinusoidal current injections at 1 Hz ( ± 30 pA ) ( Figure 1K ) , while a holding current bias was varied to mimic the tonic drive occurring in ON and OFF UBCs . We found that all UBCs ( n = 21 ) responded with a modest phase advance of 9 . 9 ± 4 . 1° ( Figure 1L ) , independently of the holding current bias , up to the sodium spike inactivation threshold ( 200 ± 85 pA ) . The sharpness of the UBC response was strongly regulated by the holding current , with hyperpolarized cells firing spikes concentrated at the peak of the sinusoidal current injection ( Figure 1M ) . This effect is likely to account for the difference of clustering factor between ON and OFF UBCs ( Figures 1E and J ) . Hence , the strong heterogeneity in UBC phase shifts in response to MF stimulation is not due to intrinsic conductances . We therefore turned our attention to synaptic conductances to explain this heterogeneity . A subpopulation of UBCs is known to express group II mGluRs ( mGluR2 ) ( Borges-Merjane and Trussell , 2015; Jaarsma et al . , 1998; Rousseau et al . , 2012 ) which are positively coupled to a GIRK2 containing inward rectifying potassium channel ( Knoflach and Kemp , 1998; Rousseau et al . , 2012; Russo et al . , 2007 ) . Synaptic currents evoked by trains of MF stimulations were recorded from OFF-UBCs at a holding voltage of −60 mV ( n = 25; Figure 2 ) . All OFF-UBCs displayed a slow outward post synaptic current ( PSC ) . This current was blocked by the mGluR2 selective blocker LY341495 ( 1 μM , n = 25 , residual amplitude after block −5 . 3 ± 10 . 1 pA; Figure 2A ) . mGluR2 block revealed a NBQX-sensitive inward current in 48% of OFF-UBCs ( n = 12 ) ( peak 35 . 2 ± 28 . 7 pA; NBQX block of the total charge 95 ± 9%; Figure 2B ) . These AMPA currents did often show a sharp peak at the onset of stimulation , which could result in a net inward current , but the steady-state current was always largely dominated by the outward mGluR2 components . The steady-state amplitude of mGluR2 PSC increased with the stimulation frequency from 10 Hz ( 41 . 5 ± 26 . 9 pA ) to 100 Hz ( 129 . 9 ± 43 . 4 pA; n = 14 cells , p<0 . 001 ) ( Figures 2C and D ) . 10 . 7554/eLife . 15872 . 005Figure 2 . mGluR2 synaptic currents govern phase inversion in OFF-UBCs . ( A ) Synaptic currents evoked by a train of MF stimulations ( 500 ms at 50 Hz ) in an OFF-UBC in control conditions ( ctr; black ) , after mGluR2 block by LY341495 ( 0 . 5 µM , gray ) , and after block of AMPAR by NBQX ( 1 μM , light gray ) . Stimulation train is shown on top . ( B ) Quantification of the isolated AMPA component in OFF-UBCs after mGluR2 block . ( C ) Steady-state mGluR2 currents evoked by 2s stimulations at 10 , 20 , 50 , and 100 Hz display saturation . ( D ) Frequency dependence of steady-state mGlur2 synaptic currents . Average is in black . ( E ) The decay time constant of the EPSC at the offset of 500 ms trains of stimulation at 50 Hz correlates with the phase shift recorded in response to a sinusoidal MF stimulation modulation modulated at 1 Hz rate ( Figure 1 ) . The black line represents the linear regression . ( F ) The rise time constant of the outward currents evoked by 500 ms stimulations at 50 Hz correlates with the spikes concentration factor k1/2 obtained in Figure 1 . ( G ) OFF-UBC responses to sinusoidal MF stimulation rate modulations at 0 . 3 , 1 , and 3 Hz . Stimulation protocols are shown on top and one period of recording below . ( H ) Fits of the instantaneous firing rate vs phase relationship of the cell shown in ( G ) . ( I ) Firing frequency amplitude modulation and phase shifts of 25 OFF-UBCs in response to sinusoidal MF stimulation rate modulations at 0 . 3 , 1 , and 3 Hz . The cell displayed in ( G ) and ( H ) is plotted in red . Average ± SD are indicated for each modulation frequency . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 00510 . 7554/eLife . 15872 . 006Figure 2—source data 1 . Numerical data corresponding to panels B , D , E , F , I of Figure 2 . Each column of the D tab represents raw data for a cell of the experimental set . Other numbers are averages across cells . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 006 mGluR2 currents displayed highly variable activation and decay kinetics ( 20% to 80% rise time: 259 ± 95 ms; weighted decay time constant 306 ± 149 ms ) . OFF-UBC fired at the end of the period of low MF activity during 1 Hz modulations and their phase shift was positively correlated with the decay time constant of the mGluR2 current recorded in voltage-clamp ( r = 0 . 621 , n = 15 , p<0 . 05 ) ( Figure 2E ) . The total number of spikes per cycle was positively correlated with the amplitude of mGluR2 hyperpolarization ( r = 0 . 627 , n = 18 , p<0 . 01 ) , suggesting that rebound-promoting intrinsic conductances like IH or IT ( Diana et al . , 2007; Russo et al . , 2007 ) , amplify OFF-UBC responses . The spike burst was interrupted by the rise of the mGluR2 current ( see Figure 2A ) , as shown by the negative correlation of the 20 to 80% mGluR2 rise time and concentration factor k ( 0 . 668 , n = 18 , p<0 . 01 ) . To characterize the filtering properties of OFF-UBCs , we applied sinusoidal modulations of the MF stimulation rate at frequencies of 0 . 3 Hz , 1 Hz and 3 Hz to the same cells ( Figure 2G ) . Peak and trough firing frequencies of MF stimulations were adjusted to mimic constant amplitude rotations at variable angular frequencies ( see Experimental Procedures ) . As a result , the peak MF firing frequency increased from ~40 Hz at 0 . 3 Hz modulation to ~156 Hz at 3 Hz modulation . The amplitude of OFF-UBC responses increased significantly from 0 . 3 Hz to 1 Hz and dropped at 3 Hz ( p=0 . 0003; Friedman’s test; Figure 2H and I ) . Consistently with the increase of PSC amplitude at higher MF firing frequencies ( Figure 2D ) , 20% and 68% of the OFF-UBCs were completely silenced during 1 Hz and 3 Hz modulationsrespectively ( Figure 2H and 2I ) . Phase shifts increased significantly with the modulation frequency ( 193 ± 46° at 0 . 3 Hz; 274 ± 28° at 1 Hz; 367 ± 29° at 3 Hz; p<0 . 0003; Friedman’s test; Figure 2I ) . These results suggest that OFF-UBCs , signal the end of a movement in the non-preferred direction of their input MF by a brief high-frequency burst of spikes . MF EPSCs mediated by AMPARs have previously been recorded from UBCs ( Kinney et al . , 1997; Rossi et al . , 1995; van Dorp and De Zeeuw , 2014 ) . These AMPA EPSCs are characterized by a prominent slow component produced by the entrapment of glutamate in the cleft of the giant MF synapse onto UBCs ( Kinney et al . , 1997; van Dorp and De Zeeuw , 2014 ) . These slow EPSCs are likely to mediate the ON behavior and have been proposed to perform some kind of integrative operation on MF activity ( Kinney et al . , 1997; van Dorp and De Zeeuw , 2014 ) . Synaptic currents were evoked by trains of MF stimulations ( 0 . 5 s at 50 Hz ) to approximate the peak rate and duration of 1 Hz sinusoidal modulation . In all ON-UBCs tested EPSCs were blocked by the competitive AMPAR/kainate antagonist NBQX ( 1 or 5 μM; block 92 . 6 ± 8 . 8% of the charge; n = 10; Figure 3A ) . The time course of the compound AMPA EPSC evoked by MF stimulation trains varied widely between cells ( n = 31; Figures 3A , D ) , displaying classical transient currents ( Figure 3B ) as well as slow sustained components . The amplitude of the fast EPSC evoked by the first stimulation ranged from 5 to 210 pA ( 71 ± 66 pA; n = 21 ) and was inversely correlated to its decay time course ( 2 . 1 ± 1 . 5 ms ) ( r = 0 . 737; p<0 . 001; n = 21; Figure 2—source data 1A ) . Presynaptic MFs are able to sustain release at 50 Hz without depression ( Saviane and Silver , 2006 ) . Nevertheless the amplitude of fast EPSCs decreased dramatically ( by 67 . 4 ± 35 . 8% of the first peak at steady-state , n = 21; Figure 3B ) , in a way which was inversely correlated to the decay time course of the first fast EPSC ( r = 0 . 823; p<0 . 001; n = 21; Figure 3C ) . Transient EPSCs were always accompanied by a low level of sustained steady AMPAR current ( −15 . 9 ± 8 . 4 pA before the 4th stimulation; Figures 3A and B ) , indicative of the prolonged presence of glutamate in the synaptic cleft ( Nielsen et al . , 2004 ) . 10 . 7554/eLife . 15872 . 007Figure 3 . AMPAR activation dominates ON-UBC synaptic responses and presents three components . ( A ) Synaptic currents evoked by a train of MF stimulations of 500 ms at 50 Hz in an ON-UBC before and after perfusion of NBQX ( 1 μM ) . Note the transient early EPSC , followed by a slow buildup , and the rebound current at stimulation offset . The stimulation train is shown on top . ( B ) A sustained steady component persists after desensitization of the phasic response during a train of stimuli . The inset shows a superimposition of the fast EPSCs triggered by the first ( pink ) and the last ( gray ) stimulations . The exponential fit of the first EPSCs decay is displayed in purple . ( C ) Desensitization of the transient EPSCs at the end of a 500 ms stimulation at 50 Hz correlates to the τdecay of the transient EPSCs after the first stimulation ( r = 0 . 823; p<0 . 001; n = 21 ) . ( D ) Examples of slow EPSC buildup in two ON-UBCs . ( E ) The slow EPSCs buildup is not correlated to the first transient EPSC amplitude ( n = 20 ) . ( F ) Analysis of the anomalous rebound at the end of the stimulation . The last 20% of amplitude decay ( red arrowheads ) are fitted by an exponential function forced at the last point of stimulation ( red curve ) . The isolated rebound current obtained by subtraction of the exponential fit function is displayed in green in the inset . Gray area: charge of rebound current . ( G ) Correlation of the rebound current amplitude with its rise time ( r = 0 . 583 , p = , n = 23 ) . ( H ) Correlation of the rebound current amplitude with the amplitude of the transient EPSC ( r = 0 . 611 , n = 21 ) ( I ) Correlation of the rebound current amplitude with the amplitude of the slow EPSCs buildup ( r = 0 . 763 , n = 25 ) . Voltage-clamp stimulation protocol in c , d , f , g , h: 500 ms at 50 Hz . ( J ) AMPA EPSCs before ( Ctr , black ) and after 10 µM cyclothiazide ( Ctz , gray ) perfusion . Left bottom inset: isolated rebound current obtained as in ( F ) in Ctr and Ctz . Right inset: transient EPSCs at the first stimulation in Ctr and Ctz conditions . MF stimulations are shown on top . Voltage-clamp stimulation protocol: 200 ms at 50 Hz . ( K ) EPSCs charge and rebound amplitude in control condition ( Ctr , black ) and after 10 µM cyclothiazide perfusion ( Ctz , gray; n = 11 ) . Voltage-clamp stimulation protocol in ( C ) , ( D ) , ( F ) , ( G ) , ( H ) , ( I ) : 500 ms at 50 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 00710 . 7554/eLife . 15872 . 008Figure 3—source data 1 . Numerical data corresponding to panels C , E , G , H , K of Figure 3 . Each line is the result for a cell of the experimental set . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 008 In 67% of UBCs , an additional slow component developed during the train , a further element suggesting glutamate concentration buildup in the synaptic cleft ( Kinney et al . , 1997 ) . This slow EPSC buildup reached a steady-state ( 64 ± 35 pA; Figure 3D ) with variable time course ( τ = 0 . 347 ± 0 . 244 s at 50 Hz ) . The amplitude of the transient currents and of the buildup currents were not correlated ( p=0 . 12; Figure 3E ) . The presence of two independent excitatory components suggests that glutamate buildup recruits independent AMPA receptor populations in the two phases ( fast and slow ) of the synaptic responses . Overall , these data are consistent with the idea that AMPA receptors present at MF to UBC synapses are exposed to glutamate concentration transients of highly variable amplitude and duration which spatio-temporal profiles shape MF EPSCs kinetics through transient activation , steady-state activation and desensitization . In 87% of all AMPAR expressing UBCs ( n = 27 out of 31 cells ) , an anomalous rebound current developed at the end of the stimulation ( Figure 3A and B ) . The 4 remaining cells only displayed slow buildup EPSCs which decayed exponentially with a time constant of 604 ± 361 ms ( Figure 3D , cell1 ) . Rebound and decay are temporally overlapping processes . The anomalous rebound current was thus isolated by subtraction of the interpolated fitted exponential decay at the offset of the compound EPSC ( see Experimental Procedures; Figure 3F ) . The amplitude of the rebound current ( 27 . 7 ± 16 . 7 pA ) was correlated to its rise time constant ( r = 0 . 583 , p<0 . 01 , n = 23; Figure 3G ) , which varied widely from cell to cell ( 54 ± 24 ms; min = 25; max = 99; n = 23 ) . The amplitude of the rebound current was also correlated with both the amplitude of the transient EPSC integral ( r = 0 . 611 , p<0 . 01 , n = 20; Figure 3H ) and with the amplitude of the slow buildup EPSC ( r = 0 . 763 , p<0 . 001 , n = 25; Figure 3I ) , suggesting that both fast and buildup components of transmission contribute to the genesis of the rebound . Cyclothiazide ( 10–20 µM ) , a modulator of AMPARs known to suppress desensitization , potentiated the charge during stimulation ( 201 ± 54% of control; n = 11 ) and suppressed the rebound current ( 27 ± 18% of control; n = 11 ) ( Figures 3J and K ) . These results support the idea that rebound is generated by AMPAR recovery from steady-state desensitization ( Kinney et al . , 1997 ) during slow glutamate clearance from the giant MF synapse onto UBCs ( Billups et al . , 2002; Rossi et al . , 1995 ) . We investigated the link between EPSC time-course and ON-UBC phase shift during sinusoidal MF stimulation . We reasoned that fast and slow synaptic charge occurring during MF trains would favor phase-advanced firing , while rebound charge would favor phase-delayed firing ( Figure 4A , C ) . Indeed , the response phase of ON-UBCs at 1 Hz was linearly correlated with the ratio between the rebound charge and the synaptic charge during the 50 Hz trains ( see Experimental Procedures ) ( slope = 0 . 1 per 100°; r = 0 . 734 , p<0 . 001 , n = 22; Figure 4D ) . Neither the charge during the train alone ( p>0 . 1; r = 0 . 213; n = 16; Figure 2—source data 1B ) , nor the onset time course of the buildup component did correlate with the phase ( p>0 . 1; r = 0 . 036; n = 14; Figure 2—source data 1C ) . 10 . 7554/eLife . 15872 . 009Figure 4 . The AMPA rebound current defines the phase shift of UBC responses to sinusoidal input modulations . ( A ) AMPA EPSCs of an ON-UBC with early phase shift . ( B ) Voltage response of the same UBC shown in A . ( C ) Circular Gaussian fit of the firing rate distributions of the cell shown in ( B ) . ( D ) Correlation of the response phase at 1 Hz to the ratio charge at the offset of stimulation / total synaptic charge ( r = 0 . 734 , p<0 . 001 , n = 22 ) . ( E ) Correlation of the time delay to maximal firing to τrise of the rebound current ( r = 0 . 87 , p<0 . 001 , n= 16 ) . ( F ) Example of an ON-UBC’s voltage response to 1 Hz sinusoidal modulations of MF stimulation ( 4 cycles shown out of 10 ) before ( Ctr ) and after 10 µM cyclothiazide ( Ctz ) perfusion . Stimulations are shown on top . ( G ) Phase shift in control condition ( black ) and in 10 µM cyclothiazide perfusion ( gray , n = 7 ) . ( H ) ON-UBC responses to sinusoidal MF stimulation rate modulations at 0 . 3 , 1 , and 3 Hz . Stimulation protocols are shown on top and one period of recording below . Circular Gaussian fits of the firing rate distributions of the same cell . ( I ) Modulated firing amplitudes and phase shifts of 25 ON-UBCs in response to sinusoidal MF stimulation rate modulations at 0 . 3 , 1 , and 3 Hz . The cell displayed in ( H ) is plotted in red . Average ± SD are indicated for each modulation frequency . Voltage-clamp stimulation protocol in ( D ) , ( E ) : 500 ms at 50 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 00910 . 7554/eLife . 15872 . 010Figure 4—source data 1 . Numerical data corresponding to panels D , E , G , I of Figure 4 . Each line is the result for a cell of the experimental set . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 010 In all delayed cells ( positive phase shift ) , the phase of UBC firing was strongly correlated with the charge of the rebound EPSC ( r = 0 . 75 , p<0 . 001 , n = 16; Figure 2—source data 1D ) . A mechanistic explanation for this result could be that larger rebounds have slower time to peak ( Figure 3g ) . Indeed we found that the onset time constant of the rebound was strongly correlated to the phase of UBC firing ( r = 0 . 87 , p<0 . 001 , n= 16 ) . When phase was converted to a time delay to maximal firing ( 2 . 78 ms deg−1 ) the regression slope was close to 0 . 33 ( r = 0 . 87 , p<0 . 001 , n= 16 Figure 4E ) indicating that maximum firing occurs near the peak of the rebound current ( 3 onset time constants ) . Cyclothiazide ( Figure 4F ) , at a concentration which suppresses the rebound current , shifted the phase of UBC firing ( 143 ± 44° in control; −59 ± 36° in cyclothiazide; n = 7; p=0 . 01; Figure 4G ) towards values similar to control cells displaying small rebound ( p=0 . 9 , n = 6 cells with little rebound current , n = 7 cells in cyclothiazide ) . Hence anomalous rebound mediates phase inversion in the response to modulated MF inputs . The frequency dependence of the UBC response was then tested by implementing sinusoidal modulations of the MF stimulation rate at frequencies of 0 . 3 Hz , 1 Hz and 3 Hz ( Figure 4H ) . In contrast to OFF-UBCs the average amplitude of firing modulation increased at higher MF modulation frequencies ( 5 . 1 ± 5 . 9 Hz at 0 . 3 Hz , 11 . 0 ± 8 . 1 Hz at 1 Hz , and 14 . 0 ± 9 . 7 Hz at 3 Hz; p=0 . 0009; Friedman’s test ) . Interestingly , the frequency of maximal modulation varied from cell to cell . The population response of UBCs also showed significant frequency-dependent increase of phase lag ( 45 . 9° at 0 . 3 Hz , 108 . 9° at 1 Hz and 191 . 6° at 3 Hz; p<0 . 0001; Friedman’s test ) . and remained broadly distributed at all frequencies ( SD: 76 . 3° , 73 . 9° and 92 . 7° at 0 . 3 Hz , 1 Hz and 3 Hz respectively; Figure 4I ) . Phase dispersion of the ON-UBC responses is therefore a property preserved over a wide range of movement dynamics . Modeling of glutamate concentration profiles in the MF to UBC synaptic cleft has shown that the average frequency dependent behavior of the peak and steady-state conductance can be accounted for by a detailed Markov model of the AMPA receptor ( van Dorp and De Zeeuw , 2014 ) . To further investigate the role of synaptic glutamate profiles in the cell to cell variability of UBC responses we built a simplified model with reduced parameter space . First , we designed the minimal Markov scheme which can account for a bell-shaped steady-state concentration-response curve with non-zero saturating steady-state current , as recorded in UBCs ( Kinney et al . , 1997 ) . This model includes one close state , two open states and one desensitized state ( Figure 5A ) ( See Experimental Procedures ) . The on-rate of desensitization was adjusted to account for the depression of the fast receptor current during MF trains and the recovery rate was adjusted to yield a steady-state activation curve with a peak of 16% of the total conductance at 25 µM ( Figure 5B ) ( Experimental Procedures ) and a fractional steady state conductance at saturating glutamate concentration of 2 . 5% . This simple Markov model was compared to a three state model with similar affinity and maximal steady-state current , but displaying a monotonic steady-state concentration response curve ( Figures 5A and B ) . When submitted to slow synaptic glutamate buildup ( Figure 5C ) at increasing firing modulation frequencies , for which stable glutamate modulation amplitudes are reached ( Figure 5D , blue ) , the three state model saturated above 0 . 7 Hz due to basal glutamate buildup ( Figure 5D , green ) while the four state model showed a resonance above 1 Hz and greatly increased modulation bandwidth ( Figure 5D , red ) . This resonance was directly linked to the steady-state rebound , as attested by a 180° phase shift of the response compared to glutamate , as opposed to the three state model situation ( Figure 5E ) . Hence , the bell-shaped steady-state response of the AMPARs at UBC synapses plays a central role to encode the time integral of modulated inputs over a wide frequency bandwidth . 10 . 7554/eLife . 15872 . 011Figure 5 . Modeling the AMPAR responses of ON-UBCs . ( A ) Simple Markov models for desensitizing AMPARs with bell-shaped steady-state activation curve ( red ) and monotonous steady-state activation ( green ) . ( B ) Peak and steady-state dose-response curves for the red and green models in ( A ) . ( C ) Simulated glutamate concentration ( blue ) and AMPAR activation ( red and green ) during a 3 Hz sinusoidal stimulation of the presynaptic mossy fiber ( same parameters as for the experimental data ) . Each presynaptic action potential was set to produce a glutamate transient of 2 μM which rose in 15 ms and decayed exponentially with a t of 600 ms . ( D ) Peak to peak modulation amplitude of the glutamate concentration and fractional AMPAR activation ( colors as in C ) as a function of the MF firing rate modulation frequency . ( E ) Phase of the glutamate concentration and AMPAR response relative to the MF modulation . Note the phase inversion of the bell-shaped model . ( F ) Schematic depiction of glutamate concentration profiles ( in blue ) at the giant MF-UBC synapse upon multivesicular release . Simulated responses ( glutamate in blue , bell-shaped AMPAR Markov model in red ) to a train of 5 stimulation at 20 Hz are shown for the following parameters ( peak , τrise , τdecay ) of action-potential evoked glutamate transients: top inset ( 4 μM , 20 ms , 250 ms ) , middle inset ( 500 μM , 1 ms , 250 ms ) , lower inset ( 500 μM , 0 ms , 1 . 5 ms ) . ( G ) Example of the response of 4 ON-UBCs to a 500 ms train of MF stimulation at 50 Hz and of their fit ( red curve ) by the summed activation of three AMPAR populations submitted to different glutamate transients , as in ( F ) . ( H ) Response of an ON-UBC to a sinusoidal modulation of its afferent MF firing rate at 1 Hz ( black trace ) and simulated EPSC obtained with the parameters obtained through fitting in ( G ) ( middle red trace ) . The response of a simulated integrate and fire neuron to this EPSC is shown in the bottom trace . Purple trace , exponentiated cosine fit of the EPSC used in ( J ) . ( I ) Correlation between the simulated phase-dependent EPSC and the experimental phase-dependent firing response of UBCs ( n = 22 ) . Significant correlations ( p=0 . 05 ) are displayed in black bars . ( J ) Correlation between the phase shift of the simulated EPSC and the measured phase shift of the UBC response in current clamp . Line represents equality . Hollow circle: cell displayed in ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 011 To explain the phase diversity of UBCs we reasoned that AMPARs at distinct locations in the giant MF to UBC synaptic articulation will face diverse synaptic glutamate transients , thus imparting varied response dynamics . If located at the center of the synaptic area AMPARs will sense a large and prolonged glutamate transient that will build up during the train ( Figure 5F , middle trace ) . Receptors located at the periphery of the synaptic cleft will either sense low amplitude and slow glutamate buildup ( Figure 5F , upper trace ) or a fast glutamate transient if they are located in front of a synaptic release site ( Figure 5F , lower trace ) ( see Experimental Procedures for detailed model descriptions ) . We were able to fit accurately the complex time course of synaptic currents in all UBCs ( see a few examples in Figure 5G ) . The SD of the residual error was on average 2 . 55 ± 0 . 28 pA ( mean ± SEM; median = 2 . 25 ) , which was much smaller than the amplitude of the currents ( 18 . 82 ± 4 . 25 pA; median = 13 . 54 ) , and of the same order as the fluctuations in the current traces as measured in the 50 ms before stimulation ( 0 . 71 ± 0 . 08 pA; median = 0 . 63 ) . Thus , while AMPAR properties explain the presence of an anomalous rebound , the time course of synaptic glutamate transients can fully account for the diversity of synaptic responses at MF to UBC synapses . We then used the parameters obtained by fitting the EPSCs to compute the expected AMPA response to sinusoidal MF modulations ( Figure 5H ) . In most cells ( 14 out of 22 ) phase histograms of the simulated EPSCs and of the experimentally recorded UBC firing rate were significantly correlated ( p<0 . 05; t-test; Figure 5I ) . Fitting the simulated EPSC by a sinusoidal function yielded a simulated phase shift . Simulated and experimental phase shifts were linearly correlated ( R = 0 . 89 and p=2e-5; t-test and slope = 1 . 04 for cells with correlated phase histograms ) ( Figure 5J ) . In the cells displaying significant correlation between the EPSCs and firing rate , the phase difference between modeled and experimental data was approximated by a Gaussian distribution ( −24 . 6 ± 64 . 3 ) and experimental firing could be approximately reproduced using a simple integrate and fire model ( Figure 5H ) . These data confirm that the dynamics of glutamate accumulation at the MF to UBC synapse during time-modulated MF firing is the major determinant of firing phase in UBCs . We next investigated the computational consequences of the diversity of phase shifts in the UBCs responses . We thus simulated a granular layer network with 4500 Granule Cells ( GCs ) receiving inputs from 500 MFs , that were either extrinsic ( eMFs ) or from UBCs . The network structure is illustrated in Figure 6A . We compared the response of GCs to sinusoidal eMF stimulation , in the presence and absence of UBCs . In the model network , each GC receives 4 inputs ( Palay and Chan-Palay , 1974 ) . In the network without UBCs , all 4 inputs are eMFs ( Figure 6A , left ) . In the network with UBCs , each input is randomly and independently set ( Figure 6A , right ) to be from either an eMF or a UBC with equal probability ( 0 . 5 ) . 10 . 7554/eLife . 15872 . 012Figure 6 . Simulation of a Granular network with/without UBCs . ( A ) Illustration of the Granular network . Granular cells receive inputs from only eMFs without UBCs ( left ) and both eMFs and UBCs ( right ) , . UBCs have different phase shifts . ( B ) Firing rate vs phase curves of simulated eMFs ( blue ) and all recorded UBCs ( red ) at 0 . 3 Hz ( left ) , 1 Hz ( middle ) and 3 Hz ( right ) modulations . ( Ci ) Sample Granular cell outputs without UBCs in the network . ( left , upper ) Raster plots of 30 random selected GCs in one input period , and ( left , bottom ) their averaged firing rate curve; ( right ) polar plots of GC phase shifts . ( Cii ) Same as ( Ci ) , but for sample Granular cell outputs with UBCs in the network . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 012 We simulated the dynamics of the granular layer model in response to sinusoidal vestibular inputs . We used the same stimulation protocols as in our experiments , with three different frequencies ( Figure 6B , blue ) . All eMFs inputs were either in phase or in anti-phase with modeled sinusoidal movements ( Arenz et al . , 2008 ) . Parameters of model UBCs were chosen by randomly sampling parameters fitted from recorded UBCs ( Figure 6B , red ) . This led to a large diversity of phase shifts . As a consequence , a subset of UBCs fired at any phase of the oscillatory cycle . Spike trains from eMFs and iMFs were then injected into the GCs ( see Experimental Procedures for stimulation details ) . The mean conductance from eMFs/UBCs to GCs was adjusted to match the EPSC amplitudes reported in the literature ( Arenz et al . , 2008; Schwartz et al . , 2012 ) . To introduce heterogeneity in the network , all synaptic conductances were drawn randomly from a Gaussian distribution ( see Experimental Procedures ) , whereas , for the sake of simplicity , the inhibition from Golgi cells was replaced by a constant inhibitory conductance ( Billings et al . , 2014 ) ( see Discussion ) . The responses of simulated GCs are shown in Figure 6C . Without UBCs , GCs fire in phase with eMFs inputs , as shown in Figure 6Ci . The spike trains of 30 randomly selected GCs in a single trial with 1 Hz stimulation and their average firing rate over trials are shown in Figure 6Ci ( left ) . The amplitude of modulation and phase shift from the fitted GC firing rate curves were plotted in a polar plot ( Figure 6Ci ( right ) ) . We found that the preferred firing phases were in phase with the extrinsic vestibular mossy fibers , i . e . in phase or antiphase with simulated movement . When UBCs are included in the network , the response of GCs shows much more heterogeneous phase shifts . The distribution of phases in the network comprising UBCs is significantly closer to a uniform distribution compared to the network without UBCs ( KS distance between network with UBCs and uniform: 0 . 1673; between network without UBCs and uniform: 0 . 4134; both distributions are significantly different , KS test , p<1e-10 ( Figure 6Cii ) , similar to experimental in vivo data ( Barmack and Yakhnitsa , 2008 ) . Therefore , the heterogeneity of UBCs phase shifts leads to a diversity of GCs phase shifts . We then investigated the functional consequences of this GC phase shift variability on downstream Purkinje cells ( PCs ) . In particular , we asked to which degree the heterogeneity of GC phase relationships has an impact on the ability of a PC to learn arbitrary input-output relationships . We consider a single PC that receives inputs from all simulated GCs of the network . The parallel fiber to Purkinje cells synapses are endowed with a standard climber fiber ( CF ) -dependent learning rule ( see Experimental Procedures ) , such that synapses are depressed upon coincidence of presynaptic GC and CF activity , while they are potentiated when presynaptic GCs are active during CF silence ( Dean et al . , 2010 ) . We tested the ability of the PC to learn to fire preferentially at arbitrary phases of the sinusoidal eMF drive , by implementing an “error signal” in the CFs , proportional to the difference between actual and desired PC outputs . Figure 7 shows that in a network with UBCs , the PC can learn to fire at arbitrary target phases , even though eMF inputs peak only at two possible phases ( Figure 7A , top and Figure 7B ) . This ability breaks down in a network without UBCs , in which PCs can only learn to fire in phase with eMF inputs ( Figure 7A , bottom and Figure 7B ) . This is consistent with experiments showing that the response of PCs , similarly to GCs , has a large degree of heterogeneity ( Barmack and Yakhnitsa , 2008 ) . 10 . 7554/eLife . 15872 . 013Figure 7 . A downstream Purkinje cell can learn to fire at arbitrary phases in a network containing UBCs . ( A ) The responses of the Purkinje cell are tuned to the given target phase in the network with UBCs . Each target phase is indicated by each color . ( B ) The mean square error ( MSE ) between the target and the response of Purkinje cell is systematically smaller in the network with UBC ( blue ) than those without UBC ( black ) . Each color point indicates the PC’s response shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15872 . 013 We have provided here a detailed characterization of the filtering properties of UBCs in the presence of sinusoidal MF inputs . We have shown that UBCs encode these inputs with a surprising degree of heterogeneity , their responses spanning all the possible range of phase shifts , at all the frequencies tested here ( from 0 . 3 to 3 Hz ) . Our study reveals that populations of UBCs in the vestibular cerebellum behave as a bank of diverse temporal filters . By virtue of this temporal diversity , it is possible to find GCs that fire at any phase of a sinusoidal head movement , as hypothesized by adaptive filter models ( Dean et al . , 2010; Fujita , 1982; Gao et al . , 2012; Roberts and Bell , 2000 ) , and consistent with in vivo data ( Barmack and Yakhnitsa , 2008 ) . The UBC therefore extends the combinatorial notions of pattern separation ( Marr , 1969 ) and expansion recoding ( Albus , 1971 ) into the temporal domain . In turn , this diversity of temporal representations allows PCs to learn arbitrary phase response functions , as shown here by implementing a classical climbing fiber-driven learning rule that modifies the weights of parallel fiber to PC synapses . Our results are consistent with a recent study in electric fish , which showed that delayed responses necessary for predicting sensory consequences of motor commands are generated by UBCs in the eminentia granularis posterior ( EGp ) of the mormyrid electric fish , a structure similar to the granular layer of the mammalian cerebellum ( Kennedy et al . , 2014 ) . An interesting question for future research is whether the properties of UBCs in mammals are set to optimize performance of the cerebellar circuit , given the natural statistics of vestibular inputs . The differential distribution of calretinin-expressing ( type I ) and mGluR1-expressing ( Type II ) UBCs in vestibular lobules and parasagittal band within these lobules ( Consalez and Hawkes , 2012; Mugnaini et al . , 2011 ) support the idea of some functional specialization . Additionally , the presence of UBCs in the dorsal vermis and cerebellar hemispheres of higher mammals ( Dino et al . , 1999 ) suggests that the temporal filter properties described in this work may also be useful outside the context of vestibular sensory processing . We have established that the diversity of UBC responses is due to a combination of two factors: ( 1 ) the heterogeneous expression of ionotropic and metabotropic glutamate receptors ( see also Borges-Merjane and Trussell , 2015 ) and ( 2 ) the variability of AMPAR-mediated responses , due to cell-specific combinations of transient , build-up and rebound synaptic components . The heterogeneity of UBCs has already been recognized ( Mugnaini et al . , 2011 ) , particularly in relation to the expression of metabotropic glutamate receptors ( Borges-Merjane and Trussell , 2015; Kim et al . , 2012; Nunzi et al . , 2002; Rousseau et al . , 2012; Russo et al . , 2008 ) . However , their contribution to MF synaptic transmission for dynamic inputs was unknown . The paucity of ionotropic receptors at some mGluR2-expressing MF-UBC synapses is also surprising , given the size of the synaptic articulation and the number of synaptic release sites ( Mugnaini et al . , 1994 ) . As a consequence , cells expressing mGluR2 receptors ( OFF UBCs ) fire in antiphase , or in advance of the inputs . Adding further complexity to this temporal diversity , our data ( Figure 1—source data 1 ) suggest that the relevant subpopulation of mGluR1-expressing UBCs ( Borges-Merjane and Trussell , 2015; Mugnaini et al . , 2011 ) , not studied here , participate to the very slow integration of the MF inputs . The predominance of mGluR transmission at the MF synapse is not directly linked to the giant synaptic articulation , as mGluRs are exclusively found at extrasynaptic sites on the dendritic appendages of the UBC brush ( Jaarsma et al . , 1998 ) . mGluR activation , which occurs at standard synaptic articulations on spines or dendritic shafts upon repetitive stimulations , may thus implement temporal integration of population activity in many other neuronal types throughout the CNS . Cells expressing AMPARs ( ON UBCs ) fire with a large diversity of phase shifts . We show that the diversity of phase shifts in AMPAR-mediated responses could be well described by exposing AMPARs to various types of synaptic glutamate transients . To account for late-phase responses and for the wide bandwidth of UBC modulations , the AMPAR must present a bell-shaped steady-state response curve , as described experimentally ( Kinney et al . , 1997 ) . The combination of these properties and of spillover-induced desensitization at the UBC synapse generates strong rebound currents following a train of MF action potential ( see also van Dorp and de Zeeuw , 2014 ) . These properties differ dramatically from the fast reliable transmission observed at other giant synapses like the calyx of Held , where low release probability and fenestration of the apposition limit glutamate spillover and AMPAR desensitization ( Taschenberger et al . , 2002 ) . We hypothesize that a combination of factors , including the complex geometry of MF-UBCs synapse , the variety of AMPARs and auxiliary subunits ( Jackson and Nicoll , 2011 ) , and the localization of AMPARs with respect to release sites ( Jaarsma et al . , 1995 ) , could explain the variety of AMPA EPSCs profiles at MF to UBC synapses . Our results strongly support the importance of the desensitized states for the functional properties of ionotropic glutamate receptors . Other sources of differential phase responses in the vestibular system have been described in the literature . Primary sensory cells have been shown to display different levels of non-linearity in response to sinusoidal movements which appear related to cellular excitability ( Pfanzelt et al . , 2008 ) . In the vestibular nuclei , diversity of single cell properties ( Sekirnjak and du Lac , 2002 ) and synaptic dynamics of primary afferences ( Broussard , 2009 ) could in principle generate a diversity of secondary mossy fiber inputs to the vestibulo-cerebellum , though the range of phase shifts shown in those reports is fairly modest and much smaller than the one shown here . Indeed , Arenz et al . ( 2008 ) showed highly stereotyped rate modulation of mossy fiber inputs to vestibulo-cerebellar GCs during head rotation with little phase shift . In GCs , a diversity of afferent mossy fibers synaptic properties could lead to a diversity of delays ( Chabrol et al . , 2015 ) in the range of tens of milliseconds , much smaller than needed to cover the phase spectrum during head movements . In the cerebellar models of delayed eyelid conditioning ( Kalmbach et al . , 2010 ) , intrinsic dynamics of the granular layer circuit involving feed-forward and feed-back inhibitory interactions have been proposed to produce delay-specific granule cell activity patterns ( Kalmbach et al . , 2011; Medina et al . , 2000; Rössert et al . , 2015; Yamazaki and Tanaka , 2007 ) . These other sources of diversity in phase responses might allow the cerebellar circuitry to learn input-output associations with significant delays even in circuits that lack UBCs . However , we showed here that UBCs can generate much larger delays , as well as OFF-responses , within a wide range of stimulation frequencies . The variable tuning frequency of UBCs could explain the frequency-specific learning of the VOR gain modulation ( Lisberger et al . , 1983 ) , because different populations of UBCs and hence of granule cells would be optimally modulated for different rotation frequencies . Finally UBCs have the capability to generate high-frequency rebound bursts ( Diana et al . , 2007; Locatelli et al . , 2013 ) , as shown here for some of the OFF-UBCs , which may provide useful timing information , as suggested in the electro-sensory lobe of the electric fish ( Kennedy et al . , 2014; Sawtell , 2010 ) . Golgi cell inhibitory inputs to UBCs ( Dugue et al . , 2005 ) , which were pharmacologically blocked in this study , may enhance this rebound behavior ( Kennedy et al . , 2014 ) , providing for richer interactions between the granular layer dynamics and UBC intrinsic excitability . Our data support a preferential role of UBCs in tasks in which non trivial associations between inputs and outputs at different phases need to be learned , like phase-shifted VOR or high-velocity compensatory eye movements during which non-linear inverse eye dynamics have to be computed ( Green et al . , 2007; Lisberger , 2009 ) . Optogenetic suppression of vestibular UBC activity should lead to significant impairment in these motor tasks . Furthermore the higher prevalence of UBCs in the hemispheric lobules in higher mammals ( Dino et al . , 1999 ) argues for the involvement of this cell-type in complex motor and cognitive tasks involving the cerebellum . Our results fit well with a growing number of studies that have emphasized the benefits of heterogeneity for coding in neural circuits . Shamir and Sompolinsky , ( 2006 ) found that neuronal heterogeneities allow networks to overcome the drastic limits imposed by neuronal correlations on the accuracy of population codes ( see also Ecker et al . , 2011 ) . The benefits of neuronal heterogeneities have also been pointed out in several specific sensory systems . Padmanabhan and Urban , ( 2010 ) showed that the diversity of intrinsic properties of mitral cells of the olfactory bulb allows a population of such cells to double the information contained in their responses , compared with a homogeneous population . Gjorgjieva et al . , ( 2014 ) examined the benefits of `pathway splitting' in sensory systems , focusing on the emergence of ON and OFF pathways in the retina . A recent further showed the benefit of two types of OFF-cell thresholds for maximizing information transmission ( Kastner et al . , 2015 ) . Most of these studies focused on heterogeneities in static single neuron properties . Here , we have shown that the benefits of heterogeneities extend into the temporal domain , and can be exploited by a read-out that can learn arbitrary temporal input-output relationships . All experiments were performed according to the ethics rules of the Centre National de la Recherche Scientifique and protocols were approved under number 02235 . 02 of the general agreement C750520 . Wistar rats ( 22 to 30 days old ) were deeply anesthetized with isoflurane and the vermal part of the cerebellum was isolated in a cold bicarbonate based solution ( BBS ) containing ( in mM ) : 125 NaCl , 3 . 5 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 20 glucose , 1 . 6 CaCl2 , 1 MgCl2 , and 0 . 00005 minocycline , ( oxygenated with 95% O2 , 5% CO2 ) . Parasagittal slices ( 290 mm ) were immediately obtained with a vibratome ( HM 650 V; Microm ) while preparation was kept in an ice-cold cutting solution composed of ( in mM ) : 130 K-gluconate , 15 KCl , 2 EGTA , 20 HEPES , 25 glucose , 0 . 05 D- ( - ) -2-Amino-5-phosphonovaleric acid ( D-AP5; Tocris ) , and 0 . 00005 minocycline ( pH adjusted to 7 . 4 with NaOH ) . Slices recovered initially at 33°C for 30 s in a modified BBS solution ( in mM ) : 225 D-mannitol , 2 . 34 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 25 Glucose , 0 . 513 CaCl2 , 7 . 671 MgCl2 , 0 . 05 D-AP5 , and 0 . 00005 minocycline ( oxygenated with 95% O2 , 5% CO2 ) . Then slices were incubated in oxygenated BBS at 33°C for up to 6 hr . Slices were placed in a recording chamber mounted on an upright microscope ( BX51W , Olympus ) , visualized with deep red light ( 750 ± 25 nm ) and a CoolSnap SF CCD camera ( Photometrics ) . The preparation was elevated on a nylon grid to promote continuous perfusion of a bubbled BBS solution ( 3 ml/min ) under the slice . Electrophysiological recordings were obtained at 34–36°C from UBCs in the lobule X of the cerebellar vermis . Voltage- and current-clamp recordings were performed in the whole-cell configuration . Data were sampled at 40–50 kHz and filtered at 5 kHz acquired using an EPC10 amplifier and the Patchmaster software ( HEKA ) . Patch pipettes ( 3 . 5–4 MΩ ) were pulled from borosilicate glass capillaries ( Hilgenberg ) with a vertical puller and filled with an intracellular solution containing ( in mM ) : 135 KMeSO4 , 3 NaCl , 1MgCl2 , 0 . 1 EGTA , 10 phosphocreatine-K2 , 10 HEPES , 4 ATP-Mg , and 0 . 4 GTP-Na2 ( pH adjusted to 7 . 35 with KOH; osmolarity adjusted to 295 mOsm ) . EGTA was replaced with 10 mM BAPTA and KMeSO4 concentration was decreased to 114 mM in a first set of experiments ( n=25 ) aimed at recording the synaptic currents only . Alexa Fluor-488 ( 20 μM; Invitrogen ) was added to the pipette solution to visualize the cell morphology and easily identify UBCs . Cell membrane capacitance ranged from 7 to 20 pF . Step current injections ( 5 to 50 pA ) from a membrane potential of −80 mV were used to test the capability of UBCs to generate high frequency bursting activity . Pipette for mossy fiber stimulation was filled with HEPES-buffered solution and it was placed in the granule cell layer at more than 100 μm from the cell body . The stimulation protocols were obtained by modeling eMF activity as reMF[1+Asin ( 2πft ) ]+ with reMF=26 Hz , and the modulation A=53f is obtained from in vivodata at frequency f=0 . 3 ( Arenz et al . , 2008 ) . The stimulation consists in a constant fire rate 26 Hz for 10 s , and then a modulated rate for another 10 s at three different frequencies 0 . 3 , 1 and 3 Hz , respectively . GABAergic and glycinergic components were blocked during the experiments by perfusion of 6-imino-3- ( 4-methoxyphenyl ) -1 ( 6H ) -pyridazinebutanoic acid hydrobromide ( SR95531; 2 μM; Abcam Biochemicals ) , ( 2S ) -3-[[ ( 1S ) -1- ( 3 , 4-dichlorophenyl ) ethyl] amino-2-hydroxypropyl] ( phenylmethyl ) phosphinic acid hydrochloride ( CGP 55845; 0 . 5 mM; Tocris ) , and strychnine ( 1 μM ) . Synaptic currents were identified by bath application of antagonists . mGluR1 component was blocked with ( 3 , 4-Dihydro-2H-pyrano[2 , 3-b]quinolin-7-yl ) - ( cis-4-methoxycyclohexyl ) -methanone ( JNJ 16259685; 1 mM; Tocris ) . mGluR2 component was blocked with ( 2S ) -2-amino-2-[ ( 1S , 2S ) -2-carboxycycloprop-1-yl]-3- ( xanth-9-yl ) propanoic acid ( LY 341495; 0 . 5 μM; Tocris ) . AMPA component was blocked with 2 , 3-Dioxo-6-nitro-1 , 2 , 3 , 4-tetrahydr obenzo[f]quinoxaline-7-sulfonamide disodium salt ( NBQX; 1 mM; Abcam Biochemicals ) . AMPA receptor desensitization was modified by the allosteric modulator 6-Chloro-3 , 4-dihydro-3- ( 5-norbornen-2-yl ) -2H-1 , 2 , 4-benzothiazidiazine-7-sulfonamide-1 , 1-dioxide ( cyclothiazide; Abcam Biochemicals ) at the specified concentrations . Data were analyzed with Igor PRO ( WaveMetrics ) and spikes detection was performed with the custom-made threshold-detection algorithm SpAcAn ( Dugué et al . , 2005; http://www . spacan . net/ ) . Duration of the membrane potential depolarization after the end of sinusoidal MF modulation in complex cells was calculated as the time between the last stimulation and the time at which the membrane potential decays at the resting membrane potential value . The phase shift of UBC firing rate modulation was obtained by fitting single spikes phase distribution over one period with a circular normal function ( von Mises function ) f ( θ ) =rmin+ ( rmax−rmin ) ( ek2cos ( θ−ϕ ) −e−k2 ) / ( ek2−e−k2 ) , where rmax ( rmin ) is the maximal ( minimal ) firing rate , ϕ is the preferred phase shift of the cell and k is a parameter measuring the concentration of spikes around the preferred phase ( k = 0 indicates no modulation by phase , while a large k indicates highly concentrated spikes ) . The phase shift was relative to the maximum MF stimulation at each cycle , and it could be positive , consistent with a delay of UBC maximum firing rate , or negative , consistent with a strong delay or an anticipation of phase . CV2 is the mean of 2\|ISI1−ISI2\|/\|ISI1+ISI2\| , where ISI1 and ISI2 are successive inter-spikes intervals ( Ruigrok et al . , 2011 ) . Average ratio of depolarization / hyperpolarization and the steady-state firing rate in current-clamp protocols were estimated in the last second of 26 Hz steady stimulation . The peak of fast AMPA EPSC was obtained by fitting their decay with an exponential function and is given as the amplitude of the exponential decay . This measure avoids contamination by slow/buildup components of the response . For the same reason the charge of fast EPSCs is given as the product of the amplitude and of the decay time constant of the exponential fit . The buildup component of EPSC developed in the majority of the case with a delay of a few stimulations ( 3rd or 5th stimulation ) . We therefore distinguished the slow/steady-state component which follows fast EPSCs from the buildup component . The amplitude of the steady-state components was estimated as the amplitude of the base current of the exponential fit of the first EPSC . For comparison with other components its charge was obtained by multiplying this amplitude by the duration of the stimulation ( 500 ms ) and subtracted from the charge of the current envelope to yield the buildup charge . Finally the rebound EPSC at the end of the stimulations was estimated by subtraction of an extrapolated exponential decay on which it superposes . The timecourse of this exponential decay was obtained by fitting the last 20% of the decay ( amplitude wise ) , after the rebound , with an exponential function forced in amplitude and origin to the steady amplitude at the time when stimulation was stopped . Peak and τON ( rise time ) were obtained from the exponential fit of the onset of the subtracted slow rebound EPSC . Statistical analysis was performed with Rstudio using Mann-Whitney U test , Friedman’s test and Shapiro-Wilk test for normality of data . Results are quoted as mean ± SD unless stated otherwise . p<0 . 05 indicates statistical significance . UBC AMPAR-mediated synaptic currents exhibit complex dynamics including fast transients , slow buildup and post-stimulation slow rebound . We could quantitatively describe this synaptic dynamics with a linear combination of three types of AMPARs: ‘close’ receptors close to a release site , with instantaneous rise and exponential decay of the glutamate for the fast transients;; ‘intermediate’ receptors with a non-zero rise time for the rising of the rebound; ‘far’ receptors to give rise to the buildup and associated rebound . AMPAR were modeled by a 4-state Markov modelC⇌β2α2[ x ]O2⇌β1α1[ x ]O1⇌βDαDD , with one closed state C , two open states O1and O2 , and a desensitized state D . Transitions between states were described by rate constants α and β . The transition rates from the close to the open state O1 , and from O1 to O2 , were proportional to the glutamate concentration x . The fractions of receptors in each state ( r2 in O2 , r1 in O1 , and d in D ) obey the following dynamics:r2′=α2x ( 1−r1−r2−d ) − ( β2+α1x ) r2+β1r1r1′=α1xr2− ( β1αD ) r1+βDdd′=αDr1−βDd Each type of receptor is subjected to its own glutamate concentration dynamics x . The ‘close’ glutamate has the following dynamics:dx/dt=x/[τdecay ( 1+x/u ) ]+s∑kδ ( t−tk ) , At the spike time tk , the glutamate concentrations increase by a step s , then decay with a concentration-dependent time constant τdecay ( u=30 μM ) . The concentration-dependence of the glutamate decay accounts for the non-exponential concentration decay found at proximity of the glutamate release sites , as a consequence of 2D or 3D diffusion ( Barbour , 2001 ) . Both ‘intermediate’ and ‘far’ glutamates with additional rise time to buildup have the following dynamics:dx/dt= ( y−x ) /τdecay ( 1+x/u ) , dy/dt=−y/τrise+s∑kδ ( t−tk ) , where the rise variable y increases by a step s when a spike occurs , and decays with a time constant τrise . The total synaptic current isIAMPA=[gclose ( xclose , 1+rxclose , 2 ) +gint ( xint , 1+xint , 2 ) +gfar ( xfar , 1+xfar , 2 ) ]Vhold , with Vhold=−60 mV , and gs are conductance strengths for each receptor . We fitted the recorded UBC synaptic traces with the model above . To reduce the number of parameters , we used the following values for the rate constants: α1=0 . 03 μM−1 ms−1 , α2=0 . 15 μM−1 ms−1 , αD=2 ms−1 , β1=β2=10 ms−1 , and fit other 11 parameters: g , τdecay , s , for all three types of receptors and τrise for both ‘intermediate’ and ‘far’ receptors . The model was used to fit UBC synaptic currents triggered by a train of stimulation of 0 . 5 s at 50 Hz , close to the 1 Hz modulation parameters . After fitting , we fixed the parameters of each UBC , and then use these parameters to produce a simulated response to the modulated stimulation at 1 Hz used in the experiments for each UBC , as in Figure 5 . We first calculated the correlation coefficients of the histograms of simulated EPSCs and experimental spike rate responses , and verified that most cells have a positive correlation , which implies that we emulated synaptic currents well enough to explain the spiking dynamics of most UBCs . We used the p-values to define well correlated cells in Figure 5I . We then fitted simulated EPSCs with a circular normal function to obtain the predicted phase shift of UBC spiking . We set up a simple network model to study the effects of neuronal and network mechanisms on the information transfer from the input , represented by the eMF activity , to the output , represented by the GC activity . The network includes 4500 GCs with 500 eMFs and 500 UBCs . Each GC receives 4 inputs , which are all eMF in the network with no UBCs , and eMFs and UBC in the network with UBCs . In networks with UBCs included , each synaptic input to GC could be either eMF or UBC with equal probability of 50% , as observed experimentally in organotypic cultures ( Nunzi and Mugnaini , 2000 ) and as expected from the number ( Dugue et al . , 2005 ) and divergence of UBCs ( Berthie and Axelrad , 1994 ) . Previous experimental studies have shown that the angular velocity of the rat head rotation is encoded linearly by the mossy fiber's firing rate ( Arenz et al . , 2008 ) . Here we consider two types of eMFs , one that is in phase , the other in anti-phase with the velocity ( Arenz et al . , 2008 ) . Therefore , we modeled the input velocity information as a sinusoid function of the eMF firing rate νeMF=reMF[1+Asin ( 2πft ) ]+ , where [ . ]+ rectifies the firing rate to be non-negative , reMF=26 Hz and the modulation A=53fk obtained from in vivo data with movements at f=0 . 3 Hz ( Arenz et al . , 2008 ) , and k is drawn uniformly within ( 0 , 1 ) . Similarly , we modeled the dynamics of UBC with a circular normal function with parameters rmax , rmin , ϕ and k were obtained from the fitting of all 47 recorded UBCs in Figure 6 . Half of the UBCs were modeled by sampling randomly from 47 recorded UBCs , and the other half was also sampled from recorded UBCs , but phase reversed ( i . e . adding a phase shift of 180degrees ) , since eMFs can display phase and anti-phase behavior in vivo ( Arenz et al . , 2008 ) . We also performed simulations where the parameters of all modeled UBCs were randomly chosen from the range of values of fitting parameters , to get more heterogeneous phase shifts in the firing rate curves . Results were indistinguishable from the results presented here . With the given firing rate νeMF and νUBC , we generated the spike trains by the time-rescaling method ( Brown et al . , 2002 ) , then fed these input spike trains to GCs . We modeled the Purkinje cell ( PC ) as a firing rate unit , whose firing rate was an instantaneous linear function of the weighted inputs from all GCs , i . e . rPC ( t ) =∑ iwiriGC ( t ) , where riGC ( t ) , is the average GCs firing rate at a given phase within an oscillatory cycle . We defined a 'target' firing rate of the PC , rPC , target ( t ) =r¯ ( 1+cos⁡ ( 2πft−ϕ ) ) , where ϕ is the preferred target phase to learn , and r¯ is the average rate of 32 Hz ( typical rate of PCs in vivo ) . To achieve this target firing rate , the synaptic weights of the Purkinje cell evolved according to the following learning rule: Δwi=η riGC ( t ) E , where E= ( rPC , target ( t ) −rPC ( t ) ) is the 'error signal' , which could be implemented by climbing fibers . Weights wi were initialized as 0 . 5 , and updated by Δwi , and are set to zero if they become negative . The learning rate was η=0 . 001 . The learning process was simulated until convergence of both the weights and the error signal after 10000 cycles . Codes for all numerical simulations are available on https://sites . google . com/site/jiankliu .	Whether walking , riding a bicycle or simply standing still , we continually adjust our posture in small ways to prevent ourselves from falling . Our sense of balance depends on a set of structures inside the inner ear called the vestibular system . These structures detect movements of the head and relay this information to the brain in the form of electrical signals . A brain area called the vestibulo-cerebellum then combines these signals with sensory input from the eyes and muscles , before sending out further signals to trigger any adjustments necessary for balance . One of the main cell types within the vestibulo-cerebellum is the unipolar brush cell ( or UBC for short ) . UBCs pass on signals to another type of neuron called Purkinje cells , which support the learning of motor skills such as adjusting posture . Zampini , Liu et al . set out to test the idea that UBCs transform inputs from the vestibular system into a format that makes it easier for cerebellar Purkinje cells to drive this kind of learning . First , recordings from slices of rodent brain revealed that UBCs respond in highly variable ways to vestibular input , with both the size and timing of responses varying between cells . This is because vestibular signals trigger the release of a chemical messenger called glutamate onto UBCs , but UBCs possess a variety of different types of glutamate receptors . Vestibular input therefore activates distinct signaling cascades from one UBC to the next . According to a computer model , this variability in UBC responses ensures that a subset of UBCs will always be active at any point during vestibular input . This in turn means that Purkinje cells can fire at any stage of a movement , which boosts the learning of motor skills . The next steps will be to test this hypothesis using mutant mice that lack specific receptor subtypes in UBCs or UBCs completely . A further challenge for the future will be to build a computer model of the vestibulo-cerebellar system that includes all of its component cell types .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Mechanisms and functional roles of glutamatergic synapse diversity in a cerebellar circuit
In the inner plexiform layer ( IPL ) of the mouse retina , ~70 neuronal subtypes organize their neurites into an intricate laminar structure that underlies visual processing . To find recognition proteins involved in lamination , we utilized microarray data from 13 subtypes to identify differentially-expressed extracellular proteins and performed a high-throughput biochemical screen . We identified ~50 previously-unknown receptor-ligand pairs , including new interactions among members of the FLRT and Unc5 families . These proteins show laminar-restricted IPL localization and induce attraction and/or repulsion of retinal neurites in culture , placing them in an ideal position to mediate laminar targeting . Consistent with a repulsive role in arbor lamination , we observed complementary expression patterns for one interaction pair , FLRT2-Unc5C , in vivo . Starburst amacrine cells and their synaptic partners , ON-OFF direction-selective ganglion cells , express FLRT2 and are repelled by Unc5C . These data suggest a single molecular mechanism may have been co-opted by synaptic partners to ensure joint laminar restriction . In many regions of the nervous system , neurons and their arbors are organized in parallel layers . This organization provides an architectural framework that facilitates the assembly of neural circuits in a stereotyped fashion , a crucial feature that underlies function of the structure . Laminated structures are composed of multiple different classes and subtypes of neurons that form distinct connections in specific stratified layers . During development , the cell bodies and/or neurites of these different neuronal subtypes become restricted to one or more distinct strata . Costratification of arbors promotes synaptic specificity by placing appropriate synaptic partners in close proximity to one another . As such , understanding how lamination occurs is essential to uncovering the molecular basis of how highly-specific neural circuits form . The mouse retina is an excellent system to study lamination . The inner plexiform layer ( IPL ) of the retina is a stratified neuropil composed of axons and dendrites belonging to ~70 different subtypes of neurons . These neurons synapse selectively on specific partners , forming a complex set of parallel circuits , so a high degree of specificity is required during the wiring process ( for review see Sanes and Zipursky , 2010; Hoon et al . , 2014 ) . The IPL has been well-characterized structurally and functionally . Three major class of neurons ( bipolar , amacrine , and retinal ganglion cells ( RGCs ) ) form connections with each other in five IPL synaptic sublayers , termed S1-S5 ( Figure 1B ) . Most neurons project selectively to just one or a few of these sublayers . There are many genetic and cell biological tools available to study neurons with lamina-specific projections and retinal neurons are amenable to culture ex vivo allowing in-depth analysis of the receptor-ligand interactions that underlie laminar organization . For all these reasons we chose the IPL region of the mouse retina as a model system to study lamination . 10 . 7554/eLife . 08149 . 003Figure 1 . Methodology to identify recognition proteins for an extracellular receptor-ligand binding screen . ( A ) Flow chart describing the process of conducting candidate-based binding screen . A flow chart depicting the process of predicting the cell surface and secreted proteins in the mouse genome prior to candidate selection is outlined in Figure 1—figure supplement 1 . A table of the 65 candidate genes is included as Figure 1—source data 1 and a description of the 15 previously-unreported cDNAs that encode new isoforms is presented as Figure 1—source data 2 . ( B ) Schematic representation of the IPL showing the five sublayers ( S1-S5 ) , three major classes of neurons: amacrines ( Am , blue ) , bipolars ( Bp , green ) , retinal ganglion cells ( RGCs , magenta ) and the function of the sublayers in visual processing ( OFF and ON ) . Neurite stratifications provide an example of differential laminar organization . ( C ) Schematic representation of the ELISA-based binding assay . Receptor proteins ( blue ) tagged with alkaline phosphatase ( AP; yellow ) are tetramerized on the ELISA plate via an anti-AP antibody ( yellow ) . Binding of tetramerized ligand ( purple ) tagged with the Fc region of IgG1 ( Fc; green ) to receptor is detected by inclusion of an anti-Fc antibody conjugated with horseradish peroxidase ( HRP; orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 00310 . 7554/eLife . 08149 . 004Figure 1—source data 1 . Table lists the 65 candidate genes selected for the binding screen , the 121 proteins encoded by different isoforms or cleavage products , EntrezGene identifiers and Accession numbers , primer sequences used for cDNA cloning of the extracellular domain , protein type ( secreted , GPI-linked or transmembrane ) and the protein concentrations for both the AP- and Fc-tagged proteins used in the binding screen . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 00410 . 7554/eLife . 08149 . 005Figure 1—source data 2 . Previously-unreported cDNAs encoding new isoforms . Table lists the gene symbols , the name assigned to each new isoform and a description of how the new isoform differs from previously reported cDNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 00510 . 7554/eLife . 08149 . 006Figure 1—figure supplement 1 . Flow-chart for predicting cell surface and secreted proteins in mouse genome . The full repertoire of cell surface and secreted proteins encoded in the mouse genome was predicted using a variety of bioinformatics programs as follows . The Mouse Genome 430 2 . 0 microarray ( Affymetrix , CA ) contains 45 , 101 probeset IDs . Of these , 35 , 469 have UniProtKB/Swiss-Prot identifiers and , as such , correspond to protein-encoding genes . We downloaded the protein sequence for each gene from the UniProtKB/Swiss-Prot database . Protein sequences were submitted to the SignalP server which predicts the presence of a signal peptide ( Petersen et al . , 2011 ) and the TMHMM server which predicts the presence of a transmembrane domain ( Krogh et al . , 2001 ) . Proteins containing a signal peptide and/or a transmembrane domain were analyzed 1 ) for the presence of domains known to be present in proteins expressed at the cell surface or secreted using SMART ( Schultz et al . , 1998; Letunic et al . , 2012 ) , Pfam ( Finn et al . , 2014 ) and InterPro ( Hunter et al . , 2012 ) and 2 ) for gene ontology ( GO ) cellular component terms consistent with cell surface or secreted proteins ( Ashburner et al . , 2000 ) . Probeset IDs for genes encoding these proteins were analyzed using dChip software ( Li and Hung Wong , 2001 ) for differential expression amongst the 13 different retinal neuron subtypes . Probeset IDs with ≥3-fold differences in expression amongst the cell subtypes were selected . Genes were ranked according to published data demonstrating that the proteins are known to be involved in cell adhesion , recognition and neuronal guidance or targeting . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 006 Extracellular interactions between neighboring neurons or between neurons and their environment mediate molecular recognition events that direct laminar organization by providing instructions to neurons regarding where to grow ( through attraction or repulsion ) , how to organize neurites and with whom to form synaptic connections ( for review see Tessier-Lavigne and Goodman , 1996; Kolodkin and Tessier-Lavigne , 2011; Lefebvre et al . , 2015 ) . In this way , molecular recognition specificity ( i . e . receptor-ligand interactions ) translates into wiring specificity . To date , only a small number of interacting proteins and the instructions they provide to neurites during laminar organization of the mouse IPL has been identified ( Matsuoka et al . , 2011; Sun et al . , 2013; Duan et al . , 2014 ) . A global understanding of how laminar organization of the ~70 different subtypes develops in the IPL requires four systems-level criteria: 1 ) knowledge of all the secreted and cell surface proteins present within the developing structure that are available to mediate recognition events; 2 ) an inclusive description of which of these recognition proteins can engage in receptor-ligand interactions ( the 'interactome' ) ; 3 ) a comprehensive understanding of the functional consequence each interaction has on developing neurites ( i . e . attraction or repulsion ) ; and 4 ) a complete atlas detailing the expression of every ligand and its cognate receptor in each neuronal subtype to know which cells are capable of recognizing and responding to one another . Together these data will provide a platform for understanding the molecular basis of how complex neural circuits form between many different subtypes of neurons within an entire structure . Here we employed a combination of systems biology approaches to address these four criteria and begin the process of studying IPL lamination on a global level ( Figure 1A ) . To address the first criteria , we analyzed microarray data from 13 different subtypes of IPL neurons and selected genes encoding cell surface and secreted proteins that were differentially expressed – these are good candidates for mediating cell-cell recognition across subtypes . To address the second criteria , we used a modified version of a technology we previously developed ( Wojtowicz et al . , 2007 ) to perform a high-throughput , receptor-ligand biochemical screen that tested every pairwise combination of these candidate recognition proteins for binding . This screen identified ~50 previously-unreported receptor-ligand pairs , several between seemingly-unrelated proteins and others between new members within families of proteins previously known to interact . To investigate whether the receptor-ligand interactions we identified have functional relevance for IPL development , we focused on one family of type I transmembrane receptor-ligand interactions , those between a set of three FLRTs ( Fibronectin Leucine-Rich Transmembrane , FLRT1-3 ) and four Unc5s ( Uncoordinated5 , Unc5A-D ) . Some interactions among these molecules have previously been described ( Karaulanov et al . , 2009; Sollner and Wright , 2009; Yamagishi et al . , 2011; Seiradake et al . , 2014 ) , while others are newly identified in our screen . Members of both the Unc5 and FLRT families exhibit multiple roles in development in a variety of different systems with various interaction partners ( Bottcher et al . , 2004; Dakouane-Giudicelli et al . , 2014; Finci et al . , 2015; Akita et al . , 2015 ) . Using immunostaining and single cell ex vivo stripe assays , we found FLRTs and Unc5s exhibit distinct sublaminar expression patterns in the IPL and elicit repulsion and/or attraction in subsets of retinal neurons . Together these findings are consistent with a role for these families of proteins in mediating differential recognition events between neurons during laminar organization . We propose that , like Contactins , Sidekicks and Dscams in the chick retina ( Yamagata et al . , 2002; Yamagata and Sanes , 2008; Yamagata and Sanes , 2012 ) , FLRTs and Unc5s are positioned to provide a code for mediating laminar organization in the developing mouse IPL . Differential expression of extracellular proteins provides a molecular mechanism by which neuronal subtypes distinguish amongst one another . We therefore reasoned that good candidates for mediating neuronal subtype-specific recognition in the IPL are cell surface and secreted proteins that are differentially expressed in different subtypes of amacrine , bipolar and retinal ganglion cells . As no published list of all cell surface and secreted proteins in the mouse genome exists , we first predicted all of the cell surface and secreted proteins using a variety of bioinformatics approaches . A detailed description of this process is outlined in Figure 1—figure supplement 1 . To identify differentially-expressed recognition proteins ( Figure 1A ) , we analyzed microarray data collected from 13 different subtypes of neurons that arborize within different combinations of IPL sublaminae ( Kay et al . , 2011b; Kay et al . , 2012 ) . The microarray analyses were performed using neurons harvested at P6 , a developmental time when extensive neurite extension , arbor refinement , laminar organization and synapse formation are occurring in the IPL . We identified ~200 genes encoding extracellular proteins that exhibited ≥3-fold difference in microarray expression levels amongst the neuronal subtypes . Based on the domains present in each protein and known players involved in cell-cell recognition , we selected 65 genes as primary candidates and cloned them from retinal cDNA ( Figure 1—source data 1 ) . Because many of the genes encode more than one protein isoform as a result of alternative splicing or proteolytic cleavage , these primary candidates comprised 121 distinct cDNAs , including 15 splice variants that have not been previously reported ( Figure 1—source data 2 ) . New splice variants were identified for Ncam1 , Netrin5 , several Semaphorins and all four Unc5s ( i . e . Unc5A-D ) . The candidate proteins fall into three categories: secreted ( 26/121; 22% ) , GPI-linked ( 17/121; 14% ) and type I transmembrane ( 78/121; 64% ) . Proteins with multiple transmembranes were not included because their extracellular region is not contiguous and , as such , recombinant protein comprising the entire extracellular domain cannot be readily produced . We cloned the extracellular region of our 121 candidate proteins into two expression plasmids that C-terminally tag the proteins with 1 ) alkaline phosphatase ( AP ) or 2 ) the Fc region of human IgG1 ( Fc ) . Additionally , there is a 6X-His epitope tag on the C-terminus of both AP and Fc . Recombinant AP- and Fc-tagged proteins were produced by transient transfection of HEK293T cells . As these proteins have a signal peptide but no transmembrane domain or GPI-propeptide , they are secreted into the culture media . For AP-tagged proteins , 106 out of 121 ( 88% ) proteins were produced at optimal concentrations; for Fc-tagged proteins , 110 out of 121 ( 91% ) proteins were produced at optimal concentrations ( see Materials and methods ) ( Figure 1—source data 1 and Figure 2—figure supplement 1 and Figure 2—figure supplement 2 ) . The amount of recombinant protein present in the culture media was quantified using an endpoint kinetic enzymatic assay ( AP-tagged proteins ) or quantitative Western blots ( Fc-tagged proteins ) and the levels of protein in the media were normalized . We prefer to use normalized protein concentrations so that the levels of binding can be directly compared between receptor-ligand pairs and interacting pairs with high levels of binding can be identified . However , some proteins were expressed at levels lower than the optimized concentrations ( Figure 1—source data 1 ) . Nevertheless , these proteins were included in the screen . We next screened for interactions between candidate proteins utilizing a high-throughput , extracellular protein ELISA-based binding assay ( Figure 1C ) . The screen is a modified version of an assay we previously described that is quantitative over a 70-fold range ( Wojtowicz et al . , 2007 ) ( see Materials and methods ) . For this study , the workflow was converted from an insect cell strategy to one that would accommodate mammalian proteins . It is largely the case that interactions at the cell surface exhibit low affinities ( KD ~ µM ) and fast dissociation rates ( Vandermerwe and Barclay , 1994 ) , kinetic properties that allow transient , contact-dependent interactions to occur between recognition proteins expressed on neighboring cells in vivo but often make biochemical detection in vitro difficult . Our ELISA-based binding assay surmounts this limitation because it utilizes a strategy that tetramerizes the AP-tagged receptor and Fc-tagged ligand proteins ( see Materials and methods ) . By inducing tetramers , which provides additive or avidity effects , the assay is highly sensitive allowing proteins with micromolar affinities to be detected at nanomolar concentrations . Such clustering of cell surface proteins ( through dimerization , trimerization , tetramerization and pentamerization ) is standard practice for detecting ligand-receptor interactions in vitro ( Bushell et al . , 2008; Ramani et al . , 2012; Ozkan et al . , 2013 ) as well as in culture experiments where cellular responses to ligands are investigated ( Davis et al . , 1994 ) . As extracellular interactions are refractory to detection by standard interactome methodologies such as yeast-two-hybrid ( Braun et al . , 2009 ) , our ELISA-based binding assay provided the first platform for performing high-throughput screening of extracellular proteins ( Wojtowicz et al . , 2007 ) . The high-throughput nature of the assay is due , in large part , to the ability to test AP- and Fc-tagged extracellular domain proteins for binding directly in conditioned culture media following transient transfection , thereby obviating the requirement for arduous protein purification . Furthermore , by employing secreted , recombinant proteins , the assay monitors direct protein-protein interactions so it does not suffer the caveat that interactions may reflect indirect binding . As such , this assay , along with two similar , independently-developed ELISA-based binding methods ( Bushell et al . , 2008; Ozkan et al . , 2013 ) , provides a significant advancement for the study of extracellular protein-protein interactions over low-throughput techniques such as co-immunoprecipitation that , additionally , cannot distinguish between direct and indirect interactions . To assess which of the 121 candidate recognition proteins can engage in protein-protein interactions as cognate receptor-ligand pairs , we tested them ( and five Drosophila Dscam1 controls , i . e . 126 proteins ) for binding using the ELISA-based assay . The Dscam1 controls were included because some Dscam1-Dscam1 interacting pairs exhibit high levels of binding while others exhibit very low levels , thereby serving as a positive control for the sensitivity of the screen ( Wojtowicz et al . , 2007 ) . We tested the 126 proteins for binding in a matrix which reciprocally tests every pair-wise combination ( i . e . 126 x 126 = 15 , 876 binding reactions ) ( Figure 2; Visser et al . , 2015 ) . This includes 126 homophilic pairs and 7 , 875 unique heterophilic pairs . We included reciprocal pairs because sometimes a receptor-ligand interaction will occur in one orientation but not the other . Therefore , by testing each binding pair in both orientations , we decrease our false negative rate . 10 . 7554/eLife . 08149 . 007Figure 2 . High-throughput binding screen results and FLRT-Unc5 interactions . ( A ) 126 x 126 binding matrix . The 126 Fc- and AP-tagged extracellular domain proteins are arrayed along the x and y axes , respectively , in the same order such that homophilic interactions lie on the diagonal . The matrix is colored with a heat map such that high levels of binding are shown in white and no binding is shown in black . Values on the heat map scale represent HRP activity reported as absorbance at 650 nm . Background-subtracted data were deposited in the Dryad database Visser et al . , 2015 . Western blots of the proteins used in the screen are shown in Figure 2—figure supplement 1 and Figure 2—figure supplement 2 . ( B ) Subset of binding matrix showing FLRT-Unc5 interactions along with Ncam1 homophilic and Lrrc4c-NetrinG1 heterophilic interactions . Heat maps were generated using Image J ( Schneider et al . , 2012 ) . ( C ) Titration binding curves to monitor FLRT-Unc5 interactions using purified Unc5 protein binding to FLRT attached to an ELISA plate . FLRT1 , blue; FLRT2 , magenta; FLRT3 , green . Three independent experiments were performed in duplicate and average values are plotted . Error bars represent Standard Deviation . ( D ) Cell aggregation assays . CHO . K1 cells expressing full length Unc5 ( magenta ) and FLRT ( green ) were mixed together and incubated with shaking . Mixed aggregates of magenta and green cells represent trans heterophilic binding . Two independent experiments were performed and representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 00710 . 7554/eLife . 08149 . 008Figure 2—figure supplement 1 . Western blots of proteins for biochemical screen . α-6X-His Western blots of the AP-6X-His tagged proteins used in biochemical screen were used to assess that recombinant proteins were produced and full-length . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 00810 . 7554/eLife . 08149 . 009Figure 2—figure supplement 2 . Western blots of proteins for biochemical screen . α-6X-His Western blots of the Fc-6X-His tagged proteins used in biochemical screen were used to assess that recombinant proteins were produced and full-length . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 009 Interacting proteins identified in the screen were defined as those that exhibited ≥5-fold binding above background levels . Background was determined using absorbance readings at 650 nm ( Abs650nm ) for the 126 control wells that included ligand Fc-tagged culture media ( + anti-Fc-HRP antibody ) with mock culture media rather than AP-tagged receptor media ( background: mean Abs650nm = 0 . 064 , standard deviation = 0 . 009 ) . Using this criteria , we identified 192 unique interaction pairs , ~50 of which , to our knowledge , have not been reported in the literature ( Figure 3 and Figure 4; Visser et al . , 2015 ) . To assess the quality of our screen , prior to conducting it we generated a list of 109 receptor-ligand interactions that we expected to see based upon published data . Of these 109 positive control interaction pairs , we identified 91 giving us a false negative rate of 17% . This frequency is lower than published values for the yeast-two-hybrid screen which gives rise to false negative rates between 28 and 51% ( Huang and Bader , 2009 ) . 10 . 7554/eLife . 08149 . 010Figure 3 . New interactions identified in biochemical screen . ( A ) Interactions observed between a subset of proteins included in the screen . Lines indicate direct protein-protein interactions ( red line , not previously reported; gray line , previously known ) . Families of proteins are represented by color . Only one member of the Semaphorin family ( Sema3A , brown ) and one member of the Plexin family ( PlxnA4 , yellow ) are shown . The complete binding data for all Semaphorins , Plexins and Neuropilins ( Nrp , purple ) are shown in Figure 4 . For space considerations , gene names are used for proteins ( e . g . Cntn1 for Contactin1 ) . Figure 1—source data 1 includes full protein names and aliases . ( B ) ( Top panel ) Previous studies have demonstrated that Nrp1 ( purple ) can form a holoreceptor complex for Sema3A ligand ( brown ) through cis interactions with PlxnA4 ( yellow ) , Cntn2 ( green ) and a variety of other proteins in the cell membrane ( for review see Yazadani and Terman , 2006 ) . ( Bottom panel ) Our binding screen identified that Sema3A can engage in direct protein-protein interactions with both PlxnA4 and Cntn2 in the absence of Nrp1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 01010 . 7554/eLife . 08149 . 011Figure 4 . Summary of interactions between Sema-Nrp and Sema-Plxn proteins , highlighting new interactions observed in our screen as well as previously known interactions . A complete grid of known interactions was compiled from results reported in ten Semaphorin review articles ( Yazdani and Terman , 2006; Neufeld and Kessler , 2008; Wannemacher et al . , 2011; Hota and Buck , 2012; Neufeld et al . , 2012; Yoshida , 2012; Gu and Giraudo , 2013; Roney et al . , 2013; Worzfeld and Offermanns , 2014; Masuda and Taniguchi , 2015 ) and in independent primary literature searches conducted by several members of our laboratory . We included data from ten review articles because there is considerable variability in the interactions reported ( see Figure 4—source data 1 and Figure 4—source data 2 ) . All interactions reported in the reviews were corroborated in the primary literature and are denoted in the table by colored boxes that indicate the type of experiment supporting the interaction . Pink = evidence from cell binding assays , surface plasmon resonance , coimmunoprecipitation , transwell suppression and ex vivo explant outgrowth or growth cone collapse . Blue = genetic interactions . Gray , failure to find interaction by one or more of the above methods ( i . e . published negative interaction ) . A black dot ( • ) indicates a positive interaction observed in our screen . The reference and a description of the supporting data for each previously-known interacting pair are presented in Figure 4—source data 2 . It is important to note that there are multiple aliases for most Sema , Plxn and Nrp genes and , as such , our literature searches included these alternative names ( e . g . several Sema proteins were initially called collapsins and Sema3B was once called Sema5 ) . These aliases are listed in Figure 4—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 01110 . 7554/eLife . 08149 . 012Figure 4—source data 1 . Sema-Nrp and Sema-Plxn interactions published in review articles . A separate binding grid is shown for the interaction pairs reported in each of ten review articles ( Yazdani and Terman , 2006; Neufeld and Kessler , 2008; Wannemacher et al . , 2011; Hota and Buck , 2012; Neufeld et al . , 2012; Yoshida , 2012; Gu and Giraudo , 2013; Roney et al . , 2013; Worzfeld and Offermanns , 2014; Masuda and Taniguchi , 2015 ) . Interaction pair boxes are colored in dark gray . The review reference and PubMed ID is listed above each grid . The upper left table with the colored boxes presents a compilation of the interactions reported in all ten review articles . The number in each box represents how many of the ten review articles report the interaction . The boxes are colored using a heat map such that interactions reported by all 10 review articles are colored maroon and those reported by only 1 review article are colored blue . Numbers in yellow font represent interactions that were unverifiable in the primary literature . Unverifiable means that 1 ) no primary paper was cited for the interaction by the review article and our exhaustive search of the primary literature could not identify a paper reporting the interaction or 2 ) the interaction was cited by the review article but the paper cited did not test this binding interaction . Note that the unverifiable interactions were reported by only one or two of the ten review artcles ( one case , Sema3G-Nrp1 , was reported by three out of ten review articles ) . Unverifiable interactions are determined to be unpublished and are denoted as such in main text Figure 4 but are described in Figure 4—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 01210 . 7554/eLife . 08149 . 013Figure 4—source data 2 . Literature search results for Sema-Nrp and Sema-Plexin interactions . Colored boxes depict interactions reported in ten review articles ( Yazdani and Terman , 2006; Neufeld and Kessler , 2008; Wannemacher et al . , 2011; Hota and Buck , 2012; Neufeld et al . , 2012; Yoshida , 2012; Gu and Giraudo , 2013; Roney et al . , 2013; Worzfeld and Offermanns , 2014; Masuda and Taniguchi , 2015 ) . Review-reported interactions that we were able to verify in the primary literature ( pink ) , review-reported interactions that we were unable to verify in the primary literature ( yellow; see thorough description in Figure 4—source data 1 legend ) , reported genetic interactions ( blue ) , reported negative results ( gray; yellow font in gray box indicates that this interaction was also reported in one or more review articles but we were unable to verify in the primary literature ) . A description of the data that determines the color of each box is presented along with the reference for those data ( PubMed ID in blue font ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 01310 . 7554/eLife . 08149 . 014Figure 4—source data 3 . Gene name aliases for Sema , Nrp and Plxn . Aliases were obtained from NCBI Gene and include Mus musculus as well as orthologes in Homo sapiens , Rattus norvegicus , Danio rerio and Gallus gallus . These names were used for conducting primary literature searches to identify published Sema-Plxn and Sema-Nrp interacting pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 014 Some of the new receptor-ligand pairs identified involve proteins from families previously not known to associate with one another ( e . g . FLRT1-Cntn3 , Sema3A-Cntn2 and Ncam1-Dscam ) illustrating the importance of conducting unbiased pairwise screens ( Figure 3A-B ) . Other new interactions were observed between proteins previously believed to engage exclusively in homophilic , but not heterophilic , binding ( e . g . amongst Dscam , Dscaml1 and Sdk2 ) ( Yamagata and Sanes , 2008 ) . In addition , new binding pairs were found between members of protein families previously known to interact with one another ( e . g . FLRTs-Unc5s and Dscam-Netrin5 ) ( Andrews et al . , 2008; Ly et al . , 2008; Liu et al . , 2009; Karaulanov et al . , 2009; Sollner and Wright , 2009; Yamagishi et al . , 2011 ) . Three of the families included in the screen are the Semaphorins ( Sema ) , Plexins ( Plxn ) and Neuropilins ( Nrp ) . Previous studies have shown that five classes of Sema ligands ( Sema3-7 ) interact directly with four classes of Plxn receptors ( PlxnA-D ) or indirectly through binding to the Plxn co-receptors , Nrp1 and Nrp2 ( for review see Yoshida , 2012; Gu and Giraudo , 2013 ) . The specificity of Sema-Plxn interactions is largely restricted within distinct classes ( e . g . Sema4s bind PlxnBs and Sema5s bind PlxnAs ) with crosstalk occasionally observed ( e . g . Sema4C binds PlxnD1 ) . These broadly-defined principles of binding specificity have collectively emerged from a large number of studies that each investigated interactions between limited subsets of Semas and Plxns . Our screen included all members of these families ( 20 Sema , nine Plxn and two Nrp proteins ) and , as such , is the first comprehensive study of Sema-Plxn and Sema-Nrp binding specificity ( Figure 4 ) . Notably , we observed 1 ) that Nrp1 and Nrp2 can directly interact with some members of both the Sema4 and Sema6 families; 2 ) that some Sema3s can interact directly with Plxns in the absence of Nrp1 or Nrp2 ( previously only Sema3E was known to interact with PlxnD1 directly and signal in the absence of Nrp ) ( Gu et al . , 2005 ) ; and 3 ) new Sema4/5/6-Plxn interaction pairs . In total , we identified twenty-four previously-unreported Sema-Nrp or Sema-Plxn interactions and confirmed four others that had been suggested by genetic interactions ( see also Figure 4—source data 1 and Figure 4—source data 2 ) . Together , the results of our screen reveal a wide variety of new interactions among cell surface proteins , which we expect will provide a useful resource to the community of investigators studying cell-cell recognition in a variety of different systems . To validate a subset of hits in our screen , we performed additional binding experiments on two families of interacting type I transmembrane proteins , the FLRTs and Unc5s . Interactions between all three FLRT ( FLRT1-3 ) and all four Unc5 ( Unc5A-D ) family members were observed in the screen; and all pairs exhibited high levels of binding at or near the level of saturation of detection ( mean Abs650nm value = 2 . 14 ) . These families were selected for further study because they were some of the strongest hits , with binding levels comparable to positive controls such as Ncam1 homophilic binding and NetrinG1-Lrrc4c heterophilic binding ( Figure 2B ) . Furthermore , of the 12 possible FLRT-Unc5 interactions ( i . e . 3 FLRTs x 4 Unc5s ) , prior to our screen , four had been described in the literature ( three in mouse and one in zebrafish ) ( Karaulanov et al . , 2009; Sollner and Wright , 2009; Yamagishi et al . , 2011 ) suggesting that the eight new FLRT-Unc5 binding pairs we identified were likely to represent biologically-relevant interactions rather than false positives . To test the additional FLRT-Unc5 interactions observed in our screen , we performed titration binding experiments ( Figure 2C ) using purified protein . We utilized a fixed concentration of FLRT receptor on an ELISA plate and varied the concentration of purified Unc5 ligand . In all cases , we observed concentration-dependent binding curves . Because the extracellular region of the proteins used in these titration curves is tetramerized , the FLRT-Unc5 binding constants we observed ( i . e . on the order of ~1–10 nM ) are much higher than published affinities using monomeric protein in surface plasmon resonance experiments ( 0 . 3-21 μM ) ( Seiradake et al . , 2014 ) . This observation is similar to findings by Wright and colleagues which showed that pentamerization of extracelluar domains in their ELISA-based binding platform , AVEXIS , can improve the sensitivity of detection over monomeric proteins by at least 250-fold ( Bushell et al . , 2008 ) . To assess whether all FLRTs and Unc5s can interact between opposing cell surfaces , we performed cell aggregation assays . Full-length versions of FLRT1-3-myc and Unc5A-D-FLAG were co-transfected into CHO . K1 cells along with a plasmid expressing GFP or RFP , respectively . Western blots confirmed that the full-length proteins were produced and immunostaining for the C-terminal epitope tag showed staining around the periphery of the cell consistent with surface expression ( data not shown ) . Using the cell aggregation assay , we tested every combination of FLRTs and Unc5s and found that all pairs interact between opposing cells as evidenced by cell aggregation ( Figure 2D ) . By contrast , no clusters were observed between mock transfected cells , FLRT-FLRT or Unc5-Unc5 expressing cells . Together these data confirm that , as observed in our binding screen , trans interactions occur between all FLRT-Unc5 pairs . We next wanted to know what effect FLRTs and Unc5s have on retinal neuron outgrowth . To investigate the cell biological response of primary retinal neurons ( i . e . attraction or repulsion ) , we performed ex vivo stripe assays ( Vielmetter et al . , 1990; Delamarche et al . , 1997 ) . Because the IPL contains arbors from ~70 different subtypes of neurons , each of which may respond differently ( or not at all ) to the same protein ligand , it was necessary for us to use a stripe assay that would provide single-cell resolution . The tremendous value of single-cell stripe assays is that they allow the response of an individual subtype of neuron to be observed within a mixed population . As such , we designed and fabricated microfluidic devices ( Figure 5—figure supplement 1 and Materials and methods ) to pattern 30 µm stripes , a width appropriate for the growth of single IPL neurons whose cell bodies average between 10-30 µm ( data not shown ) . Our design is similar to others that have been used to monitor the effect of a purified ligand on neurite outgrowth of single dissociated neurons ( Weinl et al . , 2003; Yamagishi et al . , 2011; Singh et al . , 2012; Beller et al . , 2013; Sun et al . , 2013 ) . We dissected and dissociated neurons from wild-type P6 retinas and cultured individual neurons on FLRT or Unc5 stripes . We reasoned that proteins involved in mediating laminar organization , or other recognition events that play a role in neural circuit formation , would elicit a response ( i . e . attraction or repulsion ) in only a subpopulation of neurons . While the majority of neurons did not respond to FLRT or Unc5 stripes , growing indiscriminately across them , we observed small populations of neurons ( 5-18% ) that responded to FLRT1 ( n=61/375 , 16% attractive; n=19/375 , 5% repulsive ) , FLRT2 ( n=63/344; 18% repulsive ) , FLRT3 ( n=37/438 , 8% attractive; n=33/438 , 8% repulsive ) , Unc5C ( n=45/396 , 11% repulsive ) and Unc5D ( n=49/407 , 12% repulsive ) stripes ( Figure 5A-I ) . No significant response of neurons was observed to Unc5B stripes ( n=3/380 , 1% repulsive ) relative to control laminin stripes ( n=1/88 , 1% repulsive ) . There also were no attractive or repulsive responses to Unc5A stripes ( n=257/257 , 100% permissive ) but we did observe a modest population-wide reduction in neurite outgrowth and decreased viability ( data not shown ) . Together these data demonstrate that Unc5C , Unc5D , and all three FLRTs mediate recognition events between subtypes of retinal neurons and suggest that FLRTs and Unc5s may contribute to development of the retinal circuit . 10 . 7554/eLife . 08149 . 015Figure 5 . Subpopulations of primary retinal neurons respond to FLRT and Unc5 protein in stripe assays . Individual retinal neurons harvested from wild- type retinas at P6 were cultured for 4– 6 days on glass coverslips containing alternating stripes of laminin and a purified candidate recognition protein . ( A ) Quantification showing the percent of neurons that exhibited a repulsive ( green ) , attractive ( magenta ) or permissive ( gray ) response to stripes of the candidate recognition protein . n = total number of neurons scored . Raw data are reported in the main text . ( B-I ) Example images showing responses of neurons to stripes of the indicated FLRT or Unc5 protein ( magenta ) . Stripes were prepared using microfluidic devices as outlined in Figure 5—figure supplement 1 and were visualized by addition of BSA-TRITC ( magenta ) to the purified FLRT or Unc5 protein patterned . As coverslips were coated with the growth-promoting protein , laminin , prior to application of the stripes , the black ( unstriped ) regions of the coverslip contain laminin . Neurons were immunostained with an antibody against beta-tubulin ( Tuj1; green ) . ( J-K ) Example neurons co-stained for Tuj1 ( green ) and FLRT2 ( cyan in J ) or Unc5C ( cyan in K ) . Neurons that express FLRT2 are repelled by Unc5C stripes ( J ) , while neurons that express Unc5c are repelled by FLRT2 stripes ( K ) . See main text for quantification . ( L-M ) Gain-of-function stripe assay . Neurons transfected with full-length FLRT2-myc ( green ) are repelled by Unc5C stripes ( L ) whereas , neurons transfected with full-length Unc5C-FLAG ( green ) are not repelled by Unc5C stripes ( M ) . Scale bar , 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 01510 . 7554/eLife . 08149 . 016Figure 5—figure supplement 1 . Microfluidic device design for patterning protein stripes for stripe assay . Top-down view of the microfluidic channels ( red ) in the PDMS devices . See Materials and methods for additional details regarding channel dimensions and fabrication . Scale bar for upper panel , 150 μm . Scale bar for lower-panel , 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 016 To investigate which subpopulations of retinal neurons are using FLRTs and Unc5s to mediate recognition events involved in wiring , we next assessed the expression of FLRTs and Unc5s in the developing retina using immunostaining of P2 , P4 and P6 retinal sections ( Figure 6 and Figure 6—figure supplement 1 ) . All FLRT and Unc5 antibodies were highly specific with little to no cross-reactivity as assessed by ELISA using purified protein ( Figure 6—figure supplement 2 ) . To visualize the boundaries of the five IPL sublaminae ( S1-S5 ) , we stained retinal sections with an antibody against vesicular acetylcholine transporter ( VAChT ) . VAChT stains the dendrites of two subtypes of amacrine cells called OFF and ON starburst amacrine cells ( SACs ) that arborize within functionally-distinct sublaminae S2 and S4 , respectively ( Stacy and Wong , 2003 ) . As such , the positions of the other sublaminae ( i . e . S1/3/5 ) can be inferred relative to the VAChT stain in S2/4 ( Haverkamp and Wassle , 2000 ) . 10 . 7554/eLife . 08149 . 017Figure 6 . Expression of FLRT and Unc5 proteins in the developing IPL . ( A-G ) Retinal sections from C57Bl/6 P6 mice immunostained with an antibody against vesicular acetylcholine transporter ( VAChT; magenta ) , which is expressed by SAC dendrites and thus serves as a marker for sublaminae S2 and S4 , and an antibody against one of the FLRTs or Unc5s ( green ) as indicated in each panel . DAPI ( blue ) labels cell bodies in the inner nuclear layer ( INL ) and ganglion cell layer ( GCL ) flanking the IPL ( for schematic see Figure 1B ) . FLRT and Unc5 antibodies were highly specific as demonstrated by ELISA and shown in Figure 6—figure supplement 1 . Expression patterns at P2 and P4 are shown in Figure 6—figure supplement 2 . Scale bar , 50 μm . Relative fluorescence of each marker across IPL sublayers S1-S5 is quantified in the histograms plots provided in the right panels . All images were processed together so that the relative fluorescence intensity levels of the staining can be compared amongst different FLRT and Unc5 antibodies . Histogram images produced using ImageJ ( Schneider et al . , 2012 ) . ( H ) Schematic summarizing expression pattern of each FLRT and Unc5 protein across IPL sublayers . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 01710 . 7554/eLife . 08149 . 018Figure 6—figure supplement 1 . ELISA to test binding specificity of FLRT and Unc5 antibodies . ( A ) RGMA-Fc-6X-His ( control ) and Unc5-Fc-6X-His proteins were captured on a 96-well ELISA plate and stained with each Unc5 antibody in a matrix followed by a secondary antibody conjugated to HRP . Abs650 nm values at 60 min are shown . ( B ) RGMA-Fc-6X-His ( control ) and FLRT2-Fc-6X-His proteins were captured on a 96-well ELISA plate and stained with each FLRT2 antibody in a matrix followed by a secondary antibody conjugated to HRP . Abs650 nm values at 60 min are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 01810 . 7554/eLife . 08149 . 019Figure 6—figure supplement 2 . Developmental analysis of FLRT and Unc5 expression in the IPL . Retinal sections from C57Bl/6 wild type P2 , P4 and P6 ( P6 images same as Figure 6 ) immunostained with an antibody against the FLRTs or Unc5s ( green ) as indicated in each panel . Co-staining of FLRTs and Unc5s with anti-against vesicular acetylcholine transporter ( VAChT; magenta ) , which labels SAC dendrites in sublaminae S2 and S4 , is shown in the right panels . DAPI ( blue ) labels cell bodies in the inner nuclear layer ( INL ) and ganglion cell layer ( GCL ) flanking the IPL ( for schematic see Figure 1B ) . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 019 At P6 we observed laminar-restricted expression patterns for all FLRTs and three out of the four Unc5s ( Figure 6 ) . FLRT1 expression was largely restricted to neurites that arborize in S1 ( Figure 6A ) , FLRT2 was most highly expressed in S2/4 ( Figure 6B ) and FLRT3 expression was largely restricted to S3 ( Figure 6C ) . Unc5A was highly expressed in the cell body layers flanking the IPL and , within the IPL , was expressed in neurites that arborize in S1/2/3/5 ( Figure 6D ) , Unc5C was most highly expressed in S1/3/5 ( Figure 6F ) and Unc5D expression was largely restricted to S1/5 . Unc5B did not show laminar restriction—it was expressed at low levels uniformly across the IPL ( Figure 6E ) . Comparison of the expression patterns observed at P6 with the patterns observed at P2 and P4 ( Figure 6—figure supplement 2 ) demonstrates that laminar-restricted expression of FLRT1-3 and Unc5A , C , D is spatio-temporally regulated . Three patterns of developmental regulation were observed . One subset of proteins , FLRT2 and Unc5C , showed broad expression across the IPL at P2 that gradually became sublamina-restricted by P6 . A second group , FLRT1 and FLRT3 , showed sublaminar bias already at P2 that changed only slightly as the IPL expanded with age . The final group , Unc5A and Unc5D , added new sublayers at later ages: Unc5A was not observed in the IPL until P6 , even though immunoreactivity was detected in neuronal somata at earlier ages , suggesting that IPL innervation by Unc5A-positive cells happens later than other family members . Unc5D , meanwhile , exhibited S1 restriction at P2-4 and then added expression in S5 at P6 . Interestingly , the expression pattern of Unc5D may remain dynamic after P6 , as immunostaining published by Feldheim and colleagues suggests that , while S5 expression is maintained , S1 expression is lost by P8 ( Sweeney et al . , 2014 ) . The three patterns of laminar restriction we observed – termed 'initially diffuse , ' 'initially precise , ' and 'stepwise' lamination – have been seen in previous studies of IPL laminar targeting ( Mumm et al . , 2006; Kim et al . , 2010 ) . The spatio-temporal and laminar-specific expression patterns of the FLRTs and Unc5s suggest that members of both families may contribute to specific cell-cell interactions that mediate these developmental strategies for laminar organization . Between P2 and P4 , Unc5C and FLRT2 expression patterns become restricted to complementary sublaminae in the IPL with Unc5C concentrated in S1/3/5 and FLRT2 predominantly expressed in S2/4 ( Figure 6—figure supplement 2 ) . Complementary expression suggests that these lamina-specific stratifications may arise due to repulsive interactions between neuronal subtypes expressing FLRT2 and Unc5C . Consistent with this model , our ex vivo stripe assays revealed subpopulations of neurons that are repelled by FLRT2 and subpopulations of neurons that are repelled by Unc5C ( Figure 5A , D , H ) . We hypothesized that repulsion by Unc5C stripes is due to interactions with FLRT2 expressed on repelled neurons . To investigate this possibility we performed immunostaining on neurons repelled by Unc5C stripes with antibodies against FLRT2 ( as well as FLRT1 and FLRT3 ) . Neurons repelled by Unc5C stripes expressed FLRT2 ( n=26/26 ) ( Figure 5J ) but not FLRT1 or FLRT3 ( data not shown ) . Conversely , neurons repelled by FLRT2 stripes expressed Unc5C ( n=30/30 ) ( Figure 5K ) . Together these data are consistent with a model wherein interactions between FLRT2 and Unc5C induce mutual repulsion via bidirectional signaling in both the ligand- and receptor-expressing cells . Repulsive signaling of Unc5 in response to ligand binding has been well-established ( for review of Netrin1-induced repulsion see Moore et al . , 2007; for FLRT2-induced repulsion via Unc5D see Yamagishi et al . , 2011 ) . In our stripe assays we observe FLRT2-expressing retinal neurons that are repelled by Unc5C which is consistent with a model whereby Unc5C binding to FLRT2 induces repulsion in the FLRT2-expressing neuron; however , repulsive signaling downstream of FLRTs has not been reported . So we next asked whether Unc5C-FLRT2 interactions can induce repulsion in FLRT2-expressing retinal neurons by performing gain-of-function stripe assays . Using transient transfection , we ectopically expressed either full-length FLRT2-myc or full-length Unc5C-FLAG ( control ) in retinal neurons cultured on Unc5C stripes and monitored the response of neurons that expressed these exogenous proteins as assessed by anti-myc and anti-FLAG immunostaining , respectively . Importantly , this gain-of-function experiment was possible because only 11% of wild-type retinal neurons are repelled by Unc5C stripes ( Figure 5A ) and , as such , the vast majority of neurons are available to exhibit a gain-of-function phenotype . We tested several commercially-available transfection reagents and found one that was capable of giving rise to ~10% transfection efficiency in our retinal neuron cultures ( n=67/691 neurons transfected , see Materials and methods ) . We obtained 39 FLRT2-myc transfected neurons and observed that all 39 neurons were repelled by Unc5C stripes ( n=39/328 neurons transfected; 15 coverslips ) ( Figure 5L ) . In our control transfections , we obtained 28 neurons that expressed Unc5C-FLAG and observed that 27/28 neurons grew permissively across the Unc5C stripes ( n=28/363 neurons transfected; 13 coverslips ) ( Figure 5M ) . One neuron that ectopically expressed Unc5C-FLAG was repelled by Unc5C stripes . We hypothesize that this neuron is one of the 11% of wild-type neurons that is endogenously repelled by Unc5C . These data demonstrate that FLRT2 is sufficient to mediate repulsion in response to Unc5C and , as such , repulsive signaling can occur downstream of FLRT2 . We next sought to identify which of the ~70 different subtypes of IPL-projecting neurons are the ones that express FLRT2 and are repelled by Unc5C . In retinal sections , FLRT2 expression co-localized with VAChT expression in S2/4 at P4 and P6 ( Figure 6B and Figure 6—figure supplement 2 ) . As such , we hypothesized that the FLRT2-expressing neurons are the same neurons that express VAChT – i . e . the starburst amacrine cells ( SACs ) which arborize in S2/4 between P0 and P3 ( Stacy and Wong , 2003 ) . To determine whether SACs express FLRT2 during and following arborization within S2/4 , we performed in situ hybridization against Flrt2 in sections at both P1 and P6 along with calbindin immunostaining which selectively stains SACs at these ages ( Kay et al . , 2012 ) . Calbindin immunostaining was used to label SACs because VAChT immunoreactivity does not persist through the in situ hybridization protocol ( nor does it label SAC cell bodies at P2-6 ) . This analysis revealed that Flrt2 is expressed by a subset of cells that includes: 1 ) SACs; 2 ) a sparse non-SAC population in the inner nuclear layer ( INL ) ( presumably amacrines due to their laminar position close to the IPL and the fact that bipolar cells are not yet born at P1 ) ; and 3 ) a non-SAC population in the ganglion cell layer ( GCL ) that , based upon their large soma size , are likely to be retinal ganglion cells ( Figure 7A ) . Notably , at P1 , Flrt2 expression is predominantly detected in ON SACs whose cell bodies reside in the GCL while , at P6 , Flrt2 expression is predominantly detected in OFF SACs whose cell bodies reside in the INL . 10 . 7554/eLife . 08149 . 020Figure 7 . SACs and Drd4-GFP ooDSGCs express FLRT2 and are repelled by Unc5C . ( A ) Flrt2 is expressed by SACs , a second amacrine population , and a subset of RGCs . In situ hybridization for Flrt2 RNA ( magenta ) was combined with immunostaining for calbindin ( green ) , a selective SAC marker at the ages shown ( P1 and P6 ) . Yellow arrows indicate Flrt2+ SACs . Cells in the inner nuclear layer ( INL ) expressing Flrt2 but not calbindin ( purple arrows ) define a non-SAC Flrt2+ amacrine population . Non-SACs in the ganglion cell layer ( GCL ) are likely RGCs , based on their large soma size ( purple arrows ) . Among SACs , Flrt2 is detected predominantly in ON SACs ( which reside in the GCL ) at P1 whereas it is detected more readily in OFF SACs ( which reside in the INL ) at P6 . However , ON SACs positive for Flrt2 are observed at P6 ( yellow arrow in GCL ) , suggesting that Flrt2 is not selective for one SAC population over the other . ( B ) RGCs expressing Flrt2 include direction-selective ganglion cells ( DSGCs ) . Double staining for Flrt2 and CART , an ooDSGC marker , at P1 and P6 . Double-labeled cells ( yellow arrows ) are observed in the GCL . Not all ooDSGCs express Flrt2 , however , as CART+ Flrt2– cells are also apparent ( green arrows ) . Purple arrows indicate Flrt2+ cells that are not ooDSGCs; this group likely includes SACs . Scale bar , 10 µm . ( C-E ) SACs express FLRT2 protein . Dissociated SACs from P2 Chat-Cre::RosaLSL-tdTomato mice that specifically express tdTomato ( magenta ) in SACs . Neurons were co-stained with an antibody against Tuj1 ( green ) and ( C ) FLRT1 , ( D ) FLRT2 , ( E ) FLRT3 ( cyan ) . Only FLRT2 co-localized with tdTomato-positive SACs . SACs were also negative for Unc5s as shown in Figure 7—figure supplement 1 . ( F-G ) tdTomato SACs ( magenta ) grown on Unc5C ( F ) or FLRT2 ( G ) stripes ( green ) . Stripes were visualized by addition of PLL-FITC to the purified Unc5C or FLRT2 protein patterned . Unc5C ( F ) but not FLRT2 ( G ) repelled SACs . ( H-I ) Dissociated Drd4-GFP ooDSGCs ( green ) in culture harvested from P3 mice that specifically express GFP in ooDSGCs . ( H ) Drd4-GFP neurons on Unc5C stripes co-stained with an antibody against Tuj1 ( green ) and FLRT2 ( cyan ) . ( I ) Drd4-GFP neurons on FLRT2 stripes stained with an antibody against Tuj1 ( green ) . Neurons cultured 8 DIV . Scale bar , 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 02010 . 7554/eLife . 08149 . 021Figure 7—figure supplement 1 . Expression of Unc5s in SACs . Dissociated SACs ( tdTomato , magenta ) harvested from P2 Chat-Cre::RosaLSL-tdTomato mice that specifically express tdTomato in SACs . Neurons were co-stained with an antibody against Tuj1 ( green ) and ( A ) Unc5A , ( B ) Unc5B , ( C ) Unc5C and ( D ) Unc5D ( cyan ) . None of these co-localized with tdTomato in SACs . DOI: http://dx . doi . org/10 . 7554/eLife . 08149 . 021 To confirm that FLRT2 protein is expressed in SACs , we performed FLRT2 immunostaining on cultured retinal neurons from a mouse strain that genetically expresses tdTomato specifically in SACs ( Chat-Cre::RosaLSL-tdTomato ) ( Sun et al . , 2013 ) . It was necessary to use these transgenic mice to visualize SACs in culture because the VAChT antibody that stains SACs in retinal sections does not stain cultured SACs ( J . N . K . , unpublished observations ) . Furthermore , it was necessary to perform FLRT2 immunostaining in dissociated cultured neurons because , in retinal sections , FLRT2 stains neurites in the IPL but not cell bodies in the adjacent INL and GCL ( Figure 6B and Figure 6—figure supplement 2 ) thereby preventing identification of the cell ( s ) to which the FLRT2-positive neurites belong . Immunostaining of dissociated SACs harvested at P2 demonstrated that tdTomato-positive SACs express FLRT2 ( n=47/47 ) but not FLRT1 ( n=0/55 ) or FLRT3 ( n=0/67 ) ( Figure 7C-E ) . Consistent with our in situ hybridizations , we also observed non-SAC neurons that expressed FLRT2 ( Figure 7D ) . As Unc5C expression localizes to S1/3/5 where SACs do not arborize ( Figure 6F and Figure 6—figure supplement 2 ) , we expected that SACs would not express Unc5C . Indeed , while a subset of tdTomato-negative neurons were immunoreactive for Unc5C , no Unc5C expression was observed in SACs ( n=0/39 ) ( Figure 7—figure supplement 1 ) . Furthermore , none of the other Unc5s were expressed in SACs ( Figure 7—figure supplement 1 ) . If the FLRT2-Unc5C interaction induces repulsion of SACs , we would expect FLRT2-expressing SACs to be repelled by Unc5C stripes in the ex vivo stripe assay . Indeed , we observed robust repulsion of SACs from Unc5C stripes ( n=49/53 , 92% repelled ) ( Figure 7F ) . In contrast , SAC processes crossed FLRT2 stripes indiscriminately ( n=71/71 , 0% repelled ) ( Figure 7G ) . Together these findings demonstrate that SACs express FLRT2 both during and after the developmental time when their neurites are becoming restricted to S2/4 and that SACs are repelled by Unc5C . Since SACs do not express FLRT1 and FLRT3 , SAC repulsion by Unc5C could be due to interactions with FLRT2 . These data suggest that repulsive FLRT2-Unc5C interactions may contribute to laminar organization of SAC neurons in the developing IPL . By in situ hybridization we found that Flrt2 is expressed in a non-SAC population in the GCL ( Figure 7A ) . Direction-selective ganglion cells ( DSGCs ) arborize dendrites in S2/4 and are the post-synaptic partners of SACs ( Demb , 2007; Wei and Feller , 2011; Vaney et al . , 2012; Masland , 2012 ) . We therefore wondered whether DSGCs might also express Flrt2 . To test this idea , we combined Flrt2 in situ hybridization with immunostaining against the neuropeptide CART ( cocaine- and amphetamine-regulated transcript ) , which stains the most numerous category of DSGCs , ON-OFF DSGCs ( ooDSGCs ) ( Kay et al . , 2011a ) . CART is a selective ( though not exclusive ) marker for ooDSGCs ( Kay et al . , 2011b; Ivanova et al . , 2013 ) . We observed that about half of CART-immunoreactive cells are Flrt2-positive ( n=12/23 CART+Flrt2+ , n=11/23 CART+Flrt2– ) suggesting that a subset of ooDSGCs expresses FLRT2 ( Figure 7B ) . As ooDSGCs exhibit S2/4 laminar restriction , we next asked whether ooDSGCs , like SACs , express FLRT2 protein and are repelled by Unc5C stripes . To test this we cultured neurons from a mouse strain that genetically expresses GFP under control of the dopamine receptor 4 promoter ( Drd4-GFP ) in a subtype of ooDSGCs that prefer posterior motion ( Gong et al . , 2003; Huberman et al . , 2009; Kay et al . , 2011a ) . The Drd4-GFP cells were encountered in our cultures only rarely , perhaps because our cultures were not optimized for RGC survival , or because they are a remarkably sparse cell type comprising ≤5% of ganglion cells which are themselves only 1% of retinal neurons ( Kay et al . , 2011a ) . Nevertheless , when healthy Drd4-GFP neurons were identified , we observed that they expressed FLRT2 and were repelled by Unc5C stripes ( n=7/7 , 100% repelled; 7 coverslips ) ( Figure 7H ) but not by FLRT2 stripes ( n=10/10 , 0% repelled; 7 coverslips ) ( Figure 7I ) . These data suggest that at least one subtype of DSGCs may utilize repulsive FLRT2-Unc5C interactions to achieve laminar restriction in the developing IPL . The results of our binding screen drew our attention to FLRTs and Unc5s . We discovered that members of these protein families are expressed in strikingly specific laminar patterns during early IPL development . Using stripe assays , we found that all members of these families except Unc5A and Unc5B are capable of eliciting attractive and/or repulsive behavior from subsets of retinal neurons . Notably , Unc5A and Unc5B also showed the least laminar specificity in their IPL expression patterns . These two features of Unc5A and Unc5B biology suggest that they are unlikely to play a role in IPL lamination . By contrast , the other members of these two families are excellent candidates to mediate IPL lamination based on their expression patterns , bioactivities and receptor-ligand interactions that we report here . The expression patterns of FLRT2 and Unc5C are remarkably complementary in the developing IPL , suggestive of a repulsive role for this receptor-ligand pair . Consistent with this notion , we found that neurons expressing FLRT2 are repelled by Unc5C and , conversely , neurons expressing Unc5C are repelled by FLRT2 . Using transfected primary neurons , we demonstrated that ectopic expression of FLRT2 is sufficient to mediate repulsion in response to Unc5C . While we cannot rule out the possibility that this response to Unc5C arises due to the presence of another cell surface protein ( s ) that gets recruited in cis by exogenous FLRT2 expression , taken together our data suggest that FLRT2-Unc5C interactions can induce repulsion in a subset of primary retinal neurons . Sema , Plxn and Nrp proteins comprise large numbers of diverse cell recognition proteins involved in neural circuit formation and an ever-increasing list of cell biological processes ( for review see Yoshida , 2012; Gu and Giraudo , 2013 ) . While many binding partners within these families have been described , a comprehensive study of all Sema-Nrp and Sema-Plxn pairs has never been conducted . We included the complete families because our microarray data demonstrated that many members are differentially expressed in different subtypes of IPL neurons . Additionally , at the time we were selecting candidates for our screen , Kolodkin and colleagues reported that Sema5A and Sema5B interactions with PlxnA1 and PlxnA3 play a role in laminar organization in the developing mouse IPL ( Matsuoka et al . , 2011 ) . As such , we hypothesized that other family members are involved and reasoned that understanding the complete interaction network is necessary for evaluating genetic phenotypes in vivo . Previously , Sema3s were believed to require Nrp1 for signaling through PlxnA co-receptors ( Tamagnone et al . , 1999 ) and Cntn2 was believed to interact with Sema3A only indirectly through cis interactions between Cntn2 and Nrp1 ( Dang et al . , 2012 ) . Direct protein-protein interactions observed in our screen between Sema3A-Cntn2 and Sema3A-PlxnA4 suggest that Sema3A may be able to signal directly through these receptors in the absence of Nrp1 ( Figure 3B ) . The additional interaction partners we identified will thus enable the field to better understand how the interplay among Semas-Plxns-Nrps , as well as other potential Sema receptors such as Cntn2 and PlxnA4 , contribute to laminar organization of the IPL and other cellular responses in a variety of different systems . The three FLRTs and four Unc5s represent 12 potential heterophilic receptor-ligand pairs . Prior to our screen , four pairs had been reported amongst varying combinations of Xenopus and mouse proteins ( FLRT1-Unc5B , FLRT2-Unc5D , FLRT3-Unc5B and FLRT3-Unc5D ) ( Karaulanov et al . , 2009; Sollner and Wright , 2009; Yamagishi et al . , 2011 ) . Using a variety of binding assays , we observed interactions between all FLRTs and all Unc5s . Further confirmation that the eight additional FLRT-Unc5 pairs we observed are biologically-relevant has been provided by Seiradake et al . who recently reported several of these interactions ( Seiradake et al . , 2014 ) . FLRTs and Unc5s are broadly expressed in the developing nervous system as well as in other tissues . While in some regions FLRTs and Unc5s exhibit striking cell-type-specific expression patterns ( including the cortex , hippocampus and the developing retina as we have shown here ) , in other areas multiple FLRTs and Unc5s are expressed in overlapping regions ( Haines et al . , 2006; Gong et al . , 2009; Yang et al . , 2013; Seiradake et al . , 2014 ) . As such , the promiscuous binding of all FLRTs to all Unc5s seemingly presents a conundrum . Based upon the observed binding properties , a FLRT2-expressing neuron might well interact with all neurons that express any one of the four Unc5s . As such , how can FLRT-Unc5 interactions provide recognition specificity ? Does promiscuous binding reduce the total possible number of distinct FLRT-Unc5 binding specificities from 12 ( i . e . 3 FLRTs x 4 Unc5s ) to one ( i . e . FLRT-Unc5 ) ? Our experiments ( Figure 2C ) and those of others ( Seiradake et al . , 2014 ) have demonstrated that different FLRT-Unc5 pairs exhibit differences in binding affinity ( while our binding curves plateau due to saturated levels of detection and therefore preclude the determination of binding constants , the qualitative determination that there are differences can be inferred from the shifting of curves relative to one another along the x-axis ) . We speculate that these differences in binding affinity contribute to recognition specificity . The diverse cadherin family of homophilic and heterophilic cell surface proteins provides a classic example where this is the case . As with FLRTs and Unc5s , several members of the cadherin family exhibit similar levels of promiscuous homophilic and heterophilic binding in cultured cell-based assays but , when binding constants are determined using SPR or analytical ultra centrifugation , differences in binding affinity are observed which , in turn , mediate the sorting of cells into different tissues in vivo ( Katsamba et al . , 2009 ) . Conflicting reports have been published regarding whether or not FLRTs engage in homophilic interactions ( Karaulanov et al . , 2006; Yamagishi et al . , 2011; Seiradake et al . , 2014; Lu et al . , 2015 ) . Similar to previous experiments that failed to detect binding of soluble FLRT ectodomains to FLRT-expressing cells in culture ( Yamagishi et al . , 2011 ) or FLRT-mediated cell aggregation ( Lu et al . , 2015 ) , we did not observe FLRT homophilic interactions in our biochemical screen or cell aggregation assay . Furthermore , in our stripe assays , FLRT2-expressing SACs and Drd4-GFP ooDSGCs did not respond to FLRT2 stripes . A recent study reported that FLRT homophilic binding is difficult to detect in vitro due to very low binding affinity and is highly sensitive to experimental conditions ( Seiradake et al . , 2014 ) . When measured using surface plasmon resonance , homophilic binding of FLRTs was below the sensitivity of detection ( ~100 μM ) and , in SEC-MALS experiments , a minor increase in molecular weight ( from ~70 kDa to ~80 kDa ) was seen with increasing concentration , but no well-defined FLRT dimer fraction was observed . In addition , detection of FLRT-mediated homophilic cell aggregation required five days of continuous cell shaking , a time period considerably longer than standard protocols which typically monitor cell aggregation after shaking for 1−4 hours . In crystal structures of a portion of the FLRT2 and FLRT3 extracellular domains , conserved lattice contacts were observed between cis-oriented FLRT proteins ( Seiradake et al . , 2014 ) . Mutations at this interface impaired tangential spread of pyramidal neurons between adjacent cortical columns in vivo which the authors interpreted as a resulting from a defect in attractive FLRT homophilic binding . Subsequent structural and biochemical studies by Lu et al . investigating interactions between FLRT and latrophilin , a cell surface adhesion-type G-protein-coupled receptor , demonstrated that , while the FLRT mutant exhibits a decrease in dimerization via size-exclusion gel filtration , binding of the FLRT mutant to latrophilin is completely abolished ( Lu et al . , 2015 ) . These findings , in addition to the authors' inability to detect FLRT homophilic binding between cells , led them to conclude that the FLRT homodimer likely occurs in cis and that the in vivo pyramidal neuron phenotype may be due to a defect in FLRT-latrophilin binding . In our stripe assays we observe subpopulations of primary retinal neurons that are attracted to FLRT1 and FLRT3 stripes . As latrophilins are expressed in the retina ( Arcos-Burgos et al . , 2010 ) ( J . N . K . , unpublished observations ) , it will be interesting to determine whether attraction of these neurons is mediated by FLRT interactions with neuronally-expressed latrohpilin or another yet-unidentified trans interaction partner . Repulsive signaling induced by FLRT2 ligand binding to Unc5D-expressing pyramidal neurons modulates radial migration in the developing mouse cortex ( Yamagishi et al . , 2011 ) . Furthermore , FLRT3 induces repulsion of Unc5B-expressing intermediate thalamic explants ex vivo ( Seiradake et al . , 2014 ) . In both of these cases , neurons expressing Unc5s are repelled by FLRT ligand demonstrating that signaling downstream of Unc5 induces repulsion in the Unc5-expressing cell . Consistent with these findings , we observed that Unc5C-expressing retinal neurons are repelled by FLRT2 . These data suggest that , in addition to Unc5B and Unc5D , signaling downstream of Unc5C can elicit a repulsive response . We observed that FLRT2-expressing SACs and Drd4-GFP ooDSGCs are repelled by Unc5C ligand . These observations are consistent with a mechanism whereby binding of Unc5C ligand to FLRT2 receptor induces repulsive signaling in the FLRT2-expressing cell . Using a gain-of-function stripe assay , we found that FLRT2 expression is sufficient to elicit a repulsive response to Unc5C ligand . These findings suggest the intriguing possibility that a bidirectional mechanism of repulsive signaling can occur whereby FLRT2-Unc5C interactions induce repulsion in both FLRT2- and Unc5C-expressing cells . A mechanism of bidirectional signaling has been well characterized between Eph receptors and their ephrin ligands ( for review see Park and Lee , 2015 ) . Such a mechanism of Unc5C-FLRT2 mutual repulsion would provide an elegant and efficient molecular solution for directing laminar organization/restriction of both FLRT2- and Unc5C-expressing neurons into adjacent layers , S2/4 and S1/3/5 , respectively , during development of the IPL . Our future studies will be aimed at identifying and characterizing the neuronal subtype ( s ) that arborizes in S1/3/5 and expresses Unc5C to determine whether they are repelled by FLRT2 and if they are necessary to ensure laminar restriction of SACs and ooDSGCs in S2/4 . Furthermore , as additional subtypes that we have not yet characterized also express FLRT2 , other neurons have the potential to utilize FLRT2 for laminar organization either through interactions with Unc5C or other FLRT2 binding partners . IPL sublayers contain axons and dendrites of retinal neurons devoted to specific visual processing tasks ( Masland , 2012 ) . By projecting to the same sublayer , circuit partners interact specifically with each other , facilitating appropriate synaptic partner choices . A striking example is the retinal circuit that detects image motion , the so-called direction-selective ( DS ) circuit , which comprises cofasciculated arbors of SACs and ooDSGCs stratified in IPL sublayers S2 and S4 . Precise inhibitory connections from SACs onto DSGCs regulate DSGC firing in response to motion in particular directions , producing direction-selective responses ( Demb , 2007; Wei and Feller , 2011; Vaney et al . , 2012; Masland , 2012 ) . The mechanisms mediating the initial assembly of these IPL sublayers , or the co-recruitment of SAC and ooDSGC to those layers , are not known . The laminar choices of ON and OFF SACs are influenced by repulsive interactions between Plxn2 and Sema6A ( Sun et al . , 2013 ) . However , in PlxnA2-/- and Sema6A-/-mutants , most SAC dendrites still assemble in the correct sublamina and even when SACs make errors they still target to S2 or S4 ( Sun et al . , 2013 ) . This suggests that an additional molecular mechanism ( s ) functions in parallel to mediate precise laminar restriction of SACs . Here we show that SACs and at least one subset of ooDSGCs ( the Drd4-GFP population ) express FLRT2 and are repelled by Unc5C . We propose that these ( and perhaps other ) direction-selective circuit neurons become laminar-restricted in S2/4 , and/or maintain their laminar restriction once formed , due to repulsive interactions with Unc5C expressed on neighboring neurites in S1/3/5 . Definitive evidence that SACs and/or Drd4-GFP cells require FLRT2 and Unc5C for laminar targeting in S2/4 awaits genetic loss-of-function analyses . Nevertheless , our results suggest that evolution may have co-opted the same repulsive mechanism in both pre- and post-synaptic cells as a strategy for ensuring they both arborize in close spatial proximity to one another , thereby facilitating interactions between synaptic partners and limiting opportunities for inappropriate connections with neurons devoted to different visual processing tasks . Here we present an integrated systems-level approach using cell subtype-specific gene profiling to drive candidate-based , high-throughput , biochemical receptor-ligand screening . Using this approach , we demonstrate that , in addition to genetic screens , biochemical screens provide another strategy for identifying recognition proteins that play a role in facilitating the laminar organization that underlies visual function . However , this study represents merely the tip of the iceberg . Our biochemical screen sampled only a small fraction of the recognition proteins present in a limited number of neuronal subtypes in the developing IPL . Here we present data that support a model for how a single receptor-ligand interaction contributes to the laminar organization of two subtypes of neurons . However , our ultimate goal is to understand lamination on a global scale . We are optimistic that combining 1 ) inclusive gene profiling data gathered from each of the ~70 different IPL neuronal subtypes ( for which numerous more markers are now available ) with 2 ) larger-scale biochemical screens aimed at identifying the entire IPL extracellular interactome , we can elaborate a comprehensive view of how laminar organization develops in the mouse IPL . Microarrays for 13 different subtypes of IPL neurons were performed as described ( Kay et al . , 2011b; Kay et al . , 2012 ) ( NCBI Gene Expression Omnibus; accession GSE35077 ) . A variety of on-line tools and databases were used to identify differentially-expressed genes that encode transmembrane , GPI-linked and secreted proteins . The details of these methods are described in Figure 1—figure supplement 1 . Antibodies used in this study include: mouse anti-PLAP ( Thermo Fisher Scientific; Waltham , MA ) , mouse anti-human IgG1-Fc-HRP ( Serotec; Raleigh , NC ) , mouse anti-myc ( Abcam; UK , 1:1000 ) , mouse anti-FLAG ( Abcam , 1:1000 ) , chicken anti-GFP ( Abcam , 1:6000 ) , goat anti-FLRT1 ( R&D Systems; Minneapolis , MN , 1:25 ) , rabbit anti-FLRT2 ( Abcam , 1:25 ) , goat anti-FLRT3 ( R&D Systems , 1:50 ) , goat anti-Unc5A ( R&D Systems , 1:25 ) , rabbit anti-Unc5B ( Santa Cruz Biotechnology; Santa Cruz , CA , 1:200 ) , rabbit anti-Unc5C ( Santa Cruz Biotechnology , 1:50 ) , goat anti-Unc5D ( R&D Systems , 1:100 ) , mouse anti-His-HRP ( Qiagen; Germany , 1:5000 ) , goat anti-Human IgG ( H+L ) DyLight 680 ( Rockland; Limerick , PA , 1:4000 ) , guinea pig anti-vesicular acetylcholine transporter ( VAChT ) ( EMD Millipore; Hayward , CA , 1:500 ) , mouse anti-neuronal class III beta-tubulin ( Tuj1 ) ( Covance; Princeton , NJ , 1:1000 ) , rabbit anti-cocaine- and amphetamine-regulated transcript ( CART ) ( Phoenix Pharmaceuticals; Burlingame , CA , 1:2000 ) , rabbit anti-calbindin ( Swant Inc; Switzerland , 1:5000 ) . HEK293T and CHO . K1 cells were cultured according to ATCC guidelines . C57Bl/6 mice ( Harlan ) were used for wild-type retinal section immunostaining and primary retinal neuron cultures . Chat-Cre::RosaLSL-tdTomato mice were generated by crossing a tdTomato driver line ( B6 . 129S6-Chattm1 ( cre ) lowl/J × B6 . 129S6-Gt ( Rosa ) 26Sortm9 ( CAG-tdTomato ) Hze/J , Jackson Labs; Bar Harbor , ME ) with a mouse that has an IRES-Cre recombinase downstream of the endogenous choline acetyl transferase gene ( Ivanova et al . , 2010 ) . Chat-Cre::RosaLSL-tdTomato mice express fluorescent protein in SACs . Dopamine receptor D4-GFP ( Tg ( Drd4-GFP ) W18Gsat ) mice were obtained from Mutant Mouse Regional Resource Center-University of North Carolina ( https://www . mmrrc . org/catalog/sds . php ? mmrrc_id=231 ) ( Gong et al . , 2003 ) . Genotypes were identified using genomic PCR . All animal procedures were approved by the University of California , Berkeley ( Office of Laboratory Animal Care ( OLAC ) protocol #R308 ) and they conformed to the National Institutes of Health Guide for the Care and Use of Laboratory Animals , the Public Health Service Policy and the Society for Neuroscience Policy on the Use of Animals in Neuroscience Research . Retinal genes were PCR amplified from mouse retinal cDNA . Upstream and downstream primers contained NotI and SpeI or AscI sites ( Figure 1—source data 1 ) , respectively , which were used to subclone into two pCMVi vectors ( gift of John Ngai ) , pCMVi-[extracellular region]-AP-6X-His and pCMVi-[extracellular region]-Fc-6X-His . Mouse Dscam , Dscaml1 , Sdk1 and Cntn genes were subcloned from existing plasmids ( Yamagata and Sanes , 2008 ) . Full-length versions of FLRT1-3 and Unc5A-D were cloned from retinal cDNA into a derivative of the pTT3 vector ( Bushell et al . , 2008 ) and into pUB using downstream primers that introduce C-terminal myc and FLAG epitope tags , respectively . All plasmids used in this study have been submitted to Addgene ( Cabridge , MA ) . Fc-6X-His- and AP-6X-His-tagged recombinant proteins were expressed by transient transfection of HEK293T cells grown in media containing 10% Ultra-Low IgG fetal bovine serum ( Invitrogen; Carlsbad , CA ) using linear polyethylenimine ( PEI ) transfection reagent ( Thermo Fisher Scientific ) . For 15 cm plates , 32 μg of plasmid DNA and linear PEI ( Cf=40 μg/ml ) was added to 3 . 2 ml Opti-MEM ( Invitrogen ) , vortexted briefly , incubated for exactly 10 minutes at room temperature and added dropwise onto cells . Culture media was harvested 6 days post transfection . The amount of Fc- and AP-tagged proteins in the media was quantified as described previously ( Wojtowicz et al . , 2007 ) . For stripe assays , 6X-His-tagged proteins were purified using TALON metal affinity resin ( Clontech Laboratories; Mountain View , CA ) and quantified using the Bradford assay as described previously ( Wojtowicz et al . , 2004 ) . AP and Fc tags were specifically chosen for their ability to homodimerize . This forces the attached extracellular domain to adopt a dimer conformation . Further clustering of the dimerized proteins is achieved using monoclonal anti-AP and anti-Fc antibodies at limiting concentrations , thereby forcing saturation of the antibodies with a dimer bound to each of the antibody’s two binding sites – thus inducing a tetrameric conformation . The technical aspects of the binding screen were modified from Wojtowicz et al . , 2007 as follows: AP-tagged protein was used at 33 U/ul ( where a unit [U] is equivalent to the activity of 10 pg of purified calf intestinal phosphatase ( Thermo Fisher Scientific Pierce ) ) and Fc-tagged protein was used at 140 nM . This was necessary to convert the assay from one that tested Drosophila proteins expressed in Drosophila S2 cells to one that tests mammalian proteins produced in HEK293T cells . Background ( Abs650nm = 0 . 064 ) was determined using wells containing all binding reaction components with mock culture media in place of AP-tagged culture media . Background-subtracted data were deposited to the Dryad database Visser et al . , 2015 . CHO . K1 cells were co-transfected with pTT3-FLRT-myc + pGreen or pTT3-Unc5-FLAG + dsRed plasmids at a 5:1 ratio using TransIT-CHO transfection reagent ( Mirus Bio; Madison , WI ) according to the manufacturer’s protocol . Cells were incubated at 37°C and 5% CO2 overnight , harvested with trypsin for exactly 5 minutes , resuspended in aggregation media ( CHO . K1 media containing 70 U/ml DNAse I and 2 mM EGTA ) and counted . FLRT-myc/GFP and Unc5-FLAG/RFP cells ( 0 . 5 x 105 each in 250 ul ) were mixed together in a 24-well ultra-low adhesion plate ( Corning Inc; Corning , NY ) and incubated for four hours in a 37°C , 5% CO2 incubator on a belly dancer mixer at 90 rpm . Cells were diluted 1:5 in aggregation media and 100 ul was added to two 35 mm glass-bottom dishes ( MatTek Corp; Ashland , MA ) . Clusters containing >10 cells were counted using an Axiovert S100 fluorescence microscope ( Carl Zeiss; Germany ) . Microfluidic devices were designed using the AutoCAD program ( AutoDesk; San Rafael , CA ) . The design included nine groupings of ten channels . Channels were 30 μm wide , 100 μm high and separated from one another by 30 μm . Each grouping was separated by 150 μm . Microfluidic device features were fabricated using SU8 photoresist on a silicon wafer ( Stanford Foundry; Stanford University , Palo Alto , CA ) and coated with Teflon for quick feature release . Features were then transferred into polyurethane casting masters ( Smoothcast 326 ) . Devices were produced as follows: Poly-dimethyl-siloxane ( PDMS , SYLGARD ) was mixed in a 10:1 base to crosslinker ratio , poured into casting masters , degassed overnight and let cure at 37°C for a minimum of 24 hours . After release peel from the casting master , 1 . 2 mm inlet and outlet holes were punched ( Ted Pella Inc; Redding , CA ) and devices were mounted feature side up on glass slides before wrapping in aluminum foil and autoclaving for 10 minutes . Following autoclaving , devices were allowed to dry overnight at room temperature . Glass coverslips ( 12 mm Assistant-Brand , Carolina; Burlington , NC ) were washed with 70% ethanol for 7 days with ethanol changed every day and then stored in 70% ethanol . Upon removal from ethanol , coverslips were rinsed thoroughly with water , coated sequentially with 25 µg/ml poly-D-lysine ( Sigma-Aldrich; St Louis , MO ) and 50 µg/ml laminin ( Sigma-Aldrich ) . Microfluidic devices were applied to coverslips and desiccated to strengthen seal . Stripes were prepared by pulling protein solutions through microfluidic devices using a vacuum at 7 psi . Protein solutions contained 100 µg/ml purified protein ( FLRT-Fc-6X-His , Unc5-Fc-6X-His or laminin ) , mixed with 100 µg/ml BSA-TRITC or PLL-FITC ( to visualize the stripes ) . Protein solutions were incubated in devices at 37°C in a humidified chamber overnight and then wet-peeled in autoclaved milliQ water and stored in 1X PBS until use . Dissociated retinal neurons were prepared using a modified version of a protocol developed by Ben Barres ( Barres et al . , 1988 ) . Retinas from P6 ( wild type; three independent experiments ) , P2 ( Chat-Cre::RosaLSL-tdTomato; three independent experiments ) and P3 ( Drd4-GFP; two independent experiments ) mice were quickly dissected from the eyecup into cold D-PBS ( GE Healthcare HyClone; Logan , UT ) , followed by digestion in D-PBS containing ( per 500 ml ) 165 units of papain ( Worthington Biochemical; Lakewood , NJ ) , 2 mg of N-Acetyl-L-Cysteine ( Sigma-Aldrich ) , 8 µl 1N Sodium Hydroxide ( Sigma-Aldrich ) and 0 . 4 mg DNase ( Worthington Biochemical ) for 45 minutes at 37°C . The retinas were gently triturated in low-ovomucoid ( Worthington Biochemical ) then high-ovomucoid ( Worthington Biochemical ) , each trituration step followed by a 10 minute spin at 1000 rpm . Cells were resuspended in panning buffer ( 0 . 02% BSA in D-PBS , 5 µg/ml insulin ) , passed through a 40 μm cell strainer and then incubated for 30 minutes in a 15 cm petri dish coated with lectin I from Bandeiraea simplicifolia ( BSL-1 ) ( Vector Laboratories; Burliname , CA ) to deplete macrophages ( with vigorous shaking at 15 and 30 minutes to remove non-specifically attached cells ) . The supernatant was harvested , passed through a 40 μm cell strainer and 0 . 5 x 105 cells were seeded ( 1 x 105 for Drd4-GFP ) per well of 24-well plates onto glass coverslips containing purified protein stripes . Cells were seeded into 750 μl neurobasal-based culture medium ( Invitrogen ) containing 50 U/ml penicillin , 50 µg/ml streptomycin ( Invitrogen ) , 5 µg/ml insulin ( Sigma-Aldrich ) , 1 mM sodium pyruvate ( Invitrogen ) , 100 µg/ml transferrin ( Sigma-Aldrich ) , 100 µg/ml crystalline BSA ( Sigma-Aldrich ) , 60 ng/ml progesterone ( Sigma-Aldrich ) , 16 µg/ml putrescine ( Sigma-Aldrich ) , 40 ng/ml sodium selenite ( Sigma-Aldrich ) , 160 µg/ml triiodo-thyronine ( Sigma-Aldrich ) , 2 mM L-glutamine ( Sigma-Aldrich ) , B-27 Supplement ( Invitrogen ) , 50 µg/ml N-Acetyl Cysteine ( Sigma-Aldrich ) , 50 ng/ml brain derived neurotrophic factor ( BDNF ) ( Peprotech; Rocky Hill , NJ ) , 10 ng/ml ciliary neurotrophic factor ( CNTF ) ( Peprotech ) and 10 nM forskolin ( Sigma-Aldrich ) . Cultures were incubated at 37°C , 5% CO2 . Every 2–3 days , half of the volume of the media in each well was removed and replaced with fresh media . Neurons were allowed to grow for 4–8 days . For gain-of-function stripe assays , neurons were transfected in three independent experiments approximately 24 hours post seeding as follows using Attractene transfection reagent ( Qiagen ) . 0 . 2 μg of plasmid DNA and 0 . 5 μl of Attractene was added to Opti-MEM in a final volume of 60 μl , incubated 15 minutes at room temperature and added dropwise onto cells . Following transfection , cells were allowed to grow as described above . Note that for expression in primary retinal neurons , the FLRT2-myc and Unc5C-FLAG transgenes were moved from pTT3 ( vector used for cell aggregation assays ) into the pUB vector , bearing the human Ubiquitin-C promoter . For reasons that are unclear to us , transfection of the pGreen vector gave rise to an ~10% transfection efficiency as determined by the number of Tuj1+/GFP+ vs Tuj1+/GFP- neurons but transfection with pTT3-FLRT2-myc yielded hardly any FLRT2-myc+ cells . When we moved the FLRT2-myc transgene into pUB , we obtained robust FLRT2-myc expression in ~10% of neurons . As such , expression vector choice can have a significant effect on transfection results and , in this case , was crucial for the success of the experiment . Retinas were dissected from P2 , P4 and P6 wild-type mice , fixed 1 . 5 hours ( P2 and P4 ) or 45 minutes ( P6 ) in 4% paraformaldehyde at 4°C , equilibrated in 30% sucrose until retinas sank ( 2-3 hours ) , immediately embedded in O . C . T . ( Tissue-Tek ) , frozen on dry ice and sectioned immediately or stored at −80°C until sectioning . Cryostat sectioning ( 10 µm ) was performed using a Microm HM550 ( Thermo Fisher Scientific ) . Sections were blocked 1 hour in 1X PBS containing 2% normal donkey serum , 2% BSA , 4% Triton X-100 , 0 . 4% SDS ( blocking buffer ) and incubated with primary antibodies in blocking buffer overnight at 4°C . Secondary antibodies were incubated in blocking buffer for 45 minutes at room temperature . Sections were imaged using a Nikon Eclipse E600 fluorescence microscope ( Nikon; Japan ) . Primary neurons and CHO . K1 cells were fixed in ice cold 4% paraformaldehyde/1X PBS for 15 minutes , blocked 30 minutes and incubated with primary antibodies overnight at 4°C ( blocking buffer for CHO . K1 cells was 1X PBS containing 2% normal donkey serum , 2% BSA , 0 . 05% Triton X-100 ) . Secondary antibodies were incubated 2 hours at room temperature . Primary neurons were imaged using a Nikon Eclipse E600 fluorescence microscope with the exception of triple-labeling experiments ( i . e . when far red secondary antibodies were used ) and then neurons were imaged using a Zeiss LSM 710 AxioObserver confocal microscope . CHO . K1 cells were imaged using a Zeiss Axiovert S100 fluorescence microscope . Full-length Flrt2 cDNA ( NCBI accession #BC096471 ) was obtained from GE Dharmicon ( Lafayette , CO ) in vector pCMV-Sport6 . Sequencing confirmed presence of the correct insert . Plasmid was linearized at the 5’ end of the insert and antisense digoxigenin-labeled RNA probes ( DIG RNA labeling mix ) ( Roche Diagnostics; Switzerland ) were synthesized using a T7 site present in the vector ( MAXIscript kit , Thermo Fisher Scientific ) . The probes were purified on a G50 spin column ( GE Healthcare ) and hydrolysed at 60°C in bicarbonate buffer ( 40 mM NaHCO3 , 60 mM Na2CO3 ) to an expected size of 500 bp . P1 and P6 retinas were quickly dissected from the eyecup in ice-cold Hank’s balanced salt solution buffered by 10 mM HEPES , fixed in 4% paraformaldehyde/1X PBS for 90 minutes on ice , washed twice with 1X PBS , and sunk in 30% sucrose/1X PBS for 1 hour . Immediately upon sinking , tissues were frozen in TFM ( Triangle Biomedical Sciences; Durham , NC ) and stored at −80°C until sectioning at 20 µm on a cryostat . In situ hybridization was performed on retinal sections as described ( Kay et al . , 2011b; Yamagata et al . , 2002 ) . Probes were detected with peroxidase-coupled anti-digoxigenin followed by a Cy3-tyramide color reaction . After the color reaction , slides were washed at least 4 times over 2 hours in 1X PBS . They were then subjected to antibody labeling as follows . Slides were incubated in blocking solution ( 1X PBS containing 3% donkey serum and 0 . 3% Triton X-100 ) for 30 minutes at room temperature . Primary antibodies , diluted in blocking solution , were applied overnight at 4°C . Slides were washed twice in 1X PBS and stained with donkey anti-rabbit secondary antibodies conjugated to Alexa-488 ( Jackson Immunoresearch; West Grove , PA , 1:1000 ) . Retinas from four different mice were used at each age , and were stained in two independent experiments .	A nervous system comprises complex circuits of neurons connected by junctions called synapses . These connections need to develop in a highly specific manner , which means neurons need to be able to recognize one another and ‘figure out’ with which neighboring neuron or neurons they should form connections . Neurons do this by physically interacting with one another via proteins on their cell surfaces; these proteins essentially provide instructions to each of the neurons . However , for most neurons , details remain unclear about how they recognize and ‘talk’ to one another to form the connections needed to develop into working neural circuits . To form precise connections , neurons must navigate their way to the appropriate location that places them close to the other neurons with which they need to connect ( also known as their “synaptic partners” ) . In many regions of the nervous system , neurons become organized in parallel layers during development such that synaptic partners reside within the same layer . This process is called lamination and it occurs in the retina in the back of the mammalian eye . Now Visser et al . have searched for the cell surface proteins that are involved in lamination in the mouse retina . This search involved a number of different gene expression , biochemistry and cell biology-based techniques . Visser et al . identified two families of proteins that might control the lamination of many different subtypes of neurons . The findings reveal some of the molecular mechanisms that underlie the formation of neural circuits in the developing retina and suggest that a pair of synaptic partners may use the same recognition proteins to ensure that they target to the same layer . The next step will be to confirm whether the proteins identified are indeed responsible for organizing neurons into distinct layers during the development of the mouse retina .	[ "Abstract", "Introduction", "Results", "Discussion", "Conclusions", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "neuroscience" ]	2015	An extracellular biochemical screen reveals that FLRTs and Unc5s mediate neuronal subtype recognition in the retina
Exchange of extracellular cystine for intracellular glutamate by the antiporter system xc− is implicated in numerous pathologies . Pharmacological agents that inhibit system xc− activity with high potency have long been sought , but have remained elusive . In this study , we report that the small molecule erastin is a potent , selective inhibitor of system xc− . RNA sequencing revealed that inhibition of cystine–glutamate exchange leads to activation of an ER stress response and upregulation of CHAC1 , providing a pharmacodynamic marker for system xc− inhibition . We also found that the clinically approved anti-cancer drug sorafenib , but not other kinase inhibitors , inhibits system xc− function and can trigger ER stress and ferroptosis . In an analysis of hospital records and adverse event reports , we found that patients treated with sorafenib exhibited unique metabolic and phenotypic alterations compared to patients treated with other kinase-inhibiting drugs . Finally , using a genetic approach , we identified new genes dramatically upregulated in cells resistant to ferroptosis . Transporters for small molecule nutrients , including sugars , nucleotides , and amino acids , are essential for cellular metabolism and represent potential targets for drug development ( Hediger et al . , 2013 ) . System xc− is a cell-surface Na+-independent cystine–glutamate antiporter composed of the 12-pass transmembrane transporter protein SLC7A11 ( xCT ) linked via a disulfide bridge to the single-pass transmembrane regulatory subunit SLC3A2 ( 4F2hc ) ( Sato et al . , 1999; Conrad and Sato , 2012 ) . System xc− is required for normal mammalian blood plasma redox homeostasis , skin pigmentation , immune system function , and memory formation ( Chintala et al . , 2005; Sato et al . , 2005; De Bundel et al . , 2011 ) . Aberrant system xc− function is implicated in tumor growth and survival , cancer stem cell maintenance , drug resistance , and neurological dysfunction ( Okuno et al . , 2003; Buckingham et al . , 2011; Ishimoto et al . , 2011; Yae et al . , 2012; Timmerman et al . , 2013 ) ; inhibition of system xc− may prove useful in a number of therapeutic contexts . Efforts to treat gliomas and lymphomas in human patients by modulating system xc− activity with the low potency , metabolically unstable small molecule , sulfasalazine ( SAS , Gout et al . , 2001 ) , were unsuccessful ( Robe et al . , 2009 ) . While some progress has been made toward developing more potent compounds based on the SAS scaffold ( Shukla et al . , 2011 ) , the identification of system xc− inhibitors based on alternative scaffolds remains a pressing need and would be useful to test the hypothesis that system xc− inhibition is therapeutically beneficial in glioma and other contexts . We previously demonstrated that the small molecule erastin prevents Na+-independent cystine uptake ( Dixon et al . , 2012 ) , suggesting that erastin may inhibit system xc− function and represent a novel scaffold targeting this transport system . Intriguingly , treatment of some cell lines with erastin or SAS triggers an iron-dependent , non-apoptotic form of cell death , termed ferroptosis ( Dixon et al . , 2012 ) . Ferroptosis is characterized by the accumulation of intracellular soluble and lipid reactive oxygen species ( ROS ) , a process that is counteracted by the glutathione-dependent activity of the enzyme glutathione peroxidase 4 ( GPX4 ) ( Dixon and Stockwell , 2013; Yang et al . , 2014 ) . Erastin , and other ferroptosis-inducing compounds of this class , are therefore of interest both for their effects on amino acid transport and their ability to induce a novel cell death pathway . In this study , we show that erastin and its analogs specifically inhibit cystine uptake via system xc− , trigger ferroptosis in a variety of cellular contexts and act much more potently than SAS . Surprisingly , we found that the clinically approved multi-kinase inhibitor sorafenib can also inhibit system xc− and trigger ferroptosis under some conditions , an observation that may be relevant to both the anti-cancer properties and the profile of adverse events associated with this drug . We further show that small molecule inhibition of system xc− function leads to endoplasmic reticulum ( ER ) stress , as indicated by the transcriptional upregulation of genes linked to the ER stress response . The upregulation of the ER stress response gene CHAC1 ( ChaC , cation transport regulator homolog 1 ) serves as a useful pharmacodynamic marker of system xc− inhibition . Finally , we found that resistance to system xc− inhibition is correlated with dramatically increased expression of AKR1C family members that regulate the detoxification of oxidative lipid breakdown products , providing potential insight into the downstream consequences of system xc− inhibition , and the execution mechanism of ferroptosis . Erastin and SAS were previously shown to trigger ferroptosis in human HT-1080 fibrosarcoma cells grown on two-dimensional substrates with atmospheric levels of oxygen ( i . e . , 21% oxygen ) ( Dixon et al . , 2012 ) . We endeavored to generalize and validate the lethality of erastin towards cancer cells in several ways . First , we tested whether the same effects were observed in other cell types using a ‘modulatory profiling’ strategy ( Wolpaw et al . , 2011; Dixon et al . , 2012 ) . This method allows for the simplified detection and presentation of small molecule combination effects on cell viability ( modulatory effect , Me < 0 , sensitization; Me = 0 , no effect; Me > 0 , rescue ) . We observed that in five different human cancer cell lines , cell death induced by either erastin or SAS was rescued by the same canonical ferroptosis inhibitors: the iron chelator ciclopirox olamine ( CPX ) , the lipophilic antioxidants trolox and ferrostatin-1 ( Fer-1 ) , the MEK inhibitor U0126 , the protein synthesis inhibitor cycloheximide ( CHX ) and the reducing agent beta-mercaptoethanol ( β-ME ) ( Dixon et al . , 2012; Figure 1A , B ) . Thus , the ferroptotic death phenotype , whether induced by erastin or SAS , was similar in all cell lines tested . The inhibition of cell death by β-ME indicates that cell death most likely involves inhibition of system xc− function , as β-ME treatment can generate mixed disulfides taken up by other transporters , thereby circumventing the need for system xc− function ( Ishii et al . , 1981 ) . 10 . 7554/eLife . 02523 . 003Figure 1 . Cell death is triggered by erastin and related compounds in different cell lines under a variety of physiological conditions . ( A and B ) Modulatory effect ( Me ) profiles of erastin- and SAS-induced death in five different cell lines ( 143B , BJeHLT , BJeLR , Calu-1 , and HT-1080 ) in response to six different cell death inhibitors ( U0126 , Trolox , Fer-1 , CPX , CHX , β–ME ) or the vehicle DMSO . Me >0 indicates rescue from cell death . ( C and D ) Relative viability of MCTSs formed over 72 hr from HT-1080 ( C ) or Calu-1 ( D ) cells in response to erastin , RSL3 or staurosporine ( STS ) ±β-ME or ferrostatin-1 ( Fer-1 ) . Viability was assessed by Alamar blue and represents mean ± SD from three independent biological replicate experiments . Data were analyzed by two-way ANOVA with Bonferroni post-tests , p<0 . 05 , p<0 . 05 , *p<0 . 001 , ns = not significant . ( E and F ) Viability of HT-1080 ( E ) and DU145 ( F ) cells cultured under 1% or 21% O2 levels in response to erastin ( 5 μM ) ±Fer-1 ( 1 μM ) or CPX ( 5 μM ) . Viability was assessed by Alamar blue and represents mean ± SD from three independent biological replicate experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 003 Next , we sought to test whether the lethal mechanisms of action of erastin and SAS were influenced by cell growth architecture . Specifically , we tested whether the ferroptotic lethal mechanism could be activated in multicellular tumor spheroids ( MCTSs ) , three-dimensional cellular aggregates proposed to recapitulate key aspects of the structural and metabolic heterogeneity observed in tumor fragments and micrometastases ( Friedrich et al . , 2009 ) . We grew MCTSs from HT-1080 and Calu-1 cells for 72 hr and then investigated the effects of erastin ±β-ME or ±Fer-1 on MCTS growth and viability . For comparison , we also tested the growth inhibitory effects of ( 1S , 3R ) -RSL3 ( hereafter RSL3 ) , a small molecule that triggers ferroptosis by inhibiting GPX4 , which is downstream of system xc− in the ferroptotic cascade ( Yang et al . , 2014 ) , as well as staurosporine ( STS ) , which triggers apoptosis . We observed that HT-1080 and Calu-1 MCTSs were killed by erastin and RSL3 ( Figure 1C , D ) . The effects of both erastin and RSL3 were rescued by Fer-1 , while β-ME suppressed the lethality of erastin , but not of RSL3 , as expected ( Figure 1C , D ) . Neither β-ME nor Fer-1 modulated the effects of STS on MCTS growth or viability ( Figure 1C , D ) . These observations indicate that erastin , as well as RSL3 , are able to trigger ferroptosis in a similar manner in both two- and three-dimensional culture conditions . Finally , given that erastin triggers an oxidative form of cell death , we tested whether the lethality of erastin was affected by growth in low ( 1% ) vs high ( 21% ) levels of O2 . Cells from two different erastin-sensitive cancer cell lines ( HT-1080 and DU-145 ) were grown for 24 hr under low or high O2 levels and then treated for a further 24 hr with various agents , prior to the analysis of cell death . We observed that compared to DMSO-treated cells , erastin ( 5 μM ) -treated cells were killed under both high and low O2 conditions with little ( DU-145 ) or no ( HT-1080 ) difference in lethality ( Figure 1E , F ) . In both cell lines , erastin-induced death was suppressed by both Fer-1 ( 1 μM ) and CPX ( 5 μM ) ( Figure 1E , F ) , indicating that the same lethal mechanism ( i . e . , ferroptosis ) was responsible for cell death under both high and low O2 conditions . Thus , even under relatively low O2 conditions , it is still possible for erastin to activate the ferroptotic mechanism . The ability to modulate system xc− activity may be clinically useful , but requires small molecule inhibitors with suitable pharmacological properties that are also specific for this antiporter ( Gorrini et al . , 2013 ) . Erastin treatment ( 5 μM ) completely abolished the Na+-independent uptake of radiolabelled [14C]-cystine in both HT-1080 fibrosarcoma and Calu-1 lung carcinoma cancer cells , as did sulfasalazine ( SAS ) at 100-fold higher concentrations ( 500 μM ) ( Figure 2A ) . Conversely , erastin and SAS had no effect on Na+-independent [14C]-phenylalanine uptake ( Figure 2B ) . An excess of cold D-phenylalanine did suppress [14C]-phenylalanine uptake , confirming that Phe transport was inhibitable under these assay conditions ( Figure 2B ) . Thus , in HT-1080 and Calu-1 cells , erastin and SAS block system xc− ( SLC7A11 + SLC3A2 ) -mediated cystine uptake selectivity over other transport systems and amino acids , such as system-L- ( SLC7A5 + SLC3A2 ) -mediated Phe uptake . 10 . 7554/eLife . 02523 . 004Figure 2 . Erastin inhibits system xc− function potently and specifically . ( A and B ) Na+-independent uptake of 14C-cystine ( A ) and 14C-L-phenylalanine ( Phe ) ( B ) over 5 min in HT-1080 and Calu-1 cells treated with erastin or SAS . D-Phe was included as a positive control in B . ( C ) Structure and lethal potency ( EC50 in HT-1080 cells ) of erastin and the inactive erastin analog erastin-A8 . ( D ) Dose-dependent inhibition of glutamate release by erastin and erastin-A8 ( Era-A8 ) . ( E ) Glutamate release ±erastin in HT-1080 cells in which SLC7A11 was silenced for 48 hr using two independent siRNAs . ( F ) SLC7A11 mRNA levels assayed using RT-qPCR in si-SLC7A11-transfected cells . ( G ) Glutamate release in response to erastin , SAS , RSL3 artesunate and PEITC , ±beta-mercaptoethanol ( β-ME ) . ( H ) Dose-response analysis of glutamate release from HT-1080 and Calu-1 cells in response to erastin and SAS . All data are from three independent biological replicates . Data are presented as mean ± SD . Data in A and B are normalized to DMSO controls ( set to 100% ) . Data in A , B , E and G were analyzed by ANOVA with Bonferroni post-tests , p<0 . 05 , **p<0 . 001 , ns = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 00410 . 7554/eLife . 02523 . 005Figure 2—figure supplement 1 . Monitoring system xc− activity by following glutamate release . ( A ) Overview of the assay design . Glutamate released by the cell is detected by an enzyme-linked reaction . Erastin and SAS inhibit glutamate release from system xc− , but not other transporters . System xc− is one of several transporters that can release glutamate , including system XAG and others . ( B ) SLC7A5 expression was silenced in HT-1080 cells for 48 hr using two independent siRNAs and then glutamate release was assayed ±erastin . ( C ) SLC7A5 mRNA levels in HT-1080 transfected as in ( B ) . Data in B and C represent mean ±SD from three independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 005 We confirmed the ability of erastin and SAS to inhibit system xc− using an enzyme-coupled fluorescent assay that detects glutamate release into Na+-containing culture medium ( Figure 2—figure supplement 1A ) . We validated this assay in three ways . First , we showed that erastin ( 1 ) inhibited glutamate release , while a non-lethal ( Yagoda et al . , 2007 ) erastin analog lacking the p-chlorophenoxy moiety ( erastin-A8 , 2 ) did not ( Figure 2C , D ) . Second , we showed that both erastin treatment and silencing of SLC7A11 with either of two independent siRNAs resulted in a significant , quantitatively similar inhibition of glutamate release ( Figure 2E , F ) ; silencing of the system L transporter subunit SLC7A5 using two independent siRNAs had no effect on basal or erastin-mediated inhibition of glutamate release ( Figure 2—figure supplement 1B , C ) . Third , we found that only erastin and SAS inhibited glutamate release , while , as expected , RSL3 , artesunate and PEITC did not; while artesunate and PEITC induce iron-dependent cell death , neither are known to inhibit system xc− or induce ferroptosis ( Trachootham et al . , 2006; Hamacher-Brady et al . , 2011; Dixon et al . , 2012; Figure 2G ) . Thus , the above results suggest that both erastin and SAS specifically inhibit SLC7A11-dependent system xc− function . The ability of erastin to specifically inhibit cystine uptake via system xc− is further supported by recent metabolomic profiling data ( Skouta et al . , 2014; Yang et al . , 2014 ) and gene expression experiments described below . In light of disappointing clinical results using SAS ( Robe et al . , 2009 ) , it is desirable to identify potent inhibitors of system xc− with favorable pharmacological properties . Using the glutamate release assay to quantify inhibition of system xc− activity , we found that erastin was ∼2500 times more potent than SAS as an inhibitor of system xc− function in both HT-1080 and Calu-1 cells ( HT-1080: erastin IC50 = 0 . 20 µM , 95% C . I . 0 . 11–0 . 34 µM , SAS IC50 = 450 µM , 95% C . I . 280–710 µM; Calu-1: erastin IC50 = 0 . 14 µM , 95% C . I . 0 . 081–0 . 21 µM , SAS IC50 = 460 µM , 95% C . I . 350–590 µM ) ( Figure 2H ) . Thus , the erastin scaffold may afford a more suitable starting point than SAS for the development of potent and selective inhibitors of system xc− function . We hypothesized that it would be possible to improve further the potency of the erastin scaffold through targeted synthesis . Towards this end , we undertook a search for more potent analogs , beginning with an achiral analog ( 3 , Yang et al . , 2014 ) that lacked the methyl group at the chiral center , and that had an isoproproxy substituent in place of the ethoxy group on erastin ( 1 ) ( Figure 3A ) . This compound ( 3 ) was more synthetically accessible , but otherwise exhibited a similar lethal potency as erastin in HT-1080 cells . We synthesized 19 analogs of 3 ( Supplementary file 1 , Figure 3A . Data also available as ‘Extended Materials and Methods’ from Dryad data Repository [Dixon et al . , 2014] ) , and tested each in HT-1080 cells in a 10-point , twofold dose-response assay for lethal potency and efficacy ( Figure 3B ) . To assess in each case whether lethality involved inhibition of system xc− and induction of ferroptosis , as opposed to the induction of another form of death , experiments were performed ±β-ME . To assess in a parallel assay the correlation between lethal potency and inhibition of system xc− activity , we examined glutamate release using a high throughput , 96-well Amplex Red assay system in human CCF-STTG1 astrocytoma cells ( Figure 3B ) . Overall , the lethal potency in HT-1080 cells was found to correlate significantly with the degree of system xc− inhibition observed in CCF-STTG1 cells ( Pearson R2 = 0 . 86 , p<0 . 0001 ) . Of note , we observed that glutamate release in CCF-STTG1 cells was in general more sensitive to erastin and analogs than HT-1080 cells ( compare Figure 3B to Figure 2H ) . These results support the hypothesis that the ability of erastin analogs to trigger ferroptosis is quantitatively linked to their ability to inhibit system xc− function . 10 . 7554/eLife . 02523 . 008Figure 3 . Structure activity relationship ( SAR ) analysis of erastin . ( A ) Structures of 20 erastin analogs . ( B ) Lethal EC50 for each analog determined in HT-1080 cells in a 10-point , twofold dilution assay , starting at a high concentration of 20 μM , ±β-ME ( 18 μM ) . Data represent mean and 95% confidence interval ( 95% C . I . ) from three independent biological replicate experiments . Also reported are IC50 values for inhibition of glutamate release as determined in CCF-STTG1 cells . These data represent the average of two experiments . All values are in μM . ND: not determined . ( C ) Dose-response curves for selected erastin analogs ( 3 , 14 and 21 ) in BJeLR and BJeH cells . Data represent mean ± SD from three independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 008 We investigated the above data set in more detail for insights into the erastin structure activity relationship . Erastin's quinazolinone core ( Region A ) is found in a number of biologically active compounds and is considered to be a ‘privileged’ scaffold ( Welsch et al . , 2010 ) . Modifications to this region ( 4–10 ) , including substitution of the quinazolinone for quinolone ( 4 ) or indole ( 5 ) , obtained using a Meth-Cohn quinoline synthesis ( Supplementary file 1 ) , resulted in moderate to severe losses of lethal potency compared to 3 , suggesting that the quinazolinone core scaffold is essential for the lethality of erastin . Modifications to the piperazine linker ( Region B , 11–12 ) were not tolerated , with 12 being completely inactive in both the HT-1080 lethality and CCF-STTG1 glutamate release assays . We speculate that rigidification of this portion of the scaffold is essential for activity and that an increase in the number of rotatable bonds in this region results in a higher entropic cost of binding , decreasing lethal potency . Single atom changes to the acetoxy spacer ( Region C , 13–15 ) were likewise poorly tolerated , resulting in significant losses in potency that correlated with reduced inhibition of system xc− activity . Strikingly , subtle modifications to Region D , including replacement of the chlorine with a fluorine ( 16 ) , replacement of the para-chloro substituent with a meta-chloro ( 17 ) or elimination of it altogether ( 18 ) reduced or abrogated both lethality and system xc− inhibitory activity . As suggested by the weakened potency of 16 and 17 , and the inactivity of 18 , the chlorine atom may make a key halogen bonding interaction with the surrounding environment ( Wilcken et al . , 2012 ) that is essential for binding . Finally , modifications to Region E , including addition of a bromo group ( 20 ) , a phenyl ( 21 ) or a furanyl substituent ( 22 ) resulted in fivefold or greater improvements in lethal potency that were mirrored by fivefold to 10-fold improvements in the inhibition of system xc− activity compared to 3; the most potent compound , 21 , inhibited glutamate release with below 5 nM potency in the CCF-STTG1 assay . Crucially , these more potent compounds potently triggered lethality in HT-1080 cells via ferroptosis , as death was fully suppressed by β-ME . Previously , we have shown that erastin and lethal analogs thereof demonstrate selective lethality towards human BJ fibroblasts engineered to express human telomerase , SV40 large and small T antigen , and oncogenic HRASV12 ( BJeLR ) compared to isogenic cells expressing only telomerase ( BJeH ) ( Dolma et al . , 2003; Yang et al . , 2014 ) . We tested the most potent lethal analog ( 21 ) , along with the parent compound ( 3 ) and a representative non-lethal analog ( 14 ) , in these cell lines . While 14 was inactive , we found that both lethal analogs ( 21 and 3 ) retained selectivity towards BJeLR vs BJeH cells ( Figure 3C ) . Consistent with the pattern of lethality observed in HT-1080 cells , 21 was a more than 20-fold more potent lethal molecule compared to 3 ( BJeLR EC50 of 22 nM [95% C . I . 20-25 nM] vs 490 nM [95% C . I . 350-690 nM] , respectively ) . Together , these results demonstrate that it is possible to improve the potency of the erastin scaffold substantially while retaining oncogenic RAS-selective lethality , especially via modifications to Region E . The combination of these new structural insights , together with complementary results concerning modifications that enhance the metabolic stability of erastin ( Yang et al . , 2014 ) , may result in suitable compounds for clinical studies . Given the above results , we hypothesized that the effects of erastin were due entirely to inhibition of system xc− function and the consequent depletion of cystine ( and ultimately cysteine ) from the intracellular milieu . If so , co-treatment with β-ME should reverse all effects resulting from erastin treatment . To test this hypothesis in a global manner , we examined patterns of changes in the transcriptome using RNA sequencing ( RNA-Seq ) of mRNA harvested from HT-1080 cells treated for 5 hr with DMSO , erastin ( 10 µM ) , β-ME ( 18 µM ) or erastin + β-ME . From two independent biological replicates , we obtained an average of ∼30 . 5 million unique mapped reads and 11 , 867 unique transcripts ( with Fragments Per Kilobase of exon per Million reads [FPKM] ≥1 in both replicates ) per condition . After data processing and averaging of replicates , we identified 33 mRNAs with two-fold more counts ( ‘up-regulated’ ) and four mRNAs with twofold fewer counts ( ‘down-regulated’ ) in erastin-treated samples vs DMSO-treated controls ( Figure 4A , B; Data available as ‘Data Package 1’ from Dryad data Repository [Dixon et al . , 2014] ) . In support of the hypothesis , erastin-induced changes in mRNA expression were reversed by co-treatment with β-ME for all 33 up-regulated genes ( Mann–Whitney test , p<0 . 0001 ) and for each of the four down-regulated genes . These results suggest that the effects of erastin on cellular physiology detectable at the level of mRNA expression are due to depletion of intracellular cystine , arising as a consequence of inhibition of system xc− function . 10 . 7554/eLife . 02523 . 006Figure 4 . Analysis of erastin effects using RNA-Seq . ( A and B ) List of genes upregulated ( A ) and downregulated ( B ) by erastin treatment , as detected in HT-1080 cells using RNA Seq . The number of fragments per kilobase of exon per megabase of sequence ( FPKM ) was counted and is expressed as a fold-change ratio between the different conditions . E/D: Erastin/DMSO expression ratio . E+β-ME/D: Erastin+β-ME/DMSO ratio . ATP6V1G2: ATP6V1G2-DDX39B read-through transcript . Data represent the average of two independent biological replicates for each condition . ( C and D ) mRNA expression level of CHAC1 determined by RT-qPCR in HT-1080 and Calu-1 cells in response to erastin ±β-ME treatment for 5 hr . Data are from three independent biological replicates and represented as mean ± SD and were analyzed by one-way ANOVA with Bonferroni post-tests , p<0 . 01 , p<0 . 001 , ns = not significant . In D significance is indicated relative to the DMSO control . ( E ) CHAC1 mRNA levels in 13 different erastin-sensitive cell lines treated with erastin or STS ( 6 hr ) . Results in E were analyzed using the Kruskal–Wallis test , *p<0 . 001 , ns = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 00610 . 7554/eLife . 02523 . 007Figure 4—figure supplement 1 . Analysis of ER stress in response to erastin treatment . ( A ) Western blot analysis of erastin ( 5 μM , 7 hr ) —sorafenib ( 10 μM , 12 hr ) —and thapsigargin ( 10 ng/ml , 6 hr ) -treated cells reveals changes associated with ER stress including phosphorylation of eIF2alpha and up-regulation of ATF4 protein levels relative to unphosphorylated forms and total protein respectively . : non-specific band in the ATF4 blot . ( B ) Splicing of XBP-1 mRNA analyzed by PCR and agarose gel electrophoresis . Treatment conditions are as in ( A ) . ( C ) Viability of HT-1080 cells over time treated with DMSO or erastin ± actinomycin D or cycloheximide ( CHX ) , as indicated . Cell counts were measured using a Vi-Cell analyzer . ( E and F ) CHAC1 mRNA levels measured in HT-1080 cells in response to the indicated erastin analogs . ( E ) AE , PE and MEII were reported previously in Yang et al . ( 2014 ) . Results presented in A and B are representative of three independent experiments . Data shown in ( D and E ) is the average of three independent biological replicates , and in ( C and F ) the average of two replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 007 We noted that several of the genes upregulated by erastin were associated with activation of the eIF2alpha-ATF4 branch of the ER stress response pathway ( e . g . , ATF3 , DDIT3 , DDIT4 [Jiang et al . , 2004; Whitney et al . , 2009] ) . Consistent with this , we observed that the 33 up-regulated genes were significantly enriched for GO Biological Process terms related directly to the ER stress and unfolded protein responses ( GO:0034976 , response to endoplasmic reticulum stress , p=8 . 0 e−11; GO:0006987 , activation of signaling protein activity involved in unfolded protein response , p=1 . 3 e−9; GO:0032075 , positive regulation of nuclease activity , p=1 . 5 e−9 ) . The eIF2alpha-ATF4 branch of the ER stress/unfolded protein response can be upregulated by amino acid depletion ( Harding et al . , 2003 ) , which we hypothesize is linked to intracellular cysteine depletion downstream of system xc− inhibition by erastin . We investigated further the connection between erastin treatment and activation of the eIF2alpha-ATF4 pathway and observed that , relative to DMSO-treated controls , erastin treatment ( 5 μM , 7 hr ) resulted in phosphorylation of eIF2alpha and up-regulation of ATF4 at the protein level ( Figure 4—figure supplement 1A ) . We saw no evidence for enhanced splicing of the XBP1 mRNA , which provides a readout for activation of a parallel ER stress response pathway ( Figure 4—figure supplement 1B ) . In HT-1080 cells , co-treatment with the transcriptional inhibitor actinomycin D ( 1 μg/ml ) inhibited erastin-induced changes in gene expression ( see below ) and delayed but did not prevent erastin-induced cell death ( Figure 4—figure supplement 1C ) . Thus , while blockade of system xc− by erastin ( and other agents , see below ) can trigger a robust transcriptional signature indicative of ER stress , it is doubtful that this transcriptional response is essential for the lethality observed following erastin treatment . Pharmacodynamic ( PD ) markers would be useful to determine when cells are responding to system xc− inhibition , such as in response to erastin treatment . We therefore explored the RNA-Seq profiles for suitable candidate PD makers; the most highly up-regulated gene observed in erastin-treated HT-1080 cells by RNA-Seq was CHAC1 ( ∼24-fold , Figure 4A ) , an ER stress-responsive gene known to be upregulated downstream of ATF4 ( Gargalovic et al . , 2006; Mungrue et al . , 2009 ) . We validated these results by RT-qPCR using fresh samples prepared from HT-1080 and Calu-1 cells , and confirmed that CHAC1 up-regulation was fully reversible by co-treatment with β-ME ( Figure 4C ) . CHAC1 mRNA upregulation was observed in response to seven different active erastin analogs described above ( 3 , 20–22 ) and in a recent publication ( AE , PE and MEII , [Yang et al . , 2014] ) ; low levels of CHAC1 upregulation were also observed in response to two less potent analogs ( 14 , 16 , both of which nonetheless retain some ability to inhibit system xc− function , see Figure 3B ) , suggesting that the induction of ER stress and CHAC1 upregulation may be more sensitive to the inhibition of system xc− than cell viability ( Figure 4—figure supplement 1E , F ) . We examined the specificity of the above response—we observed transcriptional upregulation of CHAC1 following treatment with system xc− inhibitors ( erastin and SAS ) , but not in response to RSL3 , artesunate , rotenone or buthionine sulfoximine ( BSO ) , agents that induce oxidative stress , but that do not inhibit system xc− ( Figure 4D , Figure 4—figure supplement 1D ) , suggesting that CHAC1 upregulation can specifically indicate agents that inhibit system xc− function vs those that trigger redox stress by other means . Upregulation of CHAC1 in erastin-treated HT-1080 cells was prevented by co-treatment with the transcriptional inhibitor actinomycin D as well as the protein synthesis inhibitor CHX , suggesting that CHAC1 mRNA upregulation downstream of erastin treatment requires new transcription and translation ( Figure 4—figure supplement 1D ) . We next tested the generality of CHAC1 upregulation in response to erastin , and observed that across a panel of 13 cancer cell lines , treatment with erastin , but not the apoptosis-inducer STS , resulted in a significant increase in CHAC1 expression ( Figure 4E ) . Thus , erastin can trigger a number of changes in cell physiology specifically linked to cystine depletion , and CHAC1 up-regulation could be a useful transcriptional PD marker for exposure to erastin and other agents that deplete cells of cystine or cysteine . This may be useful in testing other potential inhibitors of system xc− function . Inhibition of system xc− activity and/or glutathione depletion may be useful in combination with other therapies to selectively target specific tumor types or sensitize them to other agents ( Dai et al . , 2007; Ishimoto et al . , 2011; Timmerman et al . , 2013 ) . We therefore used a modulatory profiling strategy to test whether the lethality of 20 mechanistically diverse compounds could be enhanced in both A549 and HCT-116 cells by system xc− inhibition using erastin ( 10 μM ) , or by glutathione depletion using BSO ( 2 . 5 mM ) . Overall , we observed that the modulatory effect ( Me ) values for the test compounds clustered around zero ( i . e . , additive enhancement of death ) , with the exception of RSL3 , phenylarsine oxide ( PAO ) and sorafenib ( Figure 5A , B ) . RSL3 induces ferroptosis through a mechanism independent of system xc− ( see above ) , while PAO binds vicinal thiols and has lethal activity that is opposed by glutathione-dependent enzymes ( Lillig et al . , 2004 ) , rationalizing the synergistic effects observed with these two compounds . The observation that sorafenib ( BAY 43-9006 , Nexavar ) , a multi-kinase inhibitor clinically approved for the treatment of renal cell carcinoma ( Wilhelm et al . , 2006 ) , could synergize with erastin or BSO treatment was unexpected , and suggested that sorafenib could affect the ferroptosis pathway . 10 . 7554/eLife . 02523 . 009Figure 5 . Identification of sorafenib as an inhibitor of system xc− . ( A and B ) Modulatory profiling of ( A ) A549 and ( B ) HCT-116 cells in response to either buthionine sulfoximine ( BSO ) or erastin ±20 different lethal compounds ( see ‘Materials and methods’ for the full list ) . ( C ) Viability of HT-1080 cells treated for 24 hr with ferroptosis inhibitors ( β-ME , Fer-1 , DFO ) ±sorafenib , erastin or STS . ( D ) Quantification of the inhibition of glutamate release by sorafenib , erastin and imatinib ±Fer-1 . ( E and F ) HT-1080 cells treated with control vehicle ( DMSO ) or sorafenib ( 10 μM ) for 18 hr prior to the assay . ( E ) Total glutathione levels measured using Ellman's reagent . ( F ) Lipid ROS levels assayed using C11-BODIPY 581/591 . ( G ) Viability of HT-1080 cells treated for 24 hr with erastin , sorafenib , nilotinib , masitinib or imatinib ±β-ME or Fer-1 . Data in C–G represent mean ± SD from at least three independent biological replicates . Cell viability in C and G was quantified by Alamar blue . Data in C , D and G were analyzed by one- and two-way ANOVA with Bonferroni post-tests; data in E and F were analyzed using Student's ttest , p<0 . 05 , p<0 . 01 , **p<0 . 001 , ns = not significant relative to the indicated treatments . In ( D ) , none of the comparisons between DMSO and Fer-1 treated samples were significant ( p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 00910 . 7554/eLife . 02523 . 010Figure 5—figure supplement 1 . Effect of sorafenib on cell viability . ( A ) Glutamate release from same cell lines in response to erastin or sorafenib . ( B ) Viability of five cell lines ( HT-1080 , TC-32 , 143B , Calu-1 and U2OS ) in response to 48 hr treatment with increasing doses of erastin or sorafenib ± DMSO ( squares , black ) or 5 μM Fer-1 ( triangles , purple ) . Data shown is the average of three independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 010 We investigated this hypothesis as follows . First , we found that in HT-1080 cells sorafenib ( 10 μM , 24 hr ) treatment-induced cell death was significantly inhibited by the known small molecule ferroptosis suppressors β-ME , Fer-1 and DFO ( Dixon et al . , 2012 ) ; these same inhibitors suppress erastin ( 10 μM , 24 hr ) -induced ferroptotic death , but not STS-induced apoptotic death ( Figure 5C ) . Thus , sorafenib alone can trigger ferroptosis . The observed suppression of sorafenib-induced death by β-ME was striking and immediately suggested that sorafenib , like erastin and SAS , could be acting to inhibit system xc− function . Indeed , we observed that sorafenib , like erastin , but not the kinase inhibitor imatinib , resulted in a dose-dependent inhibition of system xc− function , as assessed using the glutamate release assay in HT-1080 cells ( Figure 5D ) . The ability of sorafenib and erastin to suppress system xc− activity was not inhibited by co-treatment with Fer-1 , demonstrating that this effect is upstream of Fer-1-sensitive ROS accumulation ( Figure 5D ) , as expected . Likewise , similar to erastin and SAS , we observed a robust transcriptional upregulation of CHAC1 in HT-1080 cells in response to sorafenib treatment ( Figure 4D ) . We also observed upregulation of the same biochemical markers of ER stress observed previously with erastin , namely phosphorylation of eIF2alpha and increased levels of ATF4 , without any change in XBP1 splicing ( Figure 4—figure supplement 1A , B ) . Finally , as observed previously with erastin treatment ( Dixon et al . , 2012; Yang et al . , 2014 ) , we found that sorafenib treatment ( 10 μM , 18 hr ) of HT-1080 cells significantly depleted total glutathione and resulted in the accumulation of lipid peroxides as detected by flow cytometry using C11-BODIPY 581/591 ( Figure 5E , F ) . Together , these results suggest that , like erastin , sorafenib inhibits system xc−-mediated cystine import , leading to ER stress , glutathione depletion and the iron-dependent accumulation of lipid ROS . To test the generality of these results , we examined the ability of sorafenib to inhibit system xc− activity and trigger ferroptosis in additional cell lines . Consistent with the initial results , in all five cell lines examined , we observed that sorafenib and erastin treatments ( 20 μM ) caused a comparable inhibition of system xc− function , as assessed by glutamate release ( Figure 5—figure supplement 1A ) . However , unlike erastin , we observed that sorafenib triggered Fer-1-suppressible ferroptosis in HT-1080 cells only within a narrow concentration window ( sorafenib EC50 = 18 μM , sorafenib+Fer-1 EC50 = 43 μM ) , before causing Fer-1-insensitive death at higher concentrations ( Figure 5—figure supplement 1B ) . In the four other cells lines , we observed only slight ( 143B ) or non-existent ( TC32 , Calu-1 , U2OS ) differences in sorafenib EC50 values either with or without Fer-1 ( Figure 5—figure supplement 1B ) . Thus , while sorafenib can inhibit system xc− activity robustly , this manifests as a ferroptotic cell death phenotype only over a narrow range of concentrations; sorafenib appears to trigger additional lethal mechanisms that act in parallel to the ferroptotic response at higher concentrations . Sorafenib could conceivably inhibit system xc− activity by modulating the activity of a kinase that controls system xc− function , or through ‘off-target’ modulation of a non-kinase target ( e . g . , SLC7A11 itself , a system xc− regulatory protein , or a more indirectly related target ) . We took two approaches in an attempt to address this question . First , we examined the effects of functionally-related kinase inhibitors . The global pattern of kinase inhibition by sorafenib against 300 purified kinase domains is highly similar to that of nilotinib , masitinib , and imatinib ( Anastassiadis et al . , 2011 ) , yet none of these agents appear to trigger ferroptosis , as defined by sensitivity to ferroptosis-specific cell death inhibitors β-ME and Fer-1 ( Figure 5G ) . Thus , even kinase inhibitors with targets similar to those of sorafenib do not necessarily trigger ferroptosis . Second , we attempted to dissociate the ability of sorafenib to trigger ferroptosis vs other lethal mechanisms . To do this we synthesized and tested a set of 87 sorafenib analogs for the ability to trigger β-ME- and Fer-1-suppressible death in HT-1080 cells in twofold , 10-point dilution series assays , starting at a highest concentration of 20 or 40 μM . In summary , while many analogs retained lethal activity ( three examples of lethal compounds are shown in Figure 6A ) , none of the 87 analogs could trigger ferroptosis with enhanced selectivity for β-ME- and Fer-1-suppressible death over other lethal mechanisms compared to the parent compound itself . Several of the analogs that we synthesized were unable to trigger death ( four example compounds are shown in Figure 6B ) . These analogs ( SRS13-67 , SRS14-98 , SRS13-48 , and SRS15-11 ) contain modifications predicted to disrupt atomic interactions essential for the binding of sorafenib to kinase targets such as BRAF , including burial of the -CF3 group in a hydrophobic pocket and hydrogen bonding with the urea group ( Lowinger et al . , 2002; Wan et al . , 2004 ) . Conversely , the active analogs shown here ( Figure 6A ) mostly retain these features . Thus , one hypothesis is that sorafenib triggers ferroptosis by inhibiting an unknown kinase whose activity is necessary for constitutive system xc− activity . Alternatively , sorafenib could modulate system xc− activity by interacting with a non-kinase target that harbors a binding pocket resembling that found within the active site of sorafenib-sensitive kinases . 10 . 7554/eLife . 02523 . 011Figure 6 . Analysis of sorafenib analog function . ( A and B ) Sorafenib and 87 sorafenib analogs were prepared and tested in HT-1080 cells for the induction of cell death ( EC50 ) and the suppression of this cell death by β-ME ( 18 μM ) and Fer-1 ( 1 μM ) over 48 hr . Representative data are shown for three analogs that retained lethal activity ( SRS13-45 , SRS13-60 , SRS13-47 ) and four analogs that lost lethal activity ( SRS13-67 , SRS13-48 , SRS14-98 , SRS15-11 ) in the cell-based assay . ( C ) Glutamate release in response to sorafenib and select analogs . ( D ) DEVDase ( caspase-3/7 ) activity in response to sorafenib and select analogs as measured by the cleavage of a fluorescent rhodamine-linked substrate . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 011 To further bolster the working model , we evaluated four of the above sorafenib analogs ( two lethal , two non-lethal ) for their effects on system xc− function , using the glutamate release assay , and on the induction of apoptotic death , using a fluorogenic caspase-3/7 substrate cleavage assay . Consistent with the above data , two lethal sorafenib analogs ( SRS13-45 and SRS13-60 ) significantly inhibited glutamate release , while two non-lethal analogs ( EC50 >40 μM ) , SRS13-67 and SRS14-98 , did not ( Figure 6C ) . The ability of these analogs to induce caspase-3/7 ( DEVDase ) activity did not vary significantly from one another or from DMSO-treated controls , as compared to cells treated with a positive control inducer of apoptosis , the proteasome inhibitor MG132 ( Figure 6D ) . Together , these results may help account for other reports of caspase-independent , sorafenib-induced death ( Panka et al . , 2006; Katz et al . , 2009 ) and support the hypothesis that sorafenib triggers ferroptotic cell death via , possibly indirect , inhibition of system xc− . Sorafenib is a clinically-used drug for the treatment of renal cell carcinoma and other indications . We speculated that the ability of sorafenib to trigger both ferroptotic and non-ferroptotic death would result in a unique spectrum of clinical observations in patients treated with sorafenib compared to other kinase inhibitors . Specifically , a subset of patients for any drug typically experience adverse events that are dependent on the drug mechanism . Thus , we speculated that the pattern of adverse events could report on the similarity or differences of mechanisms across drugs . Previously , we applied a large-scale statistical analysis to the Food and Drug Administration Adverse Event Reporting System ( FAERS ) to systematically identify drug effects and interactions ( Tatonetti et al . , 2012 ) . Here , we sought to use this approach to discover correlations between sorafenib exposure and human health unique to this drug . First , we identified those reports of patients with exposure to sorafenib , and a set of reports that could serve as controls; for this , we used a high dimensional propensity-score model previously validated for this use in the FAERS that has been shown to mitigate confounding bias and to improve the accuracy of statistical estimates ( Tatonetti et al . , 2012 ) . Using disproportionality analysis ( Bate and Evans , 2009 ) , we identified adverse drug effects for sorafenib and for a set of comparison kinase-targeted drugs for which sufficient data was available in our data ( dasatinib , erlotinib , gefitinib , imatinib , lapatinib , and sunitinib ) , none of which ( at 20 μM ) were found to trigger ferroptosis in HT-1080 cells or inhibit system xc− activity as assayed using the glutamate release assay ( Figure 7—figure supplement 1 ) . We then filtered out effects that could be attributed to chemotherapy and grouped the drug-effect associations by the physiological system that the adverse event affected . For example , cardiovascular-related adverse events were grouped into the cardiovascular system category . For each kinase inhibitor , we counted the number of reports in each of 20 physiological system categories that involved the above drugs . We then compared this number to the counts obtained from the selected control cohorts . Using a Fisher's exact test , we evaluated significance of associations of each kinase inhibitor to each physiological system category , and then we plotted these data as a heatmap , clustering the data in an unsupervised , hierarchical manner using the computed p values ( Figure 7 ) . 10 . 7554/eLife . 02523 . 012Figure 7 . Summary of adverse events reported with sorafenib and other kinase inhibitors . Analysis of adverse events across 20 physiological system categories associated with kinase inhibitor treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 01210 . 7554/eLife . 02523 . 013Figure 7—figure supplement 1 . Test of the ability of various kinase inhibitors to induce ferroptosis and inhibit system xc− . ( A ) Viability of HT-1080 cells in response to 48 hr treatment with erastin ( as a control ) or various kinase inhibitors ± β-ME and Fer-1 . ( B ) Ability of kinase inhibitors to inhibit glutamate release in HT-1080 cells . Data represent the average of three experiments for all compounds except dasatinib in ( A ) , which was tested only once . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 013 In this analysis , we observed that sorafenib treatment was associated with a significant number of adverse events in 15/20 physiological system categories , the most observed for any drug . A subset of the adverse events uniquely associated with sorafenib compared to all other kinase inhibitor drugs included musculoskeletal , nervous system , and pathological disorders as well as hemorrhage; this pattern was not observed for patients treated with sunitinib , which is approved for the same indication as sorafenib ( Stein and Flaherty , 2007 ) , making it unlikely that these events are confounded by the particular patient population under study . These results suggest that , compared to other clinically-approved kinase inhibitors , sorafenib treatment has an increased propensity to generate unexpected adverse events in a variety of physiological systems . While these data are merely correlative at this point , one possibility , given the unique ability of sorafenib to inhibit system xc− among tested kinase inhibitors , is that this effect contributes to increase the chance of an adverse event in combination with one or more underlying modifying factors . Our results suggested that inhibition of system xc− function by erastin , SAS and sorafenib ( within a narrow concentration range ) triggers ferroptosis . We sought to identify genetic modifiers of this process in erastin-resistant cell lines . First , we isolated from DU-145 prostate cancer cells five clonal cell lines that displayed significant ( >threefold ) resistance to the lethal effects of erastin but not the multidrug resistance pump substrate Taxol ( paclitaxel ) ( Figure 8A–C ) . These five resistant cell lines displayed significant resistance to additional ferroptosis inducers including SAS , sorafenib and the more potent erastin analogs identified above ( 20–22 ) ( Figure 8D , E , Figure 8—figure supplement 1 ) . These lines also displayed resistance to RSL3 , which triggers ferroptosis not through inhibition of system xc− but through the inhibition of the glutathione peroxidase GPX4 ( Yang et al . , 2014; Figure 8—figure supplement 1 ) . This observation suggested that resistance was unlikely to be due to any effect on upstream cystine import or glutathione production . Indeed , using the glutamate release assay , we found that system xc− was equally sensitive to the inhibitory effects of erastin in the parental DU-145 line and the five resistant cell lines ( Figure 8F ) . Likewise , the parental and resistant cell lines exhibited largely equivalent levels of basal total glutathione and depletion of glutathione following erastin treatment ( Figure 8G ) . 10 . 7554/eLife . 02523 . 014Figure 8 . Isolation and analysis of erastin-resistant clones identifies AKR1C genes as mediators of resistance to system xc− inhibition . ( A ) Outline of the isolation of DU-145 erastin-resistant clones . ( B–H ) Comparison of the parental DU-145 cell line and five erastin-resistant clonal lines to the indicated lethal compounds: ( B–E ) , response to different lethal compounds , ( F ) Glutamate release , ( G ) Glutathione concentration , ( H ) DCF fluorescence . ( I ) Summary of RNA Seq analysis of the five erastin-resistant clones vs the parental DU-145 cells . Fold-change in expression ( y-axis ) and absolute change in FPKM ( x-axis ) is computed from the average of the five resistant cells lines vs parental . ( J ) Summary of AKR1C family member activity possibly relevant to ferroptosis . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 01410 . 7554/eLife . 02523 . 015Figure 8—figure supplement 1 . Test of the ability of additional ferroptosis inducers to cause death in parental DU-145 and erastin-resistant DU-145 cell lines . Dose-response curves for parental DU-145 and erastin-resistant DU-145 clones treated with active ( 3 , 20 , 21 , 22 ) and inactive ( 14 , 16 ) erastin analogs and 1S , 3R-RSL3 . The color coding for all graphs is as shown in the legend of the 1S , 3R-RSL3 graph . Data represent mean ± SD from three independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02523 . 015 To explore further potential mechanisms of resistance , we examined basal and erastin-stimulated ROS levels in parental DU-145 cells and a subset of the erastin-resistant clones . We observed that the resistant cell lines exhibited substantially lower levels of basal and erastin-induced ROS accumulation , as detected by H2DCFDA using flow cytometry ( Figure 8H ) . This result suggested that resistance to various ferroptosis inducers was likely due to an inhibition of the accumulation of lethal oxidative species . To identify candidate genes involved in this process , we used RNA Seq to identify changes in gene expression associated with resistance . In this analysis , we focused on genes that were transcriptionally upregulated in resistant clones vs the parental cell line . In total , we identified 73 genes that were upregulated at least 10-fold on average across the five resistant cell lines compared to the parental clones ( Data available as ‘Data Package 2’ from Dryad data Repository [Dixon et al . , 2014] ) . The two genes that exhibited the highest average fold-change in expression and that were upregulated to highest average absolute levels within the cell ( e . g . , FPKM >100 ) were AKR1C1 ( 586-fold up-regulation ) and AKR1C2 ( 528-fold up-regulation ) ( Figure 8I ) . A third family member , AKR1C3 was upregulated 84-fold . The AKR1C1-3 enzymes have been shown to participate in the detoxification of toxic lipid metabolites ( such as 4-hydroxynonenal ) generated downsteam of the oxidation of various polyunsaturated fatty acid species ( Figure 8J; Burczynski et al . , 2001 ) . Thus , overexpression of AKR1C family members ( and potentially other genes ) may confer partial resistance to erastin by enhancing the detoxification of reactive aldehydes generated downstream of the oxidative destruction of the plasma membrane during ferroptosis . Amino acid transporters , such as system xc− , are potentially attractive drug targets , as these proteins are crucial for cell survival and growth . In this study , we have demonstrated that erastin is a potent and specific inhibitor of system xc−-mediated cystine uptake and further elucidated the mechanism of action of this and other system xc− inhibitors that are able to trigger ferroptosis . Previous metabolomics analysis had suggested that erastin inhibited system L ( SLC3A2 + SLC7A5 ) -mediated amino acid transport in Jurkat T cells ( Dixon et al . , 2012 ) . It was therefore surprising to observe that in HT-1080 and Calu-1 cells erastin inhibited system xc− ( SLC3A2 + SLC7A11 ) -mediated cystine uptake , but not system-L-mediated phenylalanine uptake . These results rule out the possibility that erastin inhibits SLC3A2-dependent transporters non-specifically . Further , given evidence that Jurkat cells do not express system xc− ( Kakazu et al . , 2011 ) , we hypothesize that in cells lacking system xc− , erastin can inhibit structurally-related transporters ( e . g . , system L ) . An alternative hypothesis is that erastin binds to some indirect target that , in Jurkat cells , favors the inhibition of system L , while in HT-1080 , Calu-1 and possibly other cells , favors the inhibition of system xc− . A definitive resolution of this matter will require further study . An important goal is to identify scaffolds capable of inhibiting system xc− with greater potency than existing compounds typified by SAS and derivatives ( Gorrini et al . , 2013 ) . We found that erastin is a substantially more potent inhibitor of system xc− function than SAS . Further optimization of the erastin scaffold yielded analogs with improved potency against this antiporter in the low nanomolar range that retained a degree of selectivity towards oncogenic mutant HRAS-expressing cells . Our work suggests that the ability of erastin , SAS and sorafenib to induce ferroptosis is tied to the inhibition of system xc−-mediated cystine import , and the consequent depletion of glutathione and loss of GPX4 activity ( Yang et al . , 2014 ) . However , these compounds are also predicted to inactivate the cystine/cysteine redox cycle ( Banjac et al . , 2008 ) and , by restricting the intracellular supply of cysteine , to inhibit new protein synthesis . Together , these effects may further reduce cell growth or cause cell death in certain contexts where the induction of ferroptosis per se is not possible . The ability of system xc− inhibitors such as erastin to trigger ferroptosis at similar concentrations in both monolayer and three-dimensional MCTS cultures , and at both high ( 21% ) and low ( 1% ) O2 levels , suggests that these compounds are capable of overcoming the ‘multicellular resistance’ phenomenon observed with many lethal molecules ( Desoize and Jardillier , 2000 ) and do not necessarily require high levels of ambient O2 to be lethal . Consistent with the hypothesis that erastin treatment deprives cells of cystine , RNA Seq expression profiling and subsequent follow-up studies of erastin-treated cells revealed transcriptional upregulation of DDIT3 ( CHOP ) , DDIT4 ( REDD1 ) and ATF3 , canonical targets of the eIF2alpha-ATF4 branch of the unfolded protein response ( UPR ) /ER stress pathway , which we also showed to be activated biochemically . These results are consistent with previous work showing upregulation of these genes in mammalian cells cultured in the absence of cysteine ( Lee et al . , 2008 ) . Intriguingly , ATF4 is thought to be an important regulator of SLC7A11 expression and deletion of Atf4 in mouse embryonic fibroblast cells results in an oxidative , iron-dependent death phenotype that is highly reminiscent of ferroptosis ( Harding et al . , 2003 ) , suggesting that the basal level of ATF4 may set the threshold for ferroptotic death . CHAC1 is a downstream target of the eIF2alpha-ATF4 pathway ( Mungrue et al . , 2009 ) and our results suggest that CHAC1 upregulation may be useful as a PD marker for cystine or cysteine-starved cells . Whether CHAC1 plays a role in the execution of ferroptosis remains unclear . ChaC-family proteins were recently reported to function in yeast as intracellular glutathione-degrading enzymes ( Kumar et al . , 2012 ) . One possibility is that CHAC1 upregulation following system xc− inhibition may actively contribute to glutathione depletion in cells deprived of cysteine , although inhibition of CHAC1 transcription had only minimal effects on the viability of erastin-treated cells . The finding that sorafenib can inhibit system xc− was unexpected . Despite sorafenib's multi-kinase inhibitory activity , there is disagreement about whether the sorafenib lethal mechanism of action in cells involves kinase inhibition or binding to an alternative target ( Wilhelm et al . , 2008 ) . It has previously been shown that sorafenib treatment inhibits translation ( Rahmani et al . , 2005 ) , induces ER stress and the expression of DDIT4 ( REDD1 ) via the eIF2alpha-ATF4 pathway ( Rahmani et al . , 2007; Kim et al . , 2011 ) , and causes caspase-independent death ( Panka et al . , 2006; Katz et al . , 2009 ) . Most recently , it was suggested that in hepatocellular carcinoma cells , sorafenib can trigger iron-dependent death ( Louandre et al . , 2013 ) . Here , we show that inhibition of system xc−-mediated cystine import by sorafenib can lead to both the induction of an ER stress response ( as indicated by phosphorylation of eIF2alpha and upregulation of both ATF4 and CHAC1 ) and ferroptotic cell death . Thus , our results provide a satisfying mechanistic explanation for previous observations: namely , that sorafenib can inhibit system xc− function , leading to ER stress and in some cells the induction of non-apoptotic , iron-dependent , ferroptotic cell death . Data collected in Phase I clinical trials of sorafenib-treated patients demonstrate that at clinically recommended doses ( 400 mg ) , it is possible to achieve maximum plasma concentration of sorafenib from 5 . 2–21 μM ( Awada et al . , 2005; Strumberg et al . , 2005 ) . This encompasses the range within which we observe the ferroptosis-inducing effects of sorafenib . Notably , sera collected from sorafenib-treated patients display evidence of protein oxidation , with higher levels of protein oxidation being correlated with improved patient outcomes ( Coriat et al . , 2012 ) . Thus , it is conceivable that sorafenib could be having effects in vivo related to the inhibition of system xc− function and the subsequent generation of reactive oxygen species . Indeed , given that renal cell carcinomas are among the most sensitive of all cancer cell lines to the lethal effects of erastin ( Yang et al . , 2014 ) , it may be of interest to re-evaluate whether the efficacy of sorafenib observed in patients could be due , at least in part , to inhibition of system xc−-mediated cystine uptake . Likewise , adverse events that are observed in a minority of patients treated with sorafenib may be due to inhibition of system xc− . While not all patients treated with sorafenib experience adverse events , we suspect that specific underlying ( and unknown ) sensitizing factors will render a minority of individuals more sensitive to these events . It would be important to account for this potential toxicity in the design of future therapeutics . As with many molecularly targeted compounds ( Holohan et al . , 2013 ) , the ultimate clinical utility of system xc− inhibition will be influenced by the ability of target cells to evolve resistance to these inhibitors . We found that exposure to erastin can result in the emergence of cell populations partially resistant to this compound due to dramatic overexpression of multiple AKR1C family members . These enzymes have a number of substrates but have been shown to detoxify toxic lipid metabolites such as 4-HNE ( Burczynski et al . , 2001 ) , which are likely produced by the oxidative lipid fragmentation processes that occur during the execution of ferroptosis ( Dixon et al . , 2012; Skouta et al . , 2014; Yang et al . , 2014 ) . The expression of the AKR1C genes is controlled by the antioxidant master regulatory transcription factor NRF2 , which itself is under the negative regulation of KEAP1 ( Lou et al . , 2006; Agyeman et al . , 2012; Jung and Kwak , 2013 ) . Mutations of both these genes are observed in numerous cancers ( Jaramillo and Zhang , 2013 ) , and we would predict that these changes enhance AKR1C expression and possibly render cells resistant to the induction of ferroptosis downstream of system xc− inhibition . Erastin was synthesized as described ( Yagoda et al . , 2007 ) . Additional erastin and sorafenib analogs were prepared as described in the Supplementary file 1 . Data also available as ‘Extended Materials and Methods’ from Dryad data Repository ( Dixon et al . , 2014 ) . The synthesis of ( 1S , 3R ) -RSL3 was described ( Yang et al . , 2014 ) . Sorafenib , imatinib , erlotinib , lapatinib , nilotenib , dasatinib , sunitinib , and gefetinib were from SelleckChem ( Houston , USA ) . Unless otherwise indicated , all other compounds were from Sigma-Aldrich ( St . Louis , USA ) . BJeH , BJeHLT and BJeLR cells were obtained from Robert Weinberg ( Whitehead Institute ) . 143B cells were obtained from Eric Schon ( Columbia University Medical Center ) . HT-1080 and Calu-1 cells were obtained from American Type Culture Collection . BJeH , BJeHLT and BJeLR cells were grown in DMEM High-Glucose media ( Gibco/Life Technologies Corp . , Grand Island , NY ) plus 20% M199 ( Sigma , St . Louis , MO ) and 15% heat-inactivated fetal bovine serum ( FBS ) . HT-1080 cells were grown in DMEM High-Glucose medium ( Gibco ) supplemented with 10% FBS and 1% non-essential amino acids ( Gibco ) . Calu-1 and U2OS cells were grown in McCoy's 5A media ( Gibco ) supplemented with 10% fetal bovine serum . MEFs were grown in DMEM supplemented with 10% fetal calf serum . 143B cells were grown in DMEM High-Glucose supplemented with 10% FBS . All cell lines were grown in humidified tissue culture incubators ( Thermo Scientific ) at 37°C with 5% CO2 . Except where indicated , all media were supplemented with penicillin and streptomycin ( Gibco ) . For low oxygen experiments cells were grown under normal ( 21% ) oxygen conditions , then split into two 6-well dishes , one of which was cultured at 21% O2/5% CO2 in a regular tissue culture incubator , and one of which was transferred to a HypOxygen H35 incubator for growth under 1% O2/5% CO2 conditions for 24 hr . The next day compounds were added directly to the plates . In the case of the 1% O2 , condition , compound addition was done within the confines of the chamber to prevent media re-oxygenation . 24 hr later , cells were removed from the chamber and viability was assessed immediately by Vi-Cell . Multicellular tumor spheroids ( MCTSs ) were grown in 96-well Corningware Ultra Low Attachment ( ULA ) Plates ( CLS 3474 ) . 200 μl of cell suspension containing 104 cells/ml were added to each well of the ULA plate , after which they were incubated at 37°C/5% CO2 for 72 hr to allow for MCTS formation . MCTSs were then treated with lethal compounds ( vehicle control [DMSO] , 10 μM Erastin , 1 μM RSL3 , or 1 μM STS ) ±inhibitors ( vehicle control [DMSO] , 1 μM Ferrostatin-1 , or 25 μM β-mercaptoethanol ) by carefully aspirating 50 μl of media from each well , and replacing with 50 μl each of media containing 4 × desired treatment concentration of the lethal or inhibitor . After 72 hr of treatment , MCTS images were acquired using an EVOS fl microscope ( Advanced Microscopy Group/Life Technologies Corp . ) equipped with a 10× phase contrast objective . Three independent fields were acquired for each experimental condition . Representative samples from one field of view are shown . Viability was then measured using Alamar blue as described above and measured on a Victor3 platereader . 200 , 000 HT-1080 or Calu-1 cells/well were seeded overnight in 6-well dishes ( Corning Life Sciences , Tewksbury , MA ) . The next day , cells were washed twice in pre-warmed Na+-free uptake buffer ( 137 mM choline chloride , 3 mM KCl , 1 mM CaCl2 , 1 mM MgCl2 , 5 mM D-glucose , 0 . 7 mM K2HPO4 , 10 mM HEPES , pH 7 . 4 ) , then incubated for 10 min at 37°C in 1 ml of uptake buffer , to deplete cellular amino acids . At this point , in each well the buffer was replaced with 600 µl uptake buffer containing compound and 0 . 12 µCi ( 80–110 mCi/mmol ) of L-[3 , 3'-14C]-cystine or 0 . 2 µCi of L-[14C ( U ) ]-phenylalanine ( PerkinElmer , Waltham , MA ) and incubated for 3 min at 37°C . Cells were then washed three times with ice-cold uptake buffer and lysed in 500 µl 0 . 1 M NaOH . To this lysate was added 15 ml of scintillation fluid , and radioactive counts per minute were obtained using a scintillation counter . All experiments were repeated in three independent biological replicates for each condition . To control for differences in the absolute counts of radioactivity between replicates , data were first normalized to DMSO ( set to 100% ) within each replicate , then averaged across three biological replicates . The release of glutamate from HT-1080 cells into the extracellular medium was detected using an Amplex Red glutamate release assay kit ( Molecular Probes/Life Technologies Corp . , Eugene , OR ) . For compound treatment experiments , 200 , 000 cells/well were seeded overnight into 6-well dishes ( Corning ) . The next day , cells were washed twice in PBS and then incubated for one hour in Na+-containing , glutamine-free media ( Cellgro/Corning ) containing various compounds at different concentrations . For siRNA experiments , cells were transfected with siRNAs for 48 hr ( see above ) , then washed twice in PBS and incubated for an hour in Na+-containing , glutamine-free media . 50 μl of medium per well was removed and transferred to a 96-well assay plate ( Corning ) and incubated with 50 μl of a reaction mixture containing glutamate oxidase , L-alanine , glutamate-pyruvate transaminase , horseradish peroxidase , and Amplex Red reagents as per the manufacturer's protocol . Glutamate release was first normalized to total cell number determined by Vi-Cell counting at the end of the experiment , then values were expressed as a percentage of no treatment ( DMSO ) controls . In some experiments , a glutamate standard curve was used to quantify the exact amount of glutamate release . Of note: as the medium contained Na+ , the total amount of glutamate release reflects the activity of both system xc− ( Na+-independent ) and non-system xc− glutamate transporters and therefore never reaches 100% inhibition , as system xc− accounts for only a portion of total glutamate release . This glutamate release assay was used during the testing of erastin analogs . Human astrocytoma cells ( CCF-STTG1 ) were used to assay cystine–glutamate antiporter ( xc− ) activity . Cells , cultured in RPMI + 10% FBS , were grown in 96-well plates; when confluent , cells were washed with Earle's Balanced Salt Solution ( EBSS , Sigma ) containing Ca2+ and Mg2+ to remove residual glutamate . Cells were then incubated for 2 hr at 37°C with either EBSS/glucose ( blanks ) or EBSS/glucose containing 80 µM cystine ( totals ) ± compounds ( 30 nM–100 µM ) . The known xc− inhibitor , ( S ) -4-carboxyphenylglycine ( S-4CPG ) , was used as positive control . Following incubation , glutamate released into medium was detected using the Amplex Red system ( Life Technologies ) , as per the manufacturer's instructions . HT-1080 cells were reverse transfected with siRNAs ( Qiagen , Germantown , MD ) using Lipofectamine RNAiMAX ( LFMax , Invitrogen/Life Technologies Corp . ) . Briefly , 1–10 nM ( final concentration ) of siRNAs was aliquoted into 250 µl Opti-MEM media ( Gibco ) in the bottom of each well of a 6-well dish ( Corning ) . An additional 250 µl medium + LFMax was added to each well and incubated for 15 min . At this point , 150 , 000 HT-1080 cells were added to each well in regular HT-1080 medium . The plates were swirled to mix and incubated for 48 hr at 37°C in a tissue culture incubator prior to analysis . RNA was extracted using the Qiashredder and Qiagen RNeasy Mini kits ( Qiagen ) according to the manufacturer's protocol . 1–2 μg total RNA for each sample was used as input for each reverse transcription reaction , performed using the TaqMan RT kit ( Applied Biosystems/Life Technologies Corp . , Foster City , CA ) . Primer pairs for were designed for target transcripts using Primer Express 2 . 0 ( Applied Biosystems ) . Quantitative PCR reactions were performed using the Power SYBR Green PCR Master Mix ( Applied Biosystems ) . Triplicate samples per condition were analyzed on an Applied Biosystems StepOnePlus qPCR instrument using absolute quantification settings . Differences in mRNA levels compared to ACTB internal reference control were computed between control and experimental conditions using the ΔΔCt method . Cell viability was typically assessed in 384-well format by Alamar Blue ( Invitrogen , Carlsbad , CA ) fluorescence ( ex/em 530/590 ) measured on a Victor3 platereader ( PerkinElmer ) . In some experiments , Trypan Blue dye exclusion counting was performed using an automated cell counter ( ViCell , Beckman–Coulter , Fullerton , CA ) . Cell viability in test conditions is reported as a percentage relative to the negative control treatment . Total intracellular glutathione ( GSH+GSSG ) was measured using a glutathione assay kit based on Ellman's reagent ( Cayman Chemical #703002; Ann Arbor , USA ) according to instructions . 200 , 000 HT-1080 cells per well were seeded overnight in 6-well dishes ( Corning ) . The next day , cells were treated with compounds ( erastin for 5 hr , sorafenib for 18 hr ) , then washed once in 500 μl PBS and harvested by scraping into phosphate buffer ( 10 mM X , 1 mM EGTA ) . Cells were then lysed by sonication ( 7 cycles , 2 s on , 1 s off ) and spun at 4°C for 15 min at 13 , 000 rpm to pellet membranes . Supernatants were mixed with 500 μl of a 10% solution of metaphosphoric acid ( wt/vol ) , vortexed briefly , and centrifuged for 3 min at 4000 rpm . The supernatant was transferred to a new tube and to this was added 50 μl of triethanolamine solution ( 4 M ) . This was vortexed and 50 μl per sample was aliquoted to each well of a 96-well plate . 150 μl of assay buffer containing 5 , 5'-dithiobis ( 2-nitrobenzoic acid ) ( DTNB , Ellman's reagent ) was added to each well and the reaction was incubated for 25 min at room temperature with rotation , at which point absorbance was measured at 405 nM . The glutathione concentration was calculated in reference to a glutathione standard curve and normalized to total cell number per well , as determined from parallel plates . Flow cytometry was performed using an Accuri C6 flow cytometer equipped with a 488 nm laser . Reactive oxygen species accumulation was assessed using H2DCFDA and C11-BODIPY 581/591 ( both from Molecular Probes/Life Technologies ) as described in Dixon et al . ( 2012 ) . RNA was isolated from compound-treated HT-1080 cells , or from DU-145 parental and erastin-resistant cell lines , as described for RT-qPCR reactions . Poly-A pull-down was then used to enrich mRNAs from total RNA samples ( 1 μg per sample , RIN >8 ) and libraries were prepared using Illumina TruSeq RNA prep kit ( San Diego , CA ) . Libraries were then sequenced using an Illumina HiSeq 2000 instrument ( Columbia Genome Center , Columbia University ) . For the analysis of gene upregulation by erastin treatment , five samples were multiplexed in each lane , to yield the targeted number of single-end 100 bp reads for each sample ( 30 million ) , as a fraction of 180 million total reads for the whole lane . For the analysis of erastin-resistant cell lines , paired-end 100 bp reads for each sample ( 60 million ) were collected . Short reads were mapped to the human reference genome using Tophat ( Trapnell et al . , 2009 ) . The relative abundance of genes and splice isoforms was determined using Cufflinks ( Trapnell et al . , 2010 ) . We then looked for differentially expressed genes under the various experimental conditions using Cuffdiff , a program included in the Cufflinks package . For the HT-1080 experiments , only genes with FPKM not equal to zero in any condition and FPKM ≥1 for both replicates in the DMSO-treated condition were considered in the analysis . To further restrict the analysis to high quality data , we only examined those genes where the difference between replicate values in the two DMSO and the two erastin-treated samples was <2 . 5x . Modulatory effect ( Me ) profiling was performed as described ( Wolpaw et al . , 2011; Dixon et al . , 2012 ) . In Figure 1A , B , the following ferroptosis inhibitors were tested ( maximum concentration in 10 point , twofold dilution series listed here ) : cycloheximide ( CHX , 50 μM ) , ferrostatin-1 ( Fer-1 , 2 μM ) , trolox ( 300 μM ) , U0126 ( 15 μM ) , ciclopirox olamine ( CPX , 50 μM ) , and beta-mercaptoethanol ( β-ME , 20 μM ) . In Figure 5A , B , the following lethal compounds were tested ( maximum concentration in 10 point , twofold dilution series listed here ) : methotrexate ( 100 μM ) , bortezomib ( 5 μM ) , doxorubicin ( 50 μM ) , chlorambucil ( 500 μM ) , irinotecan ( 5 μM ) , SAHA ( 50 μM ) , actinomycin D ( 1 μg/ml ) , gefitinib ( 50 μM ) , taxol ( 5 μM ) , sorafenib ( 10 μM ) , erlotinib ( 5 μM ) , GX15 ( 10 μM ) , staurosporine ( STS , 2 μM ) , phenylarsine oxide ( PAO , 1 μM ) , RSL3 ( 5 μM ) , H2O2 ( 5 mM ) , ABT-263 ( 50 μM ) , 6-thioguanine ( 5 μM ) , 17-AAG ( 5 μM ) and imatinib ( 50 μM ) . Cells were lysed in lysis buffer consisting of 50 mM HEPES , 40 mM NaCl , 2 nM EDTA , 0 . 5% Triton X-100 , 1 . 5 mM Na3VO4 , 50 mM NaF , 10 mM sodium pyrophosphate , 10 mM sodium beta-glycerophosphate and 1 tablet of protease inhibitor . Cell lysates were separated on a 4–20% Tris-gel , transferred to nitrocellulose membrane and blocked in PBST +5% milk for an hour . Membranes were incubated with primary antibody at the following concentrations: ATF4 ( sc-200; Santa Cruz , Santa Cruz , CA ) 1:100 , phosphor-eIF2α ( 5324; Cell Signaling , Danvers , MA ) 1:200 , eIF2α ( 3597; Cell Signaling ) 1:1000 , p-PERK 1:200 ( sc-32577; Santa Cruz ) , PERK 1:200 ( sc-13073; Santa Cruz ) , BiP ( 3177; Cell Signaling ) 1:1000 , eIF4E ( 610270; BD Biosciences , San Jose , CA ) 1:1000 in PBS + 5% BSA overnight at 4°C . Next day washed and incubated with 1:2000 HRP-conjugated secondary before development with SuperSignal West Pico Substrate ( #34080; Pierce , Rockford , IL ) . HT-1080 cells were compound treated then mRNA was harvested and 2 μg of mRNA used as a template for first-strand cDNA synthesis , also as described above for RT-qPCR . The cDNA was used as input for a PCR reaction using XBP-1-specific primers to detect splicing ( Forward [5′-3′]: TTACGAGAGAAAACTCATGGCC , Reverse [5′-3′]: GGGTCCAAGTTGTCCAGAATGC ) and actin B-specific primers as a control . PCR conditions were as follows: 94°C for 1 min , followed by 35 cycles of 94°C for 30 s , 60°C for 30 s , 72°C for 1 min . PCR products were separated on a 2 . 5% agarose gel and visualized using an Syngene G:Box imaging station . Cells were seeded ( 1500 cells/well ) in a volume of 40 μl medium in 384-well plates ( Corning ) for 24 hr prior to treatment of lethals and/or inhibitors . Following compound treatment for 24 hr , 10 μl of a 1:100 vol/vol dilution of Apo-One Homogeneous Caspase 3/7 substrate solution/assay buffer ( Promega , Madison , WI ) was added to samples , and the plate was vigorously agitated for 30 s . Plates were then incubated for 8 hr in the dark at room temperature , allowing for caspase-3/7 cleavage of the fluorogenic substrate , before measuring fluorescence at excitation/emission wavelengths of 498/521 nm using a Victor3 plate reader ( Perkin Elmer ) . Except where indicated , all experiments were performed at least three times on separate days as independent biological replicates . The data shown represents the mean ± SD of these replicates . All statistical analyses and curve fitting were performed using Prism 5 . 0c ( GraphPad Software , La Jolla , CA ) . Gene Ontology ( GO ) process enrichment was computed using the web-based GOrilla tool with default settings ( Eden et al . , 2009; http://cbl-gorilla . cs . technion . ac . il ) . The Food and Drug Administration ( FDA ) collects and maintains spontaneously submitted adverse event reports in the Adverse Event Reporting System ( FAERS ) . We downloaded 2 . 9 million adverse events from FAERS representing reports through the fourth quarter of 2011 . To complement FAERS , we also downloaded the OFFSIDES drug effect database ( Tatonetti et al . , 2012 ) . In addition , we extracted the laboratory values , clinical notes , prescription orders , and diagnosis billing codes from the electronic health records ( EHR ) at Columbia University Medical Center/New York Presbyterian Hospital for 316 patients with at least one prescription order of sorafenib , dasatinib , erlotinib , gefetinib , imatinib , lapatinib , or sunitunib . These data were employed in the analysis ( described in detail in the main text ) to identify adverse events uniquely associated with sorafenib treatment . This analysis was covered under the Columbia Institutional Review Board ( IRB ) protocol number AAAL0601 . To generate resistant clones , DU-145 prostate cancer cells were seeded in 10-cm dishes with complete medium containing ∼3 × EC50 erastin ( 2 . 4 μM or 2 . 6 μM ) , which was found to be effective at initially reducing the population on any given plate to a small number of individual surviving cells . The erastin-supplemented medium was replaced every 3 days for 2–3 weeks to allow for clonal expansion . As summarized in Figure 7A , 60 resistant clones were initially isolated by ring cloning , with selection limited to non-diffuse cell clusters to minimize risk of selecting drifted populations . Clones were subsequently maintained in complete medium without erastin . Of 36 clones found to be resistant to erastin in an initial re-testing , 20 were cross-resistant to Taxol ( paclitaxel ) , a mechanistically unrelated drug . Of the remaining 16 clones , five exhibited strong resistance to erastin , but not to Taxol , and were further characterized .	Sugars , fats , amino acids , and other nutrients cannot simply diffuse into the cell . Rather , they must be transported across the cell membrane by specific proteins that stretch from one side of the cell membrane to the other . One such ‘transporter’—system xc−—is of special interest . This transporter imports one molecule of cystine from outside the cell in exchange for one molecule of glutamate from inside the cell . Cystine , a variant of the amino acid cysteine , is essential for synthesizing new proteins and for preventing the accumulation of toxic species inside the cell . Not surprisingly , many cancer cells are dependent upon the transport activity of system xc− for growth and survival . Drugs that can inhibit system xc− could therefore be part of potential treatments for cancer and other diseases . Dixon , Patel , et al . have found that the compound erastin is a very effective inhibitor of system xc− function . Certain versions of erastin are over 1000 times more potent than the previously known best inhibitor of system xc− , sulfasalazine . Dixon , Patel et al . found that using erastin and sulfasalazine to inhibit system xc− in cancer cells grown in a petri dish results in an unusual type of iron-dependent cell death called ferroptosis . By inhibiting the uptake of cystine , erastin and other system xc− inhibitors interfere with the cellular machinery that folds proteins into their final , three-dimensional shape . The accumulation of these partially-folded proteins in the cell causes a specific kind of cellular stress that can be used as a readout , or biomarker , for the inhibition of system xc− . Such a biomarker will be essential for identifying cells in the body that have been exposed to agents that inhibit system xc− and that are undergoing ferroptosis . Unexpectedly , Dixon , Patel et al . also found that the FDA-approved anti-cancer drug sorafenib inhibits system xc− . Other drugs in the same class as sorafenib do not share this unusual property . Dixon , Patel , et al . synthesized variants of sorafenib and identified sites on the drug that are necessary for it to be able to interfere with system xc− . Alongside the erastin derivatives , these new molecules may help to develop new drugs that can inhibit this important transporter in a clinical setting .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	Pharmacological inhibition of cystine–glutamate exchange induces endoplasmic reticulum stress and ferroptosis
LINE-1 ( L1 ) is an autonomous retrotransposon , which acted throughout mammalian evolution and keeps contributing to human genotypic diversity , genetic disease and cancer . L1 encodes two essential proteins: L1ORF1p , a unique RNA-binding protein , and L1ORF2p , an endonuclease and reverse transcriptase . L1ORF1p contains an essential , but rapidly evolving N-terminal portion , homo-trimerizes via a coiled coil and packages L1RNA into large assemblies . Here , we determined crystal structures of the entire coiled coil domain of human L1ORF1p . We show that retrotransposition requires a non-ideal and metastable coiled coil structure , and a strongly basic L1ORF1p amino terminus . Human L1ORF1p therefore emerges as a highly calibrated molecular machine , sensitive to mutation but functional in different hosts . Our analysis rationalizes the locally rapid L1ORF1p sequence evolution and reveals striking mechanistic parallels to coiled coil-containing membrane fusion proteins . It also suggests how trimeric L1ORF1p could form larger meshworks and indicates critical novel steps in L1 retrotransposition . The mammalian LINE-1 ( long interspersed element 1 , L1 ) retrotransposon has had a considerable impact on the evolution of mammalian genome organization and continues to shape the evolution of the human genome . Roughly 17% of the human genome sequence corresponds to fragments or full-length L1 copies of different evolutionary age , contrasting with only about 1 . 5% of our genome , which encodes all of the human proteins ( Lander et al . , 2001; Stewart et al . , 2011 ) . L1 is the only autonomously active mobile genetic element in the human genome , but also mobilizes non-autonomous Alu and SVA elements ( Garcia-Perez et al . , 2016; Goodier , 2016; Mita and Boeke , 2016; Richardson et al . , 2015 ) . Autonomous retrotransposition relies on two L1-encoded proteins . The L1ORF1 protein ( L1ORF1p ) is known as an RNA-binding protein ( Hohjoh and Singer , 1996; Martin , 1991 ) , whereas the L1ORF2 protein ( L1ORF2p ) harbors the necessary catalytic functions , consisting of an endonuclease and a reverse transcriptase ( Feng et al . , 1996; Kazazian et al . , 1988; Moran et al . , 1996 ) . L1 propagates via an RNA intermediate in a ‘copy-and-paste’ fashion . It does not rely on long terminal repeats ( LTRs ) for the reverse transcription and genome integration steps , in contrast to LTR retrotransposons and retroviruses ( Sultana et al . , 2017 ) . Hence classified as a non-LTR retrotransposon , L1 integrates via target-primed reverse transcription , a telomerase-like mechanism , where the reverse transcription of L1RNA occurs directly at the spot of genomic integration ( Cost et al . , 2002; Luan et al . , 1993 ) . It is poorly understood , however , how L1RNA , as a part of large L1 ribonucleoprotein particles ( L1RNPs ) , gains access to the chromatin in dividing ( Mita et al . , 2018 ) as well as non-dividing cells ( Kubo et al . , 2006; Macia et al . , 2017 ) . Retrotransposition must occur in germline cells in order to assure a lineage-specific , vertical transmission of L1 and its long-term survival in mammalian genomes . L1RNA and L1ORF1p are expressed in both gametogenesis and the early embryo ( Branciforte and Martin , 1994; Malki et al . , 2014; Packer et al . , 1993; Trelogan and Martin , 1995 ) , where early embryonic integrations lead to mosaic offspring ( Kano et al . , 2009; van den Hurk et al . , 2007 ) . Furthermore , retrotransposition also happens in somatic cells , such as in neuronal progenitor cells ( Coufal et al . , 2009; Faulkner and Garcia-Perez , 2017; Muotri et al . , 2005 ) . As a consequence of both germline and somatic insertions , L1 activity contributes to inter-individual human variation and diversity , but also causes genetic disease and cancer ( Burns , 2017; Hancks and Kazazian , 2016; Scott and Devine , 2017 ) . Importantly , human L1 expression and retrotransposition appears to be triggered in certain cancer types ( Carreira et al . , 2014; Scott and Devine , 2017 ) as well as in induced pluripotent stem cells ( Klawitter et al . , 2016; Wissing et al . , 2012 ) , as detected by the expression of L1ORF1p ( Klawitter et al . , 2016; Rodić et al . , 2014; Wissing et al . , 2012 ) . Hence , considering the possible implications of L1 retrotransposition for human health and for the applications of stem cells in medicine and research , it is surprising how little we know about the mechanistic details of L1 retrotransposition . Intriguingly , not only does L1 retrotransposition depend on the catalytic activity of the L1ORF2p ( Feng et al . , 1996; Moran et al . , 1996 ) , but also on an intact open reading frame encoding L1ORF1p ( Moran et al . , 1996 ) . Multiple copies of L1ORF1p associate ‘in cis’ ( Basame et al . , 2006; Kulpa and Moran , 2005; Sokolowski et al . , 2017; Taylor et al . , 2013; Wei et al . , 2001 ) with their encoding L1RNA molecule , and the resulting L1RNP is considered as a functional intermediate in the retrotransposition process ( Hohjoh and Singer , 1996; Kulpa and Moran , 2005; Martin , 1991; Taylor et al . , 2013 ) . Furthermore , L1ORF1p was shown to facilitate the rearrangement of nucleic acid structure and hence might be important as a ‘nucleic acid chaperone’ for remodeling the L1RNP ( Martin and Bushman , 2001 ) . Indeed , most of the published experimental data characterizes functions of L1ORF1p that are related to its interaction with RNA , whereas little is known about other roles of this protein in L1 retrotransposition . Mammalian L1ORF1p has a unique architecture , even among the ORF1ps encoded by non-LTR retrotransposons ( Kapitonov and Jurka , 2003; Khazina and Weichenrieder , 2009; Schneider et al . , 2013 ) . It consists of three structural domains , connected by short linkers . These domains are first , a coiled coil domain , which causes the protein to form homotrimers ( Martin et al . , 2003 ) , second , an RRM ( RNA recognition motif ) domain , and third , a C-terminal domain ( CTD ) , which cooperates with the RRM domain in binding single stranded nucleic acid substrates ( Januszyk et al . , 2007; Khazina et al . , 2011; Khazina and Weichenrieder , 2009 ) . In the case of the human protein , the coiled coil domain is preceded by a 51 residue long N-terminal region ( NTR ) , harboring two serine-proline motifs that are known phosphorylation sites ( Cook et al . , 2015 ) . Finally , there is also a short 15 residue tail at the C-terminal end of L1ORF1p that can be partially truncated without functional consequences ( Alisch et al . , 2006 ) . A series of crystal structures has uncovered the three-dimensional arrangement of the individual domains in the context of the L1ORF1p trimer ( Khazina et al . , 2011 ) . The structures were obtained from an N-terminally truncated protein , lacking the NTR and the N-terminal half of the coiled coil domain , but they revealed L1ORF1p to be a highly structured and remarkably flexible RNA-binding protein . It became clear how the coiled coil domain mediates the trimerization of the protein and how it allows for the flexible attachment and organization of the RRM and CTD domains , such that between 27 and 45 nucleotides of single stranded RNA are bound and covered by one trimer ( Khazina et al . , 2011 ) . However , although the structures rationalized trimerization and RNA binding of L1ORF1p , the N-terminally truncated protein was not able to promote L1 retrotransposition when tested in HeLa cells ( Khazina et al . , 2011 ) . We therefore decided to investigate the structural and mechanistic properties of the poorly conserved N-terminal sequences in L1ORF1p , and to which degree they contribute to L1 retrotransposition . To this aim , we combined biophysical with cell-based techniques and determined crystal structures for the entire coiled coil domain of the human L1ORF1p , enabling us to construct a composite model for the complete trimer . Surprisingly , in order to function in retrotransposition , the coiled coil apparently needs to be able to switch between fully structured and partially unstructured states . This requirement for metastability can explain the presence and delicate balance of both stabilizing and destabilizing elements in the structure of the coiled coil and the strong sensitivity to mutation . Finally , we also identified the positively charged amino terminus of L1ORF1p as an independent and novel determinant for L1 retrotransposition , a feature that is preserved in the mammalian homologs . Consequently , L1ORF1p emerges as a delicate but remarkably autonomous protein regarding its host cell molecular environment , and with functions that clearly extend beyond RNA packaging . It shows striking parallels to other dynamic coiled coil proteins , which act in membrane fusogenic processes ( Skehel and Wiley , 1998 ) , hinting at presently uncharacterized steps in the L1 retrotransposition cycle . Non-LTR retrotransposons encode ORF1 proteins with highly diverse architectures and distinct structural domains , frequently suggestive of RNA binding but possibly also of lipid or membrane interaction ( Khazina and Weichenrieder , 2009; Schneider et al . , 2013 ) . The most commonly shared feature is , however , the apparent presence of coiled coil forming regions , suggesting self-association and oligomerization into dimeric , trimeric or higher order assemblies ( Figure 1A , Figure 1—figure supplement 1A ) . Coiled coils are superhelical bundles of α-helices , where each helix is built from repeats of seven amino acids ( heptads ) . In each heptad , the amino acid positions are labeled a to g , where positions a and d point towards the center of the bundle and are typically occupied by small , hydrophobic residues . This results in a usually hydrophobic core of the coiled coil with alternating a- and d-layers . Consequently , the residues in the a- and d-positions of each heptad are the most critical ones to define and stabilize a coiled coil . Furthermore , charged residues in positions e and g frequently form stabilizing salt bridges on the surface of the coiled coil ( Lupas et al . , 2017 ) . In the case of the human L1ORF1p , the presence of a coiled coil forming region was previously identified by sequence analysis ( Hohjoh and Singer , 1996 ) , but it was difficult to define the boundaries of this coiled coil domain and to align its sequence among mammalian orthologs ( Boissinot and Furano , 2001; Boissinot and Sookdeo , 2016 ) . The identification and crystallization of the RRM domain ultimately revealed that the coiled coil domain extends right to the start of the RRM domain and that mammalian L1ORF1 proteins share seven alignable heptads preceding the RRM domain ( Khazina and Weichenrieder , 2009 ) . These heptads are numbered I to VII in the C- to N- terminal direction and include two conserved ‘RhxxhE’ trimerization motifs spanning heptads V and VI , where ‘h’ designates hydrophobic a- and d-layers and ‘x’ stands for any residue ( Kammerer et al . , 2005 ) . Trimerization motifs stabilize the parallel , trimeric structure of a coiled coil through salt bridges that form between glutamates in position e and arginines in position g’ of the preceding heptad . They probably also help to initiate coiled coil formation and to define the correct register for coiled coil assembly ( Ciani et al . , 2010; Kammerer et al . , 2005 ) . Together with the RRM and CTD domains , heptads I to VII therefore define the alignable or conserved portion of L1ORF1p ( Figure 1A , Figure 1B , Figure 1C , Figure 1—figure supplement 1B , Supplementary file 1 ) . The conserved portion of human L1ORF1p trimerizes as the full length protein and binds and releases nucleic acid substrates , but it is not sufficient to support retrotransposition ( Khazina et al . , 2011 ) . The remaining , N-terminal portion of L1ORF1p is variable among mammalian orthologs and cannot be consistently aligned ( Figure 1—figure supplement 1B , Supplementary file 1 ) . It is therefore also missing from recently published alignments ( Boissinot and Sookdeo , 2016; Yang et al . , 2014 ) . The N-terminal portion of L1ORF1p consists of the presumably disordered NTR , followed by additional heptad repeats that complete the predicted coiled coil domain ( Figure 1D ) . It is possible though to unambiguously align the presently active human L1ORF1p sequence with ancestral L1ORF1p sequences reconstructed from the human genome ( L1PA1 up to L1PA5 ) ( Khan et al . , 2006 ) and with closely related L1ORF1p sequences from the great apes and macaques ( Figure 1—figure supplement 2 ) . This alignment predicts a coiled coil domain with seven additional heptad repeats ( VIII to XIV ) and with an insertion of three amino acids in or around heptad IX . Such an insertion disturbs the periodicity of the coiled coil and is called a ‘stammer’ , in comparison to ‘stutters’ , which are insertions of four residues ( Brown et al . , 1996 ) . Most importantly , the alignment also illustrates the rapid evolution of the N-terminal portion of human L1ORF1p as compared to the rest of the sequence and especially the accumulation of non-conserved residues in the N-terminal half of the coiled coil domain ( Figure 1B , Figure 1—figure supplement 2 ) . This part of the coiled coil domain has previously been claimed to be under positive selection , because it appears to evolve more rapidly than expected from a neutral rate of evolution ( Boissinot and Furano , 2001; Khan et al . , 2006 ) . Furthermore , among mammalian L1ORF1ps , the number and regularity of the N-terminal heptads varies considerably ( Figure 1—figure supplement 1B , Supplementary file 1 ) , especially in mice , where heptad duplications and deletions are well documented for the three active L1 lineages ( Sookdeo et al . , 2013 ) ( Figure 1—figure supplement 1C , Supplementary file 1 ) . To characterize the molecular properties and functional requirements of the essential but poorly conserved N-terminal portion of L1ORF1p , we therefore took an individual , structure-based approach with the human L1ORF1p . Sequence analysis suggests the coiled coil domain to begin with residue Y52 of the human L1ORF1p ( Figure 1C , Figure 1D ) . Considering the high sequence variation even among primate orthologs , it was unclear , however , whether the entire sequence could form one continuous coiled coil , where and how the three amino acid insertion would be accommodated and what would be the structural and functional consequences of the numerous non-canonical residues in the predicted a- and d-layers . We therefore tried to obtain a detailed crystal structure and indeed , a bacterially expressed fragment of human L1ORF1p , encompassing the entire coiled coil domain ( hL1ORF1p-cc ) crystallized with two trimers ( T1 and T2 ) present per asymmetric unit , yielding two slightly differing structures of the trimeric human L1ORF1p coiled coil domain at 2 . 65 Å resolution ( Figure 2A , Figure 2—figure supplement 1A , Figure 2—figure supplement 1B , Table 1 ) . Trimer T1 forms an extended rod with an overall length of 150 Å . For the polypeptide chains A and B , all 14 heptads are found in the electron density in a continuously helical , extended conformation . The variable coiled coil sequences therefore indeed extend the previously characterized C-terminal heptads ( I-VI ) and retain threefold symmetry up to heptad XI . In heptad XII , chain C begins to deviate and breaks the threefold symmetry , and then becomes untraceable in the electron density in heptad XIII . Chains A and B instead continue a helical packing with likely support from crystal contacts ( Figure 2A ) . Trimer T2 is highly similar to trimer T1 for heptads II to XI ( r . m . s . d . for Cα atoms = 0 . 781 Å ) , but , in comparison to trimer T1 , heptad XII also remains roughly threefold symmetric . Furthermore , heptads XIII and XIV are untraceable in electron density in the case of chains B and C . In the case of chain A , heptad XIII locally unwinds and loses its α-helical structure , whereas heptad XIV is still α-helical but bent by ~90° with respect to the threefold axis , making crystal contacts with a T1 trimer from a neighboring asymmetric unit ( Figure 2A ) . Apparently therefore , the N-terminal heptads of the coiled coil domain are deformable and can switch between an α-helical and an unwound state . In particular , the non-canonical K62 and F69 in the d-layers of heptads XIII and XII might be difficult to maintain in a three-fold symmetric state . In solution , these heptads hence might preferably engage in alternating binary interactions between two of the three chains , resulting in a dynamic structure at the N-terminal end of the coiled coil domain rather than in the formation of a stable rod . A thorough analysis of the molecular contacts and a computational analysis of coiled coil parameters reveals a mixture of stabilizing and destabilizing interactions along the sequence of the coiled coil domain ( Figure 3 , Figure 3—figure supplement 1 ) . Most strikingly , in heptad IX , there is a sharply localized distortion in the helical geometry of both the coiled coil bundle and of the individual polypeptide chains . The distortion is caused by the stammer , which can be precisely assigned to residues M91 , E92 and L93 . These three residues form an extra 310-helical turn between positions d and e of heptad IX and create an additional hydrophobic core layer ( d* ) at L93 . ( Figure 2A , Figure 3A , Figure 3D , Figure 3—figure supplement 1A , Figure 3—figure supplement 1D ) . Consequently , the individual helices are locally overwound and stretched in concert with a strong increase in the left-handed supercoiling of the bundle ( Figure 3B , Figure 3—figure supplement 1B ) . Trimeric stammer structures have previously been discussed only in synthetically designed coiled coil environments ( Hartmann et al . , 2016; Hartmann et al . , 2009 ) and occur much less frequently in natural coiled coils than stutters , which , in structural terms , are easier to accommodate ( Lupas et al . , 2017 ) . Also in the coiled coils of mammalian L1ORF1 proteins , stutters occur more often , such as in murine L1ORF1p ( Figure 1—figure supplement 1B , Figure 1—figure supplement 1C , Supplementary file 1 ) . In general , stammers are considered to have an unfavorable , destabilizing effect on the respective coiled coil structure ( Lupas et al . , 2017 ) , consistent with the local increase in the averaged atomic B-factors of the two crystallized trimers of the human L1ORF1p coiled coil ( Figure 3C , Figure 3—figure supplement 1C ) . Additionally , this coiled coil hosts a series of non-canonical and non-ideal a- and d-layers , which are also considered to be destabilizing ( Figure 1C , Figure 2A , Figure 3D , Figure 3—figure supplement 1D ) . In particular , these are the distorted d-layers in heptads XIII ( K62 ) and XII ( F69 ) , the cysteine and threonine-containing d-layers of heptads XI and X , and the cysteine-containing a-layers of heptads VII and VI . Finally , there are the previously described ion-coordinating layers of heptads III and II ( Khazina et al . , 2011 ) . Chloride-binding asparagines ( heptad II , N142 ) are not uncommon in the d-layer of parallel , trimeric coiled coils ( referred to as asparagines at d- , or short , N@d- layers [Hartmann et al . , 2009] ) as they help to define both the trimeric state and the correct register of the three chains . Arginines ( heptad III , R135 ) are much more rarely observed at d-layers , and especially the combination with a glycine ( G132 ) in the preceding a-layer , where the guanidino groups of R135 coordinate a second chloride ion , is unique so far to the human L1ORF1p ( Khazina et al . , 2011 ) ( Figure 3D , Figure 3—figure supplement 1D ) . The destabilizing effects of the stammer and of the non-ideal core layers are balanced , however , by numerous peripheral interactions between pairs of neighboring polypeptide chains and involving polar residues in positions b , e , and g ( Figure 1C , Figure 2A , Figure 2—figure supplement 1B , Figure 3A , Figure 3—figure supplement 1A ) . Next to the two consecutive trimerization motifs in heptads V and VI , which are conserved in all mammals , there are two additional , non-conserved trimerization motifs in heptads II and X and a peripheral interaction with inverse polarity in heptad VII , that is with an arginine in position e and a glutamate in position g’ . The trimerization motifs differ at position b , where various alternative residues contribute to the motif in three of the four cases ( S119 , D112 , T81 in heptads V , VI , X , respectively , Figure 3D , Figure 3—figure supplement 1D ) . As a result , the stammer is flanked by stabilizing motifs both on its C-terminal and on its N-terminal side , and the non-canonical layers in heptads II , VI , VII and X are hedged by peripheral interactions . It is clear as well that this mixture of stabilizing and destabilizing interactions results in the observed distribution of the crystallographic B-factors along the sequence of the coiled coil domain , reflecting a more malleable structure of the coiled coil on its N-terminal side ( Figure 3C , Figure 3—figure supplement 1C ) . However , given the high sequence variability in the coiled coil region , there appear to be many functional combinations of stabilizing and destabilizing interactions , raising the question of how crucial it is to balance the respective effects along the sequence . The presently determined structures of the coiled coil domain match extremely well with the previously determined structures of the conserved portion of human L1ORF1p ( Khazina et al . , 2011 ) over heptads II-VI ( r . m . s . d . = 0 . 422 Å over 105 Cα atoms in residues C111-S145 , Figure 2B ) . Heptad VII is not completely traceable in the electron density of the conserved portion ( Khazina et al . , 2011 ) and the C-terminal residues of the coiled coil domain ( W150-Y152 ) are distorted due to crystal packing interactions . Using the overlap for a structural superposition , it is possible to generate a composite model for the human L1ORF1p trimer , which has overall dimensions of 77 Å by 179 Å and comprises the complete coiled coil , RRM and C-terminal domains , that is comprises the conformationally defined region of L1ORF1p ( Figure 2B ) . Coloring the model according to sequence variability illustrates the striking frequency of variable residues in the N-terminal half of the coiled coil , including residues both from core layers and from the surface of the coiled coil ( Figure 2—figure supplement 1C ) . Furthermore , the N-terminal half of the coiled coil reveals an alternation of positively charged , neutral and negatively charged surfaces , where an acidic patch at the transition between heptads XI and XII is the most prominent feature . In contrast , the conserved portion of the model is strongly positively charged , especially in the RNA binding clefts between the RRM and C-terminal domains ( see also Khazina et al . , 2011 ) . Notable exceptions are heptads V and VI with their conserved and highly acidic surface ( Figure 2—figure supplement 1D ) . The composite model of the human L1ORF1p trimer ( Figure 2B ) lacks residues M1-N51 and E324-M338 , because these residues were missing from the expressed constructs or disordered in the available crystal structures . The C-terminal residues are highly variable or absent in mammalian orthologs ( Figure 1—figure supplement 1B , Supplementary file 1 ) and can be partially removed ( Alisch et al . , 2006 ) or also extended with artificial peptide tags without blocking retrotransposition activity ( Goodier et al . , 2007; Kulpa and Moran , 2005; Taylor et al . , 2013 ) . This suggests the C-terminal residues are not functionally required . The 51 N-terminal residues , however , contain functionally relevant phosphorylation sites ( Cook et al . , 2015 ) , but protein constructs including the NTR failed to crystallize . We therefore used circular dichroism ( CD ) spectroscopy and analytical size exclusion chromatography to investigate the structure and potential interactions of the NTR ( Figure 4 , Figure 5 ) . CD spectroscopy is an excellent method to detect the presence of secondary structure in solution and reveals a purely α-helical spectrum for the hL1ORF1p-cc coiled coil construct ( Figure 4A ) . In contrast , a peptide corresponding to the NTR ( hL1ORF1p-NTRH6 ) lacks α-helices or β-strands ( Figure 4B ) , consistent with the disorder prediction analysis ( Figure 1D ) . Furthermore , the coiled coil sequence forms extended trimers in solution as confirmed by multiangle laser light scattering ( MALLS , Figure 4C ) , whereas the NTR remains monomeric ( Figure 4D ) . Finally , the NTR also fails to interact with the remainder of human L1ORF1p ( hL1ORF1p-ΔNTR ) when added ‘in trans’ and tested by size exclusion chromatography ( Figure 4E , Figure 4F , Figure 4—figure supplement 1 ) . The NTR also fails to interact with hL1ORF1p-ΔNTR when residues S18 and S27 are substituted by aspartates , mimicking the phosphorylated state of the NTR ( Figure 4—figure supplement 2 ) . As a result , and in the absence of additional interaction partners , the unstructured NTR peptides appear to be hanging from the deformable and potentially dynamic N-terminal end of the coiled coil domain of the fully assembled trimer , without stable attachment to any of the structured domains . The present crystal structures and solution studies show that the coiled coil can form over the entire length of the 14 heptads . However , the seven C-terminal heptads , which are already sufficient for trimer formation , also are clearly better defined in the electron density map than the seven N-terminal heptads . These show increasingly elevated B-factors and begin to deviate from the threefold symmetry the closer the sequence is located to the amino terminus ( Figure 2A , Figure 3C , Figure 3—figure supplement 1C ) . We therefore tested whether the variable , N-terminal portion of L1ORF1p would still be able to trimerize in the absence of the conserved portion , but this is clearly not the case . The respective construct ( hL1ORF1p-Δcons ) remained monomeric at concentrations up to 1 . 3 mM ( Figure 5A ) and , most surprisingly , is unstructured according to CD spectroscopy ( Figure 5B ) . Apparently , folding of the seven N-terminal heptads only occurs when it is triggered by the C-terminal heptads as in the context of the full length L1ORF1p , or possibly , by external binding partners . Consequently , the C-terminal heptads are required for a formation of a continuous coiled coil structure . Alternatively , and at sufficiently high concentration ( 5 . 2 mM ) , the N-terminal portion of L1ORF1p begins to dimerize ( Figure 5C ) . Clearly however , the molecular contacts in this dimer must be structurally distinct from the binary interaction of two helices in the trimer , and they could occur in either parallel or anti-parallel orientation . Given the dependence of homo-trimerization on the C-terminal heptads , we also wondered how the structural stability is affected along the sequence of the coiled coil domain and whether upon thermal denaturation the coiled coil domain would come apart in separate steps or rather cooperatively . Hence , we monitored the loss of α-helical content by CD spectroscopy as a function of increasing temperature and found that indeed , the coiled coil domain ( hL1ORF1p-cc ) unfolded in a stepwise fashion with two transitions at 37° C and 70° C ( Figure 5D ) . In summary , it is therefore reasonable to assume that the first transition reflects the unfolding of exclusively the N-terminal heptads . This leads to a model of the L1ORF1p trimer , where the N-terminal heptads are in a subtle equilibrium between structured and unstructured states and can switch between these states at physiological temperature . To answer the question whether and how much the presence and the biophysical properties of the non-conserved L1ORF1p sequences matter for L1 retrotransposition , we tested a series of L1ORF1p mutants in a well-established , plasmid-based L1 retrotransposition assay in HeLa cells ( Moran et al . , 1996 ) . In this assay , the retrotransposition of a tagged L1 copy into HeLa cell genomic DNA confers an antibiotic resistance . This allows resistant cells to form colonies on a dish , which can then be counted and normalized to wildtype levels ( Figure 6 , Figure 7 ) . Expression of the respective L1ORF1p mutants was monitored by western blotting ( Figure 6—figure supplement 1 , Figure 7—figure supplement 1 ) . Because the conserved portion of L1ORF1p was known to be insufficient for activity , we first tested a further extension up to heptad IX , which produces a regular , uninterrupted heptad pattern . However , this construct remained inactive ( Figure 6A ) . Next , we reasoned that the NTR with its apparent phosphorylation sites might need to be present , and we generated a series of internal heptad deletions extending over the first seven , five and two of the N-terminal heptads . None of these constructs was active , although at least the latter two were also well expressed ( Figure 6A ) . This result is somewhat surprising , because heptad deletions frequently occurred in the evolution of the mammalian L1 element ( Figure 1—figure supplement 1B , Figure 1—figure supplement 1C , Supplementary file 1 ) . It suggests that the variable , deformable and non-ideal parts of the coiled coil domain are functionally required in their entirety and consequently , that their ability to alternate between a structured and an unstructured state likely plays a role in the L1 retrotransposition cycle . In a final step , we therefore exclusively deleted the three stammer residues from heptad IX , generating an uninterrupted , fourteen heptad coiled coil domain with presumably increased stability . This construct too completely failed to retrotranspose ( Figure 6A ) . Thermal melting of the respective coiled coil domain construct revealed that , despite the deletion of the stammer ( hL1ORF1p-ccΔ ( 91–93 ) ) , the unfolding still was biphasic and hence still not cooperative ( Figure 5E ) . However , both unfolding transitions were shifted to higher temperature , indicating that the local deletion of the stammer causes a widespread stabilization over the entire coiled coil domain . Consequently , the human L1ORF1p seems to have evolved to operate in retrotransposition in a rather narrow window of ( in ) stability . To further probe the permissive window of coiled coil stability , we tested additional L1ORF1p variants . We primarily targeted unusual core layers , generating single or multiple point mutations at a time ( Figure 6B , Figure 6C ) . First , we addressed the ion-containing heptads II and III ( Figure 6B ) . Replacement of the unusual R135 at position IIId with an asparagine ( R135N ) had the goal to preserve the hydrophilic properties and to support the trimeric state of the coiled coil by allowing for an additional N@d layer . This mutation had a rather negligible effect on retrotransposition , whereas the regularizing R135I substitution clearly reduced activity . Also the regularizing G132V substitution in position IIIa detectably reduced retrotransposition , whereas an N142I substitution in position IId had a lesser effect . Surprisingly however , the double mutation G132I/R135I and the triple mutation G132I/R135I/N142I ( Khazina et al . , 2011 ) did not only completely abolish retrotransposition , but also reduced protein levels markedly ( Figure 6—figure supplement 1B ) . Presumably , the low protein abundance is caused by a faster degradation of these variants and either due to an improved recognition of the rigidified trimer conformation by the proteolytic machinery or , alternatively , due to an increased misassembly of the coiled coil domain in a wrong register . Second , we addressed heptads VI to IX , which contain a previously investigated series of leucine-containing d-layers ( Figure 6C ) . These leucines had been tested by various combinations of destabilizing alanine or regularizing valine substitutions , which completely aborted retrotransposition ( Doucet et al . , 2010; Goodier et al . , 2007 ) . We included the leucine positions in our analysis but used hydrophilic asparagines for substitutions , with the goal to support a trimeric coiled coil but without further stabilization of the structure . However , both an L93N/L100N double mutation in positions IXd* and VIIId and an L107N/L114N double mutation in positions VIId and VId remained inactive . The respective proteins were expressed at reduced levels and moreover migrated slightly abnormally on gels ( Figure 6—figure supplement 1B ) . In contrast , a C104I/C111I double mutation in positions VIIa and VIa was active in retrotransposition , although detectably reduced . In eukaryotes , cysteines are not very frequent in the a-layers of trimeric coiled coils ( Woolfson and Alber , 1995 ) . They form tris-thiolate sites , which are predisposed for heavy metal ion binding ( Ruckthong et al . , 2016 ) and which might have special preferences for the neighboring d-layers , such as , for example , the absence of β-branched residues . Indeed , this could possibly explain why the d-layer leucines cannot easily be exchanged in the human sequence although they are not even conserved among primates ( Figure 1—figure supplement 2 ) . More generally , our results demonstrate that both non-canonical and canonical layers of the coiled coil are very sensitive to mutation and interdependent , and that an idealization and stabilization of the coiled coil structure is rather counterproductive with regard to the ability to retrotranspose . Third and finally , we also mutated a surface residue in the variable part , C86 in position Xg , which was converted to a serine ( C86S ) . Surprisingly , even this peripheral single atom substitution of a poorly conserved residue strongly reduced retrotransposition ( Figure 6D ) . This suggests that the interdependence of coiled coil residues extends beyond the core layers and to the non-conserved N-terminal half of the coiled coil . The high sequence variability among mammalian L1ORF1ps therefore could well result from an internal coevolution of the residues within the coiled coil domain , where a mutation at one position would trigger a compensatory mutation elsewhere in the coiled coil . To learn whether the NTR is similarly sensitive to mutations , we also started with a deletion analysis ( Figure 7A ) . Not surprisingly , a deletion of the entire NTR , including the apparent phosphorylation sites , resulted in an inactive L1 element . However , also consecutive extensions of the L1ORF1p toward the original N-terminus did not rescue L1 retrotransposition activity , indicating that either the overall length of the NTR or the first five amino acids behind the first methionine were crucial for activity . We therefore generated an internal deletion , spanning the second half of the NTR ( Figure 7B ) . This construct showed reduced but clearly detectable activity and consequently , the N-terminal amino acids in L1ORF1p are key to L1 retrotransposition activity . Closer inspection of the N-terminal sequence shows an accumulation of positively charged residues , but also a remote similarity to an N-terminal myristoylation signal . We therefore first altered the sequence to convert it to a strong myristoylation signal , substituting M1GKKQNRK with M1GARASRK ( Bologna et al . , 2004 ) . However , the respective protein construct was only poorly detectable and showed no activity at all , indicating that an N-terminal myristoylation does probably not play a positive role in retrotransposition ( Figure 7C ) . Since the N-terminal methionine is very likely removed from the wildtype sequence ( Frottin et al . , 2006 ) , the remaining GKKQNRK sequence turns into a strongly positively charged patch as long as the main chain amino terminus and the lysine side chains remain non-acetylated and free of any other kind of potential modification . We therefore tested single alanine substitutions in the first three positions behind the methionine ( G2A , K3A and K4A ) as well as a G2A/K3A double mutation ( Figure 7D ) . Strikingly , each of the single point mutations strongly reduced L1 retrotransposition , whereas the G2A/K3A double mutation completely abolished it . Notably , however , the single G2A mutation was accompanied by a very low protein level ( Figure 7—figure supplement 1B ) . Since L1ORF1p is a known ubiquitination target , the low protein abundance might be rationalized by a context-induced and K3-dependent ubiquitination and degradation ( MacLennan et al . , 2017 ) . But because the G2A/K3A double mutation was expressed at normal levels , we did not follow up on this possibility here any further . Instead , we generated two alternative single point mutations , G2R and K3R , which were well tolerated and , for K3R , even led to a detectable increase in retrotransposition activity ( Figure 7E ) . This result shows that it is not the identity of the N-terminal amino acids or the presence of a specific post-translational modification , but rather the presence and accumulation of the positive charges that matter for retrotransposition in this case . Finally , it is important to note that the positive charges need to be present near the N-terminus of the NTR , because internal substitutions of positive charges ( K13A and R48A ) did not have a strong effect ( Figure 7F ) and because the addition of N-terminal tags to the natural amino terminus of L1ORF1p is known to abolish retrotransposition activity ( Goodier et al . , 2007; Taylor et al . , 2013 ) . Therefore , the positively charged N-terminal end of human L1ORF1p emerges as a previously unknown retrotransposition requirement and appears to be a feature that is conserved all the way through the evolution of the mammalian L1 element ( Figure 1—figure supplement 1B , Supplementary file 1 ) . Non-LTR retrotransposition is still poorly understood on a mechanistic level . In particular , it is unclear what are the molecular properties of the diverse ORF1 proteins and how these properties promote essential steps in the retrotransposition cycle . Moreover , ORF1 proteins do not have cellular or viral homologs from which their mechanics and function could be deduced , requiring an individual analysis . The present work leads to a deeper mechanistic understanding of the human L1ORF1 protein and especially of its previously only poorly characterized and rapidly evolving N-terminal sequences . There are two key findings . First , L1 retrotransposition apparently requires a long , non-ideal and metastable coiled coil with the ability to switch between structured and partially unstructured states . Second , retrotransposition activity also requires a flexible NTR with a strongly positively charged amino terminus . We therefore speculate that adjacent phosphorylation , conformational changes in the coiled coil domain , and/or the bound L1 RNA could regulate the availability of the amino-terminal residues in a cellular context and hence control crucial steps in the L1 retrotransposition cycle . Our findings reinforce the picture of the L1ORF1p as a delicate and highly flexible protein with functions that clearly go beyond its previously investigated RNA binding and chaperoning functions ( Figure 8 , Figure 8—video 1 , Figure 8—figure supplement 1A ) . The conserved portion of the coiled coil domain plays a central role for the assembly of the trimer . It is necessary and sufficient to specify and promote trimerization , and it serves as a scaffold for the oriented but flexible attachment of the RRM and CTD domains , which cooperate in binding single stranded RNA substrates ( Khazina et al . , 2011 ) . Importantly , however , the conserved portion of the coiled coil domain also triggers the assembly of the non-conserved portion , which , together with the positively charged amino terminus on the unstructured NTR , fulfills crucial but hitherto poorly contemplated roles in the retrotransposition cycle . These are outlined in the following . A mechanistic requirement to switch between structured and transiently unstructured states imposes opposing constraints on the amino acid sequence of the coiled coil and can explain the presence of untypical core layers or heptad expansions , and hence the conservation of an irregular rather than a canonical coiled coil structure in the evolution of the L1ORF1p . The need to switch between conformational states can also explain the rapid sequence evolution in the N-terminal half of the coiled coil . A faster-than-neutral amino acid substitution rate can arise when an initial mutation that disturbed the finely calibrated balance of stabilizing and destabilizing interactions gets compensated and fixed by another mutation elsewhere in the sequence of the coiled coil . Such an intrinsic cause for the rapid evolution is also supported by engineered coiled coil chimeras generated from reconstructed ancestral and modern human L1ORF1p proteins ( Naufer et al . , 2016 ) . These chimeras functioned only in one of two possible combinations , whereas the original proteins both are fully functional in the retrotransposition assay . External causes for the rapid evolution may independently exist in the form of a coevolving restriction factor or of an evasive interaction partner from the host ( Daugherty and Malik , 2012 ) . The fact , however , that highly diverged L1ORF1ps from the mouse or a reconstructed L1ORF1p from the megabat promote human L1 retrotransposition in HeLa cells ( Wagstaff et al . , 2011; Yang et al . , 2014 ) argues against the existence of an evasive interaction partner and indicates a remarkable autonomy of L1ORF1p to promote retrotransposition independently of the host cell’s molecular environment . Our findings reinforce parallels of the L1ORF1p to other coiled coil proteins , where coiled coil formation also allows for the switch between two ( or more ) conformational states , including the exposure or capture of functional peptide sequences . Classical examples are viral membrane fusion proteins ( Chen et al . , 1999; Kobe et al . , 1999; Weissenhorn et al . , 1997 ) , best studied for the influenza hemagglutinin . Here , refolding and homotypic trimeric coiled coil formation exposes N-terminal and hydrophobic peptides that mediate membrane fusion ( Lin et al . , 2014; Skehel and Wiley , 2000 ) . L1ORF1p lacks any such hydrophobic sequences , but the positively charged amino terminus might also serve to target lipid bilayers due to their negative surface charge ( Hoernke et al . , 2012; Kim et al . , 1991 ) . Other examples are the eukaryotic SNARE proteins ( Sutton et al . , 1998 ) , where heterotypic tetrameric coiled coil formation specifies vesicle targeting to membranes and causes signal-dependent vesicle fusion by zipping up in a stepwise fashion ( Gao et al . , 2012; Jahn and Fasshauer , 2012; Südhof and Rothman , 2009 ) . A final example is the bacterial protein M1 ( McNamara et al . , 2008 ) , which can form dimeric coiled coils in two alternative registers , but where the transient and destabilized intermediate is functionally important for pathogenicity , allowing the capture of fibrinogen-derived peptide sequences ( Stewart et al . , 2016 ) . The thermal melting experiment with the L1ORF1p-derived coiled coil indicates that its N-terminal half can come apart at a physiological temperature , whereas its C-terminal half remains trimeric . Furthermore , when tested in isolation at a high local concentration , the N-terminal sequence of L1ORF1p dimerizes . These observations raise the possibility that trimers of L1ORF1p directly interact with each other at high concentrations , for example during assembly of L1RNPs . The consequence of dimerizing trimers is not only a linear array on the RNA , but rather a three-dimensional meshwork with probably variable regularity ( Figure 8—figure supplement 1B ) . Meshwork formation may explain the cytoplasmic ‘aggregates’ of L1ORF1p that had been observed early on by ultracentrifugation ( Hohjoh and Singer , 1996; Martin , 1991 ) and by fluorescence light microscopy ( Goodier et al . , 2007; Martin and Branciforte , 1993 ) , and also why L1ORF1p appears to ‘polymerize’ when artificially assembled on long , single stranded DNA ( Naufer et al . , 2016 ) . Electron micrographs of what presumably are perinuclear clusters of L1ORF1p in mutated mouse spermatocytes show an irregular dotted pattern , where 5–6 dots occasionally form semi-closed circles ( Soper et al . , 2008 ) . Intriguingly , when L1ORF1p trimers are assembled into hexameric rings by simple modeling , one obtains a similar diameter of roughly 150 Å ( Figure 8—figure supplement 1B ) . A single L1RNA could theoretically accommodate up to 130 trimers ( Khazina et al . , 2011 ) , but this number appears to be considerably lower in purified L1RNPs , obtained from HEK293T cells under stringent salt conditions ( Taylor et al . , 2013 ) . Functionally , meshwork formation might allow L1ORF1p to sequester L1RNA from the host cell environment and to shield it from processes such as deadenylation and decay ( Wahle and Winkler , 2013 ) until the L1RNP finally gains access to nuclear chromatin for the reverse transcription and integration steps of the L1 retrotransposition cycle . The requirement of a positively charged N-terminal peptide sequence came as an unexpected and novel finding here and merits future investigation . At the current stage , we can only speculate on possible functions , but the presence of an essential peptide at the N-terminus in conjunction with an irregular and dynamic coiled coil reinforces mechanistic parallels to viral membrane fusion proteins , where conformational changes regulate the exposure of their fusion peptides ( Skehel and Wiley , 1998; White et al . , 2008 ) . In the case of the L1ORF1p , the positively charged N-terminus could act as a cellular localization or transport signal and/or to target chromatin , especially since certain insect retrotransposons encode L1ORF1p-like proteins with an N-terminal PHD domain ( Metcalfe and Casane , 2014 ) . PHD domains are frequently found in chromatin reader proteins ( Musselman and Kutateladze , 2011; Sanchez and Zhou , 2011 ) and also occur in other non-LTR retrotransposon-encoded ORF1ps of different architectural types ( Kapitonov and Jurka , 2003; Khazina and Weichenrieder , 2009 ) . Another possible function of the positively charged amino terminus might be the modulation of RNA binding on the RRM and CTD domains , in particular when it comes to facilitating binding and/or release of L1RNA in the context of remodeling a larger L1RNP . Finally , positively charged peptides can also target and perturb negatively charged lipid bilayers ( Hoernke et al . , 2012; Kim et al . , 2002; Kim et al . , 1991 ) , further extending the analogies with the viral membrane fusion and the eukaryotic SNARE proteins from a purely mechanistic to a truly functional level . Intriguingly indeed , the perinuclear clusters of L1ORF1p in mouse spermatocytes appear to be surrounded by a double membrane ( Soper et al . , 2008 ) , and Horn et al . have recently shown a dependence of L1 retrotransposition on an interaction of L1ORF1p with components of the ALIX/ESCRT membrane budding complex ( Horn et al . , 2017 ) . It is therefore not unreasonable to speculate that L1ORF1p also functions in membrane-related processes and particularly in overcoming the nuclear barrier in non-dividing cells ( Kubo et al . , 2006; Macia et al . , 2017 ) , where a classical , nuclear pore-mediated entry of the large L1RNPs is rather difficult to conceptualize . L1ORF1ps from the mouse and from the megabat can functionally replace the human L1ORF1p in human cells , despite considerable sequence divergence and despite the extreme mutational sensitivity of the protein ( Wagstaff et al . , 2011; Yang et al . , 2014 ) . Similarly , human L1 sequences function in non-human cell-lines and transgenic mice and rats ( Kano et al . , 2009; Moran et al . , 1996; Morrish et al . , 2002; Muotri et al . , 2005; Ostertag et al . , 2002 ) . This suggests that L1ORF1ps act rather autonomously and in a fundamental fashion , which does not require a highly specific adaptation to the host species . Furthermore , these findings also lead to the intriguing question whether ORF1ps with an entirely different type of architecture , such as found in other non-LTR retrotransposons , could functionally replace the human L1ORF1p as well . Although direct experimental evidence is still missing , our early observations ( Khazina and Weichenrieder , 2009; Schneider et al . , 2013 ) and recent large-scale sequence analyses of non-LTR retrotransposons ( Heitkam et al . , 2014; Ivancevic et al . , 2016; Metcalfe and Casane , 2014 ) increasingly support the hypothesis of functional redundancy and of a ‘reticulate’ ( Metcalfe and Casane , 2014 ) rather than a tree-like evolution of ORF1ps . This means that RNA packaging , multimerization and membrane-targeting could be functions which are shared among most ORF1ps encoded by non-LTR retrotransposons ( Schneider et al . , 2013 ) . Together with previously published structures and analyses ( Januszyk et al . , 2007; Khazina et al . , 2011; Khazina and Weichenrieder , 2009 ) , the human L1ORF1p clearly emerges as the currently best understood ORF1p among non-LTR retrotransposons . The combined structural and mechanistic insight , and the large number of functional mutations presented in this study will enable future research to identify , distinguish and analyze novel steps in the L1 retrotransposition cycle . Furthermore , it is becoming increasingly clear that there are multiple lines of defense to protect the human genome from the uncontrolled propagation of the L1 element . These include mechanisms to control L1RNA transcription and post-transcriptional mechanisms aiming at L1RNA ( Goodier , 2016; Pizarro and Cristofari , 2016 ) , but also processes that directly target the L1ORF1p and merit further investigation ( MacLennan et al . , 2017 ) . On an entirely different note , the present work also leads to a deeper understanding of the fundamental principles underlying the evolution , stability and dynamics of a coiled coil in a physiological context . Coiled coils are among the most intensively studied protein folds ( Hartmann , 2017 ) , can be characterized and described from first principles ( Crick , 1953; Lupas et al . , 2017 ) and have become a preferred target for protein design ( Woolfson , 2017 ) . The present L1 retrotransposition assay could therefore serve as one of the most sensitive assay systems for testing coiled coil designs in a cellular environment . Finally , for conditions such as certain human cancers with elevated L1 retrotransposition ( Burns , 2017; Hancks and Kazazian , 2016; Scott and Devine , 2017 ) , it might become feasible and desirable to develop synthetic small molecules or synthetic peptides ( Modis , 2008 ) , with the goal to target the stability and function of the coiled coil and thereby to prevent further damage to the genomic DNA by L1 insertion . L1 sequences were retrieved , translated and aligned from the following sources . The human L1 . 3 sequence ( Dombroski et al . , 1993; Sassaman et al . , 1997 ) corresponds to the NCBI accession L19088 . 1 . Ancestral human L1 sequences are from Khan et al . ( Khan et al . , 2006 ) and the currently active mouse L1 lineages ( A1 , Tf1 , Gf1 ) are from Sookdeo et al . ( Sookdeo et al . , 2013 ) . Mammalian L1 sequences are found in Boissinot et al . ( Boissinot and Sookdeo , 2016 ) , and the reconstructed megabat sequence ( NCBI KF796623 . 1 ) is from Yang et al . ( Yang et al . , 2014 ) . Individual accession numbers for the reconstruction of primate L1ORF1p sequences are listed in Supplementary file 2 . PCoils ( Lupas , 1996 ) as integrated in the MPI Bioinformatics Toolkit ( Alva et al . , 2016 ) and IUPred ( Dosztányi et al . , 2005 ) were used for assigning coiled coil propensity and the probability of disorder , respectively . The DNA sequences encoding purified fragments of the human L1ORF1 protein , hL1ORF1p-NTRH6 ( M1–N51-HHHHHH ) , hL1ORF1p-Δcons ( GPHM1–E103 ) , hL1ORF1p-cc ( GPHMS53–Y152 ) , hL1ORF1p-ccΔ ( 91–93 ) ( GPHMS53–Y152 lacking residues 91–93 ) and hL1ORF1p-ΔNTR ( GPHMS53–M338 ) are derived from the L1 . 3 sequence ( Dombroski et al . , 1993; Sassaman et al . , 1997 ) . They were PCR-amplified from a plasmid harboring a M121A/M125I/M128I triple mutation . The residues substituting the three methionines correspond to the respective residues in the murine sequence , do not reduce human L1 retrotransposition activity , but avoid aberrant initiation of bacterial translation ( Khazina et al . , 2011 ) . The sequence encoding hL1ORF1p-NTRH6 was inserted into the pET15b expression plasmid ( Novagen ) . The sequence encoding hL1ORF1p ( DD ) -NTRH6 with phospho-mimicking aspartates ( S18D/S27D ) was obtained by site-directed mutagenesis . The sequences encoding hL1ORF1p-Δcons , hL1ORF1p-cc , hL1ORF1p-ccΔ ( 91–93 ) and hL1ORF1p-ΔNTR were inserted into the pnEA-pH expression plasmid , which provides an N-terminal and removable hexa-histidine tag ( Diebold et al . , 2011 ) . All plasmids are listed in Supplementary file 3 . Proteins were expressed in the Escherichia coli strain Rosetta 2 ( DE3 ) ( Novagen ) at 20°C overnight . All constructs were purified from cleared cell lysates apart from hL1ORF1p-ccΔ ( 91–93 ) , which was solubilized from inclusion bodies with the addition of 6M guanidinium hydrochloride . After an initial Ni2+-ion affinity step , the removable hexa-histidine tags were cleaved overnight with recombinant human rhinovirus 3C ( HRV3C ) protease , and hL1ORF1p-ΔNTR was further purified by a heparin affinity step . Finally , all constructs were purified by size exclusion chromatography using a Superdex 75 column ( GE Healthcare , Chicago , Illinois ) for hL1ORF1p-NTRH6 , hL1ORF1p ( DD ) -NTRH6 , hL1ORF1p-Δcons , hL1ORF1p-cc and hL1ORF1p-ccΔ ( 91–93 ) , and a Superdex 200 column ( GE Healthcare ) for hL1ORF1p-ΔNTR . Concentrated protein samples were flash-frozen in gel filtration buffer ( 10 mM HEPES , pH = 7 . 5 , 300 mM NaCl , 1 mM DTT ) and stored at −80°C for further use . Initial crystals of hL1ORF1p-cc ( 45 mg/ml in gel filtration buffer ) were obtained in several conditions by sitting drop vapor diffusion ( 18° C ) mixing 0 . 2 μl of protein solution with 0 . 2 μl of reservoir solution over an 80 μl reservoir . Crystals were optimized by manual screening around several initial conditions and flash frozen in liquid nitrogen with additional cryoprotection . The best-diffracting crystal ( 2 . 65 Å resolution , Table 1 ) was obtained over a reservoir of 0 . 1 M HEPES ( pH = 7 . 0 ) , 0 . 15 M ( NH4 ) 2SO4 and 12% PEG 2000 . It was grown in a sitting drop by mixing 0 . 5 μl reservoir solution and 0 . 5 μl protein solution at a concentration of 22 mg/ml , suspended over a reservoir of 66 μl . Cryoprotection was achieved by shortly soaking the crystal in reservoir solution supplemented with glycerol to a final concentration of 20% . Diffraction data were collected at 100 K on a Pilatus 6M detector ( DECTRIS , Baden-Daettwil , Switzerland ) on beamline PXII ( X10SA ) of the Swiss Light Source ( SLS ) , Villigen , Switzerland . Data were processed and scaled in spacegroup P21212 , using XDS and XSCALE ( Kabsch , 2010 ) . The structure was solved by molecular replacement using PHASER ( McCoy et al . , 2007 ) from within the CCP4 package ( Winn et al . , 2011 ) and with a search model containing nine heptads of a trimeric coiled coil . The search model was created by N-terminally extending the known structure of the six C-terminal L1ORF1p heptads ( PDB-ID: 2ykp , residues 111–152 ) ( Khazina et al . , 2011 ) with an additional three heptads of polyalanine sequence . Two copies of the search model were found in the asymmetric unit . This structure was then improved and extended by iterative cycles of model building in COOT ( Emsley et al . , 2010 ) and refinement using REFMAC ( Murshudov et al . , 2011 ) from the CCP4 package . Final refinement rounds were done using BUSTER ( Bricogne et al . , 2016 ) . The diffraction data and refinement statistics are summarized in Table 1 . The stereochemical properties for the structures were verified with MOLPROBITY ( Chen et al . , 2010 ) , and coiled coil parameters were analyzed using TWISTER ( Strelkov and Burkhard , 2002 ) . Sequence conservation was mapped to the protein structure using ProtSkin ( Denisov et al . , 2004 ) and illustrations were prepared in PyMOL ( http://www . pymol . org ) with the APBS plugin ( Baker et al . , 2001 ) to visualize the electrostatic surface potential . Analytical size exclusion chromatography coupled to multiangle static laser light scattering ( MALLS ) was done in gel filtration buffer and essentially as previously described ( Khazina et al . , 2011; Khazina and Weichenrieder , 2009 ) . Protein concentrations ranged from 0 . 3 mM to 5 . 2 mM in the case of hL1ORF1p-Δcons . Size exclusion chromatography was done on a Superdex 200 ( 10/300 GL ) column , apart from hL1ORF1p-Δcons , which was analyzed on a Superdex 75 ( 10/300 GL ) column . MALLS was done using miniDAWN TREOS and Optilab rEX instruments ( Wyatt Technologies , Santa Barbara , California ) and the associated software ( Astra from Wyatt Technologies ) for molecular weight determination . Circular dichroism ( CD ) measurements were done at a protein concentration of 0 . 15 mg/ml in gel filtration buffer without DTT , on a JASCO J-810 spectropolarimeter ( JASCO , Easton , Maryland ) equipped with a thermoelectric temperature controller . Spectra were recorded using a 0 . 1 cm path cuvette at a 1 nm band width with response of 2 s . A scanning speed of 100 nm/min and a data pitch of 0 . 1 nm were used . Thermal denaturation was monitored at 222 nm with a temperature ramp of 1°C/min and a data pitch of 0 . 5°C . Ellipticity calculation , buffer subtraction and smoothing was done in the software provided by JASCO . The mean residue ellipticity ( MRE ) was then calculated accounting for protein concentration and sequence length . In Figures 4 and 5 , the MRE is expressed in units of degrees / ( cm x M ) , where the molar concentration refers to the number of amino acids rather than protein molecules . One degree / ( cm x M ) equals 100 degrees x cm2/dmol . To score the L1 retrotransposition frequency of L1ORF1p mutants , we adapted a well established cell culture assay ( Moran et al . , 1996 ) that relies on a plasmid-based L1 reporter construct ( pJM101/L1 . 3 ) ( Moran et al . , 1996; Sassaman et al . , 1997 ) and yields G418-resistant HeLa cell colonies only upon a successful retrotransposition . Mutants of the L1 reporter construct were generated by site-directed mutagenesis and are listed in Supplementary file 3 . DNA sequencing was used to verify that the desired mutations were the only changes in the L1 reporter construct . Depending on the L1ORF1p variant and its pre-scored activity , HeLa cells were grown and transfected either in standard six-well plates or in 6 cm dishes . Transfection efficiency was monitored with the help of a luciferase reporter vector ( pCIneo-Rluc-ΔSV40neo , Supplementary file 3 ) ( Lazzaretti et al . , 2009 ) that was co-transfected with each L1 construct . Each series of experiments always included the wildtype L1 reporter construct as a reference . Cells were split 48 h after transfection . In the case of six-well plates , one half of the cells was grown for 12–13 days in DMEM containing G418 , and the other half of the cells was used to measure luciferase activity levels on day 3 after transfection . In the case of the 6 cm dishes , a third of the cells was seeded into 10 cm dishes for G418 selection , and another third of the cells was seeded into six-well plates for a subsequent luciferase activity measurement . The G418-resistant HeLa cell colonies were fixed and stained with Giemsa , colony numbers were scored , and the retrotransposition frequency was determined as the number of G418-resistant colonies per number of transfected cells . In Figures 6 and 7 , the L1 retrotransposition activity is calculated with respect to the wildtype reporter plasmid , with the mean and standard deviations calculated from three independently replicated series of experiments . HeLa cells were provided by Elisa Izaurralde and tested for the absence of Mycoplasma using a ‘MycoAlert’ kit ( Lonza , Basel , Switzerland ) . To monitor protein expression levels of L1ORF1p mutants , HeLa cells were transfected with modified L1 reporter plasmids encoding C-terminal HA-tags on the respective L1ORF1p variants ( Supplementary file 3 ) . HA-tags were inserted by site-directed mutagenesis and DNA sequencing was used to verify that the HA-tag was the only change in the L1 reporter construct . HeLa cells were seeded in six-well plates at a density of 0 . 75 × 106 cells per well and transfected after 24 h . L1 reporter plasmids were co-transfected with plasmid pT7-EGFP-C1-MBP ( Supplementary file 3 ) ( Lazzaretti et al . , 2009 ) to express a GFP-MPB fusion protein as a transfection control . As a reference , each series of experiments always included the wildtype L1 reporter construct with an HA-tagged L1ORF1p . Empty plasmid ( pcDNA3 . 1 ) served as a negative control and endogenous tubulin was detected as a gel loading control . Cells were lysed 48 h post-transfection in a lysis buffer containing 20 mM HEPES ( pH = 7 . 6 ) , 150 mM NaCl and 0 . 4% Igepal-CA630 . The protein concentration in the lysates was quantified using the Bradford reagent ( Bio-Rad , Hercules , California ) . Equivalent amounts of total protein from the lysates were loaded on a polyacrylamide gel for electrophoresis , followed by transfer to a nitrocellulose membrane and probing with antibodies . Monoclonal HRP-conjugated anti-HA antibody ( Roche , Basel , Switzerland , RRID:AB_390917 , 1:5000 ) was used to probe for HA-tagged L1ORF1p . Monoclonal anti-GFP antibody ( Roche , RRID:AB_390913 , 1:2000 ) and monoclonal anti-tubulin antibody ( Sigma Aldrich , St . Louis , Missouri , RRID:AB_477583 , 1:5000 ) were used to probe for GFP-MBP and tubulin , respectively . Polyclonal anti-mouse IgG-HRP ( GE Healthcare , RRID:AB_772193 , 1:10000 ) was used as a secondary antibody . Western blots were developed with the ECL Western Blotting Detection System ( GE Healthcare ) according to the manufacturer's recommendations and protein expression levels were classified as normal ( +++ , more than 70% of wildtype ) , reduced ( ++ , between 70% and 30% of wildtype ) or poor ( + , less than 30% of wildtype ) . The atomic coordinates and structure factors have been deposited in the Protein Data Bank under accession number 6FIA .	Almost half of the human genome consists of DNA strings that have been copied and pasted from one part of the genome to another many thousands of times . These strings of DNA are called mobile genetic elements . Mobile elements can disrupt important genes , causing disease and cancer , but they can also drive evolution . Presently , only one type of mobile element , called LINE-1 , is active in the human genome and able to multiply without help from other mobile elements . LINE-1 DNA is ‘transcribed’ to form molecules of LINE-1 RNA , which can then be ‘translated’ into two distinct proteins . These bind to LINE-1 RNA , which then gets back-transcribed into DNA and inserted as a new LINE-1 element in a new region of the genome . One of the two proteins , called L1ORF1p , forms complexes where three copies of the protein come together . These ‘trimers’ cover and protect LINE-1 RNA and are required for LINE-1 mobility . Different versions of L1ORF1p are found in different animals . Part of the protein is the same across all mammals , and this ‘conserved’ part controls the ability of L1ORF1p to bind to RNA . The non-conserved part of L1ORF1p differs even between humans and their closest animal relatives and little was known about its structure or role . However , this rapidly evolving part of L1ORF1p is essential for LINE-1 mobility . Using X-ray crystallography , Khazina and Weichenrieder obtained a molecular snapshot of the part of L1ORF1p that interacts with other copies of the protein to form trimers . Combined with earlier snapshots of L1ORF1p’s conserved part , this generated a complete structural model of the L1ORF1p trimer . Additional biophysical characterizations suggest that L1ORF1p trimers form a semi-stable structure that can partially open up , indicating how trimers could form larger assemblies of L1ORF1p on LINE-1 RNA . Indeed , the need to maintain a semi-stable structure could explain why L1ORF1p is evolving so rapidly . A second important finding is that the beginning of L1ORF1p needs to be positively charged – a requirement that warrants further exploration . The structural and mechanistic insight into L1ORF1p points to critical new steps in LINE-1 mobilization . It will help to design inhibitor molecules with the goal to halt the mobilization process at various points and to dissect such steps in great detail . Understanding how to control LINE-1 mobility could help to improve stem cell therapies and reproduction assistance techniques , due to the fact that LINE-1 mobility is a potential source of mutation in stem cells , egg and sperm cells , and newly formed embryos .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2018	Human LINE-1 retrotransposition requires a metastable coiled coil and a positively charged N-terminus in L1ORF1p
Recent developments in detector hardware and image-processing software have revolutionized single particle cryo-electron microscopy ( cryoEM ) and led to a wave of near-atomic resolution ( typically ∼3 . 3 Å ) reconstructions . Reaching resolutions higher than 3 Å is a prerequisite for structure-based drug design and for cryoEM to become widely interesting to pharmaceutical industries . We report here the structure of the 700 kDa Thermoplasma acidophilum 20S proteasome ( T20S ) , determined at 2 . 8 Å resolution by single-particle cryoEM . The quality of the reconstruction enables identifying the rotameric conformation adopted by some amino-acid side chains ( rotamers ) and resolving ordered water molecules , in agreement with the expectations for crystal structures at similar resolutions . The results described in this manuscript demonstrate that single particle cryoEM is capable of competing with X-ray crystallography for determination of protein structures of suitable quality for rational drug design . Single-particle cryo-electron microscopy ( cryoEM ) is currently undergoing a revolution due to the recent development of a new generation of detectors using the complementary metal-oxide semiconductors ( CMOS ) technology ( Kuhlbrandt , 2014 ) . These cameras directly detect incoming electrons without the need for a scintillator converting the electrons into photons and are characterized by improved detective quantum efficiencies at all spatial frequencies compared to traditional charge-coupled device cameras ( CCDs ) or photographic films ( Ruskin et al . , 2013; McMullan et al . , 2014 ) . The fast read-out rate of these devices also allows for recording movies composed of multiple frames during typical image exposure times enabling correction of beam-induced sample motion and stage drift to reduce image blurring ( Brilot et al . , 2012; Shigematsu and Sigworth , 2013 ) . Several algorithms have been developed for tracking beam-induced motion and stage drift in movies recorded with direct detectors ( Campbell et al . , 2012; Bai et al . , 2013; Li et al . , 2013 ) . These data processing strategies have been used to produce a multitude of high-resolution reconstructions of samples in the MDa mass range ( Amunts et al . , 2014; Fernandez et al . , 2014; Voorhees et al . , 2014; Wong et al . , 2014; Allegretti et al . , 2014 ) as well as of protein complexes once considered too small for single-particle cryoEM ( ≤300 kDa ) ( Cao et al . , 2013; Liao et al . , 2013; Lu et al . , 2014 ) . These achievements constitute a tremendous leap forward for single-particle cryoEM . This method can now compete with X-ray crystallography for determination of protein structures at near-atomic resolution while also offering the unique advantage of enabling the characterization of heterogeneous or flexible protein complexes ( Scheres et al . , 2007 ) . High-resolution cryoEM can thus provide information about protein dynamics , which is key to understanding the functions of many macromolecular complexes . We report here the structure of the Thermoplasma acidophilum 20S proteasome ( T20S ) , determined at 2 . 8 Å resolution by single-particle cryoEM . This enzyme is a key component of the cell metabolism carrying out degradation of ubiquitinylated polypeptides to regulate the concentration of specific proteins as well as eliminating misfolded products . The T20S proteasome forms a 700 kDa complex comprising 14 α and 14 β subunits organized with D7 symmetry . Our reconstruction enables identification of the conformational preference of some amino-acid side chains ( rotamers ) and resolves ordered water molecules , in agreement with the expectations for crystal structures at similar resolutions . These outcomes constitute a significant step forward in overcoming the resolution limit achievable by single particle cryoEM , which can now provide protein structures of suitable quality for rational drug design . Careful alignment of the microscope was performed before data collection to ensure Thon rings were visible up to ∼2 . 5 Å resolution in the power spectrum of micrographs collected over amorphous carbon using the same conditions as for data acquisition . Prior to beginning data collection , coma-free alignment was carried out to align the beam to the column optical axis with the assistance of a Zemlin tableau ( Glaeser et al . , 2011 ) . Automated data collection was carried out using Leginon ( Suloway et al . , 2005 ) to control both the FEI Titan Krios microscope and the Gatan K2 Summit camera operated in ‘super-resolution’ mode . A subset of micrographs was automatically selected using a filter built into the Appion pipeline ( Lander et al . , 2009 ) that accepts images based on the resolution limit to which Thon rings can be confidently identified in the image power spectra . This is achieved by computing cross-correlation coefficients ( CC ) between the 1-D radially averaged power spectrum of each micrograph and the calculated Contrast Transfer Function ( CTF ) . Only those micrographs showing a CC ≥ 80% at a resolution of 4 Å or better were retained for further processing . One of these micrographs and its corresponding power spectrum are shown in Figure 1 . Using this selection criterion , 196 micrographs were selected , from which a total of 87 , 066 particles were automatically picked using FindEM ( Roseman , 2004 ) . These movies have been deposited in the EMPIAR database with the accession number EMPIAR-10025 . Particle images were sorted and selected based on mean and standard deviation pixel values within Appion ( Lander et al . , 2009 ) and using Xmipp cl2d reference-free alignment and clustering ( Sorzano et al . , 2010 ) to yield a stack of 59 , 864 particles containing both top and side views that were then used for refinement and reconstruction . 10 . 7554/eLife . 06380 . 003Figure 1 . Typical micrograph of ice-embedded T20S proteasome and corresponding power spectrum . ( A ) Micrograph of T20S after movie-frame alignment . Inset: Reference free 2D class averages showing a side view ( left ) and a top view ( right ) . ( B ) Thon rings are visible well beyond 4 Å−1 resolution in the power spectrum of the micrograph shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06380 . 003 The selected substack of T20S particles was refined starting from a low-pass filtered initial model to high resolution using the gold standard procedure implemented in Relion ( Scheres and Chen , 2012; Scheres , 2012b; Bai et al . , 2013 ) . This projection-matching refinement step provided a reconstruction at 2 . 98 Å resolution ( estimated angular accuracy 1 . 3° ) using all 38 movie frames and the full exposure of 53 e−/Å2 . Previous measurements using catalase crystals have suggested that exposures as low as 11 e−/Å2 might be necessary to attain 3 Å maps ( Baker and Rubinstein , 2010; Baker et al . , 2010 ) . However , the loss of signal due to radiation must be balanced against the need for sufficient contrast in the images to allow for accurate alignment of the individual particles . Collecting data in movie mode using direct detectors has profoundly changed the way data is acquired and processed: many reported reconstructions have been obtained after eliminating sub-optimal frames ( Li et al . , 2013 ) or applying a frequency-dependent weighting scheme to the frames ( Scheres , 2014 ) . We emphasize that we were able to obtain a T20S reconstruction at 2 . 98 Å resolution using the full exposure of 53 e−/Å2 and prior to weighting individual movie frames . We obtained similar results in determining the structure of an icosahedral virus to 3 . 7 Å resolution using an exposure of 38 e−/Å2 with a microscope operated at 200 kV ( Campbell et al . , 2014 ) . These results contrast with previous measurements and demonstrate that useful high-resolution information may survive much higher exposures than previously thought for single particle cryoEM . We subsequently used the particle polishing procedure implemented in version 1 . 3 of the Relion software to account for individual beam-induced particle translations and to calculate a frequency-dependent weight for the contribution of individual movie frames to the reconstruction ( Scheres , 2014 ) . Plots of the relative B factor and Cf intersect , derived from the Guinier plots for each single frame reconstruction , along with the corresponding frequency-dependent weights are presented in Figure 2 . The first frame of each movie featured a large movement in comparison with the following ones , which we and others interpret as the initial settling of beam-induced motion during exposure . This processing scheme improved the resolution of the reconstruction to 2 . 83 Å ( estimated angular accuracy 1 . 0° ) , and this was supported by a significant enhancement of map quality . 10 . 7554/eLife . 06380 . 004Figure 2 . Fourier shell correlation curves and radiation damage weighting plots . ( A ) Gold-standard FSC curves for the T20S reconstructions before and after particle polishing as well as after excluding particles based on the uncertainty of the angular assignments . Estimated resolutions are 2 . 98 Å , 2 . 83 Å , and 2 . 81 Å , respectively . The FSC curve computed between the atomic model and the test map is shown in blue . The FSC reaches 0 . 5 at 2 . 86 Å resolution , in agreement with the resolution estimated by gold standard refinement in Relion . ( B ) Estimated values for Bf ( top ) and Cf ( bottom ) during the particle polishing procedure . ( C ) Frequency-dependent relative weights for all movie frames . The first , third , and final movie frames are highlighted in green , red , and blue , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06380 . 004 In a final round of refinement , we discarded particle images with the greatest angular uncertainty based on the height of the probability distributions at their maximum . An empirically determined cutoff of 0 . 07 was used , which gave a stack of 49 , 954 particle images ( 699 , 356 asymmetric units ) . Refinement of this smaller dataset under otherwise identical conditions produced a map with a slightly improved resolution of 2 . 81 Å ( estimated angular accuracy 0 . 8° ) . This result suggests that analyzing the probability distributions over all orientations is a convenient way of discarding particles that are not contributing positively to the reconstruction ( ∼10 , 000 particles in this case ) . This map has been deposited in the EMDB with the accession number EMD-6287 . We fitted the T20S atomic coordinates obtained from a crystal structure at 1 . 9 Å resolution ( Forster et al . , 2005 ) into our cryoEM reconstruction . The density is continuous from residues 13 to 233 and residues 1 to 203 for α and β subunits , respectively , with most side chains being resolved ( Figure 3 ) . In line with previous studies ( Allegretti et al . , 2014; Bartesaghi et al . , 2014 ) , many negatively charged residues had lost their carboxylate groups , likely due to radiation sensitivity . Our T20S reconstruction shows many features supporting the resolution claim of 2 . 8 Å . Differences between the crystal structure and the cryoEM reconstruction are easily discernable at the level of the backbone and can be adjusted accordingly using real-space refinement procedures . The rotameric conformations of many amino-acid side chains are clearly visible in the electron potential map ( Figure 4 ) and reveal differences with a crystal structure of the T20S determined at 3 . 4 Å resolution ( Lowe et al . , 1995 ) . The quality of our reconstruction allows distinguishing between Phe and Tyr amino-acid residues based on the appearance of the density for their side chains ( Figure 4 ) , as shown previously ( Zhang et al . , 2010 ) . The presence of a hydroxyl group para to the phenyl group of Tyr residues unambiguously establishes the presence of the Tyr residue as opposed to the Phe residue in our reconstruction . This level of resolvability is decisive for de novo building of atomic models of proteins of unknown structures . 10 . 7554/eLife . 06380 . 005Figure 3 . CryoEM reconstruction of the T20S at 2 . 8 Å resolution . ( A ) Isosurface representation of the T20S map . ( B ) An α-helical segment from one β subunit is shown in ribbon representation docked into the corresponding region of the reconstruction . ( C ) Same α-helical segment as in ( B ) shown in atom representation docked into the corresponding region of the reconstruction . Several water molecules are visible . DOI: http://dx . doi . org/10 . 7554/eLife . 06380 . 00510 . 7554/eLife . 06380 . 006Figure 4 . Identification of the rotameric conformation of amino acid side chains and resolving of ordered water molecules in the T20S cryoEM reconstruction at 2 . 8 Å resolution . ( A ) Different rotameric conformations adopted by the Met-14 side chain ( β-subunit ) between a crystal structure of the T20S determined at 3 . 4 Å resolution ( left , PDB 1PMA ) and the EM density ( right ) . ( B ) Unambiguous establishment of the rotameric conformations of two different isoleucine residues: Ile70 ( left , α-subunit ) and Ile37 ( right , β-subunit ) . ( C ) The additional density accounting for the hydroxyl group of tyrosine side chains ( left , Tyr132 , α-subunit ) is prominent when compared to phenylalanine side chains ( right , Phe91 , α-subunit ) . ( D ) Numerous water molecules are resolved in the T20S cryoEM map . Left: a water molecule hydrogen-bonded to the carbonyl group of Val87 ( α-subunit ) . Right: a water molecule hydrogen-bonded to the side chain hydroxyl of Thr102 ( α-subunit ) . The cross-validation of the assignment of these water molecules using gold-standard half maps can be found in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06380 . 00610 . 7554/eLife . 06380 . 007Figure 4—figure supplement 1 . Cross-validation of water molecule assignment through comparison of gold standard half maps . ( A ) Water molecule hydrogen-bonded to Val87 . Left: The EM density reconstructed using all particles ( 49 , 954 ) . Middle and right: gold standard half-maps . ( B ) Water molecule hydrogen-bonded to Thr102 . Left: EM density reconstructed using all particles . Middle and right: gold standard half-maps . In each case , the density attributed to water is present in both half maps confirming that it is not due to random noise . Furthermore the position of these water molecules matches those observed in a crystal structure of the proteasome determined to 1 . 9 Å resolution ( PDB 1YAR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06380 . 007 In X-ray crystallography , when a structure is determined at a resolution significantly better than 3 Å , one expects to observe ordered water molecules interacting with amino-acid residues located at the protein surface ( Carugo and Bordo , 1999 ) . The higher the resolution the more ordered water molecules are visible . Analysis of our T20S reconstruction at 2 . 8 Å resolution revealed that many ordered water molecules are resolved , in agreement with the estimated resolution . They are positioned in appropriate chemical environments and within hydrogen-bonding distance ( 2 . 8–3 . 5 Å ) of the surrounding atoms with which they interact . All the reconstructions described in this manuscript used the RELION gold standard refinement procedure ( Scheres and Chen , 2012 ) . In this protocol , two half datasets are refined completely independently throughout , and after each cycle the new reference volumes are low-pass filtered to the resolution where the Fourier shell correlation ( FSC ) between the two volumes drops to 0 . 143 ( Rosenthal and Henderson , 2003 ) , thus preventing overfitting of noise . As a result , the noise present in each independent map is not correlated . We were therefore able to cross-validate our assignment of water molecules as they are accounted for by the density in both maps derived from half datasets despite the reduced number of particle images contributing to each map ( Figure 4—figure supplement 1 ) ( DiMaio et al . , 2013 ) . Finally , the 1 . 9 Å crystal structure of the T20S proteasome also unequivocally supports the assignment of the water molecules identified in the cryoEM map as an independent cross-validation metric ( Forster et al . , 2005 ) . The structure presented here demonstrates that cryoEM is now capable of producing structures at better than 3 Å resolution , which is a prerequisite for structure-based drug design ( Anderson , 2003 ) . The possibility to resolve the conformational preference of some amino-acid side chains ( rotamers ) and identify ordered water molecules opens new horizons for cryoEM and structural biology in general . This level of information is required for drug lead design as these chemical entities mediate key biological processes such as catalysis , protein/protein and protein/substrate recognition . State-of-the-art cryoEM is now in a position to expedite structure-based drug discovery and will likley become increasingly important for pharmaceutical industries . We believe several factors contributed to obtaining the resolution reported here . The proteasome itself is a rigid and homogenous sample , which is ideal for high-resolution reconstruction . For data collection , we used a relatively high electron dose , which may have been key to properly aligning particles . Furthermore , we used the mechanical stage instead of beam-tilt to move to targets for high magnification exposure in order to avoid introducing any phase shift in the images . Images of each square and each hole were examined by-eye for perceived ice thickness and only those which were judged to have very thin ice were selected for final high magnification exposures . Finally , we benefited from the availability of algorithmic advances ( projection matching , particle polishing ) , which allows optimization of the doses used for alignment and final reconstruction . As the potential of cryoEM has been steadily growing in the past few years , the prospect of achieving reconstructions at ∼2 Å resolution is very attractive . Remarkably , we used only 10% of the particle images present in the dataset discussed in this manuscript . Including more particle images could further improve the resolution of the T20S reconstruction and experiments are ongoing to determine if this is the case . Finally , 3D classification approaches have also proven useful for sorting particle images ( Scheres , 2012a; Lyumkis et al . , 2013 ) and implementing such approaches may also help to improve the final resolution of our reconstruction . The T . acidophilum 20S proteasome was expressed and purified from Escherichia coli according to established protocols ( Rabl et al . , 2008; Yu et al . , 2010 ) . Three microliters of sample were applied to a 1 . 2/1 . 3 C-flat grid ( Protochips , Raleigh , North Carolina ) , which had been plasma-cleaned for 6 s at 20 mA using a Gatan Solarus ( Pleasanton , California ) . Thereafter , grids were plunge-frozen in liquid ethane using a Gatan CP3 and a blotting time of 2 . 5 s . Data were acquired using an FEI Titan Krios ( Hillsboro , Oregon ) transmission electron microscope operated at 300 kV and equipped with a Gatan K2 Summit direct detector . The extraction voltage was 4500 , the gun lens setting 3 and the spotsize 8 . A condenser aperture of 70 µm and an objective aperture of 100 µm were used . Coma-free alignment was performed using the Leginon software ( Glaeser et al . , 2011 ) . Automated data collection was carried out using Leginon ( Suloway et al . , 2005 ) to control both the FEI Titan Krios ( used in microprobe mode at a nominal magnification of 22 , 500× ) and the Gatan K2 Summit operated in ‘super-resolution’ mode ( super-resolution pixel size: 0 . 6575 Å ) at a dose rate of ∼9 counts/physical pixel/s which corresponds to ∼12 electrons/physical pixel/s ( when accounting for coincidence loss ) . We waited 40 s after physically moving the stage to each new position to allow settling before acquiring a new video . A single movie was taken per hole . Each movie had a total accumulated exposure of 53 e−/Å2 fractionated into 38 frames of 200 ms ( yielding movies of 7 . 6 s duration ) . A dataset of ∼1000 micrographs was acquired for the T20S in a single session using a defocus range between 0 . 9 and 2 . 4 µm . Whole frame alignment was carried out using the software developed by Li et al . ( 2013 ) , which is integrated into the Appion pipeline ( Lander et al . , 2009 ) , to account for stage drift and beam-induced motion . We used a frame offset of seven along with a B factor of 1000 pixels2 for aligning the movie frames ( Li et al . , 2013 ) . The parameters of the microscope contrast transfer function were estimated for each micrograph using ctffind3 ( Mindell and Grigorieff , 2003 ) . Particles were automatically picked using FindEM ( Roseman , 2004 ) integrated into the Appion pipeline ( Lander et al . , 2009 ) before sorting and selection based on mean and standard deviation pixel values and using Xmipp cl2d reference-free alignment and clustering ( Sorzano et al . , 2010 ) . Extraction of particle images for projection-matching refinements was performed using Relion 1 . 3 with an initial box size of 448 pixels2 and applying a windowing operation in Fourier space to downsize the particles images to yield a final box size of 300 pixels2 ( corresponding to a pixel size of 0 . 98 Å ) . The final pixel size was chosen to optimize the balance between the Nyquist frequency limit and the memory requirements for computational steps . Projection-matching refinements were performed with the Relion software ( Bai et al . , 2013; Scheres , 2012a , 2012b , 2014 ) imposing D7 symmetry . Reported resolutions are based on the gold-standard FSC = 0 . 143 criterion ( Scheres and Chen , 2012 ) and Fourier shell correction curves were corrected for the effects of soft masking by high-resolution noise substitution ( Chen et al . , 2013 ) . The T20S crystal structure ( PDB 1YAR ) was rigid body fitted into the cryoEM map using UCSF Chimera ( Goddard et al . , 2007 ) and then iteratively refined using Rosetta ( DiMaio et al . , 2009 ) and Coot ( Emsley et al . , 2010 ) . We used Rosetta tools recently developed for cryoEM ( DiMaio et al . , 2015; Wang et al . , 2015 ) to perform fragment rebuilding , torsion angles and cartesian minimization of atomic positions and B factor refinement . At each iteration , the best model was visually inspected in Coot and a few amino acid side chain rotamers were manually adjusted to best fit the density . Water molecules were also manually added using Coot after the first iteration of Rosetta refinement . Rosetta and Coot refinements were performed using a training map corresponding to one of the two maps generated by the gold-standard refinement procedure in Relion . The second map ( testing map ) was used only for calculation of the FSC compared to the atomic model ( DiMaio et al . , 2013 ) . The quality of the final model was analyzed with Molprobity ( Chen et al . , 2010 ) .	Proteins perform many critical tasks within cells , and to do so , they must first fold into specific shapes . Being able to visualize these shapes can help scientists to understand how proteins work , and help them create drugs that can interact with the proteins to treat diseases . The past few years have seen the rapid development of an imaging technique called single-particle cryo-electron microscopy ( or cryoEM for short ) , and this technique is now increasingly used to investigate protein structures . First , proteins are embedded in a thin film of non-crystalline ice by rapidly cooling to around the temperature of liquid nitrogen ( below −180°C ) . This traps the protein in the shape it has in solution . High-energy electrons are then transmitted through the protein sample and their interaction with the atoms in the protein is recorded by a direct electron camera . The analysis of a large series of images recorded in this way can be used to determine the approximate positions of the atoms in the protein . Previously , single-particle cryoEM techniques have not produced a detailed enough protein structure to be useful to scientists interested in drug development . By refining these techniques , Campbell , Veesler et al . have now obtained the most detailed cryoEM protein structure to date—a structure of an enzyme complex that helps get rid of proteins that are misfolded or that have become too abundant . The structure is so detailed that it reveals the shapes of some small groups of atoms that stick out from the sides of amino acids in the enzyme complex . ( Amino acids are the building blocks of enzymes and all other proteins . ) Moreover , the structure shows where individual water molecules are positioned around the protein . The level of detail in the structure produced by Campbell , Veesler et al . is high enough to be useful to drug researchers . Furthermore , because only 10% of the images Campbell , Veesler et al . collected were used to produce the structure , future work will investigate whether incorporating more of the images could reveal structures in even greater detail .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "structural", "biology", "and", "molecular", "biophysics" ]	2015	2.8 Å resolution reconstruction of the Thermoplasma acidophilum 20S proteasome using cryo-electron microscopy
Identifying individuals who are at high risk of cancer due to inherited germline mutations is critical for effective implementation of personalized prevention strategies . Most existing models focus on a few specific syndromes; however , recent evidence from multi-gene panel testing shows that many syndromes are overlapping , motivating the development of models that incorporate family history on several cancers and predict mutations for a comprehensive panel of genes . We present PanelPRO , a new , open-source R package providing a fast , flexible back-end for multi-gene , multi-cancer risk modeling with pedigree data . It includes a customizable database with default parameter values estimated from published studies and allows users to select any combinations of genes and cancers for their models , including well-established single syndrome BayesMendel models ( BRCAPRO and MMRPRO ) . This leads to more accurate risk predictions and ultimately has a high impact on prevention strategies for cancer and clinical decision making . The package is available for download for research purposes at https://projects . iq . harvard . edu/bayesmendel/panelpro . The workflow of the package includes four main parts: the input , including user and model input; pre-processing of the inputs , including user input checks and a database build; running the peeling-paring algorithm; and outputting the results . Figure 1 shows the workflow . Additionally , Appendix 1—figure 1 shows detailed sub-routines within the package . The checked pedigree and PanelPRODatabase subset , as well as any user options , are then passed to the ‘peeling-paring’ algorithm , which approximates Equation 2 in the Genotype probabilities section . It is based on the ‘peeling’ algorithm as introduced by Elston and Stewart , 1971 with its implementation based on Fernando et al . , 1993 . The ‘paring’ aspect of the algorithm limits the number of simultaneous mutations allowed . This is called the paring parameter and has a default value of 2 , which results in an approximation which has been shown to be adequate for clinical purposes ( Madsen et al . , 2018 ) . When the paring parameter is set equal to the number of distinct genes to be considered , the calculation is exact ( assuming no other missing information about the pedigree ) . Future cancer risks are then calculated based on the law of total probability , using the previously calculated posterior carrier probabilities , as described in the Methods: Mendelian modeling section . These two calculations are performed in PanelPROCalc , as listed in Table 3 . The underlying algorithm is written in Rcpp using the RcppArmadillo package ( Eddelbuettel and Sanderson , 2014; Eddelbuettel and Francois , 2011 ) . It uses , as much as possible , optimized data structures , vectorized operations and in-place modifications to be both time and memory efficient . See the Discussion section for benchmarks on the run-time of the implementation . The recursive nature of the peeling-paring algorithm allows for multiple counselees to be specified in the function call without significant increase in the computational time . This is an advantage when multiple family members are at high risk and would benefit from knowing their carrier probabilities and future cancer risks . For each proband in the pedigree , the output consists of: Messages or warnings generated from checkFam have been omitted in the example below for brevity . output <- PanelPRO ( pedigree = test_fam_1 , cancers = c ( ‘Breast’ , ‘Ovarian’ ) , genes = c ( ‘BRCA1’ , ‘BRCA2’ , ‘ATM’ , ‘MSH2’ ) , max . mut = 2 , parallel = FALSE ) ## Your model has two cancers - Breast , Ovarian and four genes - BRCA1_hetero_anyPV . . . The package includes the function visRisk to visualize the output graphically . Figure 4 demonstrates this usage for test_fam_1 . The visRisk function was implemented using the plotly ( Sievert , 2020 ) package , so that the output can be rendered interactively and display the exact probabilities upon hovering . In this section , we display some of the other test pedigrees included in PanelPRO and their corresponding output from the visRisk function , as well as a comparison to some other platforms and models . We compared PanelPRO to: BRCAPRO and MMRPRO in BayesMendel ( Chen et al . , 2004 ) , CanRisk ( Lee et al . , 2019; Carver et al . , 2021 ) , IBIS ( Tyrer et al . , 2004 ) , and PREMM-5 ( Kastrinos et al . , 2017 ) , which all support different cancers , genes and model assumptions . Table 4 summarizes the supported inputs and outputs of each model . If a particular model or platform does not support certain cancers , the irrelevant family history is simply omitted from input . For brevity , Figures 5–9 provide visualizations of the pedigrees of these additional examples , and the corresponding model outputs are reported in as figure supplements . In all cases , we used the default model settings ( if any ) . Many of the aforementioned platforms have long reports as outputs , so we have only included the portions concerned with carrier probabilities and future risks . The same information contained in the PanelPRO sample pedigrees is input to the other models; however not all of the features are used by these other platforms . Conversely , there are some inputs for the other models that PanelPRO does not include . Notably , all the cancers and genes supported by these other models are a subset of those supported in PanelPRO , except for PREMM-5 , which takes into consideration other cancers associated with Lynch syndrome which are not currently included in PanelPRO ( bile duct and sebaceous gland ) . However , PREMM-5 does not provide the associated future risk for these additional cancers , only carrier probabilities for MLH1 , MSH2 , MSH6 , PMS2 , and EPCAM . Comparing PanelPRO with BRCAPRO and MMRPRO , we see that PanelPRO offers carrier probability estimates for a larger set of genes , as well as a graphical output of the future risk . IBIS does not give any estimates for carrier probabilities; however , it gives a summary of the future risks in text format , relative to population averages . Finally , PREMM-5 gives an estimate of carrying any of five genes ( MLH1 , MSH2 , MSH6 , PMS2 , or EPCAM ) , whilst PanelPRO is able to give estimates for each of those individual genes . PREMM-5 also does not give estimates for future risks of cancer . Two pedigrees that illustrate the differences between PanelPRO and PREMM-5 are test_fam_7 ( Figure 5 ) and test_fam_11 ( Figure 9 ) . For test_fam_7 , PanelPRO estimates a 42 . 3% probability of carrying an MLH1 , MSH2 , MSH6 , PMS2 , or EPCAM mutation ( when excluding the possibility of multiple simultaneous gene mutations ) , compared to 32% for PREMM-5 . This pedigree only contains family history of colorectal and endometrial cancers , which PREMM-5 uses as key risk factors , leading to similar results . In contrast , the mutation probability estimates between the same two models for test_fam_11 are quite different . This pedigree is an extreme example that contains history of endometrial , small intestine , ovarian , and pancreatic cancers , but PREMM-5 groups the latter three cancers into a single risk factor for any other Lynch syndrome-associated cancers . The differences in the model approaches and assumptions result in PanelPRO giving a 93% estimate for a mutation in any of the aforementioned genes ( without multiple simultaneous gene mutations ) , while PREMM-5 returns 3 . 2% . test_fam_11 is an extreme pedigree , but it nonetheless illustrates the flexibility of PanelPRO for incorporating very detailed pedigree information with a high clinical impact . We list the key functions with their input ( s ) and output ( s ) in Table 3 . The PanelPRO function calls the pre-processing functions and the algorithm engine in the back-end , so we expect that most users will only need to use this main function . However , the other functions can be called separately if desired . For example , users can call buildDatabase to inspect the database of model parameters or run checkFam to examine the pedigree after it has been checked . In this section , we give the mathematical details of the main PanelPROCalc engine , which encompasses approximating genotype distributions of counselees and their future cancer risks . PanelPRO predicts an individual’s probability of having a specified genotype . We use the notation in Table 5 . Without loss of generality , let the subscript i=1 represent the counselee ( i . e . the individual who is counseled ) . For simplicity , we only consider one counselee , although the model can handle multiple counselees in a computationally efficient manner . The counselee’s genotype probability is: ( 1 ) P ( G1\|H , U ) . Using Bayes’ rule , the law of total probability and the assumption of independence of family phenotypes given genotypes and sex , this can be written as ( 2 ) P ( G1\|H , U ) ∝P ( G1 ) ∑G2 , … , GI∏i=1IP ( Hi\|Gi , Ui ) P ( G2 , … , GI\|G1 ) =P ( G1 ) ∑G2 , … , GI∏r=1R∏i=1IP ( Hri\|Gi , Ui ) P ( G2 , … , GI\|G1 ) . From this representation of the posterior probability , we can clearly see the model and user inputs to PanelPRO . P⁢ ( 𝐆1 ) represents the allele frequencies for each gene in the model . P⁢ ( 𝐇r⁢i\|𝐆i , Ui ) are derived from the cancer penetrances P ( Tr⁢i=t\|𝐆i , Ui ) . Explicitly , P⁢ ( 𝐇r⁢i\|𝐆i , Ui ) ={1-∑s=1CiP ( Tr⁢i=s\|𝐆i , Ui ) if ⁢δr⁢i=0P ( Tr⁢i=Tr⁢io⁢b⁢s\|𝐆i , Ui ) if ⁢δr⁢i=1where Tr⁢i is the random variable and Tr⁢io⁢b⁢s is the observed cancer age . By default , the allele frequencies and penetrances are obtained from existing peer-reviewed studies and estimates , but are completely customizable within PanelPRO . Since the genotype space { ( 𝐆2 , … , 𝐆I ) :𝐆i∈{0 , 1}K , i=2 , … , I} is large for large values of K , we use the peeling-paring algorithm ( Madsen et al . , 2018 ) as an approximation , only allowing a pre-specified number of mutations to be simultaneously present in the same individual . The pedigree structure from the user input is used to derive the P⁢ ( 𝐆2 , … , 𝐆I\|𝐆1 ) term in Equation 2 using Mendelian laws of inheritance . PanelPRO also estimates future cancer risk , based on the previously calculated genotype distribution of the individual . Suppose the counselee has not developed the r th cancer by their current age . Then the risk of developing the r th cancer in t0 years is ( 3 ) P ( Tr⁢1≤C1+t0 , Jr⁢1=1\|𝐇 , 𝐔 ) =∑𝐆1P ( Tr⁢1≤C1+t0 , Jr⁢1=1\|𝐆1 , 𝐔 ) P ( 𝐆1\|𝐇 , 𝐔 ) . Equation 3 produces so-called ‘crude’ risk , since competing risks of death from causes other than the specified cancer are accounted for . Thus , the reported future risk is the probability that the counselee develops the r th cancer within the next t0 years and does not die from other causes beforehand , given the cancer history and sexes of the family . P ( Tr⁢1*≤C1+t0 , Jr⁢1=1\|𝐆1 , 𝐔 ) is the crude penetrance and is also a model input with default values estimated from the literature . PanelPRO also provides the option to report ‘net’ future risk , which is the probability that the counselee develops the r th cancer in a hypothetical world where they cannot die from other causes , given the cancer history and sexes of the family . This risk type is not as realistic but some clinicians find it useful , as it focuses on the specified cancer and allows them to factor qualitatively the patient-specific covariates that may affect the patient’s risk . To report net future risk , PanelPRO uses the net penetrances P ( Tr⁢i=t\|𝐆i , Ui ) . Note that the genotype probabilities in Equation 2 were calculated using net penetrances , as we do not collect death from other causes as a user input . PanelPRO is a highly flexible package which provides an interface to efficiently calculate carrier probabilities for a wide array of cancer susceptibility genes , as well as future cancer risks . It is designed for R users . Similarly to the BayesMendel package , it can provide the computational engine behind clinical and counseling decision support tools . It excels in being fully customizable . Any combination of the 24 genes and 18 cancers currently in version 0 . 2 . 0 of the package can be included in the model . New genes and cancers can easily be added , and in fact the code allows for an arbitrary number of genes and cancers . Risk modifiers have been included for certain procedures , and more can be added as additional information becomes available . The user can also change the internal database of parameter values . The package includes a comprehensive check on the input pedigree to ensure users are informed of potentially inconsistent or infeasible data entries . When it is possible to do so safely , the data is automatically remedied and the user is then notified . Otherwise , the program will halt with an informative error message . Once the pedigree is pre-processed , the posterior probabilities are calculated efficiently . For example , test_fam_1 , which has 19 members and family history of 2 cancers , runs with all the default settings in a few seconds as shown in Figure 10 . The polynomial run-time of the peeling-paring algorithm is alleviated with PanelPRO’s Rcpp implementation . Even when relaxing the maximum mutations ( paring ) parameter , the C++ implementation is able to handle the calculations efficiently . Run-times in these ranges are certainly appropriate for clinical use , as well as use in a research setting where possibly hundreds of pedigrees have to be processed through PanelPRO . Moreover , the peeling-paring algorithm run-time scales linearly in the number of family members in the pedigree and can handle hundreds of members in an inter-generational configuration easily . PanelPRO has two main limitations . Firstly , the initial release does not handle pedigrees which contain loops . This additional functionality would be desirable in future releases , although loops in pedigrees do not happen frequently . Several studies suggest either exact or approximate computations for pedigrees with loops , see Stricker et al . , 1995 and Totir et al . , 2009 . Secondly , the polynomial scaling of peeling-paring as a function of the number of genes considered becomes significant when many genes are incorporated . This issue is of concern because we strive for future releases to contain far more genes than 24 as data becomes available . Alternative algorithms which have different time complexity properties , such as the Lander-Green family of algorithms ( Lander and Green , 1987 ) , should be explored . These algorithms scale linearly in terms of the number of genes considered , but are exponential in the number of family members in the pedigree ( Gao et al . , 2009 ) . A future objective for this package is to contain a choice of the carrier probability calculation method , and ideally an automatic selection of the one which is most efficient , depending on family size and total number of genes . Appropriate thresholds of these two parameters need to be determined by a comprehensive benchmarking exercise .	Genetic mutations that increase cancer risk can be passed down from parents to their children , which can affect families across many generations . In these families , multiple members may be affected by different types of cancer , and these cancers often develop at an early age . Unaffected family members are often referred to genetic counselling , where they can explore their own risk of cancer . Clinicians and genetic counselors can provide recommendations to minimize cancer risk and inform personal choices on how to manage that risk , such as opting for preventative surgeries or participating in regular screening . In genetic counselling sessions , highly trained clinicians and specialists use software that takes an individual’s family history of cancer and uses it to estimate their individual risk of carrying certain genetic mutations . These estimates can in turn help to predict their future risk of cancer . Many existing software packages are limited to estimating risks based on mutations in well-known cancer-related genes , such as BRCA1 and BRCA2 in breast and ovarian cancer . However , emerging evidence suggests that many of the genes associated with cancer risk work as part of a complex and overlapping network . Since current risk-profiling software packages are only designed to consider such genes in isolation , they cannot generate the most robust , accurate or comprehensive cancer risk profiles . To address this challenge , Lee , Liang et al . have developed a new risk-profiling software that can integrate a large number of gene mutations and a wide range of potential cancer types to provide more accurate estimates of individual cancer risk . This software , called PanelPRO , uses evidence identified from extensive literature reviews to model the complex interplay between genes and cancer risk . The software not only calculates risks based on known genes , but also allows other developers to integrate new cancer-related genes that may be identified in the future . Importantly , the software is compatible with genetic counselling applications , since it returns answers within seconds when reasonable family and gene database sizes are used . PanelPRO is a new , modern , flexible and efficient software package that provides an important advance towards modelling the vast genetic and biological complexity that contributes to inherited cancer risk . This software is designed to provide a more accurate and comprehensive estimate of cancer risk for individuals with family histories of cancer . As an open-source software , it is freely available for research purposes , and can be licensed by software companies and healthcare organizations to integrate electronic patient records and rapidly identify at-risk individuals across larger patient groups . Ultimately , this software has the potential to improve cancer prevention strategies and optimize the personalized decision-making processes around cancer risk .	[ "Abstract", "Introduction", "Discussion" ]	[ "tools", "and", "resources", "genetics", "and", "genomics", "cancer", "biology" ]	2021	Multi-syndrome, multi-gene risk modeling for individuals with a family history of cancer with the novel R package PanelPRO
When spinal circuits generate rhythmic movements it is important that the neuronal activity remains within stable bounds to avoid saturation and to preserve responsiveness . Here , we simultaneously record from hundreds of neurons in lumbar spinal circuits of turtles and establish the neuronal fraction that operates within either a ‘mean-driven’ or a ‘fluctuation–driven’ regime . Fluctuation-driven neurons have a ‘supralinear’ input-output curve , which enhances sensitivity , whereas the mean-driven regime reduces sensitivity . We find a rich diversity of firing rates across the neuronal population as reflected in a lognormal distribution and demonstrate that half of the neurons spend at least 50 % of the time in the ‘fluctuation–driven’ regime regardless of behavior . Because of the disparity in input–output properties for these two regimes , this fraction may reflect a fine trade–off between stability and sensitivity in order to maintain flexibility across behaviors . Rhythmic movements , such as walking , scratching , chewing and breathing , consist of a recurrent sequence of activity , which is generated by neuronal networks primarily in the spinal cord and medulla . Although , this sequential activity is formed by collective communication among the neurons , it is unknown how the participation is shared versus divided within the population . Distinct motor tasks have been reported to be divided among dedicated microcircuits in zebrafish ( Ampatzis et al . , 2014; Bagnall and McLean , 2014; Fetcho and McLean , 2010 ) . Nevertheless , do all neurons , which are dedicated to a particular motor activity , spike at approximately the same rate ? Or do only some neurons spike at high rate , while most others spike at lower rates ? An arrangement with a spectrum of different firing rates could be beneficial by adding the possibility of increasing the overall activity , for instance during uphill walking where a stronger force is needed . In this way the spinal circuit could enhance flexibility by adopting a diversity of firing rates across the population . Other networks in the central nervous system face a similar challenge of how to distribute the activity across the population in order to collectively increase the dynamic range ( Wohrer et al . , 2013 ) . In sensory processing , neural circuits must be able to retain sensitivity both to weak and strong input . Weak stimuli are amplified whereas strong stimuli are attenuated in order to reduce saturation . If there is too much activity , the circuit reaches saturation and therefore loses the ability to resolve differences in sensory input . Furthermore , amplification of weak signals by recurrent excitation pose the risk of unstable activity , which can spin out of control ( Vogels et al . , 2005 ) . This computational challenge of how networks maintain both stability and sensitivity is an open question especially for spinal networks . Stability has primarily been investigated in cortical networks and much evidence suggest that local excitation is carefully balanced by inhibition to assure stability and to widen the range of operation ( Galarreta and Hestrin , 1998; Shu et al . , 2003 ) . It is well–established that unstable states such as epileptiform activity can easily be achieved by shifting the balance in favor of excitation , e . g . by blocking inhibition ( Dichter and Ayala , 1987; Bazhenov et al . , 2008 ) . The concept of balanced excitation ( E ) and inhibition ( I ) ( balanced networks in short ) was introduced two decades ago ( Shadlen and Newsome , 1994; van Vreeswijk and Sompolinsky , 1996 ) and has sparked numerous studies both theoretical ( Amit and Brunel , 1997; Ozeki et al . , 2009; van Vreeswijk and Sompolinsky , 1998; Kumar et al . , 2008 ) as well as experimental ( Berg et al . , 2007; Okun and Lampl , 2008; Higley and Contreras , 2006; Wehr and Zador , 2003; Kishore et al . , 2014 ) . The primary purpose of theoretical models of balanced networks was initially to understand irregular spiking , which was widely observed in experiments ( Bell et al . , 1995; Shadlen and Newsome , 1994 ) . Irregular spiking was puzzling because it could not be explained by random arrival of excitatory input alone , since this randomness was effectively regularized by temporal integration ( Denève and Machens , 2016; Softky and Koch , 1993 ) . Models of balanced networks not only were able to explain irregular spiking , but also revealed other interesting phenomena , such as emergent linearity ( van Vreeswijk and Sompolinsky , 1996 ) , multifunctionalism ( Sussillo and Abbott , 2009; Hennequin et al . , 2014 ) and self–sustained stable network activity ( Amit and Brunel , 1997; Hansel and Mato , 2001; Ikegaya et al . , 2013 ) . The consensus view thus became that irregular spiking results from a mean membrane potential , which is lurking just below threshold , where it is restrained by inhibition concurrent with excitation ( Shadlen and Newsome , 1998; Bell et al . , 1995; Salinas and Sejnowski , 2000 ) , although synchrony of random excitation is sometimes needed when individual synaptic potentials are small ( Stevens and Zador , 1998 ) . This view was essentially predicted much earlier in random walk models ( Gerstein and Mandelbrot , 1964 ) . The concept of balanced E/I is now an integrated part of understanding network processing in cortex and elsewhere , but for some reason it has been forgotten in understanding spinal motor networks , with the exception of a few isolated studies ( Berg et al . , 2007; Petersen et al . , 2014 ) . The balanced E/I allow a subthreshold fluctuating membrane potential , where the spikes are evoked by synaptic transients and therefore belong to the fluctuation–driven regime ( Kuhn et al . , 2004; Tiesinga et al . , 2000 ) . This is in contrast to the more traditional mean–driven spiking ( Figure 1 ) , where the mean membrane potential ( Vm ) is well above threshold and spike timing is controlled by after–hyperpolarization ( Gerstner et al . , 2014; Renart et al . , 2007 ) . These two regimes have contrasting manifestations ( Table 1 ) : The fluctuation–driven regime has a skewed/lognormal firing rate distribution whereas the mean–driven regime has regular spiking and a symmetric distribution . A simple mechanism has been proposed to explain the lognormal firing in the fluctuation–driven regime by Roxin et al . ( 2011 ) : The skewness in distribution arises out of a supralinear transformation of the synaptic input , which is Gaussian by virtue of the central limit theorem ( Figure 1A ) . A response to multiple input , which is larger than the sum of their individual responses ( i . e . supralinear ) , will enhance sensitivity ( Rubin et al . , 2015 ) and therefore this mechanism may constitute an important physiological purpose . 10 . 7554/eLife . 18805 . 003Figure 1 . Skewness of the rate distribution reveals two regimes of neuronal spiking . ( A ) In the fluctuation–driven regime the mean input is below the spiking threshold and the IO-curve has a nonlinear shape . A normally distributed input current ( shown below x–axis ) is transformed into a skewed firing rate distribution ( y-axis ) . ( B ) In contrast , if the mean input is above threshold , the transformation is linear and the firing rate distribution is symmetric . ( C ) IO–function for both regimes: Linear for suprathreshold region and nonlinear for subthreshold region . The noise level affects the curvature of the nonlinearity ( 3 curves illustrate different levels of noise ) . ( D ) Sample recordings during motor activity from two spinal neurons in the subthreshold region , where the spiking is irregular and driven by fluctuations , and the supra–threshold region ( E ) , where the mean input is above threshold and spiking is regular . Highlighted area shown at bottom . Spikes in bottom panel are clipped . Tick marks: −50 mV , scale bars: 5 mV . ( A–B ) adapted from ( Roxin et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 00310 . 7554/eLife . 18805 . 004Table 1 . Two regimes of neuronal spiking and their definition , properties and causes . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 004Fluctuation–drivenMean–drivenKey referencesDefinitionRmItotal < VthresRmItotal > Vthres ( Gerstner et al . , 2014; Brunel , 2000 ) PropertiesLower firing ratesHigher firing ratesIrregular spikingRegular spiking ( Amit and Brunel , 1997; Shadlen and Newsome , 1998; van Vreeswijk and Sompolinsky , 1998 ) Lognormal/Skewed distributionSymmetric distribution ( Buzsáki and Mizuseki , 2014 ) ( Roxin et al . , 2011; Mizuseki and Buzsáki , 2013 ) CauseBalanced E/IIntrinsic currents , unbalanced E/I ( Bell et al . , 1995; Shadlen and Newsome , 1994; Softky and Koch , 1993 ) Synchronized excitation ( Stevens and Zador , 1998 ) This is in contrast to the mean–driven regime where the summation is linear or even sublinear , which will transform a normally distributed input to a normally ( as opposed to lognormally ) distributed firing rate ( Figure 1B ) . Such linear ( or sublinear ) transformation will reduce rather than enhance sensitivity and therefore the mean–driven regime will curb network activity ( Ahmadian et al . , 2013 ) . These two transformations work together into an S-shaped IO-curve , where weak input are amplified yet the network is kept stable for strong activity ( Figure 1C ) . Sample neurons in the two regimes are shown ( Figure 1D–E ) . If this mechanism is true , then the shape of the firing rate distribution will reveal the spiking regime of a given neuron . The degree to which neurons operate in one versus the other regime may hold the key to understanding stability , dynamic range and other important properties of network operations . Yet this still remains to be investigated , especially in spinal networks . Here , we investigate the regimes of operation of spinal neurons during different rhythmic motor behaviors , which are generated in the lumbar spinal circuits of turtles . We test the theoretical scheme put forward by Roxin et al . ( 2011 ) , by assessing the synaptic input , the spike response function in subthreshold domain , and determine the shape of the firing rate distribution . The mechanical stability of the turtle preparation allows electrophysiological recordings of unprecedented quality , such that we can combine intracellular recording with multi–electrode arrays , and thus determine the fraction of the population in the two regimes at all times . The high resistance to anoxia of turtles allows using adult animals with fully developed spinal circuitry , which have healthy network activity and which can perform multiple complex motor behaviors ( Stein , 2005 ) . Thus , we can investigate the population activity during , not just one behavior , but multiple motor behaviors . Custom designed high–density silicon electrodes recorded the population activity from hundreds of cells in the dorsoventral and rostrocaudal axes along with the intracellular Vm of single neurons and multiple relevant motor nerves ( Figure 2 ) . This is a unique experimental investigation , because it explores the link between neuronal ensemble data , which in itself is rare in spinal motor research , and the forefront of theoretical neuroscience . 10 . 7554/eLife . 18805 . 005Figure 2 . Parallel neuronal activity in the lumbar enlargement during rhythmic motor activity . ( A ) Illustration of experiment with three silicon probes inserted into the lumbar spinal cord of a turtle . Histological verification: transverse ( B ) and sagittal ( C ) slices , 200 μm thick , showing the location of the silicon probes in the spinal cord ( red traces and location illustrated on right , electrodes stained with DiD ) . ChAT staining in green and Nissl stain in blue . Scale bars: 500 μm ( D ) Vm of a single neuron ( top ) concurrently recorded with five motor nerves ( traces below ) during scratching behavior induced by a somatic touch ( onset indicated , 10 s duration ) . ( E ) Rastergram showing the parallel-recorded single units ( ∼200 neurons ) sorted according to hip flexor phase . ( F ) Firing rate distribution is positively skewed and normally distributed on a log–scale , i . e . lognormal ( inset ) . Vm resting level in ( D ) is −60 mV . For details , see Figure 2—figure supplement 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 00510 . 7554/eLife . 18805 . 006Figure 2—figure supplement 1 . Experimental setup . ( A ) Preparation with electrodes inserted into the spinal cord of a turtle , which is lying on its back with the caudal part of the carapace and spinal cord intact . The scratch reflex motor pattern is activated by the mechanical touch of the carapace with a rod attached to an actuator . ( B ) Close–up from ( A ) with nerve suction electrodes ( with silver wires ) , an intracellular electrodes and the 3 silicon probes ( green ) inserted into the spinal cord . ( C ) Post–hoc histological reconstruction of the location of three Berg64–probes . The tissue is immunostained for ChAT-positive motoneurons ( green ) and Nisslstained neurons ( red ) to differentiate motoneurons from interneurons . The probes were painted with DiI prior to insertion leaving a fluorescent trace ( blue ) , although unspecific ChAT staining at the shank location gives a cyan appearance in slice 6 . Inset illustration indicates parasagittal locations of slice 4 , 5 and 6 . Scale bar: 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 00610 . 7554/eLife . 18805 . 007Figure 2—figure supplement 2 . Sorted sample units , quality measures , and probe layout . ( A ) Average waveforms ( black ) and ± SD ( green ) of 17 units ( columns ) recorded by 8 electrodes ( rows ) situation on the same shank of a Berg64–probe . Vertical scale bar 100 μV . ( B ) Correlogram matrix for the same 17 units with the autocorrelograms in diagonal ( green ) . The quality of the spike sorting is verified by the L-ratio ( C ) and Isolation distance ( D ) for all units from the same session . ( E ) The Berg64–probe ( Neuronexus inc ) consists of 8 shanks with 8 electrodes on each shanks , located at the edge to sample over the largest volume of tissue . Dimensions are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 007 Two mechanisms have previously been proposed to explain the skewned lognormal firing rate distribution , which is also observed in other parts of the nervous system ( Buzsáki and Mizuseki , 2014 ) . Lognormal distributions could either arise from a nonlinear transformation of normally distributed inputs ( Roxin et al . , 2011 ) ( Figure 1A ) or from a linear transformation of a lognormally distributed synaptic input ( Wohrer et al . , 2013 ) . The latter mechanism was considered in connection with the sparse spiking activity in auditory cortex ( Koulakov et al . , 2009; Hromádka et al . , 2008 ) and since synaptic weights within neocortex have a heavy tail lognormal distribution rather than a Gaussian distribution ( Ikegaya et al . , 2013; Song et al . , 2005 ) . Models also show that the Vm distribution can be either skewed or Gaussian depending on the synaptic input intensity ( Ostojic , 2011 ) . Therefore , to distinguish between the proposed mechanisms , it is important to first assess whether the synaptic current is normally versus lognormally distributed . Secondly , to test whether the transformation of the synaptic input to spiking output is linear versus supralinear . We started by addressing the first requirement by investigating the synaptic input in intracellular recordings . The most relevant part of the data was found during the peak of a locomotor cycle where the Vm was in vicinity of Vt⁢h⁢r⁢e⁢s and was dominated by synaptic potentials ( Figures 1D and 3A ) . The motor activity was clearly non–stationary , which means that the spike activity was likely to move between the fluctuation– and mean–regime . Nevertheless , the rhythmic activity possessed a separation of timescales in the sense that the activity between cycles ( ∼1 s ) contained much larger excursions in Vm than within cycles ( ∼2-400 ms ) . Here , the mean Vm did not change much and for practical purposes it could be considered constant within the cycle . In the following analysis of the intracellular data we regarded the dynamics in Vm as stationary within a cycle – well aware that the comparison to theoretical models , which are based on assumption of stationarity , should be taken with a grain of salt . We intended to investigate the symmetry of the distribution of synaptic current using this assumption . The synaptic current within a cycle is difficult to assess , but rather than the mean current , we were primarily interested in the fluctuations in current , which we could approximate from Vm via Ohm’s law under the following conditions . Within a cycle , the mean Vm was just below threshold and did not change its value much . Therefore the voltage–activated conductances were approximately constant such that there was an Ohmic relationship between synaptic current and Vm . This is likely justified for neurons in fluctuation–driven regime , since the conductance is often high and dominated by balanced E/I synaptic input ( Destexhe et al . , 2003; Kumar et al . , 2008 ) . The high conductance suppresses the coupling between Vm and intrinsic conductance in a divisive manner ( Kolind et al . , 2012; Tiesinga et al . , 2000 ) . Thus , in the fluctuation–driven regime the non–Ohmic contributions were likely smaller and the I⁢Vm-relationship more linear than in the mean–driven regime . We intended to test the hypothesis of normally distributed input , but since the approximation of using the variability in Vm as a proxy for the variability in synaptic current is most valid for the neurons in fluctuation–driven regime , we needed a way to distinguish neurons that were primarily in the fluctuation–driven regime . We therefore propose a novel metric , the return map ratio , which quantifies the degree of fluctuations leading up to a spike ( Figure 3—figure supplement 1 ) . The return map ratio ( RMR ) quantifies how direct the subthreshold Vm–trajectory is between spikes and this forms the basis for selecting neurons in our analysis . An RMR close to 0 . 5 has fluctuation–driven spiking whereas a value close to 1 has mean–driven spiking ( Figure 3—figure supplement 1A , B ) . Therefore , we defined a neuron as fluctuation-driven if its RMR <0 . 7; in our sample of intracellular recordings we found 50/68 neurons in this regime . A sample neuron , which was found in the fluctuation–driven regime based on this metric illustrates how we obtained the distribution of sub–threshold Vm ( Figure 3A ) . The distribution was estimated both by selecting the Vm in between spikes ( temporal distribution ) and by collecting instances of Vm prior to spike peak in a spike triggered overlay ( ‘sigma’ in Figure 3B ) . These two estimates are in agreement with one another for the sample cell ( Figure 3C ) . This agreement is also found across the population as quantified by the mean and SD ( Figure 3D ) . The skewness for the distributions across the population is small and scattered around zero as expected for normal ( symmetric ) distributions ( Figure 3E ) . From these data we conclude that the subthreshold Vm–distributions are not skewed , but rather symmetrical and Gaussian–like ( cf . inset distributions , Figure 3E ) . Nevertheless , the minimal requirement for confirming the two–regime hypothesis for the single neuron is that the synaptic current ( not the synaptic potentials ) is Gaussian ( Figure 1 ) . As we argued earlier , if there is an Ohmic relationship between current and potential , which is likely during high–conductance states , then this requirement would be granted . More importantly , now that we do find a Gaussian Vm–distribution , it is difficult to contemplate a non-linear I⁢Vm-relationship , which would result in such a symmetric distribution . The synaptic input current would have to have a finely matched inverse distribution to cancel out this non–linearity in order to achieve a symmetric Vm–distribution . A more parsimonious explanation therefore is that , since the synaptic potentials are normally distributed , they are a result of a linear transformation of synaptic currents , which are also normally distributed . 10 . 7554/eLife . 18805 . 008Figure 3 . Subthreshold Vm–distributions are symmetric . ( A ) Sample cell spiking in the fluctuation–driven regime , and ( B ) its spike–triggered overlay to determine the Vm–distribution of trajectories 18 ms prior to spike–onset ( ‘sigma’ ) . ( C ) The Vm–distribution is estimated in two ways: via samples of Vm–instances prior to the spike peak ( top , vertical line ‘sigma’ in B ) and over time via the interspike intervals ( bottom ) . ( D ) Mean temporal– vs . spike–triggered–estimates ( top ) are closely related ( orange unity–line ) and have a near normal distribution of means ( inset ) . For details , see Figure 3—figure supplement 1 and 2 . Similarly , the variability of the two estimates ( SD ) are closely related ( bottom ) . ( E ) Sorted skewness for all neurons in fluctuation–driven regime indicate symmetric Vm–distributions ( temporal ) . Inset distributions with skewness of ± 1 illustrate no discernible asymmetry . The extreme skewness observed in the data set is around ± 0 . 5 ( broken lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 00810 . 7554/eLife . 18805 . 009Figure 3—figure supplement 1 . Quantifying the degree of fluctuations and selecting neurons in fluctuation–driven regime using the return map ratio metric . ( A ) The inter–spike Vm–trajectory of three sample neurons in mean–driven ( left ) , intermediate ( middle ) and fluctuation–driven spiking regime ( right ) . In the mean–driven regime the inter–spike trajectory moves directly from AHP resetting towards threshold , whereas in the fluctuation–driven regime the trajectory is convoluted and indirect ( right ) . ( B ) The degree of convolution in the trajectory can be quantified using return mapping , i . e . plotting Vm⁢ ( t ) versus Vm⁢ ( t+Δ⁢t ) , and quantifying the fraction of points above versus below the unity-line ( y=x ) , which we refer to as the return map ratio . An even ratio close to 0 . 5 represents convoluted path ( right ) , whereas a uneven ratio ( close to 1 ) represent a direct path ( left ) . Ratios are indicated in% . ( C ) The distribution of return map ratios for all ISIs , shown for two sample neurons , one having distribution mean close to 0 . 5 , i . e . a fluctuation–driven regime and one having the mean close to 0 . 8 , i . e . in the mean–driven regime ( blue and green arrows ) . The mean return map ratio of all neurons ( n=68 ) has a significant anti–correlation with skewness of firing rate distribution ( D ) , spike irregularity ( C⁢V2 ) ( E ) , as well as the least time spent below threshold ( LTBT ) of Vm ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 00910 . 7554/eLife . 18805 . 010Figure 3—figure supplement 2 . Population–distribution of mean Vm is Gaussian . ( A ) The mean Vm for the population of neurons is symmetrically near Gaussian–distributed ( blue ) . The mean threshold for the same population is depolarized ( red ) . ( B ) The mean thresholds correlate with the values of mean Vm and the thresholds are more depolarized as indicated by a rightward shift compared with the unity line ( red ) . ( C ) Scatter plot of all the histograms of Vm with Gaussian fits ( red ) . ( D ) Histogram of the Vm distributions with the individual mean thresholds ( Vt⁢h⁢r⁢e⁢s ) subtracted ( broken line indicates the relative location of Vt⁢h⁢r⁢e⁢s ) . Note that the Vm distributions , which have their mean far from threshold , also have a larger SD . ( E ) Normalizing each distribution with SD ( σ ) to assess the distance in terms of the size of fluctuations , i . e . ( Vm-Vt⁢h⁢r⁢e⁢s ) /σ . ( F ) Scatter plot of all the distributions of ( Vm-Vt⁢h⁢r⁢e⁢s ) /σ , has a near Gaussian distribution as indicated with the sliding population mean ( blue ) . The mean distance to threshold is approximately 3σ ( arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 010 So far , we have only looked at Vm–distributions of single neurons , which operate primarily in the subthreshold domain , and found that the synaptic input is most likely normally distributed . We do not know whether the synaptic input is also normally distributed in the mean–driven regime , but since the synaptic input is normally–distributed in the subthreshold region , it is likely also normally–distributed in the suprathreshold region . Otherwise , the input statistics from the presynaptic neurons would have to depend on the threshold of the post–synaptic neuron , which is unlikely . Above , we established that the synaptic input to a given neuron is likely normally distributed , and if this input is transformed in a supralinear fashion , the output firing rate distribution will be skewed . Nevertheless , the foundation of the skewness in population rate distribution ( Figure 2F ) is not necessarily directly linked to the skewness of the instantaneous rate distribution of single neurons . In principle , it is possible to have a population with a normal distribution of mean firing rates , where the cells themselves have lognormally distributed firing rates and vice versa . Therefore , we needed to address the distribution of mean Vm across the population and test whether this was skewed or normal . Further , since the sub–threshold IO-curve is linked to threshold , it is important to establish the distance of mean Vm from threshold with respect to the size of synaptic fluctuations , i . e . standard deviation of Vm ( σ ) . This distribution , i . e . ( Vm-Vt⁢h⁢r⁢e⁢s ) /σ , turns out to also be normally distributed with a mean around 3 σ from threshold ( Figure 3—figure supplement 2 , plotted for all n=68 neurons ) . The value used for Vt⁢h⁢r⁢e⁢s here is the mean of the estimated thresholds for all spikes ( see below ) . If we assume , when normalizing Vm this way , the IO-curve has approximately the same nonlinearity across all neurons , the population distribution of firing rates will also be skewed due to the nonlinear transformation of the normally–distributed input ( Figure 3—figure supplement 2F ) to a lognormally–distributed output . These results are in qualitative accordance with the scheme proposed previously ( Roxin et al . , 2011 ) . As another piece of the puzzle , we need to establish the shape of the neuronal response function , which rarely has been done in the subthreshold domain . The link between a normally distributed input and a lognormally distributed output is a supralinear transformation . To test whether this is a hallmark of the fluctuation–driven regime , we needed to estimate the input–output ( IO ) –function for the subthreshold domain . The IO–function of neurons is a fundamental property of the nervous system , and therefore it is well-characterized both theoretically ( Gerstner et al . , 2014 ) and experimentally ( Silver , 2010 ) . Nevertheless , it has rarely been established for fluctuation–driven spiking . Here , we estimated the IO-function for subthreshold spiking via the probability of eliciting a spike as a function of Vm in the following way . First , we collected instances of Vm shortly before the spike–onset , where Vm is depolarized yet still not part of the deterministic spike trajectory . The probability that a given value of Vm will cause a spike was estimated as the histogram of Vm–instances ( gray histogram , Figure 4A ) divided by the total time spent at all values of Vm ( green histogram ) . This gives the empirical relationship between Vm and the firing rate ( Jahn et al . , 2011; Vestergaard and Berg , 2015 ) . The IO–function had a strong non–linear shape ( Figure 4B ) . To capture the curvature we fitted both a power–law and an exponential for all n=68 neurons and the curvature had a weak negative correlation with the SD of the Vm–fluctuation ( Figure 4C–D ) as demonstrated previously ( Vestergaard and Berg , 2015 ) . Similar expansive nonlinearity has previously been characterized in sensory–driven neurons ( Anderson et al . , 2000; Hansel and van Vreeswijk , 2002; Miller and Troyer , 2002 ) . It will transform the normally–distributed synaptic potentials into a lognormally–distributed spiking output in the fluctuation-driven regime ( Figure 1A ) . For mean–driven spiking the IO-function is not supralinear , but rather linear ( or even sublinear ) , and the normally–distributed synaptic input will therefore be transformed to a normally distributed spiking output ( Figure 1B ) . In conclusion , neurons that have fluctuation–driven spiking also have a non–linear IO-transformation of synaptic potentials to spiking output . 10 . 7554/eLife . 18805 . 011Figure 4 . Fluctuation–driven spike–response curve is supralinear . ( A ) The empirical probability of evoking a spike in a small window as a function of Vm is determined using spike–triggered overlays . The probability distribution is estimated as the Vm–distribution of trajectories prior to spike–onset ( gray histogram , 1 . 7 ms prior to peak ) normalized with the total ( temporal ) Vm–distribution ( green histogram ) . Dividing this probability by the sampling interval gives the firing rate ( see Materials and methods ) . ( B ) The firing rate versus Vm for a sample neuron is strongly nonlinear . A power–law ( broken line ) and an exponential ( blue line ) are fitted to capture the nonlinearity . Note that the mean threshold ( ⋆ ) is below the largest subthreshold fluctuation ( ⋄ ) , likely due to a depolarization of threshold associated with a higher firing rate ( see also Figure 6—figure supplement 1 ) . ( C ) Power–law exponent ( α ) for different neurons are weakly anti–correlated with the fluctuations ( SD ) in their Vm ( ‘sigma’ , Figure 3B , R=-0 . 34 , p<0 . 01 ) . Linearity is indicated by horizontal broken line . ( D ) Exponential coefficient ( β ) for different neurons are also anti–correlated with the fluctuations in Vm albeit not significantly ( R=-0 . 22 , p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 011 The normally distributed input combined with the nonlinear IO–transformation should result in a skewed lognormal firing rate in the single neuron . To confirm this , we measured the distribution of the instantaneous firing rate , i . e . the inverse of ISIs . The quiet period in between burst cycles were not included in the analysis ( Figure 1D–E ) , since in these periods Vm was far from Vt⁢h⁢r⁢e⁢s and therefore in an irrelevant part of the IO–function . The firing rate distribution of many cells was positively skewed and resembled a normal distribution with near zero skewness on a log-scale ( sample cell shown in Figure 5A ) . This is expected for poisson–like spiking in the fluctuation–driven regime ( Ostojic , 2011 ) . Nevertheless , distributions for all the intracellularly recorded neurons ( n=68 ) were skewed to a varying degree from strong positive to zero skewness on a linear axis and similarly shifted downwards on log axis ( cf . gray and green histograms , Figure 5B ) . This suggests that neurons were found in a spectrum between fluctuation– and mean–driven spiking . More negative log–skewness were associated with higher mean rates ( Figure 5C ) . This is probably due to a larger presence in the mean–regime at higher firing rates , where the distribution skewness is expected to be negative on a log–scale , i . e . Gaussian on a linear scale . Note that the spectrum of skewness was substantially larger than it was for the Vm distributions above ( Figure 3E ) . Skewed Gaussian distributions are shown to illustrate the range of skewness in the data ( Figure 5D ) . In conclusion , these results suggest that the skewness in firing rates is an indicator of the degree of participation in the fluctuation–driven regime . 10 . 7554/eLife . 18805 . 012Figure 5 . Firing rate distributions are skewed to a variable degree depending on mean firing rate . ( A ) Distribution of instantaneous firing rates for a sample neuron is positively skewed on a linear axis and lognormal–like ( green histogram , inset ) . Mean indicated by broken vertical line . ( B ) Sorted distribution skewness on linear ( gray ) and logarithmic axes ( green ) for each neuron in the population . ( C ) The log–skewness across neurons is negatively correlated ( R=-0 . 5 , p<0 . 001 ) with mean firing rate , which indicates that higher firing rates are found in the mean–driven regime and less lognormally–distributed , i . e . departing from broken line . ( D ) Illustration of firing rate distributions that have positive skewness ( top ) , zero skewness ( Gaussian , middle ) and negative skewness ( bottom ) representing the range observed in the data ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 012 A neuron is not just spiking in either the fluctuation– or the mean–driven regime , rather , it likely spends time in both regimes during motor activity . To estimate the amount of time a given neuron spends in either of the two regimes we calculated the fraction of time that the smoothed Vm was above versus below threshold . We first look at two heuristic neurons , one in the fluctuation–driven regime and one in the mean–driven regime . The fluctuation–driven neuron spent most of the time below threshold ( Figure 6A ) and had more irregular spiking as quantified by a local measure of irregularity , the C⁢V2 ( green line ) . C⁢V2 is the difference of two adjacent ISIs divided by their mean ( Holt et al . , 1996; Bruno et al . , 2015 ) . In contrast , the mean–driven neuron spent most time above threshold and had more regular spiking , i . e . C⁢V2 closer to zero ( Figure 6B ) . Since the threshold was firing rate–dependent due to the inactivation of the N⁢a+–conductance ( Figure 6—figure supplement 1 ) we used the most hyperpolarized value of threshold ( broken line ) . The distribution of C⁢V2 for all trials had higher mean for the fluctuation–driven cell than the mean–driven ( cf . arrows , Figure 6C ) . Also , the cumulative time spent below threshold was higher for the fluctuation–driven cell ( 96% ) than the mean–driven cell ( 35% , Figure 6D ) . This fraction of time spent below threshold was quantified for every neuron ( n=68 ) and the population distribution had a strong mode at 1 ( top , Figure 6E ) suggesting many neurons spent much time in the fluctuation–driven regime . To compress the diversity within the population into a simpler representation , we used the reverse cumulative distribution of neurons versus time spent below threshold ( bottom , Figure 6E ) . This indicates how many neurons ( y-axis ) spent at least a given fraction of time ( x-axis ) below threshold . The intercept with the 50%–line ( broken line ) indicates what fraction of time half the population at least spent below threshold . This fraction is remarkably high ( 84% ) suggesting a prominent presence within the fluctuation–driven regime . 10 . 7554/eLife . 18805 . 013Figure 6 . Two contrasting sample neurons found in the two regimes . ( A ) Sample neuron in fluctuation–driven regime , where the mean Vm ( blue line ) is below lowest threshold ( broken line ) , the spikes are irregular ( C⁢V2≈0 . 5–1 , green line ) and driven by fluctuations ( arrow ) . ( B ) Second sample cell found in mean–driven regime , where the mean Vm is above threshold during the cycle ( arrow ) . The spiking is more regular , i . e . low C⁢V2 ( green line ) . ( C ) Mean–driven neuron ( gray ) has lower C⁢V2 than the fluctuation–driven neuron ( brown ) . Means indicated ( arrows ) . ( D ) Cumulative time of Vm shows the fluctuation–driven neuron ( FD ) spends more time below threshold ( 96% ) than the mean–driven ( MD , 35% ) . ( E ) Top: Time below threshold for population of neurons ( cells from A–D indicated ) . Bottom: Least time spent below threshold versus a given fraction of neurons ( reverse cumulative distribution function ) . Half of the neurons ( broken line ) spend at least 84% of the time in fluctuation–driven regime , i . e . have Vm below threshold . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 01310 . 7554/eLife . 18805 . 014Figure 6—figure supplement 1 . Threshold depolarizes with increase in firing rate . ( A ) Sample recording with the threshold ( red dots ) for each spike ( left ) . Mean threshold ( solid line ) ± SD ( broken lines ) . Right top: the selected cycle from trace in left ( fourth cycle indicated by gray horizontal bar at top ) . Right bottom: Selected region on shorter timescale ( gray rectangle from top trace ) . ( B ) Spike–triggered overlay with the thresholds indicated ( red dots ) . ( C ) Detection of threshold via method by Sekerli et al . ( 2004 ) : Threshold is found at the maximum of the second derivative of the trajectory in phase plane plot of Vm versus d⁢Vm/d⁢t ( red dots ) . ( D ) Distribution of threshold location prior to spike peak . ( E ) Distribution of threshold values in Vm . ( F ) Return map of the spike threshold values , shows a strong correlation between neighboring threshold values ( n vs . n+1 ) . ( G ) The change in threshold , Vt⁢h⁢r⁢e⁢s⁢h-Vt⁢h⁢r⁢e⁢s , Q⁢5 , with mean firing rate , where Vt⁢h⁢r⁢e⁢s , Q⁢5 is the threshold at the 5% quantile and Vt⁢h⁢r⁢e⁢s⁢h is the threshold for individual spikes . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 014 Mean- and fluctuation-driven spiking can be distinguished by important traits such as degree of irregularity and log-skewness of the firing–rate distribution . To verify these traits , we used another sample neuron as a heuristic illustration . We injected different levels of either positive or negative bias currents in different trials while keeping all else constant . A negative constant current injection ( −1 . 0 nA ) caused a decrease in firing rate and a slight increase in irregularity ( green line ) compared with zero injected current ( Figure 7A–B ) . Similarly , a positive current injection ( 1 . 7 nA ) caused more spikes and a decrease in irregularity ( Figure 7C ) consistent with a movement between regimes ( inset in Figure 7A ) . The decrease in irregularity with increasing input was further quantified as a negative correlation between mean C⁢V2 and injected current ( R=-0 . 84 , p≪0 . 001 ) over multiple trials ( n=18 , Figure 7D ) . This is qualitatively in agreement with previous reports ( Prut and Perlmutter , 2003; Powers and Binder , 2000; Wohrer et al . , 2013 ) . The instantaneous firing rate in the control condition ( 0 nA ) was lognormal as expected for the fluctuation–driven regime ( top , Figure 7E ) . When adding input current the distribution was shifted to the right and enriched with a negative skewness as expected for mean-driven spiking ( bottom , Figure 7E ) . This relation between input and shape of rate distribution was further confirmed by a negative correlation between multiple current injections and skewness both on linear scale ( gray dots ) and log–scale ( red dots , Figure 7F ) . Hence , skewness and irregularity are indicators of the spiking regime . 10 . 7554/eLife . 18805 . 015Figure 7 . Transition between regimes can be induced by injected current . ( A ) Hyperpolarizing Vm of a sample neuron during the motor cycle with negative injected current ( −1 . 0 nA ) . Negative current hyperpolarizes mean Vm ( blue ) and increases irregularity ( C⁢V2≈1 , green line ) compared with control condition ( B ) . ( C ) Positive current injection ( 1 . 7 nA ) has the opposite effect: Depolarization , more regular spiking and higher firing rate . ( D ) Mean of C⁢V2 over a trial vs . the constant injected current for that trial has negative correlation . ( E ) Firing rate is lognormally distributed in control ( top ) , but negatively skewed ( skewness = −1 . 9 ) when added current increases mean–driven spiking ( bottom ) . ( F ) Skewness of firing rate distribution is negatively correlated with injected current . Linear skewness shown in top gray points ( R=-0 . 73 , p<0 . 001 ) and log-skewness shown in bottom red points ( R=-0 . 70 , p<0 . 001 ) . Same neuron throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 015 An alternative to injecting electrode current is to manipulate the balance of excitation and inhibition ( E/I ) by pharmacological means . This is important for understanding the cause of irregularity and the fluctuation–driven regime . Hence , we manipulated the synaptic input in a reduced preparation with micro–superfusion of strychnine , a glycinergic blocker , over the transverse cut surface of the spinal cord ( described in [Berg et al . , 2007; Vestergaard and Berg , 2015] ) . This affected only neurons at the surface ( <300 μm ) without affecting the rest of the network , which was verified by careful monitoring of flow and the network activity via the nerve recordings . Comparing the spiking during control condition ( Figure 8A ) with that during blockade of inhibition ( Figure 8B ) , we noticed a strong increase in spiking . This is consistent with a depolarization due to disinhibition , thus ‘unbalancing’ the excitatory and inhibitory input . Reducing inhibition tipped the balance of E/I toward larger inward synaptic current , which resulted in a more depolarized Vm ( blue line ) well above threshold ( arrows , Figure A–B ) . It also resulted in higher firing rates and lower irregularity on the peak ( cf . green lines ) . Generally , the irregularity ( C⁢V2 ) was higher in the control case than in the unbalanced case ( Figure 8C ) similar to the results observed with current injection ( Figure 7A–D ) . The irregularity was also negatively correlated with depolarization of the mean Vm when unbalancing the E/I although it was uncorrelated in the control condition , where the spiking occurred in the fluctuation–driven regime ( Figure 8—figure supplement 1 ) . The instantaneous firing rate was skewed and lognormal in the control case ( top , Figure 8D ) , similar to the above sample cell ( top , Figure 7E ) . This distribution became negatively skewed when adding inward current ( bottom , Figure 7E ) . Similar effect was seen when ‘unbalancing’ the synaptic input , which also result in larger inward current . The firing rate increased ( cf . broken lines , Figure 8D ) and the distribution became negatively skewed ( cf . −0 . 2 and −1 . 5 ) as expected in the mean–driven regime ( bottom ) . To quantify the increase in time spent in the mean–driven regime , we performed an analysis similar to the analysis in the above section ( Figure 6D ) . The cumulative time spent below threshold was larger in the control condition ( 78% ) compared with the unbalanced case ( 56% , Figure 8E ) . These observations are largely consistent with the consensus view that irregular fluctuation–driven spiking is due to a balance between excitation and inhibition ( Table 1 ) . 10 . 7554/eLife . 18805 . 016Figure 8 . Transition between regimes induced by unbalancing E/I . ( A ) Sample cell in control condition and after reduction of inhibition with local strychnine ( B ) . Onset of motor program indicated ( △ ) . Blocking inhibition results in a larger net inward current , which drives the mean Vm ( blue lines ) across threshold to more mean–driven regime . As a result the spiking is less irregular on the peaks as measured with C⁢V2 ( cf . green lines ) . ( C ) Irregularity ( C⁢V2 ) was smaller after application of strychnine ( arrows indicate mean , histogram truncated at 1 . 5 ) . ( D ) Firing rate distribution is symmetric on log–scale ( top , skewness = −0 . 2 ) and negatively skewed when inhibition is blocked ( bottom , skewness=−1 . 5 ) . ( E ) Strychnine induces a more depolarized Vm and a lower cumulative time spend below the threshold ( compare 78% with 56% ) . Same neuron throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 01610 . 7554/eLife . 18805 . 017Figure 8—figure supplement 1 . Unbalancing E/I induces an anti–correlation between irregularity and depolarization . ( A ) No obvious relationship between mean Vm ( blue ) and the irregularity of the spike ( C⁢V2 , green ) in the control condition of a sample cell in the fluctuation–driven regime . ( B ) The fluctuation–driven regime is manifested as a lack of significant correlation between irregularity for each pair of ISIs ( C⁢V2 ) and the mean Vm ( R=-0 . 06 , p=0 . 48 ) . The most negative threshold indicated by vertical broken line . ( C ) Spiking of same cell as in ( A ) after elimination of glycinergic inhibition by local application of strychnine , which causes depolarized in Vm , more regular spiking at higher rate and slower fluctuations in Vm . ( D ) The removal of inhibition also puts the spiking into the mean–driven regime , which is manifested as a significant anti-correlation between irregularity of the spiking and mean Vm ( R=-0 . 35 , p≪0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 017 In the above intracellular analyses we reported the spiking irregularity in terms of C⁢V2 along with the mean Vm , current injection and pharmacological manipulation of the balance of excitation and inhibition . The C⁢V2 measure is convenient to use as an indicator of the mean– versus the fluctuation–driven regimes observed in the extracellular spiking data , since it only requires spike times . Therefore it is important to validate C⁢V2 as an indicator of spiking regime . In the above sample cell analyses we note first , that when Vm spent a larger fraction of time above threshold , i . e . in mean–driven regime , the C⁢V2 was lower ( Figure 6 ) . Second , when depolarizing a neuron artificially either with constant positive current ( Figure 7D ) , or by blocking inhibition ( Figure 8C ) , such that more spikes were in mean–driven regime , the C⁢V2 was decreased . To further substantiate C⁢V2 as an indicator of spiking regimes we looked again at the return map ratio , which is an independent metric of fluctuations during inter-spike intervals . If C⁢V2 is an indicator of the spiking regime , it should be anti-correlated with the return map ratio . This was confirmed by plotting the mean C⁢V2 for all cells ( n=68 ) against the mean return map ratio , which indeed demonstrated a significant anti–correlation ( R=-0 . 34 , p=0 . 005 ) ( Figure 3—figure supplement 1E ) . A second independent indicator of fluctuation regime is the cumulative time below threshold of Vm ( Figure 6D ) , which should be correlated with the mean C⁢V2 . We tested this using the most hyperpolarized value of theshold , since it was the most conservative , but there was no significant correlation between the cumulative time below threshold and the mean C⁢V2 . Perhaps the lack of linear relationship is due to a bias from the reset voltage and after-hyperpolarization , which is different from cell to cell and therefore randomly may introduce a large fraction of time spent below threshold . Also , intense synaptic activity is known to quench the after–hyperpolarization ( Berg et al . , 2008 ) and therefore this bias may be particularly strong when the synaptic input is not balanced as in the mean–driven regime . A third indicator of spiking regime is the skewness of the instantaneous firing rate distribution ( Figure 7E and 8D ) . We estimated the skewness of the individual firing rate distributions for all neurons ( n=68 ) and plotted it against the mean C⁢V2 ( data not shown ) . There was a significant positive correlation between the two , regardless of whether the firing rate distribution was plotted on log or linear scale ( Rl⁢o⁢g=0 . 43 , p=0 . 0003 , and Rl⁢i⁢n=0 . 41 , p=0 . 0006 ) , which suggest C⁢V2 as a valid measure for spiking regimes . A last indicator is the local mean membrane potential depolarization , which should be anti-correlated with the instantaneous C⁢V2 , if the Vm is above threshold ( Figure 8 , Figure 8—figure supplement 1D ) . Here , there was a lack of correlation between C⁢V2 and Vm before blocking inhibition , in the fluctuation–driven regime . However , after removal of inhibition , Vm was in supra–threshold domain , which introduced an anti-correlation between C⁢V2 and Vm . Hence , if the neuron is in the mean-driven regime the C⁢V2 is an indicator for the depolarization above threshold . To further verify this we performed a similar test of the relationship between instantaneous C⁢V2 and local depolarization for all neurons ( without pharmacology ) . We found that all the cells with significant relationships ( p<0 . 05 , n=16/68 ) had anti-correlation between Vm and C⁢V2 ( data not shown ) . In conclusion , the C⁢V2 measure is correlated with other measures and indicators of spiking regimes ( except the cumulative time below threshold ) and therefore C⁢V2 is a useful indicator in itself . The irregularity in spiking could be caused by a noisy threshold rather than fluctuations in synaptic potentials . Nevertheless , a noisy threshold can only explain a small part ( if any ) of the spiking irregularity . First of all , if the irregularity , that we observed in spike times , was due to a noisy threshold mechanism , we should see the same irregularity regardless of the depolarization , i . e . regardless of whether the neuron was in the sub–threshold or supra–threshold domain . Yet , the spiking irregularity was strongly dependent on depolarization ( Figures 6–8 ) . There was an adaptation in threshold ( Figure 6—figure supplement 1 ) . This was not random , but rather due to a gradual inactivation of Na +–channels throughout the burst ( Henze and Buzsáki , 2001 ) . The threshold of a given spike strongly depended on the threshold of the previous spike ( panel F ) as well as the mean firing rate ( panel G ) . The same mechanism is behind spike–frequency adaptation , which is a well–described phenomenon ( Grigonis et al . , 2016 ) . The adaptation in threshold is likely to make the IO-function more sublinear in the mean–driven regime , which will generally curb network activity . In order to verify the extent of the threshold variance beyond the contribution from inactivation of Na+–channels , we looked at the threshold of only the first spike of each cycle , such that the neuron had ample time for recovery . The variance of the first–spike threshold ( n=51 ) in a sample neuron was σt⁢h⁢r⁢e⁢s2=0 . 8 mV2 whereas the variance in synaptic potentials was more than 17–fold higher ( σV⁢m2=14 . 0 mV2 ) . Therefore a randomness in the threshold had little of no effect on the irregularity of spiking compared with the randomness in synaptic input . In some recordings the threshold may appear as uncorrelated with the membrane potential prior to the spike onset . However , rather than a noisy threshold this is likely attributed to cellular morphology . If the cell is not electrically compact , the axon initial segment , where the spike is initiated , will have a different potential than what is recorded with the electrode . If this was the case , these observations would still be compatible with the two–regime hypothesis , since spikes would still be driven either by fluctuations or a large mean current , despite the disguise of a long electrotonic distance to the recording site . So far the analysis has been performed on serially acquired intracellular recordings across trials and animals . This demonstrates that some neurons spiked primarily in the fluctuation–driven regime while others spiked in the mean–driven regime . Nevertheless , it is still unclear what the parallel population activity was during a behavior and across behaviors . How many neurons were in one versus the other regime and for how long ? First , we assessed the neuronal participation in the motor patterns by their degree of spiking during motor behavior . Neurons were active during both ipsi– and contralateral scratching behaviors ( Figure 9A–D ) . Most units had a rhythmic relationship with the nerve signals and a higher firing rate for the ipsilateral scratching compared with contralateral scratching behavior ( cf . Figure 9C and D; Videos 1 and 2 ) , which indicates participation of neurons in a hemicord to a smaller degree in the contralateral movement than the ipsilateral movement . 10 . 7554/eLife . 18805 . 018Figure 9 . Skewed neuronal participation across behaviors . ( A–B ) Two distinct motor behaviors: Ipsilateral pocket scratch ( left panel ) and contralateral pocket scratch ( right panel ) shown by intracellular recordings ( top ) and motor nerve activities . ( C–D ) Rastergrams showing the unit activities during ipsilateral pocket scratch ( C ) and contralateral pocket scratch ( D ) . Green areas mark the hip flexor phase . ( E–F ) spike count firing rate distribution for the behaviors on linear and a semi-log plot ( insets ) , indicate lognormal participation . Lognormal functions are fitted ( solid green lines ) . ( G ) Skewness on logarithmic ( green bars ) and linear scale ( gray bars ) is preserved across animals . ( H ) The inequal neuronal participation is calculated using Lorenz curve and gini coefficient . ( I ) Gini–coefficients cluster around 0 . 5 across behaviors and animals . Mean ( J ) and standard deviation on ( K ) of the distribution of firing rates on log–scale across behaviors and animals . Vm resting level in ( A–B ) is −60 mV . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 01810 . 7554/eLife . 18805 . 019Video 1 . Skewness of the population firing rate is activity–dependent: Behavior 1 ( ipsilateral scratching ) . The spiking activity in three lumbar segments shown as a 24 by 8 pixel-grid , with each pixel representing a recording channel ( top left panels , segments D8 , D9 and D10 indicated ) . Columns represent probe shanks ( separated by 200 μm ) and rows the vertical positions in the dorsoventral axis ( ∼30 μm between each ) . The light intensity of a pixel indicate the local firing rate in time estimated using Gaussian kernels . The time-dependent distribution of firing rates across the population ( green histogram , top right , logarithmic x-axis ) was fitted with a lognormal function ( appearing here as a normal distribution ) with variable skewness ( solid black line ) . Skewness of fit on linear and log scale is shown on slider ( inset ) . Note the dependence on overall activity . Lower panel: spike time rastergram ( horizontal lines represent spiking of the neurons , which are sorted according to phase ) and time is indicated with a black bar . The scratch reflex was activated at the time-point of the vertical dotted line ( ‘Stim onset’ ) . Sound is the aggregate spiking activity of the population . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 01910 . 7554/eLife . 18805 . 020Video 2 . Skewness of the population firing rate is less activity–dependent: Behavior 2 ( contralateral scratching ) . Same neuronal activity as in Video 1 , except the spinal network is now generating a different behavior . The neuronal ensemble spikes at a lower overall rate , which is reflected in a weaker relationship between skewness and activity ( compare with Video 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 020 The distribution of firing rates across the neuronal population over several trials was strongly skewed , which indicate that most neurons spike relatively infrequently with a ‘fat-tail’ of higher spiking ( Figure 9E ) . The distribution covered two orders of magnitudes from 0 . 1–10 Hz and was akin to a lognormal distribution ( inset and green lines , Figure 9E ) . Similar lognormal–like distributions have been observed in other parts of the nervous system ( Buzsáki and Mizuseki , 2014 ) . The implication of the skewed distribution is that most neurons spiked at low rates , but there was relatively many neurons spiking at higher rates indicating an overall rich diversity of firing rates . Although multi–functional spinal units have been reported previously ( Berkowitz et al . , 2010 ) it is unclear how their participation is distributed and whether the asymmetry in distribution is linked to different behaviors . To address this issue we analyzed the population spiking for multiple motor behaviors . The induction of a distinct scratch behavior is location–specific ( Stein , 2005 ) . Multiple behaviors can be evoked depending on exact location and which side of the body is touched . This allowed us to induce two distinct behaviors: ipsi– and contralateral hindlimb scratching , while recording from the same neuronal ensemble ( Videos 1 and 2 ) . These behaviors were reproducible over multiple trials ( >9 trials ) . Both behaviors had similar phase relationships between the muscle synergists , although ipsilateral scratching had stronger activity ( cf . Figure 9A and B ) . The firing rate distribution was positively skewed in both behaviors with the similar qualitative shape ( Figure 9E–F ) . This skewness was also found across animals ( green bars , Figure 9G , n=5 ) and close to zero on log–scale , i . e . lognormal ( black lower bars ) . To further quantify the uneven neuronal participation we used Lorenz statistics and the Gini-coefficient ( O’Connor et al . , 2010; Ikegaya et al . , 2013 ) . The Lorenz curve characterizes the share of cumulative participation of individual neurons of the population ( Figure 9H ) . The diagonal corresponds to the case where all neurons have the same firing rate . The deviation from equality is quantified by the Gini–coefficient , which is the fraction of area a to the total area a+b ( Figure 9H ) . The higher the coefficient , the more unequal the participation across neurons is . Both scratch behaviors had a Gini–coefficient of ∼0 . 5 ( Figure 9I ) . Although the mean firing rate could change between behaviors and between animals ( Figure 9J ) , the skewness was qualitatively similar ( Figure 9K ) . This suggests that the skewed lognormal–like firing rate distribution , and hence a presence of the fluctuation–driven regime , was preserved across behaviors and animals . Neurons do not occupy either the fluctuation– or the mean– driven regime all the time . Individual neurons can move back and forth between regimes depending on the synaptic current they receive . Neurons that spike predominantly in the mean–regime will have their mean firing rates closer together and more normally distributed compared with those spiking in the fluctuation–regime . Hence , we expected the skewness of the distribution of mean firing rates across the population to become more negative ( on log–scale ) as the general network activity increases . To address this , we analyzed the spiking across neurons in parallel . First , we estimated the time–dependent firing rate of each neuron in the population using optimal Gaussian kernel ( Shimazaki and Shinomoto , 2010 ) and measured skewness of the population distribution . The time–dependent population distribution was achieved by binning the rates in 10 ms windows ( Videos 1 and 2 ) . The mean population rate and its SD are indicated as black ± gray lines ( Figure 10A ) . As the mean firing rate increased , the skewness of the distribution ( log–scale ) became negative , which is a sign of more neurons were occupying the mean–driven regime ( cf . inset histograms , Figure 10A ) . This was further confirmed by a negative correlation between the mean firing rate ( black line in A ) and the log–skewness for all time points ( Figure 10B ) . Hence , as the general activity increased , the population distribution became less lognormal and more Gaussian , which suggests more neurons occupied the mean–driven regime during a higher general activity . 10 . 7554/eLife . 18805 . 021Figure 10 . Skewness and irregularity across the neuronal population gauge occupation in both regimes across time . ( A ) Heat map of the distribution of firing rates across the population ( n=190 units , 1 animal ) on log–scale ( y–axis ) as a function of time ( x–axis ) . Lognormal mean ± SD are indicated as black and grey lines , respectively . Distribution is indicated ( gray histograms ) at two different time points ( broken vertical lines ) . ( B ) Lognormal mean population firing rate ( black line in A ) versus log–skewness are negatively correlated , indicating more neurons move into mean–driven regime as the population rate increases . Scatter due to multiple trials , which is binned in sections , red crosses . ( C ) Distribution of irregularity ( mean C⁢V2 ) across population for all ISIs ( gray ) and when excluding of inter–burst intervals ( red ) . ( D ) Fraction of neurons , which spend a given amount of time in fluctuation–driven regime ( ic⁢r⁢i⁢t=0 . 4 , 0 . 5 and 0 . 6 ) normalized to 100% ( Reverse cumulative distribution ) . The least time spent in fluctuation–driven regime by half of the neurons ( T⁢I⁢F50 ) is given by the intercept with the broken horizontal line and distribution ( indicated by arrow ) . For this sample animal and behavior T⁢I⁢F50=56% . Inset: Values across animals , sample animal indicated ( ⋆ ) . ( E ) The T⁢I⁢F50–values across animals in both behaviors as indicated by similarity in values are remarkably conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 02110 . 7554/eLife . 18805 . 022Figure 10—figure supplement 1 . Distribution of neurons having fluctuation driven spikes and S⁢I⁢F50 values . ( A ) Reverse cumulative distribution of neurons ( y–axis ) having a given number of spikes driven by fluctuations ( x–axis ) for ipsilateral scratching for a sample animal and three values of ic⁢r⁢i⁢t ( 0 . 4 , 0 . 5 , and 0 . 6 ) . The minimal fraction of spikes driven by fluctuation in half of the neuronal population , S⁢I⁢F50 , shown in inset . Sample animal indicated ( ⋆ ) . ( B ) Same as ( A ) but for contralateral scratching . ( C ) The reverse cumulative distributions similar to ( A ) for all five animals and for ic⁢r⁢i⁢t=0 . 5 for ipsilateral scratching and for contralateral scratching ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 022 To further gauge the division of neurons in the two regimes we compared the irregularity of the spiking using C⁢V2 . This metric was verified above as a reliable indicator of spiking regimes . The distribution of the mean C⁢V2 across the population of neurons was clustered around 1 if all ISIs were included ( gray histogram , Figure 10C ) . However , measuring the irregularity in the motor cycles alone i . e . excluding the inter–burst intervals ( here , ISI < 0 . 5 s ) the mean irregularity across neurons was lower and clustered around 0 . 6 ( red histogram ) . Both distributions had substantial spread around the mean , which suggests a rich diversity spiking patterns . To get a compound measure of the behavior of the entire population across time , we considered the amount of time each neuron spent in the fluctuation–driven regime . We demarcated the fluctuation–regime as having irregularity in spiking above a critical value , i . e . CV2 > icrit . Choosing ic⁢r⁢i⁢t is not entirely objective . Complete Poisson–type irregularity has C⁢V2=1 , but the spiking is still irregular for lower values ( Feng and Brown , 1999 ) . Based on our data , even when the C⁢V2≈0 . 5 , the Vm spent as much as 96% of the time below threshold ( Figure 6C–D ) indicating fluctuation–driven spiking . Further , neurons that had C⁢V2≈0 . 5 , also had lognormal firing rate distributions ( Figure 7 ) , which also indicates the fluctuation–driven regime . For these reasons , we suggest choosing ic⁢r⁢i⁢t=0 . 5 for distinguishing regular vs . irregular spiking . A similar value was previously chosen to distinguish between regular vs . irregular ‘choppers’ in the cochlear nucleus ( Young et al . , 1988 ) . Thus , the population of spinal neurons had a large diversity in time spent in the fluctuation–driven regime . Some neurons spent as little as 20% in the fluctuation–driven regime while other spent as much as 80% . To get a quantitative handle on the occupation of neurons in the fluctuation–driven regime across the population , we considered the distribution of time spent with CV2 > icrit . This was formally quantified using the reverse cumulative distribution of neurons that spend a given fraction of time in the fluctuation–driven regime ( Figure 10D ) . The reverse cumulative distribution is plotted for 3 values of ic⁢r⁢i⁢t ( 0 . 4 , 0 . 5 and 0 . 6 ) to indicate the sensitivity to parameter choice . Obviously , choosing a lower ic⁢r⁢i⁢t results in a larger fraction of time in the fluctuation–driven regime , i . e . the curve is shifted to the right . Choosing ic⁢r⁢i⁢t larger has the opposite effect . This inverted S–shaped curve gives the fraction of neurons ( y–axis ) , which spend at least a given time in the fluctuation–driven regime normalized to 100% ( x–axis ) . Hence , half of the population spent at least 58% of time in the fluctuation regime during ipsilateral scratching ( intercept of curve with the broken line , Figure 10D ) . We refer to this metric as the time in the fluctuation–regime for half the neurons ( T⁢I⁢F50 ) . Similar T⁢I⁢F50–values were obtained for all five animals ( inset histogram ) . Qualitatively similar results were achieved for a different motor behavior , namely contralateral scratching ( Figure 10E ) . The T⁢I⁢F50 metric is a time–weighted analysis of irregularity of spike trains . In addition to measuring the time in regimes , we measured how many spikes were in one regime vs . the other . Hence , we calculated the reverse cumulative distribution of neurons that had a given fraction of spikes in the fluctuation–driven regime ( Figure 10—figure supplement 1 ) . Similar to T⁢I⁢F50 , we defined a spike–weighted metric as the spikes in fluctuation regime for half the neurons ( S⁢I⁢F50 ) . Both the S⁢I⁢F50– and T⁢I⁢F50–values were relatively conserved across animals as well as behaviors ( Figure 10D–E , Figure 10—figure supplement 1 ) . The large values of T⁢I⁢F50 and S⁢I⁢F50 indicate that the fluctuation–driven regime had a strong presence during motor behaviors , and the high similarity suggests that it may represent a conserved fundamental property of network activity . In the data analyses presented so far we have not addressed the neuronal identity of the recorded units . Nevertheless , there is a spatial division subtypes of spinal neurons , which we could take advantage of . During development , a distinct laminar organization of different cellular subtypes is formed in the dorsoventral axis ( Arber , 2012; Jessell , 2000 ) . In particular , motoneurons are primarily found in the most ventral part of the horn whereas interneurons are found in more medial to dorsal territory . Since this is the same axis that our electrode arrays were located along , it was possible to infer a likelihood of cellular identity based on location . The electrode shanks have multiple distributed electrodes ( Figure 11A ) , which made it possible to approximate the soma location using trilateration . Trilateration is the geometrical process of determining the location of a source in space using multiple recording sites combined with the fact that signals decay in the extracellular space ( Manolakis , 1996 ) . Thus , the node locations were approximated based on the amplitude of spike waveforms , which clearly decayed with distance ( Figure 11B ) . Node locations were combined for all shanks , probes and animals to form a scatter ( Figure 11C ) . Combining these locations with depth of individual shanks with respect to the surface of the spinal cord , we were able to investigate the spike patterns with respect to the absolute neuronal location . The irregularity in spiking was quantified ( mean C⁢V2 ) with respect to dorsoventral depth ( Figure 11D ) . The distributions of mean firing rates ( not shown ) and the mean C⁢V2 ( Figure 11E ) had no obvious dependence on depth . In particular , the spread in means was much smaller than the SD of the distributions themselves . The most parsimonious interpretation of these data is that the fluctuation–driven spiking regime was both present and equally prominent in all the neurons , regardless of whether the cell body was in the ventral horn or in the medial horn , i . e . equally present in motoneurons and interneurons . 10 . 7554/eLife . 18805 . 023Figure 11 . Spiking irregularity is independent of cellular location . ( A ) Layout of the 8 electrodes on a shank , which span a total of 210 μm in the dorsoventral axis . ( B ) Recorded waveforms at different locations of three sample units ( colored in red , blue and green ) . The node locations are estimated via trilateration and indicated as rings . Electrode locations are indicated as black dots . ( C ) Composite of source-locations for all shanks and all animals ( total n=921 cells ) . The location of sample units from B indicated in colors . ( D ) Irregularity of the associated spiking are estimated ( mean C⁢V2 on x-axis ) versus the dorsoventral location ( y–axis ) , where the unit locations are corrected for the depth of the individual shank with respect to the spinal cord surface . ( E ) The binned distributions of C⁢V2 as a function of depth . The distribution means are remarkably similar ( broken line as fiducial ) and a KS–test indicates no significant difference . DOI: http://dx . doi . org/10 . 7554/eLife . 18805 . 023 The fact that the relative time during which a subset of neurons occupied one of the two regimes was conserved across both behaviors and animals could indicate a key principle of neuronal processing . A fundamental challenge for neuronal networks is to perform functions while keeping the population activity from falling into either of the two extreme states: ( 1 ) the quiescent state where the neuronal spiking activity cannot remain self–sustained and ( 2 ) the unstable state of run–away recurrent spiking activity ( Vogels et al . , 2005; Kumar et al . , 2008 ) . It is well known that recurrent inhibition is important for maintaining stability , but other mechanisms may participate as well , e . g . synaptic depression or active adjustment of the shape of the neuronal response function by adaptation of spiking threshold . A nonlinear response function , as we observed in the fluctuation–regime ( Figure 4B ) , will amplify input via supralinear summation ( Rubin et al . , 2015 ) . The upward curvature will enhance synaptic fluctuations , which then accelerates the recurrent excitatory activity causing a potentially unstable state . In contrast , the response function in the mean–driven regime , is linear or even sublinear , which is likely to curb strong input . We therefore propose that the close proximity of the T⁢I⁢F50–value to 50% is an indication of a self–organizing trade–off between sensitivity and stability in order to preserve at once both network homeostasis and dynamical functionality . This conjecture remains to be further substantiated in future studies . Furthermore , the T⁢I⁢F50– and S⁢I⁢F50–values remain to be determined for other part of the nervous system and in other species . The distinction between fluctuation– and mean–driven spiking is interesting because the two types of spiking may have radically different causes , and this may hold an important clue to understanding the enigmatic motor rhythm generation . The fluctuation–driven spiking is believed to be caused by concurrent and random arrival of excitatory and inhibitory potentials resulting in a fluctuating subthreshold Vm ( Table 1 ) . In the mean–driven regime , on the other hand , the net membrane current is so large that the mean Vm ± σ is above threshold , and the ISIs are therefore determined by the recharging of the membrane capacitance following the refractory period of the previous spike ( Powers and Binder , 2000 ) . This results in a deterministic trajectory of Vm and regular ISIs . More importantly , for the mean–driven spiking the membrane current can be caused by intrinsically electrical properties as well as synaptic input , whereas the fluctuation–driven spiking is exclusively caused by synaptic input . An intrinsic property , which is commonly believed to be involved in rhythm–generation , is the pacemaker property that can autonomously generate neuronal bursting in the absence of synaptic input ( Brocard et al . , 2010; Ramirez et al . , 2004; 2011 ) . The prominent presence of the fluctuation–regime therefore implies that the majority of neuronal spikes were not driven primarily by intrinsic properties such as pacemaker potentials , but rather by synaptic communication . This can be interpreted in two ways: ( 1 ) if there is a pacemaker–driven rhythmogenic core of oscillatory neurons responsible for the motor activity ( Huckstepp et al . , 2016 ) , the core only represents a small fraction of the network , or ( 2 ) since the fluctuation–regime is prominent and pacemaker neurons are difficult to find , the motor–rhythm may be generated by other means such as emergent collective processes in the network ( Yuste , 2015 ) . Generation of movements without the need of pacemaker neurons have been predicted theoretically in central pattern generators ( Kleinfeld and Sompolinsky , 1988 ) as well as more complex sequence generation ( Hennequin et al . , 2014 ) . Even in the respiratory system , which has the most stereotypic motor rhythm , pacemaker cells appear not to be essential for generation of the rhythmic breathing , although this topic is still debated ( Feldman et al . , 2013; Ramirez et al . , 2011; Carroll and Ramirez , 2013; Chevalier et al . , 2016 ) . It remains to be understood how a distributed emergent processes can generate motor rhythms on a network level if , in fact , the pacemaker bursting is not an essential component . In spinal research , neuronal identification has improved over the last decades with the development of genetic knockouts and molecular markers ( Bikoff et al . , 2016; Goulding , 2009; Britz et al . , 2015; Kiehn , 2006 ) . Pinning down cellular identity improves the search for a potential specialization in the circuit . However , the sole focus on cellular identity to address questions in spinal research carries a weakness as well as a strength . It contains the risk of missing the collective dynamics and the delicate interaction among neuronal cell types . Neural circuits operate to perform functions by collective interaction between all neurons , where it is difficult , if not impossible , to link a particular function to the individual neuron . Functional activity may very well arise on circuit level as opposed to cellular level . This caveat is known as the neuron doctrine versus emergent network phenomena ( Yuste , 2015; Grillner , 2006 ) , and the neuron doctrine has almost exclusively been adopted in previous investigations of spinal motor circuits . To the best of our knowledge this report is the first investigation of spinal motor circuits from an ensemble viewpoint . Nevertheless , since motoneurons are fundamentally different from the rest of spinal neurons it would be helpful to distinguish them from interneurons . In our experiments we sampled from neurons , which were likely to be primarily interneurons since they are more numerous than motoneurons . The fraction of motoneurons to interneurons is 1:8 ( Walloe et al . , 2011 ) , but we were also likely to sample motorneurons , since they have large somata . To explore this further , we investigated the population activity and its relation to cellular identity by taking advantage of their spatial segregation in the dorsoventral axis ( Arber , 2012; Jessell , 2000 ) . We were able to associate an absolute location of the cellular somata ( using trilateration ) , and thus test for differences in spiking activity ( Figure 11 ) . The distribution of firing rates as well as the spiking irregularity did not have any dependence on location . This suggests that the fluctuation–driven spiking regime was both present and equally prominent in all the neurons , regardless of whether the cell bodies were in the ventral or medial horn , i . e . regardless of whether they were motoneurons or premotor interneurons . Common features of network activity for different parts of the central nervous system may provide hints towards fundamental principles of neuronal operations . In the present study we identified the following features of population motor activity: ( 1 ) synaptic input to individual spinal neurons was normally distributed ( Figure 3 ) , ( 2 ) the means of these normal distributions were also normally distributed across the population . In particular , the distance to threshold in terms of fluctuations , i . e . ( Vm-Vt⁢h⁢r⁢e⁢s ) /σ had a normal distribution and a distance from mean to threshold of 3σ on average ( Figure 3—figure supplement 2F ) . ( 3 ) The neuronal response function was supralinear when the mean input was in the subthreshold region ( Figure 4 ) . ( 4 ) There was a rich diversity of regular to irregular spiking patterns . ( 5 ) The population firing rate was skewed and lognormal–like . Many of these features have been identified before in other parts of CNS . The Vm of individual neurons is often normally distributed in cortical neurons when considering either the up– or down–state ( Destexhe et al . , 2003; Stern et al . , 1997 ) and the spiking is irregular with a C⁢V clustered around 1 ( Softky and Koch , 1993; Stevens and Zador , 1998 ) . Similar irregularity is observed in invertebrates ( Bruno et al . , 2015 ) . The distribution of mean C⁢V2 values in our experiments was clustered around 0 . 6 when ignoring the inter–burst intervals ( Figure 10C ) . This is more regular than what is observed for typical cortical neurons ( although see Feng and Brown , 1999 ) , but similar to cervical interneurons in monkeys performing isometric wrist flexion–extensions ( Prut and Perlmutter , 2003 ) . We observed a skewed and lognormal–like population distribution across behaviors ( Figure 9 , Videos 1 and 2 ) . Similar lognormal distributions have been reported in other parts of CNS ( Buzsáki and Mizuseki , 2014; Hromádka et al . , 2008; O’Connor et al . , 2010; Wohrer et al . , 2013 ) and it remains an open question how the skewness arises out of neuronal ensembles . Roxin et al proposed the mechanism where the skewness arises from a nonlinear transformation of Gaussian input ( Roxin et al . , 2011 ) . Our data supports this hypothesis . First , we observed a normally distributed Vm for individual cells , which is a proxy for the requirement of normally distributed input currents ( Figure 3 ) . Second , a supralinear IO–function covering most of this input ( Figure 4 ) . Third , a firing rate distribution of individual cells which was typically highly skewed and lognormal–like although some did not have lognormal firing ( Figure 5 ) . Nevertheless , there is a distinction between the lognormal firing of individual neurons and the lognormal distribution of mean rates across the population . Whereas the lognormal population firing rate remains to be fully understood , the skewed firing rate distribution of individual neurons is fairly well understood . Here , the skewness is due to the fluctuating input and irregularity of spiking ( Ostojic , 2011 ) . Nevertheless , we argue the mechanism for the lognormal population firing is the same as that for the individual neuron . If the subthreshold IO-function is approximately similar across the population , which our data implies ( Figure 4 ) , we can explain the lognormal population firing by a supralinear transformation , if the mean Vm across the population is also Gaussian . We did in fact find the distribution of mean Vm to be Gaussian ( Figure 3—figure supplement 2F ) . Classical studies of spinal motoneurons indicate two regimes of spiking: a primary and a secondary range ( Kernell , 2006; Meehan et al . , 2010 ) , which corresponds to different parts of the mean–driven spiking regime . This characterization was associated with the intrinsic properties without synaptic input being present . Nevertheless , a different type of fluctuation–driven spiking was discovered in experiments where synaptic input were present , in what was referred to the subprimary range in mice ( Manuel and Heckman , 2011 ) and humans ( Kudina , 1999; Matthews , 1996 ) . This subprimary range conforms to the fluctuation–regime though under a different terminology . As the name indicates , the primary range has been considered to represent the dominant mode of spiking whereas the subprimary range is a peculiarity . Nevertheless , a recent study recorded for the first time the motoneuron discharge and muscle force and found that the subprimary range accounts for 90% of the contraction force ( Manuel and Heckman , 2011 ) . This indicates that the fluctuation–regime may have a more noteworthy role in covering the dynamical range in motor control than previously assumed , which is in agreement with the observations of the present study . The experimental procedures are described in more detail at Bio-protocol ( Petersen and Berg , 2017 ) . We used the integrated turtle preparation with the spinal motor network intact ( n=5 for the multi–electrode recordings and n=60 for the serially aqquired intracellular recordings ) , in order to address how the neuronal firing rates are distributed across the population of interneurons and motoneurons in the spinal cord ( Petersen et al . , 2014 ) . These sample sizes where assumed to be large enough in the experimental design and because of a consistency in results , although a specific power analysis was not conducted . To avoid the confounding factors of supraspinal input , we spinalized the turtle . The transection was performed at the spinal cord at segments ( D3-4 ) caudal to the cervical segments , where the local circuitry has only little or no involvement in generation of motor patterns ( Mortin and Stein , 1989; Hao et al . , 2014; Mui et al . , 2012 ) . The adult turtle preparation is capable of producing elaborate motor patterns such as scratching . We used the semi-intact spinal cord of adult turtles ( Keifer and Stein , 1983; Petersen et al . , 2014 ) and recorded from the segments D8-D10 . These segments contain the essential CPG circuits ( Mortin and Stein , 1989 ) . Most of the spinal cord including the sensory periphery is left intact . The blood is replaced and the spinal column is provided with oxygenated Ringer's solution so that the neurons and the network have optimal conditions . In this experimental situation the motor behavior is as close to in vivo situation as possible , and is indistinguishable from the intact condition ( Keifer and Stein , 1983 ) . The turtle preparation allow for mechanical stability and the turtle’s resistance to anoxia allow for remarkable durability of both the recording conditions and the motor pattern reproducibility ( Vestergaard and Berg , 2015 ) . Adult red-eared turtles ( Trachemys scripta elegans ) of either sex were placed on crushed ice for 2 hr to ensure hypothermic anesthesia . The turtles were killed by decapitation and the blood was substituted by the perfusion with a Ringer’s solution containing ( mM ) : 100 NaCl; 5 KCl; 30 NaHCO3; 2MgCl2; 3CaCl2; and 10 glucose , saturated with 95% O2 and 5% CO2 to obtain pH 7 . 6 , to remove the blood from the nervous system . We isolated the carapace containing the spinal cord segments D4-Ca2 by transverse cuts ( Keifer and Stein , 1983; Petersen et al . , 2014 ) and perfused the cord with Ringer’s solution through the vertebral foramen , using a steel tube and gasket pressing against the D4 vertebra . We opened the spinal column on the ventral side along D8-D10 and gently removed the dura mater with a fine scalpel and forceps . For each insertion site for the silicon probed , we opened the pia mater with longitudinal cuts along the spinal cord with the tip of a bend syringe needle tip ( BD Microlance 3: 27G3/4" , 0 . 4x 19 mm ) . We performed the cuts parallel to the ventral horn between the ventral roots . The surgical procedures comply with Danish legislation and were approved by the controlling body under the Ministry of Justice . We used a fire polished tip of a bent glass rod for mechanical stimulation , that was mounted linear actuator . The actuator was controlled with a function generator: frequency , amplitude and duration of the stimulus . Extracellular recordings were performed in parallel at 40 KHz using a 256 channel multiplexed Amplipex amplifier ( KJE-1001 , Amplipex ) . Up to four 64-channel silicon probes were inserted in the incisions perpendicular to the spinal cord from the ventral side . We used the 64-channel Berg silicon probes ( Berg64 from NeuroNexus Inc . , Ann Arbor , MI , USA ) with 8 shanks , and 8 recording sites on each shank arranged in a staggered configuration with 30 μm vertical distance . The shanks are distanced 200 μm apart . Recordings were performed at depths in the range of 400-1000 μm . The intracellular recordings were performed in current-clamp mode with an Axon Multiclamp 700B amplifier ( Molecular devices ) . Glass pipettes were pulled with a P-1000 puller ( Sutter instruments ) and filled with a mixture of 0 . 9 M potassium acetate and 0 . 1 M KCl . Data were sampled at about 20 kHz with a 12-bit analog-to-digital converter ( Axon Digidata 1440a , Molecular devices ) . We inserted the glass electrodes from the ventral side of D8-D10 perpendicularly to the spinal cord . Neurons were located at depths ranging from about 300–800 μm . Typically we had stable intracellular recordings for multiple trials . Electroneurogram ( ENG ) recordings were performed with suction electrodes . The scratch behavior was measured by the activity of the nerves: Hip Flexor , Knee Extensor , dD8 and HR-KF . The nerve activities were recorded with a differential amplifier Iso-DAM8 ( World Precision Instruments ) with bandwidth of 100 Hz–1 kHz . For histological verification , we combined several staining techniques: The silicon probes were painted with DiI ( 1–2% diluted in ethanol ) before insertion into the spinal cord ( Blanche et al . , 2005; Vandecasteele et al . , 2011 ) . Following successful experiments , we performed Nissl– and ChAT–staining of the tissue , to determine the location of respectively neurons and motoneurons . The histological processing is detailed in ( Petersen et al . , 2014 ) . We carefully removed the tissue , perfused it and put it in phosphate buffered saline ( PBS ) with 4% paraformaldehyde for 24–48 hrs and further stored it in PBS . The tissue was mounted in an agar solution and sliced into 100 μm slices using a microtome ( Leica , VT1000 S ) . The slices were washed with PBS and incubated overnight at 5°C with primary choline acetyltransferase antibodies goat anti-ChAT antibodies ( 1:500 , Milipore , USA ) in blocking buffer , which is PBS with 5% donkey serum and 0 . 3% Triton X-100 . The slices were washed three times with PBS and incubated for 1 hr at room temperature with the secondary antibody Alexa488 conjugated to donkey anti-goat antibodies ( 1:1000 Jackson ) in blocking buffer . After three washes with PBS , the slice was mounted on cover slit with a drop of ProLong Gold antifade reagent ( Invitrogen Molecular Probes , USA ) and cured overnight at room temperature before microscopy . The slice was viewed using a confocal microscope , Zeiss LSM 700 with diode lasers , on a Zeiss Axiolmager M2 using 10x/0 . 30 EC Plan-Neofluar , 40x/0 . 6 Corr LD Plan-Neofluar , and 63x/1 . 40 oil DIC Plan-Apochromat objectives ( Zeiss ) . The data analysis was primarily done in the programming languages Matlab and Python . The correlation coefficient was calculated as the Pearson product-moment correlation coefficient . We use skewness ( Press et al . , 1992 ) or the third moment as a measure of asymmetry in the distribution around the mean , sometimes referred to as Pearson’s moment coefficient of skewness . It can be estimated using the method of moment estimator asSkewness=1N∑j=1N[xj−x¯σ]3 where x1 , … , xN are all the observations ( Vm or firing rate ) and σ and x¯ are the sample standard deviation and sample mean of the distribution . The skewness is a unitless number and a value of zero indicates perfect symmetry . A positive skew has a tale pointing in the positive direction of the axis and a negative value points in the opposite direction . Spike sorting was performed in the Klustakwik-suite: SpikeDetekt , KlusterKwik v . 3 . 0 and KlustaViewa ( Kadir et al . , 2014 ) . Raw extracellular signals were bandpass filtered from 400–9000 Hz , and spikes were detected by a median based amplitude threshold with SpikeDetekt ( Takekawa et al . , 2012; Kadir et al . , 2014; Quiroga et al . , 2004 ) . An automatic clustering of the spikes was performed in KlustaKwik , followed by manual cluster-cutting and cluster verification in KlustaViewa . The cluster quality was evaluated by several measures: The shape of the autocorrelation function , the amount of contamination in the refractory period , the Isolation distance ( Harris et al . , 2001 ) and the Lr⁢a⁢t⁢i⁢o ( Schmitzer-Torbert and Redish , 2004 ) ( Figure 2—figure supplement 2 ) . Only well isolated units was used in the further data analysis . The time-dependent firing rate ν was estimated by a gaussian kernel by convolving the spike times , s⁢ ( t ) , with a Gaussian kernel k⁢ ( t ) :ν ( t ) =∫−∞∞s ( t−t′ ) k ( t′ ) dt′ where k⁢ ( t ) is defined ask⁢ ( t ) =12⁢π⁢ω⁢e-t22⁢ω2 with the bandwidth ω optimized for each spike train with the sskernel method ( Shimazaki and Shinomoto , 2010 ) . The estimated width was in the range of 100–500 μs . The G⁢i⁢n⁢i coefficient is a measure of statistical dispersion and it is defined as a ratio of the areas on the Lorenz curve diagramG⁢i⁢n⁢i=aa+b=1-2⁢b where a+b is the area below the line of no dispersion ( the diagonal , i . e . a+b=1/2 ) , and b is the Lorenz curve , i . e . the cumulative distribution of firing rates ( Figure 9H ) . The irregularity of the spiking of individual neurons can be described by several measures . The most common measures are the coefficient of variation ( C⁢V=σ/μ ) and the Fano factor ( F=σ2/μ ) , but both measures easily overestimate the irregularity when the firing rate is non-stationary ( Holt et al . , 1996; Ponce-Alvarez et al . , 2010; Softky and Koch , 1993 ) . More advanced methods of estimating the time dependent variations in the irregularity have been developed ( Shinomoto et al . , 2009; Shimokawa and Shinomoto , 2009; Miura et al . , 2006 ) , and here we use the widely used metric C⁢V2 , which has been suggested to be the most robust measure of local spiking irregularity ( Wohrer et al . , 2013; Ponce-Alvarez et al . , 2010 ) . The time dependent C⁢V2 is defined by pairs of adjacent inter-spike intervals I⁢S⁢Ii and I⁢S⁢Ii+1:C⁢V2⁢ ( i ) =2⁢\|I⁢S⁢Ii-I⁢S⁢Ii+1\|I⁢S⁢Ii+I⁢S⁢Ii+1 where C⁢V2=1 for a Poisson process and C⁢V2=0 for a regular process . C⁢V2 can take values in the range from zero to two . We noticed a small difference in the distribution of irregularity among the neurons recorded with intracellular versus extracellular electrodes ( data not shown ) . The neurons were recorded with intracellular electrodes had more regular spiking than those recorded with extracellular electrodes . This may be caused by a systematic bias in the way the intracellularly recorded neurons were collected , as there is an experimental bias towards high firing rates . Spike sorting processing of the extracellular recordings , on the other hand , is likely to both miss spikes and contain false positives , which will cause overestimation of spiking irregularity . To get a quantitative handle on the fraction of neurons found in the fluctuation–regime across the population , we consider the distribution of neurons , f⁢ ( t ) , which spends a given amount of normalized time t in the fluctuation regime , i . e . with CV2 > icrit . We consider three values of ic⁢r⁢i⁢t , 0 . 4 , 0 . 5 and 0 . 6 , as indicators for when the neurons are in the fluctuation–regime . Formally we quantify the time in fluctuation–regime for the population using the reverse cumulative distribution of neurons ( Figure 10D–E and Figure 10—figure supplement 1 ) . The reverse cumulative fraction of neurons in the fluctuation regime F⁢ ( t ) for a given fraction of normalized time t isF ( t ) =1−∫0tf ( t ) dt , 0<t≤1 This fraction F⁢ ( t ) is the fraction of neurons , which spend at least t amount of normalized time in the fluctuation regime . To compress the distribution into a single number we use the fraction of time in fluctuation regime of half of the population , T⁢I⁢F50 , which is the value of t for which F⁢ ( t ) =50% ( arrows and broken lines , Figure 10D–E ) . Since the firing rate is rarely constant , one may want to know how many spikes are elicited in the mean– versus fluctuation regime . This is calculated in similar way , using the distribution of neurons having a normalized fraction of spikes in the fluctuation regime , i . e . spikes with CV2 > icrit , f⁢ ( s ) . The reverse cumulative of f⁢ ( s ) again gives the fraction of neurons which have at least s spikes in fluctuation regime , normalized to 100% , F ( s ) =1−∫0sf ( s ) dt , 0<s≤1 Again we compress the distribution into a single number and use the fraction of spikes , which occur in fluctuation regime of half of the population , S⁢I⁢F50 , which is the value of s for which F⁢ ( s ) =50% ( arrows and broken lines Figure 10—figure supplement 1 ) . We use a definition of the action potential threshold , which is based on the phase plot of Vm versus the derivative d⁢Vm/d⁢t . This is the second method reported in Sekerli et al . ( 2004 ) . The threshold is found as the point in the trajectory in phase space , where there is a strong departure from rest prior to the cycle . Since d⁢Vm/d⁢t is proportional to the membrane current , this point represents a strong initiation of the inward current . Defining the slope of Vm in time , f=d⁢Vmd⁢t , the threshold is defined as the largest peak in second derivative with respect to Vm in phase space , i . e . the maximum of d2 fdVm2 ( red dots , Figure 6—figure supplement 1B–C ) . This is the point with the largest acceleration from baseline prior to the peak of the action potential . The Vm trace was low–pass filtering at 5000 Hz to reduce the vulnerable to electrical noise of the estimates of derivatives . The method for estimating the response rate as a function of Vm has been described previously ( Vestergaard and Berg , 2015 ) . The relationship between firing rate , ν , and membrane depolarization is based on the assumption that spikes occur as a random renewal point–process i . e . a Poisson process . The rate is directly related to the probability , P , of a spike occurring in a small time window at a certain time t:P⁢ ( t;t+Δ⁢t ) =ν⁢Δ⁢t The window Δ⁢t has to be small such that the chance of getting more than one spike in the window is negligeble . The firing rate can thus be defined in terms of the probability of achieving a spike in an infinitesimally small time window ( Gerstner et al . , 2014 ) :ν⁢ ( t ) =limΔ⁢t→0⁡P⁢ ( t;t+Δ⁢t ) Δ⁢t This definition of ν is also called the ‘stochastic intensity’ . Since the probability P is strongly dependent on the depolarization of the membrane potential , the firing rate will be similarly dependent . To determine ν as a function of Vm we have to empirically determine the probability , P , for the smallest possible value of Δ⁢t , which is the sampling interval of the intracellular recordings . To get P as a function of membrane potential , P⁢ ( Vm ) , we first empirically determine the stochastic distribution of Vm prior to the spike ( 1 . 5-1 . 7 ms prior ) , which we know will cause a spike . Then we normalize this distribution with the amount of time spent at each Vm-level at all time . This is the estimated probability of getting a spike , P , within a small time window Δ⁢t for a given Vm , i . e . the firing rate as a function of Vm . This empirical method of relating firing rate and Vm was relatively recently invented ( Jahn et al . , 2011 ) and used in determining IO properties of e . g . motoneurons ( Vestergaard and Berg , 2015 ) . The shape of the spike response function is highly non-linear with upward curvature . This has been observed in previous experiments ( using a different method ) and has often been referred to as expansive non-linearity ( Hansel and van Vreeswijk , 2002; Miller and Troyer , 2002; Murphy and Miller , 2003; Priebe and Ferster , 2005 , 2008 ) . An exponentialν⁢ ( Vm ) =c⁢eβ⁢Vm was fitted to capture the curvature , where the curvature is represented in the exponent β , which have units of 1/m⁢V , and c is a constant of units 1/s . Such expansive non-linearities have also been investigated in the visual cortex where they are often characterized as a power-law relationship , i . e . ν⁢ ( Vm ) =k⁢[Vm-Ea]α where k is a constant and α is the power >1 , i . e . supralinear , and often ranging from 2-5 ( Hansel and van Vreeswijk , 2002; Miller and Troyer , 2002 ) . This exponent is also a measure of the expansive curvature of the non-linearity . Ea represent a subthreshold level of Vm , where the spiking probability is zero , such that the values in the sampled traces are always larger than Ea , i . e . Vm > Ea . The curvature dependence on synaptic fluctuations was assessed by the standard deviation of the distribution of Vm traces prior to the spike in the diffusion regime , i . e . where there was no link to the Vm and the spike occurrence . This distribution was chosen 18 ms prior to the spike ( Figure 3B ) . The analysis and fits were performed in Matlab with generic fitting functions . In order to distinguish neurons in fluctuation– versus mean–regime , we employ a new metric for quantifying the degree of fluctuations in Vm in between action potentials . We plot the values of Vm in a return map , which is a plot of Vm⁢ ( t ) versus Vm⁢ ( t+Δ⁢t ) . If the inter–spike Vm has a direct trajectory from the reset potential to the next spike , Vm will smoothly increase and thus Vm⁢ ( t+Δ⁢t ) will always be larger than Vm⁢ ( t ) . Therefore each point will be above the line of unity ( Figure 3—figure supplement 1A–B ) . On the other hand , if Vm has fluctuations , it will have an indirect and convolved trajectory from the reset value to the threshold . This will manifest as containing values of Vm⁢ ( t+Δ⁢t ) which are actually smaller than Vm⁢ ( t ) . Thus we use the ratio of points above versus below the unity line as a metric for how convolved and fluctuating the path of Vm is from reset to threshold . If the ratio is ∼0 . 5 then Vm is highly fluctuating , whereas if the ratio is approaching 1 the path is straight without any fluctuations . We choose a mean value of the histogram of all values to 0 . 7 to classify neurons as fluctuation– or mean–driven ( Figure 3—figure supplement 1C ) . This metric of straight versus convolved trajectory had significant negative correlation with other measures of fluctuation– regime , e . g . spike rate skewness , spike irregularity ( C⁢V2 ) and least time below threhold ( LTBT , Figure 3—figure supplement 1D–F ) . The choice of Δ⁢t is not important as long as it is larger than the timescale of electronic fluctuations of the amplifiers and smaller than the timescale of synaptic fluctuations in Vm . We consistently used Δ⁢t=1 . 5 ms for all neurons . The return map ratio is intended as a metric to analyze sub-threshold activity and therefore spikes were removed from the traces , including a 6 ms window before and after the peak . Also , the Vm containing the interburst ( defined as having ISIs > 300 ms ) intervals was removed . Trilateration is a geometrical process of determining the location of a source in 2D–space using multiple recording sites scattered in space . We adapted the method to take advantage of a distance–dependent decay of the electrical signal from the action potential in the extracellular space . In this way , the amplitudes of the waveforms , which were simultaneously recorded on multiple electrodes , revealed the location of the source in space relative to the position of the electrodes . We assumed that the electrical signal decayed as 1/r2 , where r is the distance . In Figure 2 , the following trials were used: n=[6 , 4 , 9 , 5 , 6] for ipsilateral pocket scratch and n=[6 , 3 , 10 , 5 , 6] for contralateral pocket scratch . Data used in Figure 7 has already been published in a different context ( Berg et al . , 2007 ) . A small subset of the neurons used in Figure 3D–E ( n=10 out of 68 ) has been acquired in a reduced preparation ( Petersen et al . , 2014 ) and published for an investigation of a different matter ( Berg et al . , 2007; Berg and Ditlevsen , 2013 ) . The data from experiments of blockade of inhibition using superfusion of strychnine has also been published previously in the investigation of a different matter ( Vestergaard and Berg , 2015 ) . Regarding excluding spikes from the analysis in Figure 3C–E: For the temporal distribution ( panel C ) , only ISIs > 6 ms was included and for the spike triggered Vm-distribution only ISIs > 20 ms was included , all having ISIs < 300 ms . Estimating the FV-curve ( Figure 4 ) all spikes having ISIs > 1 . 7 ms was included . Neuronal spiking has traditionally been considered to occur when the mean inward current of the cellular membrane is large enough to cross the rheobase such that the mean membrane potential ( Vm ) is above threshold ( Vt⁢h⁢r⁢e⁢s ) . In practice , the mean Vm will not exceed Vt⁢h⁢r⁢e⁢s by very much due to the active spiking and after–hyperpolarization , but if this mechanism was turned off the mean membrane current ( Im ) would drive Vm across threshold , formally written as Im>Vt⁢h⁢r⁢e⁢s/Rm where Rm is the membrane resistance . Spikes elicited in this manner are in the mean–driven regime ( Gerstner et al . , 2014; Renart et al . , 2007 ) . They have shorter inter–spike intervals ( ISIs ) because of the large Im and regular spiking due to the after–hyperpolarization . In contrast , when the mean Vm is below threshold , i . e . Im < Vthres/Rm , spikes are elicited by temporary fluctuations in Vm due to synaptic bombardment . Such spiking is in the fluctuation–driven regime ( Kuhn et al . , 2004; Tiesinga et al . , 2000; Gerstner et al . , 2014; Roxin et al . , 2011 ) . The random synaptic fluctuations cause the spiking to be more irregular , which results in a higher coefficient of variation ( CV , defined as the standard deviation ( σ ) divided by the mean of ISIs ) , than for the mean–driven regime ( cf . Figure 1D–E ) . Therefore irregularity is an indicator of the spiking regime . Another indicator of the fluctuation–driven regime is positive skewness of the firing rate distribution ( Figure 1A–B ) . These indicators are used to quantify the fraction of the population that is in one versus the other regime .	Where and how are rhythmic movements , such as walking , produced ? Many neurons , primarily in the spinal cord , are responsible for the movements , but it is not known how the activity is distributed across this group of cells and what type of activity the neurons use . Some neurons produce regular patterns of “spiking” activity , while others produce spikes at more irregular intervals . These two types of activity have different origins and represent different states of the neural network . It is not clear whether they participate equally in a movement , or if there is a hierarchy among the neurons , such that some neurons have more influence than others . Petersen and Berg studied neurons in the lower spines of turtles during rhythmic movements . The experiments show that during rhythmic scratching some neurons are very active while most aren’t particularly active at all . This is known as a lognormal distribution and is seen in many other situations , such as the levels of income of people in a society . Petersen and Berg also found that neurons can move between two regimes of activity , called the mean-driven and fluctuation-driven spiking regimes . During rhythmic scratching , the neurons are almost equally divided between the two regimes , and this division is also found in other types of rhythmic movement . This even division between the two regimes is likely to be important for maintaining a balance between the sensitivity and stability of the neural network . The next steps following on from this work are to reveal the mechanisms behind the two regimes and to find out what causes these differences in activity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Lognormal firing rate distribution reveals prominent fluctuation–driven regime in spinal motor networks
Inflammatory osteolysis is governed by exacerbated osteoclastogenesis . Ample evidence points to central role of NF-κB in such pathologic responses , yet the precise mechanisms underpinning specificity of these responses remain unclear . We propose that motifs of the scaffold protein IKKγ/NEMO partly facilitate such functions . As proof-of-principle , we used site-specific mutagenesis to examine the role of NEMO in mediating RANKL-induced signaling in mouse bone marrow macrophages , known as osteoclast precursors . We identified lysine ( K ) 270 as a target regulating RANKL signaling as K270A substitution results in exuberant osteoclastogenesis in vitro and murine inflammatory osteolysis in vivo . Mechanistically , we discovered that K270A mutation disrupts autophagy , stabilizes NEMO , and elevates inflammatory burden . Specifically , K270A directly or indirectly hinders binding of NEMO to ISG15 , a ubiquitin-like protein , which we show targets the modified proteins to autophagy-mediated lysosomal degradation . Taken together , our findings suggest that NEMO serves as a toolkit to fine-tune specific signals in physiologic and pathologic conditions . The transcription factor NF-κB is expressed ubiquitously in all cell types , readily activated by numerous factors and cytokines ( Abu-Amer and Faccio , 2006; Hayden , 2004; Ravid and Hochstrasser , 2004; Ting and Endy , 2002 ) ;it plays critical roles in modulating inflammation , immunity , cell proliferation , differentiation , and survival . While baseline NF-κB activity is essential for physiologic functions such as skeletal development ( Abu-Amer , 2013; Courtois et al . , 2001; Häcker and Karin , 2006; Li et al . , 2002; Ruocco and Karin , 2005; Ruocco and Karin , 2007 ) , exacerbated and chronic unrestrained activity of this transcriptional factor during inflammation leads to undesired harmful effects with major dysfunctional consequences including osteolysis ( Abu-Amer , 2013; Boyce et al . , 2010; Pasparakis , 2008; Ruocco and Karin , 2005; Schett and David , 2010; Xing et al . , 2005; Xu et al . , 2009 ) . In this regard , we and others have shown that , whereas baseline activity of the principal NF-κB kinase IKKβ ( also known as IKK2 ) is essential for normal skeletal development , its hyper and prolonged activation is pathologic ( Otero et al . , 2010; Zhang et al . , 2013 ) . In addition , NF-κB is the primary pathway that mediates inflammatory responses of numerous bone-targeting cytokines such as TNFα , IL-1β , and IL6 ( Abu-Amer , 2009 ) . When a stimulus is provided , signaling molecules are recruited to distal domains of the appropriate receptor leading to the assembly of the IKK complex which includes IKKγ/NEMO and IKKβ , among other adaptor proteins . This process leads to phosphorylation of downstream substrates , most notably , IκBα and activation of downstream transcriptional machinery ( Boyce et al . , 2005; Boyce et al . , 2010; Franzoso et al . , 1997 ) . Gene deletion studies have shown that members of the NF-κB signal transduction pathway , which include NF-κB1 ( p50 ) , NF-κB2 ( P52 ) , RelA ( p65 ) , IκBα , IKKα , IKKβ , and NEMO , are crucial for normal development of osteoclasts and their survival , and are considered as the principal mediators of RANK signaling ( Boyce et al . , 1999; Boyce et al . , 2010; Franzoso et al . , 1997 ) . Despite the intense research efforts focusing on the role of this pathway in physiologic and inflammatory responses , little is known regarding the cell-specific response to a given signal and pairing it with corresponding function in homeostatic and pathologic settings . For example , whereas RANKL , TNFα , IL-1β , and other factors activate NF-κB in osteoclast precursors , the molecular machinery that assigns unique signaling signatures for each stimulus remains vague . The NF-κB pathway offers a toolbox that enables signal and cell-specific signaling cascades . In this regard , signal specificity can occur at proximal ( juxtaposed to receptors ) and distal ( downstream ) pathway sites . In the former case , ligation of cell surface receptors such as TNFα receptor ( TNFR ) , RANK or IL1 receptor ( IL-1R ) triggers a chain of events that leads to the formation of a signalsome that includes unique TRAF proteins , IKKα/β and NEMO . It has been further shown that this complex is regulated by post-translational machineries , chiefly phosphorylation , ubiquitination and SUMOlyation ( Chen , 2012; Ikeda et al . , 2010; Ikeda et al . , 2011; Laplantine et al . , 2009; Liu and Chen , 2011; Ni et al . , 2008; Sebban et al . , 2006; Shambharkar et al . , 2007; Yeh , 2009 ) , which are believed to contribute significantly to signal specificity by directing downstream cues . NEMO , a key player of the IKK signalsome , is a scaffold protein that lacks enzymatic activity; yet , it is essential for NF-κB signaling evident by convergence of upstream signals directed by TRAFs prior to assembly of downstream IKK signals ( Li et al . , 2001; May et al . , 2002; Prajapati and Gaynor , 2002; Yamamoto et al . , 2001 ) . Recent studies have shown that specific NEMO domains and numerous lysine residues throughout the different domains of NEMO , especially the ubiquitin and zinc finger domains , undergo extensive ubiquitination , SUMOylation , and other post-translational modifications ( PTMs ) in response to various stimuli ( Cordier et al . , 2009; Hay , 2004; May et al . , 2002; Rushe et al . , 2008; Schröfelbauer et al . , 2012; Sebban et al . , 2006; Wu et al . , 2006 ) . Specifically , these domains and lysine residues serve as specific docking sites utilizing PTM moieties to enable recruitment of unique signaling complexes and pathway substrates in one hand , and proteasome-mediated degradation , in the other hand . Indeed , the critical role of a number of lysines and other residues such as K270 , K302 , K312 , K392 , C417 in cellular signaling have been described ( Alhawagri et al . , 2012; Bloor et al . , 2008; Ni et al . , 2008; Yang et al . , 2004 ) . In this regard , series of NEMO mutants at specific lysine residues located at the coil zipper domain revealed dominant negative and constitutive activation properties of NEMO ( Bloor et al . , 2008 ) . In the current study , we tested the functional significance of key lysine residues individually in the ubiquitin and zinc finger domains of NEMO in response to RANKL . This approach was designed to test our hypothesis that certain NEMO lysine residues serve as signal-specific docking sites that facilitate the assembly of unique signal activating- or suppressing-protein complexes in cell and stimulus specific manner . To address our aforementioned hypothesis , we conducted broad lysine ( K ) screen of NEMO and substituted strategic K and D residues in tandem with alanines and asparagine , as indicated , ( Figure 1A ) , to disrupt post-translational modifications of specific NEMO lysines , and hence , impede assembly of protein complexes and alter subsequent signaling . Wild type ( WT ) NEMO ( NEMOWT ) and various NEMO K mutants ( NEMOK ) were cloned in pMx-retroviral plasmid and expression of representative clones was confirmed in HEK293 cells ( Figure 1B ) . Viral particles of these plasmids were produced using PLAT-E cells as described previously ( Swarnkar et al . , 2016 ) . Protein expression of NEMOWT and NEMOK mutants was conducted in primary bone marrow macrophages ( BMMs a . k . a osteoclast precursors ) . BMMs from WT or NEMO null cells ( NM-cKO ) expressing NEMOK270A formed ample osteoclasts ( OC ) , reminiscent of inflammatory condition , upon exposure to permissive concentration of RANKL , when compared with cell expressing NEMOWT ( NM-WT ) ( Figure 1C ) or other forms of NEMO mutants ( Figure 1—figure supplement 1A ) , suggesting sensitization of RANKL signaling in the absence of K270 residue in NEMO . These observations were further supported by increased OC surface area ( Figure 1D ) , elevated expression of osteoclast differentiation markers ( Figure 1E–J ) and NF-κB activity ( Figure 1K ) in RANKL-treated NEMOK270A expressing BMMs compared with NEMOWT ( NM-WT ) and other NEMO forms . Expression of NEMOK270A ( also referred to in figure labels as NM-KA for brevity ) in BMMs was robust and stable compared with NM-WT and other NEMO mutants ( Figure 1—figure supplement 1B ) , despite infection of BMMs with equal number of viral particles . In fact , hyper osteoclastogenesis by NEMOK270A was not due to higher expression of the transgenic NEMO protein because 1:20 dilution of viral particles which gave rise to protein expression levels approximating those of NEMOWT and of other NEMO constructs still provoked heightened osteoclastogenesis ( Figure 1—figure supplement 1A–B ) . These findings support the notion that intact K270 appears to be essential to restrain RANKL-induced excessive osteoclastogenesis . To examine the functional relevance of this in vitro abnormal signaling by NEMOK270A mutant in the in vivo state , we generated NEMOK270A transgenic mice by cloning this construct in the Gt ( ROSA ) 26Sor locus as described previously ( Figure 2—figure supplement 1A; Otero et al . , 2012; Swarnkar et al . , 2014 ) . Gt ( ROSA ) 26Sor harboring NEMOWT ( ikbkg ) was also generated as a control ( Figure 2—figure supplement 1B ) . Mice were born at mendelian ratio , yet transgenic NEMOK270A ( ikbkg ) knockin ( KA ) mice were significantly smaller in size compared with transgenic WT and control mice ( Figure 2A ) . Upon closer analysis , we observed significant joint swelling ( Figure 2A; arrows ) , splenomegaly , altered hematopoiesis ( Figure 2B and Figure 3—figure supplement 1 ) , and enhanced mortality at 6–8 weeks of age ( not shown ) . Micro-computed tomography ( uCT ) scans and X-ray images revealed dramatic bone loss reaching 50% of BV/TV in severe cases ( Figure 2C–I and Figure 2—figure supplement 1C–F ) , with no such loss in NEMOWT transgenic mice ( NM-WT-Tg ) ( Figure 2—figure supplement 1G–L ) . Notably , knee and ankle joints were severely damaged ( Figure 2D:compare WT with KA , Figure 2—figure supplement 1C , E , F:arrows ) . Histological analysis showed that bone and joint sections contain exuberant number of TRAP-positive OCs with evidence of loss of trabecular bone and destruction of articular cartilage by osteoclasts and inflammatory cells , which may also result from synovial pannus formation containing osteoclasts ( Figure 2J–K: arrows ) . This phenotype was further corroborated with increased circulating serum levels of TRAcP5b and carboxyl-terminal cross-linked telopeptide of type one collagen ( CTX-1 ) , both are well established markers of bone breakdown ( Figure 2L–M ) . The observed joint swelling , skeletal degeneration , splenomegaly , and osteolysis point to systemic inflammation and altered hematopoiesis . Indeed , multiplex ELISA revealed that NEMOK270A mice express copious amounts of circulating inflammatory cytokines and chemokines in the serum ( Figure 3A–L . At the cellular level , FACS analysis confirmed skewing of hematopoiesis toward myelopoiesis evident by abundant frequency of CMPs and GMPs , the immediate OC progenitors , in spleen and bone marrow compartments ( Figure 3—figure supplement 1A–E; arrows ) . In addition , we detected a spike in neutrophils frequency in NEMOK270A knockin mice ( Figure 3—figure supplement 1F–J; arrows ) . These observations suggest that the ensuing inflammatory microenvironment in these mice alters hematopoiesis and exacerbates osteoclastogenesis . To provide additional mechanistic support for this proposition , ex-vivo osteoclastogenesis experiments showed that BMMs derived from NEMOK270A knockin mice readily formed osteoclasts under RANKL permissive conditions ( Figure 3M ) . Furthermore , expression of osteoclastogenic markers ( Figure 3N–R ) and activation of NF-κB ( p-p65/RelA ) ( Figure 3S ) were markedly elevated in RANKL-stimulated NEMOK270A compared with WT cells . To glean more mechanistic insights , we examined localization and cellular distribution of NEMO in HEK293 ( PLAT-E ) cells and BMMs . Whereas NEMOWT was evenly diffused in the cytoplasm , NEMOK270A transgene formed puncta juxtaposed to nuclei ( Figure 4A; arrows ) in PLAT-E cells . Electron microscopy scanning of RANKL-treated BMMs ( x7 , 500 magnification ) further showed that unlike WT cells , NEMOK270A cells ( panel labeled NM-KA ) exhibit cytoplasmic aggregates ( Figure 4—figure supplement 1A; yellow arrows ) , reminiscent of cytoplasmic debris . Thus , we surmised that K270A mutation of NEMO may have altered physiologic autophagy . To this end , in vitro RANKL-primed pre-OCs expressing NEMOWT and NEMOK270A fixed and stained with mouse NEMO and rabbit LC3 antibodies and Alexa Fluor secondary conjugates . LC3 is a bona-fide marker of autophagy , which is lapidated and degraded during physiologic autophagy . In contrast , sustained elevated levels of LC3 are observed during dysfunctional autophagy . The data summarized in Figure 4B and Figure 4—figure supplement 1B , C , showed that NEMOK270A and LC3 accumulation was far greater than NEMOWT and LC3 ( arrows ) . This result was further mirrored by LC3 immunoblots ( Figure 4C ) , pointing to a possible defect in autophagy flux in NEMOK270A expressing cells . Even more convincingly , quantitative flow cytometry analysis of LC3-GFP levels in NEMOWT and NEMOK270A BMMs revealed marked accumulation of this protein in the latter cells ( black dots ) compared with WT cells ( black dots ) 6 hr post starvation ( Figure 4D ) . Further analysis depicted in Figure 4E showed that starvation ( yellow histograms ) led to accumulation of LC3 in NEMOK270A cells ( shifted to the right ) but not in WT cells ( shifted to the left ) , which was reversed in WT cells in the presence of autophagy inhibitor chloroquine ( pink histograms ) to mimic NEMOK270A cells , as expected . In fact , chloroquine exacerbated osteoclastogenesis by WT just as NEMOK270A did ( Figure 4—figure supplement 1D ) . Finally , a change in mean fluorescence intensity ( MFI ) of GFP showed pronounced decrease in LC3 in NEMOWT ( NM-WT ) compared with meager change in NEMOK270A ( NM-KA ) cells ( Figure 4F ) . Consistent with these observation , levels of mTOR , a well-documented autophagy regulator , were elevated in NEMOK270A ( NM-KA ) cells ( Figure 4—figure supplement 1E ) . Taken together , quantitative FACS scatters and histograms confirm that NEMOK270A cells have a defect in autophagy flux and aggregation of mutant NEMO that lends itself to heightened signaling . Accumulation of NEMOK270A aggregates ( puncta ) suggests a potential defect in lysosomal degradation due to restriction in autophagosomes . Thus , we examined localization of NEMO with the lysosome marker LAMP1 by immunofluorescence and by EM scanning . The data show that NEMOK270A failed to localize with lysosomes ( Figure 5A–B , Figure 5—figure supplement 1A–C ) , suggesting a defect in delivery of NEMOK270A in autophagosome to lysosome . In fact , careful examination in EM images , showed robust expression of NEMOK270A which was restricted to autophagosome ( AP ) structures compared to NEMOWT observed in lysosomes ( L ) ( Figure 5B ) . Finaly , we show that following RANKL stimulation of BMMs and subsequent starvation , NEMO puncta accumulation persisted in NEMOK270A cells following serum starvation ( akin to defective autophagy flux ) yet puncta faded in WT cells following starvation as it undergoes processing by normal autophagy flux ( Figure 5C–D ) . In sum , our data suggest that NEMO mutation at K270A disrupts the delivery of NEMOK270A from autophagosome to lysosomes leading to accumulation of NEMO signals and subsequent buildup of inflammatory and osteoclastogenic signals . Intact NEMO K270 residue is essential for post-translational modification ( PTM ) to regulate autophagy and osteoclastogenesis . To further elucidate the mechanisms underpinning inflammation and osteolysis in NEMOK270A transgenic mice , we conducted proteomics on NEMOWT and NEMOK270A immunoprecipitates . Proteomic analysis revealed that expression of autophagy and PTM proteins , especially the ubiquitin-like protein ISG15 , is altered in cells expressing NEMOK270A mutant compared with WT cells ( Figure 6A ) . Note that ISG15 is an IFN-stimulated gene and a ubiquitin-like protein that modulate cellular signals through a process termed ISGylation analogous to ubiquitination , but its function during osteoclastogenesis has not been described . Indeed , immunofluorescence images displayed NEMO ( red ) -ISG15 ( green ) co-localization in vacuolar structures in NEMOWT cells , yet such colocalization was markedly reduced in NEMOK270A cells ( Figure 6B ) . This observation was further confirmed using immunogold EM imaging demonstrating NEMO ( 18 nm ) -ISG15 ( 12 nm ) interaction in autophagosome in NEMOWT cells , which is then delivered into the lysosome structure following cell 6 hr serum starvation ( Figure 6—figure supplement 1 ) . In contrast , neither NEMOK270A-ISG15 interaction nor the delivery of NEMOK270A to lysosomes were detected in NEMOK270A cells ( Figure 6—figure supplement 1 ) . Next , we provide biochemical evidence that whereas RANKL induced robust ISGylation in WT cells , this PTM was markedly reduced in NEMOK270A lysates ( Figure 6C ) , affirming the key role of NEMO post translational ISGylation in osteoclastogenesis , a novel finding that has not be described previously . Indeed , supporting its role as regulator of osteoclastogenesis , BMMs derived from ISG15 null mice generated far more osteoclasts than their WT counterparts ( Figure 6D–E ) . This observation was further supported by in vivo data showing that bone mass ( BV/TV ) of mice lacking ISG15 is significantly lower than WT littermates , a finding further confirmed by increased levels of serum levels of TRAPc5b and CTX-1 , both markers of bone resorption ( data not shown ) . Our findings thus far suggest that ISGylation of NEMO at K270 occurs in response to stimulation of BMMs with RANKL and appears essential for proper autophagy-regulation of NEMO . In addition , exuberant osteoclastogenesis in the absence of ISG15 or in NEMOK270A , which is presumably hypo-ISGylated , events that we show lead to defective autophagy , strongly suggest that proper ISGylation of NEMO is required to turn-off NEMO signaling through autophagy to restrain osteoclastogenesis at the opportune time . To offer further support for this paradigm , retroviral expression of ISG15 inhibited osteoclastogenesis in WT cells . In contrast , exuberant osteoclastogenesis by NEMOK270A cells , which we showed irresponsive to ISG15 and exhibit defective autophagy , remained unabated ( Figure 7A–B ) . Mechanistically , we conducted flow cytometry analysis of LC3-GFP levels in NEMOWT and NEMOK270A BMMs overexpressing retroviral ( pMX ) -ISG15 in response to serum starvation-induced autophagy . The data show that LC3I/II levels were significantly reduced in ISG15-infected WT cells compared with minimal reduction in ISG15-infected NEMOK270A cells ( Figure 7C; compare black dots in top scatters of NM-WT and NM-KA; also compare shift to the left of yellow histogram ( arrow ) depicting NM-WT compared to negligible shift in orange histograms depicting NM-KA , both overexpressing ISG15 ) . These changes are further quantified in Figure 7D , confirming significant reduction of LC3 levels in ISG15 overexpressing WT cells compared with ISG15 overexpressing NEMOK270A cells , an indicative of defective autophagy in the latter cells . These observations suggest that failure of ISG15 to inhibit osteoclastogenesis in NEMOK270A is likely due to its inability to properly tether NEMOK270A to the autophagy-lysosomal machinery . To overcome this predicament , we fused ISG15 to RFP-NEMOK270A and to RFP-NEMOWT . Unlike NEMOK270A alone , the ISG15::NEMOK270A fused constructs inhibited osteoclastogenesis ( Figure 7E–F; fusion construct ) and corrected the autophagy flux evident by disappearance of punctate in NEMOK270A cells ( Figure 7G ) , reduced LC3 positive puncta and protein ( Figure 7H–I , Figure 7—figure supplement 1 ) , and co-localization of ISG15-NEMOK270A with LAMP1 ( lysosomes ) ( Figure 7J ) . Taken together , forced fusion of ISG15 to NEMOK270A facilitates autophagy and inhibits exuberant osteoclastogenesis . Previous reports by our group and others have shown that various members of the NF-κB family are essential for osteoclastogenesis and bone homeostasis , whereas their deletion disrupts these processes and leads to skeletal abnormalities ( Abu-Amer and Faccio , 2006; Boyce et al . , 2010; Jimi and Ghosh , 2005; Otero et al . , 2012; Otero et al . , 2010; Otero et al . , 2008; Ruocco and Karin , 2005; Schett and Smolen , 2005; Swarnkar et al . , 2016; Swarnkar et al . , 2014; Whyte , 2006 ) . Generally , the transcription factor NF-κB is activated in response to a multitude of signals in all cell types leading to specific functions that in most cases depend on IKK complex activation . Subsequently , the canonical IKK complex , which is dominated by IKK2 and NEMO activates an array of downstream signals . However , the precise molecular steps underpinning signal to substrate specificity orchestrated by NF-κB in these responses remains unclear . We surmised that the scaffold protein NEMO provides such specificity . Specifically , we deduced that NEMO undergoes signal-specific PTMs at specific residue ( s ) . These PTMs facilitate corresponding biological functions by pairing signal ( specifically induced by upstream molecules such as TRAFs , IKKs , etc ) with downstream substrates as has been suggested previously ( Schröfelbauer et al . , 2012 ) . This is based on ample reports documenting robust polyubiquitination , SUMOylation , and phosphorylation of NEMO at various lysine and cysteine residues that form selectively in response to different stimuli ( Hay , 2004; Liu and Chen , 2011; Mabb and Miyamoto , 2007 ) . More importantly , in most cases , these PTMs mediate destructive or constructive functions such as proteasome-mediated degradation or conversely facilitate intracellular localization and signal transduction ( Fontan et al . , 2007; Kawadler and Yang , 2006; Lamothe et al . , 2007; Sebban et al . , 2006; Wu et al . , 2006 ) . In agreement with our hypothesis , and as a proof-of-concept , we uncovered that Lys270 modulate osteoclastogenesis . According to our findings , mutating Lys270 into Ala sensitizes BMMs to RANKL signaling and sustains heightened osteoclastogenesis in vitro and in vivo evident by robust systemic bone loss , skewing toward increased myeloid progenitor frequency and extramodular hematopoiesis ( splenomegaly ) , increased inflammatory burden evident by robust secretion of a myriad of inflammatory mediators , and devastating bone erosion of the joints of mice harboring NEMOK270A . Together , these observations suggest that intact K270 in NEMO is critical to restrain and attenuate RANKL signaling and inflammation . It further suggests that Lys270 serves as a docking site for a RANKL-induced negative feedback mechanism and mutation of this residue renders this regulatory mechanism dysfunctional resulting with robust and unrestrained osteoclastogenesis in vitro and in vivo . Another significant observation was the apparent enhanced stability and accumulation of NEMOK270A in peri-nuclear cytoplasmic structures . This puncta aggregation of NEMOK270A hinted to us that the protein accumulates and localizes in sub-cellular structures inaccessible for processing reminiscent of failed proteolysis by autophagy . Indeed , using multiple thorough approaches , we show that , unlike NEMOWT , NEMOK270A localizes to autophagosomes but failed to co-localize with lysosomes . Consistently , cells expressing this NEMO mutant express high levels of LC3 which accumulated and fails to undergo degradation upon induction of autophagy . In this regard , the link between inflammation , NF-κB , and autophagy has been widely described ( Pawlowska et al . , 2018; Ravanan et al . , 2017; Wu and Adamopoulos , 2017; Yin et al . , 2018 ) . Several studies have shown that autophagy regulates osteoclast differentiation and joint destruction in experimental rheumatoid arthritis ( Cejka et al . , 2010; Jaber et al . , 2019; Lin et al . , 2013; Sanchez and He , 2009 ) . Other studies suggested that autophagy is essential to regulate inflammatory responses by reducing levels of inflammatory cytokines ( Wu and Adamopoulos , 2017 ) , and that dysfunctional autophagy exacerbates skeletal joint disease such as rheumatoid arthritis and osteoarthritis ( Bouderlique et al . , 2016; Gros , 2017; Srinivas et al . , 2009; Yin et al . , 2018 ) . Hence , autophagy function appears cell context-dependent during physiologic and pathologic conditions . To decipher the underlying mechanism of this dysfunction , we carried out a proteomic experiment and identified a novel mechanism by which NEMO signal is processed . Indeed , we identified altered expression of major autophagy proteins and some ubiquitin ( UB ) -like proteins in NEMOK270A cells compared with NEMOWT cells . Most interestingly , we uncovered that levels of the UB-like protein ISG15 were significantly lower in NEMOK270A cells . Given these changes and the similarities between the mechanisms governing the function of ubiquitin , SUMO and ISG15 , it was intriguing to identify a potential role for this protein in our system . Unexpectedly , we found that RANKL induces robust expression of ISG15 and ISGylation profile in NEMOWT OCP immunoprecipitates compared with significantly lower ISGylation in NEMOK270A immunoprecipitates . While the impact of ISGylation on most proteins remains largely unknown , it has been suggested that this modification may regulate proteins by either disrupting or enhancing their activity ( Campbell and Lenschow , 2013; Hermann and Bogunovic , 2017; Morales et al . , 2015 ) . Despite the scarce knowledge in this field , HEK293 in vitro transfection studies have shown that NF-κB pathway is negatively regulated by ISGylation ( Minakawa et al . , 2008 ) . In other studies , ISG15 was co-localized with the autophagy proteins beclin-1 ( BCLN1 ) , HDAC6 , and P62/SQSTM1 ( Nakashima et al . , 2015; Xu et al . , 2015 ) . In one case , ISG15 conjugation of HDAC6 and P62 led to their degradation , providing strong evidence that connects IFN stimulation , ISGylation and autophagy . Moreover , it has been suggested that ISGylation acts as a defense mechanism whereby ISG15 marks proteins to re-direct them towards degradation by the lysosome ( Villarroya-Beltri et al . , 2017 ) . Accordingly , we observed an overall reduction of most autophagy proteins in NEMOK270A immunoprecipitates . Most importantly , we provide direct evidence that forced fusion of ISG15 to hyperactive NEMOK270A facilitates autophagy flux and diminishes osteoclastogenesis . Our findings suggest that ISG15 directly or through other mediators such as ubiquitin chains , tethers target cargo proteins and facilitates fusion with lysosomes leading to their degradation . More specifically , we propose that this process is essential to regulate osteoclastogenesis and attenuate RANKL-induced NF-κB signaling in a timely manner , absence of which leads to unabated osteoclastogenesis and deleterious skeletal anomalies . This is supported by the critical role of NF-κB in general and NEMO specifically in osteoclastogenesis , and by the proposed role of NEMO as a hub not only for ubiquitin PTMs but also as an interactome with autophagy proteins such as beclin-1 , P62 , and Rubicon . The significance of ISG15 is further underscored by our observation that BMMs-lacking ISG15 express modestly higher levels of NEMO and LC3 , and generate more osteoclasts compared with WT cells . Consistently , ISG15 null mice have moderate osteopenia consistent with overall increased osteoclastogenesis and bone resorption , suggesting that this could be a universal regulatory mechanism . Nevertheless , given the germline deletion of ISG15 in these mice , their overall mild phenotype could be affected by compensatory responses emanating from numerous other cells and tissues which are beyond the scope of this research . In summary , this study identifies several novel observations . First , we identified NEMO K270 as a crucial regulator of RANKL-induced osteoclastogenesis , and that RANKL appears to utilize this lysine site to exert its osteoclast ‘restraining’ ( negative-feedback ) mechanism . Second , we provide novel evidence that mutation of NEMO Lys270 to Ala inflicts an uncontrolled pathologic response that exacerbates osteoclastogenesis . Third , mice harboring myeloid NMEOK270A develop severe osteopenia and joint erosion . Fourth , we identified novel RANKL-induced expression and regulation of ISG15 . Consistently , we also provide new evidence that BMMs lacking ISG15 generate more osteoclasts compared with WT littermates , suggesting that ISG15 plays a negative role in this process . Finally , we identified autophagy as key regulatory system recruited by ISG15 through NEMOK270 to dampen NF-κB signaling and maintain homeostatic osteoclastogenesis . Altogether , we conclude that ISGylation of NEMO at Lys270 is essential for recruitment of the autophagy machinery to down regulate RANKL signaling . Ikbkg ( Nemo ) -floxed ( NM-f/f ) mice on a C57BL/6 background were provided by Dr . Manolis Pasparakis ( Cologne , Germany ) . The NEMO-K270A-floxed and NEMO-WT-Tg-floxed mice were generated at the Mouse Genetics Core , Washington University ( St . Louis , MO ) . To generate NEMO-K270A and NEMO-WT-floxed transgenic mice; cDNA encoding NEMO-K270A mutation and NEMO-WT preceded by a loxP-flanked STOP cassette was cloned into the ubiquitously expressed Gt ( ROSA ) 26Sor locus ( Figure 2—figure supplement 1A–B ) . In order to conditionally delete NEMO and express NEMO-K270A or NEMO-WT-Tg in myeloid cells , the NEMO f/f , NEMO-K270A-f/f and NEMO-WT-Tg f/f mice were crossed with Lyz2 ( LysozymeM ) -cre mice to generate LysM-cre-NEMO-flox ( NM-cKO ) , LysM-cre-NEMO-K270A-f/f ( NM-KA ) and LysM-cre-NEMO-WT-f/f ( NM-WT-Tg ) respectively . RELA ( NF-ĸB ) -GFP-luciferase reporter mice were purchased from The Jackson Laboratory ( Bar Harbor , ME , USA ) . ISG15 ( ISG15 ) knock-out mice were provided by Dr . Deborah Lenschow ( Washington University in St . Louis , MO , USA ) . All the animals were housed at the Washington University School of Medicine barrier facility . All experimental protocols were carried out in accordance with the ethical guidelines approved by the Washington University School of Medicine Institutional Animal Care and Use Committee . 6–7 weeks old mice were sacrificed and Intact long bones ( femur and tibia ) from different animals were isolated . The bones were fixed overnight in 10% neutral buffered formalin . After fixation , they were washed with Phosphate Buffer Saline ( PBS ) and transferred to 70% ethanol ( v/v ) . After fixation , bones were then scanned using Scanco Medical micro-CT systems ( Scanco , Wayne , PA , USA ) at the core facility at the Musculoskeletal Research Center at Washington University ( St . Louis , MO ) . Briefly , Images were scanned at a resolution of 20 μm , slice increment 20 μm , voltage 55 kV , current 145 μA and exposure time of 200 ms . After scanning , contours were drawn from the growth plate toward trabecular regions of femur . Approximately 150 slices were analyzed . Later contours were drawn and 3D images were constructed . X-ray analysis of whole body and isolated knee joints were performed using Faxitron Ultra Focus 100 on automatic settings and at 3X and 5X magnification , respectively . 6–7 weeks old mice were sacrificed and long bones ( femur and tibia ) from different animals were isolated . The bones were fixed overnight in 10% neutral buffered formalin . After fixation , bones were then decalcified for 2 weeks in decalcification buffer ( 14% ( w/v ) EDTA , NH4OH , pH 7 . 2 ) , dehydrated in graded ethanol ( 30–70% ) , cleared through xylene , and embedded in paraffin . Paraffin sections were stained for TRAP to visualize osteoclasts in the bone sections . For exogenous expression studies , various constructs ( cDNA ) were cloned in retroviral pMX- retroviral vector ( Cell biolabs , San Diego , CA ) . For different studies we generated pMX-GFP , pMX-flag-NEMO-WT , pMX-flag-NEMO-K270A-RFP , pMX-flag-NEMO-D304N , pMX-flag-NEMO-K319A , pMX-flag-NEMO-WT-GFP , pMX-HA-ISG15 , pMX-flag-NEMO-WT-ISG15-GFP , pMX-flag-NEMO-K270A-ISG15-GFP . To generate retroviral production pMX-vectors were first transfected into PLAT-E cells ( Cell biolabs , San Diego , CA ) using xtreme gene 9 ( Roche , San Francisco , CA , USA ) , followed by collection of virus containing media for next 2 days . This virus containing media with Polybrene ( 0 . 8 mg/ml ) was used to transduce primary bone marrow cells . Total bone marrow cells were isolated from the long bones ( femur and tibia ) and cultured in α-MEM media supplemented with 100 units/mL penicillin/streptomycin and 10% FBS ( v/v ) with 10 ng/mL M-CSF overnight to separate the adherent cells . One day after isolation , the non-adherent cells were collected and used as enriched bone marrow–derived macrophage ( BMMs ) . BMMs were further cultured with M-CSF ( 20 ng/mL ) and RANKL ( 50 ng/mL ) for 4 days followed by fixation and TRAP staining using TRAP-Leukocyte kit ( Sigma , St Louis , MO , USA ) . To investigate changes in autophagy , BMMs were cultured in M-CSF ( 20 ng/mL ) and RANKL ( 50 ng/mL ) for 2 days and used as pre-osteoclast ( preOC ) for different experiments . To investigate the effect of exogenous expression of different NEMO , NEMO mutants and ISG15 , the BMMs after one day of isolation , were transduced with retroviral particles ( generated using PLAT-E cells ) and osteoclast differentiation was initiated after 2 day of viral transduction . BMMs isolated form NF-ĸB-GFP-luciferase reporter mice were transduced with different pMX-retroviral particles . One day after transduction , the cells were cultured in the presence of M-CSF for two days , followed by RANKL treatment . Post RANKL transfection , cells were lysed and RelA-luciferase activity was measured using luciferase assay system ( GoldBio , St . Louis , MO ) . The luciferase activity was normalized with total protein concentration ( BCA assay , Pierce , Invitrogen ) . BMMs and/or pre-OC ( BMMs treated with RANKL for 2 days ) were lysed in cell lysis buffer ( Cell Signaling Technology , Danvers , MA , USA ) post treatments . Protein concentration was determined using BCA ( Pierce , Invitrogen ) and equal amounts of protein was loaded onto SDS-PAGE . After transfer , and blocking in 5% BSA for 1 hr at room temperature , membranes were probed with primary antibodies in 5% BSA in PBS-Tween ( 1% v/v ) for overnight and then washed with PBS-Tween ( 3x ) and probed with secondary antibodies from LI-COR ( Odyssey Imager; donkey anti-rabbit and anti-mouse ) for 1 hr at room temperature . Membranes were then with PBST ( 3x ) and scanned by using LI-COR Odyssey Imager ( LI-COR Biosciences , Lincoln , NE , USA ) . Western blots were also performed ( for LC3 and actin ) using capillary-based immunoassay using the Wes-Simple Western method with the anti-rabbit detection module ( Protein Simple ) . Protein expression was measured by chemiluminescence . The NEMO and ISG15 antibody were purchased from Santa Cruz , Dallas , TX , USA; phos-p65 , p65 and LC3 antibodies were purchased from Cell Signaling Technology , Danvers , MA , USA; Flag and β-actin was purchased from Sigma , St . Louis , MO , USA . Single cell suspensions from bone marrow were prepared by flushing the marrow out of femur and tibia of mice injected with BrdU ( 100 µl of 10 mg/mL solution of BrdU in sterile 1X DPBS ) 1 days before sacrifice . Following red blood cell lysis , whole bone marrow cells were stained by Zombie UV dye to distinguish live/dead cells . Then bone marrow cells were resuspended in PBS with 2% FBS ( FACS buffer ) , and further stained with biotin-conjugated lineage Ab cocktail ( anti-B220 , anti-CD3e , anti-Gr1 , anti-Ter119 ) . LSK+ ( Lin-Sca1+ckit- ) cells were stained with Ab cocktail ( anti-Sca1 PerCP Cy5 . 5 , anti-c-Kit APC eFluor 780 , anti-CD34 FITC , and CD16/32 eFluor450 ) . All FACS antibodies were purchased from either eBioscience , BioLegend ( San Diego , CA , USA ) or BD Bioscience ( San Diego , CA , USA ) . Following incubation on ice for 45 min , Ab-labeled cells were washed with FACS buffer and subjected to flow cytometric analysis ( BD X-20 ) . Data were analyzed with FlowJo software ( Tree Star Inc ) . To measure autophagy flux , flow cytometry analysis of LC3-GFP levels in NEMO WT and NEMO K270A preOC was performed , in response to autophagy induction by serum starvation . preOC were transduced with PMRX-GFP-LC3-RFP retrovirus generated in PLAT-E packing cells . Cells were serum starved for 6 hr and a flow cytometry analysis was done to detect GFP signal . Data were analyzed using FlowJo V10 . 1 software . Blood from NM-WT and NM-KA mice were collected from submandibular vein and serum was separated using BD-Microtainer tubes . The serum inflammatory cytokine levels were measured using multiplex mouse cytokine kits ( R and D Systems [Minneapolis , MN , USA] and Millipore [San Diego , CA , USA] ) . Serum cross‐linked telopeptide of type I collagen ( CTX‐I ) and TRAP levels were measured using the RatLaps ( CTX-1 ) EIA and Mouse TRAP ( TRAcP 5b ) kits ( Immunodiagnostic Systems , Boldon , UK ) using manufacturer's protocol . BMMs were cultured in presence of M-CSF ( 20 ng/mL ) and RANKL ( 50 ng/mL ) for 3 or 4 days as indicated in the figures . mRNA was isolated using PureLink RNA mini kit ( Ambion , Grand Island , NY , USA ) and cDNA were prepared using high capacity cDNA reverse transcription kit ( Applied Biosystems ) . Realtime PCR was carried out on BioRad CFX96 real time system using iTaq universal SYBR green super-mix ( BioRad , Hercules , CA , USA ) . mRNA expressions were normalized using β-actin as a housekeeping gene . The following primers were used for qPCR analysis . ( ACP5 ) TRAP-F: CGACCATTGTTAGCCACATACG , TRAP-R: CACATAGCCCACACCGTTCTC , CTSK ( CathepsinK ) -F: ATGTGGGTGTTCAAGTTTCTGC , CTSK-R: CCACAAGATTCTGGGGACTC , MMP9-F: ACTGGGCTTAGATCATTCCAGCGT , MMP9-R: ACACCCACATTTGACGTCCAGAGA , NFATC1-F: CCGGGACGCCCATGCAATCTGTTAGT , NFATC1-R: GCGGGTGCCCTGAGAAAGCTACTCTC . For immunolocalization at the ultrastructural level , preOC from NM-WT and NM-KA mice were fixed in 4% paraformaldehyde/0 . 05% glutaraldehyde ( Polysciences Inc , Warrington , PA ) in 100 mM PIPES/0 . 5 mM MgCl2 , pH 7 . 2 for 1 hr at 4°C . Samples were then embedded in 10% gelatin and infiltrated overnight with 2 . 3M sucrose/20% polyvinyl pyrrolidone in PIPES/MgCl2 at 4°C . Samples were trimmed , frozen in liquid nitrogen , and sectioned with a Leica Ultra cut UCT7 cryo-ultramicrotome ( Leica Microsystems Inc , Bannockburn , IL ) . Ultrathin sections of 50 nm were blocked with 5% FBS/5% NGS for 30 min and subsequently incubated with indicated primary antibodies for 1 hr at room temperature ( Note that I tried some of the labeling with primary antibody overnight at 4°C ) . Following washes in block buffer , sections were incubated by the appropriate colloidal gold conjugated secondary antibodies ( Jackson ImmunoResearch Laboratories , Inc , West Grove , PA ) for 1 hr . Sections were stained with 0 . 3% uranyl acetate/2% methyl cellulose and viewed on a JEOL 1200 EX transmission electron microscope ( JEOL USA Inc , Peabody , MA ) equipped with an AMT eight megapixel digital camera and AMT Image Capture Engine V602 software ( Advanced Microscopy Techniques , Woburn , MA ) . All labeling experiments were conducted in parallel with controls omitting the primary antibody . These controls were consistently negative at the concentration of colloidal gold conjugated secondary antibodies used in these studies . Post-treatments preOCs were fixed using 4% para-formaldehyde and 0 . 1% glutaraldehyde for 20 min at room temperature . Post-fixation the cells were washed using PBS ( 3x ) followed by blocking and permeabilization using 0 . 5% Goat serum and 0 . 1% saponin ( in PBS ) . Permeabilized cells were later incubated with Primary antibodies ( LC3 , NEMO , ISG15 and LAMP1 at 1:200 dilution ) and Alexa-Fluor secondary antibodies diluted ( 1:2000 ) in antibody incubation buffer ( 1% BSA in 0 . 1% Saponin in PBS ) . Fluorescent images were taken at 40X magnification . The images were analyzed using Image-J software . Statistical analyses were performed by using Student t test and Mann Whitney U test . Multiple treatments were analyzed by using one-way ANOVA . For Serum-cytokines analysis outliers were identified using ROUT method . Values are expressed as mean ± SD . P values are indicated where applicable . All the statistical analyses were done using GraphPad Prism software . Double-blind analysis was performed to analyze the IF and EM images . Number of experiment repeats , biological replicates and P values are indicted in figure legends .	The human skeleton contains over 200 bones that together act as an internal framework for the body . Over our lifetime , the body constantly removes older bone tissue from the skeleton and replaces it with new bone tissue . This “bone remodeling” also controls how bones are repaired after being damaged by injuries , disease or normal wear and tear . Cells known as osteoclasts are responsible for breaking down old bone tissue and participate in repairing damaged bone . A cellular pathway known as NF-kB signaling stimulates other cells called “bone marrow macrophages” to become osteoclasts . A certain level of NF-kB signaling is required to maintain a healthy skeleton . However , under certain inflammatory conditions , the level of NF-kB signaling becomes too high causing hyperactive osteoclasts to accumulate and inflict severe bone breakdown . This abnormal osteoclast activity leads to eroded and fragile bones and joints , as is the case in diseases such as rheumatoid arthritis and osteoporosis . Previous studies have shown that a protein called NEMO is a core component of the NF-kB signal pathway , but the precise role of NEMO in the diseased response remained unclear . Adapala , Swarnkar , Arra et al . have now used site-directed mutagenesis approach to study the role of NEMO in bone marrow macrophages in mice . The experiments showed that one specific site within the NEMO protein , referred to as lysine 270 , is crucial for its role in controlling osteoclasts and the breakdown of bone tissue . Mutating NEMO at lysine 270 led to uncontrolled NF-kB signaling in the bone marrow macrophages . Further experiments showed that lysine 270 served as a sensor to allow NEMO to bind another protein called ISG15 , which in turn helped to decrease NF-kB signaling and slow down the erosion of the bone . These findings suggest that site-specific targeting of NEMO , rather than inhibiting the whole NF-kB pathway , may help to reduce the symptoms of bone disease while maintaining the beneficial roles of this essential pathway . However , additional research is required to identify NEMO sites responsible for controlling the inflammatory component .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine" ]	2020	Inflammatory osteolysis is regulated by site-specific ISGylation of the scaffold protein NEMO
The prokaryotic tubulin homolog FtsZ polymerizes into protofilaments , which further assemble into higher-order structures at future division sites to form the Z-ring , a dynamic structure essential for bacterial cell division . The precise nature of interactions between FtsZ protofilaments that organize the Z-ring and their physiological significance remain enigmatic . In this study , we solved two crystallographic structures of a pair of FtsZ protofilaments , and demonstrated that they assemble in an antiparallel manner through the formation of two different inter-protofilament lateral interfaces . Our in vivo photocrosslinking studies confirmed that such lateral interactions occur in living cells , and disruption of the lateral interactions rendered cells unable to divide . The inherently weak lateral interactions enable FtsZ protofilaments to self-organize into a dynamic Z-ring . These results have fundamental implications for our understanding of bacterial cell division and for developing antibiotics that target this key process . Bacterial cytokinesis is initiated by the formation of a ring-like structure termed the Z-ring , a polymeric assembly of the essential tubulin homolog FtsZ at future division sites ( Bi and Lutkenhaus , 1991 ) . Once formed , the Z-ring serves as a scaffold to recruit other cell division proteins that collectively constitute the divisome ( Dajkovic and Lutkenhaus , 2006 ) . During cell division , the Z-ring constricts at the leading edge of the invaginating septum , eventually causing a mother cell to divide into two daughter cells ( Bi and Lutkenhaus , 1991 ) . FtsZ subunits have been suggested to interact through two putative sets of interfaces , longitudinal interfaces that join the subunits in a head-to-tail manner thereby forming a protofilament , and lateral interfaces that occur between protofilaments . FtsZ subunits readily assemble into protofilaments in vitro ( Mukherjee and Lutkenhaus , 1994; Romberg et al . , 2001 ) , and crystal structures of FtsZ protofilaments have been determined for both straight ( Matsui et al . , 2012; Tan et al . , 2012 ) and curved conformations ( Li et al . , 2013 ) . In vitro , FtsZ protofilaments have been observed to further associate via lateral interfaces to form higher-order structures such as sheets ( Bramhill and Thompson , 1994; Erickson et al . , 1996; González et al . , 2003; Löwe and Amos , 1999; Löwe and Amos , 2000; Oliva et al . , 2003; Yu and Margolin , 1997 ) . While several studies strongly suggested that lateral interfaces across protofilaments are important for FtsZ function ( Dajkovic et al . , 2008; Lan et al . , 2009; Milam et al . , 2012; Szwedziak et al . , 2014 ) , the precise nature and the functional relevance of these lateral interfaces remain largely unclear . Although bacterial cell division has been actively investigated for decades , the in vivo nanoscale organization of the Z-ring has not been well defined thus far . Conventional fluorescence microscopy depicts the Z-ring as a smooth , closed ring , with individual protofilaments not resolvable ( Pogliano et al . , 1997; Sun and Margolin , 1998 ) . Based on in vitro assembly studies ( Erickson et al . , 1996; Löwe and Amos , 1999; Löwe and Amos , 2000; Mukherjee and Lutkenhaus , 1994 ) , the Z-ring was initially modeled as a few single continuous polymers that wrap around the cell . Later , electron cryotomography suggested that the Z-ring is composed of individual FtsZ protofilaments that do not obviously interact laterally , scattered in a narrow band around the circumference of the cell ( Li et al . , 2007 ) . Super-resolution light microscopy indicated that FtsZ protofilaments form randomly oriented , multi-layered , discontinuous clusters within the Z-ring ( Biteen et al . , 2012; Buss et al . , 2013; Coltharp et al . , 2016; Fu et al . , 2010; Holden et al . , 2014; Jacq et al . , 2015; Rowlett and Margolin , 2014; Si et al . , 2013; Strauss et al . , 2012 ) . By contrast , recent electron cryotomography studies found a small , single-layered band of FtsZ protofilaments parallel to the membrane ( Szwedziak et al . , 2014 ) , and showed that a complete ring of FtsZ is not required to initiate constriction in the early stages of cytokinesis ( Yao et al . , 2017 ) . The link between this diverse set of conformations and Z-ring dynamics is challenging to parse without structural knowledge of the full suite of inter-subunit interactions . To address the nature and in vivo role of FtsZ lateral interactions , we solved the structure of Mycobacterium tuberculosis FtsZ ( MtbFtsZ ) in a double-stranded protofilament state . Comparison of this structure with that of MtbFtsZ in a different double-stranded protofilament state that we previously determined ( Li et al . , 2013 ) revealed two different inter-protofilament lateral interfaces . Using a combination of site-directed mutagenesis and phtotocrosslinking studies , we demonstrate that these lateral interfaces occur in living cells , and are critical for mediating cell division through the assembly of protofilaments into a functional Z-ring . FtsZ proteins from phylogenetically divergent species are known to assemble into polymers with multiple morphologies in a nucleotide-dependent manner ( Erickson et al . , 1996; Löwe and Amos , 1999; Löwe and Amos , 2000; Lu et al . , 1998; Oliva et al . , 2003; Popp et al . , 2010; White et al . , 2000 ) . Our electron microscopy analysis showed that MtbFtsZ and FtsZ from Escherichia coli ( EcFtsZ ) are able to form protofilament bundles in vitro in the presence of DEAE-dextran ( Figure 1A , B ) . The fact that protofilaments of both EcFtsZ and MtbFtsZ are able to form such assemblies , as observed previously ( Erickson et al . , 1996; Löwe and Amos , 1999 ) , suggests that the lateral interface of FtsZ protofilaments is a common and conserved characteristic . FtsZ subunits were previously observed to assemble into single- and double-stranded filaments at physiological concentrations ( Chen et al . , 2007; Oliva et al . , 2003; White et al . , 2000 ) . Our previous structural analysis of MtbFtsZ also revealed the formation of double-stranded and curved filaments , arranged in an antiparallel fashion ( Li et al . , 2013 ) . From the MtbFtsZ structure ( Li et al . , 2013 ) , we observed an inter-protofilament interface located on the external faces of strands S7 and S10 in the C-terminal subdomain ( lateral interface 1 , Figure 1C ) ( Li et al . , 2013 ) . However , the existence of only a single lateral interface within such an antiparallel arrangement of protofilaments would be self-limiting and lead only to the formation of double-stranded filaments . Formation of bundles composed of more than two FtsZ protofilaments requires additional lateral interfaces between the opposite sides of the protofilaments . We have now identified candidates for these interfaces in a new hexagonal crystal of MtbFtsZ , which has been determined to an Rfree factor of 27 . 3% at a resolution of 2 . 7 Å ( Materials and methods , Table 1 ) . Compared with our earlier MtbFtsZ structure ( Li et al . , 2013 ) , our newly determined MtbFtsZ structure is similarly double-stranded and reveals curved filaments in an antiparallel arrangement . However , in this new structure , the inter-protofilament interface is located at the external faces of helices H3 , H4 , and H5 in the N-terminal subdomain ( lateral interface 2 , Figure 1D ) . In the previously identified lateral interface ( Li et al . , 2013 ) , Arg229 of one subunit and Asp301 of the other formed two pairs of salt bridges , burying a surface area of approximately 210 Å2 ( Figure 1E ) . By contrast , the lateral interface in the new MtbFtsZ structure is composed of basic residues Arg76 , Lys77 , Lys83 , Arg119 , and Lys120 and acidic residues Glu80 , Glu87 , and Glu153 from both interacting subunits , burying a larger surface area of ~870 Å2 ( Figure 1D ) . These residues form charged complementary surfaces , suggesting the existence of electrostatic interactions . The charged residues involved in both lateral interfaces are generally conserved ( Figure 1—figure supplement 1 ) , indicating that they are functionally relevant . Such an electrostatic nature was predicted in earlier studies probing the effects of pH and ionic strength on FtsZ protofilament bundling ( Beuria et al . , 2006 ) . Interestingly , in the previous MtbFtsZ structure ( Li et al . , 2013 ) , only two of the three FtsZ subunits ( A and B ) in each protofilament participated in such lateral interactions , whereas the charged residues Arg229 and Asp301 in subunit C were ~6 Å apart ( Figure 1E ) . These in vitro observations suggest the presence of weak lateral interactions between MtbFtsZ protofilaments , in agreement with earlier electron microscopy studies as well as predictions based on kinetic modeling ( Lan et al . , 2008 ) . Guided by the similarities in amino acid sequence and tertiary structure between MtbFtsZ and Staphylococcus aureus FtsZ ( SaFtsZ ) , as well as the two lateral interfaces we have identified in MtbFtsZ filaments , we attempted to construct a model for sheet-like structures of FtsZ filaments . In light of the two MtbFtsZ structures , we initially constructed two different MtbFtsZ lateral dimer structures . Each subunit in these dimeric structures was subsequently superimposed on the SaFtsZ subunit in an SaFtsZ protofilament ( Matsui et al . , 2012 ) by aligning their main-chain atoms to generate a hybrid filament in which an MtbFtsZ protofilament pairs with an SaFtsZ protofilament . The MtbFtsZ structure in such a hybrid filament was then replaced with the SaFtsZ structure to generate an SaFtsZ filament . The final model contains four SaFtsZ protofilaments that associate laterally to form an antiparallel sheet-like structure ( Figure 1F ) . This structure is very similar to that observed for EcFtsZ ( Erickson et al . , 1996 ) and Methanococcus jannaschii FtsZ ( Löwe and Amos , 1999 ) , suggesting that the lateral interfaces observed by X-ray crystallography are identical to those observed by electron microscopy . To probe inter-protofilament contacts in living cells , we utilized an in vivo photocrosslinking approach in which we replaced each of the corresponding interfacial amino acid residues with p-benzoyl-L-phenylalanine ( pBpa ) , an unnatural photoactive amino acid that , upon UV irradiation , forms a biradical that can abstract an H atom from C-H bonds at a distance of ~3–4 Å to form a covalent adduct ( Chin et al . , 2002; Chin and Schultz , 2002; Fu et al . , 2013; Sato et al . , 2011; Zhang et al . , 2011 ) . Plasmids carrying mutated ftsZ genes were first transformed into an ftsZ conditional-null strain LY928-∆ftsZ , whose genome contains the gene encoding the orthogonal aminoacyl-tRNA synthetase and tRNA needed for the incorporation of pBpa ( Wang et al . , 2016 ) . Photocrosslinking analyses were then performed for the FtsZ-pBpa variants that were able to rescue cell growth ( Figure 2A and Figure 4—figure supplement 1 , Materials and methods ) . Upon irradiation with long-wavelength UV light , we found that FtsZ-pBpa variants R78pBpa , N79pBpa , D82pBpa , R85pBpa , R89pBpa , K155pBpa , and S231pBpa produced covalently linked homodimers , as demonstrated by immunoblotting analysis ( Figure 2B ) . The same set of pBpa variants of FtsZ were expressed in an E . coli strain that also expresses the AviTagged form of wild type FtsZ , and the putative photocrosslinked dimers were then probed with either an anti-FtsZ antibody ( which recognizes both the pBpa variant and the AviTagged wild-type FtsZ forms ) or with a streptavidin-alkaline phosphatase conjugate ( which only recognizes the AviTagged wild-type FtsZ ) . When probing with the anti-FtsZ antibody , doublet bands reflecting the migration positions of both the FtsZ monomer and dimer were detected ( Figure 2B ) . By contrast , when probing with the streptavidin conjugate , only single bands at both the monomer and the dimer positions ( corresponding to the higher molecular weight band in the anti-FtsZ immunoblot ) were detected . These photocrosslinking results clearly demonstrate that both lateral interfaces mediate interactions between FtsZ subunits in living cells . To obtain unbiased confirmation of the presence in living cells of the two crystallographically observed lateral interfaces , we further designed a random screening strategy ( Figure 3A ) ( Chin et al . , 2002; Daggett et al . , 2009; Liu and Schultz , 2010; Ryu and Schultz , 2006; Stricker and Erickson , 2003 ) . Instead of rationally introducing the unnatural amino acid pBpa via site-directed mutagenesis ( Figure 2A ) , we randomly introduced it into the EcFtsZ protein by generating a plasmid-borne library such that an in-frame TAG amber codon , which will be read as pBpa , was randomly inserted throughout the ftsZ gene ( Daggett et al . , 2009 ) . This library was then transformed into the LY928-ΔftsZ strain to screen for variants that complemented the ftsZ conditional-null phenotype . These variants were then subjected to in vivo photocrosslinking analysis to identify pBpa variants of FtsZ that can form crosslinked dimers . We obtained 31 colonies that yielded crosslinked FtsZ products . We then sequenced the ftsZ genes from these 31 colonies and identified FtsZ-pBpa variants resulting from insertion of the TAG amber codon at 10 distinct sites . Our immunoblotting analysis indicated that photocrosslinked FtsZ dimers were formed for four of these ten variants ( corresponding to pBpa incorporated at residue positions R78 , D82 , R85 , or K140; Figure 3B ) . Among these four positions , residue K140 is located at the protofilament longitudinal interface and the FtsZK140A mutant was earlier demonstrated to complement an ftsZ conditional-null strain ( Li et al . , 2013; Matsui et al . , 2012 ) , while R78 , D82 , and R85 are located at lateral interface two observed in our crystal structure . As a negative control , we sequenced the ftsZ genes isolated from 42 complementing colonies that did not generate any detectable photocrosslinked FtsZ products , from which we identified 12 distinct FtsZ-pBpa variants ( corresponding to pBpa incorporated at residue positions G22 , V37 , A41 , A48 , K61 , I64 , N73 , A113 , A114 , V119 , E147 , or L172 ) . As expected , none of these 12 residues is located at either the longitudinal or lateral interfaces . Taken together , our photo-crosslinking analyses based on unbiased , random introduction of pBpa confirm the presence of at least two interfaces that are involved in FtsZ assembly in living cells , both consistent with our in vitro crystallographic analyses . Our photo-crosslinking analyses were performed for pBpa variants that could complement wild-type FtsZ . We were surprised to find that three variants ( K121pBpa and D122pBpa from lateral interface 2 , and D304pBpa from lateral interface 1 ) failed to complement ( Figure 4—figure supplement 1 ) . To exclude potential artifacts introduced by pBpa , we replaced each of the corresponding interfacial residues with hydrophobic leucine and then characterized these mutant proteins using a similar complementation approach ( Figure 4 , Table 2 ) ( Stricker and Erickson , 2003 ) . As with pBpa replacement , K121L and D304L failed to complement ( Figure 4 , Table 2 ) . However , unlike D122pBpa , the D122L mutation was sufficient for complementation . This contrasting result with D122pBpa might be due to the bulkier size of the benzophenone-moiety side chain of pBpa compared to that of leucine . Replacement of an interfacial hydrophilic residue ( K or D ) with the hydrophobic leucine could disrupt inter-protofilament interaction , or could induce protein misfolding . However , purified FtsZK121L and FtsZD304L retained similar GTPase activity to that of wildtype ( data not shown ) , and assembled into protofilaments in a GTP-dependent manner , arguing against the possibility of protein misfolding . We performed photocrosslinking studies on the non-functional pBpa variants by expressing them in cells that also expressed the AviTagged wild-type FtsZ . Unlike functional pBpa variants ( Figure 2 ) , none of the three variants K121pBpa , D122pBpa , and D304pBpa produced any crosslinked dimer ( Figure 4—figure supplement 2 ) . These results indicate that the loss of FtsZ function in these variants is likely linked to a disruption of lateral interactions . Nevertheless , the dramatically distinct complementation results of the disruptive mutations of Ser231 and Asp304 , two residues likely involved in direct interactions at lateral interface 1 , raise an obvious concern as to whether Asp304 is important for other functions . To address this possibility , we generated two double mutants across the interface ( D304L/S231E and D304L/S231Q ) and observed complementation ( Figure 5 ) , demonstrating the formation of lateral interface one in vivo . Taken together , these data suggest that the two lateral interfaces we observed in vitro are important for FtsZ function in vivo , and lack of complementation is likely due to loss of lateral contacts . We initially postulated from the electrostatic complementarity along both lateral interfaces that short-range electrostatic interaction is the main driving force for lateral interactions . However , three lines of evidence led us to revisit this interaction mechanism . First , complementation results of presumably disruptive mutants on the lateral interface were less predictable than those of disruptive mutants on the longitudinal interface ( Li et al . , 2013 ) . Second , residues on the lateral interfaces are either polar or electrostatic , and are only generally conserved . For example , the Arg229-Asp301 pair observed in MtbFtsZ becomes Ser231-Asp304 in EcFtsZ . Third , the two complementing double mutants across the lateral interface ( D304L/S231E and D304L/S231Q ) indicate that S231E or S231Q forms favorable interactions that compensate for the disruptive effect of D304L . We further mutated Asp304 to different hydrophobic residues and observed highly variable results; for example , D304V was able to complement ( Figure 5 ) . Mutagenesis of Lys121 revealed similar variability , with K121M and K121V able to complement ( Figure 5 ) . These results , together with those from double mutagenesis ( Figure 5 ) , suggest that lateral interactions are predominantly mediated by van der Waals interactions , which are sensitive to surface geometry; the charge complementarity may enhance these associations . Moreover , these results also suggest that lateral interactions between FtsZ protofilaments are much weaker on a per subunit basis in comparison with hydrophobic longitudinal interactions . The free energy of protein-protein association is a balance between the intrinsic bond energy and the subunit entropy . The former favors association , while the latter disfavors association due to immobilizing a subunit ( Erickson , 1989 ) . This balance prompted us to examine whether protofilament formation is a prerequisite for lateral interactions to occur . To this end , we introduced A181E , a mutation known to disrupt the longitudinal interface ( Li et al . , 2013 ) , into a set of pBpa variants of FtsZ , including R78pBpa , N79pBpa , D82pBpa , R85pBpa , R89pBpa , K155pBpa , and S231pBpa , all of which produced covalently linked FtsZ dimers upon UV irradiation ( Figures 2B and 3B ) . We then performed in vivo photocrosslinking analysis with this set of A181E-containing pBpa variants , and found that photocrosslinked dimers were no longer detectable for all such variants ( Figure 6 ) . Thus , protofilament preassembly is required for lateral interactions to occur , consistent with the hypothesis that the lateral interactions are generally weak on a per-subunit basis . Nevertheless , the combined strength of all lateral interactions is presumably significant given that many interfaces are present along the protofilaments . FtsZ subunits readily assemble into protofilaments in vitro ( Mukherjee and Lutkenhaus , 1994; Romberg et al . , 2001 ) . Given that the intracellular concentration ( ~5 . 6 μM ) ( Li et al . , 2014 ) is much higher than the critical concentration ( ~1 μM ) ( González et al . , 2003 ) , it is reasonable to assume that most FtsZ molecules assemble into protofilaments in vivo . Since our complementation studies revealed the importance of both lateral interfaces for FtsZ function , we next investigated whether FtsZ mutant proteins defective in lateral interactions can integrate into the Z-ring in living E . coli cells that also express wild-type FtsZ . We first confirmed that these FtsZ mutant proteins ( D304L and K121L ) are still capable of forming GTP-dependent protofilaments ( Figure 7A ) . This capacity indicates that , when the laterally disruptive FtsZ is co-expressed with wild-type FtsZ , they can stochastically copolymerize to form hybrid protofilaments . We assume that the fraction of laterally disruptive subunits incorporated into protofilaments follows a Binomial distribution with a mean corresponding to the cellular proportion of laterally disruptive FtsZ ( Figure 7—figure supplement 1 ) , and that this fraction will determine the number of effective lateral bonds that could form between protofilaments . Given that the lateral interactions are weak , we expect that there exists a critical fraction of laterally disruptive subunits within a protofilament , above which the combined lateral interactions are insufficient to exceed the entropic cost of immobilizing the protofilament . In this case , we expect a dramatic reduction in the probability of such a protofilament interacting with other protofilaments to incorporate into the Z-ring . Thus , when co-expressed with wild type FtsZ , if the cellular proportion of laterally disruptive FtsZs is low , most protofilaments will tolerate the small degree of lateral disruption and incorporate into the Z-ring , whereas a high proportion of laterally disruptive FtsZ will interfere with Z-ring formation . Our complementation studies have already suggested that without wild-type FtsZ , laterally disruptive FtsZ mutants are lethal . For a pool of intermediate size , the protofilaments whose fraction of laterally disruptive subunits is above the threshold will be excluded from the Z-ring , leaving those hybrid protofilaments with small fraction of mutant subunits to form a functional Z-ring . As a consequence , when the cellular proportion of laterally disruptive FtsZ increases , the fraction of such subunits in the Z-ring decreases ( Figure 7—figure supplement 2 ) . This aforementioned rationale prompted us to co-express wild type FtsZ and fluorescent protein-fused mutant FtsZ , and use the midcell fluorescence signal as a proxy for Z-ring incorporation . We observed that the laterally disruptive mutants K121L and D304L and the laterally nondisruptive mutant R78L were all efficiently incorporated into the Z-ring ( Figure 7C ) , when the cellular proportions of mutant FtsZ proteins were ~40% ( Figure 7B ) . We then sought to increase the ratio of mutant FtsZ to wild-type FtsZ by using a stronger promoter to express mNeonGreen-tagged FtsZ variants . We introduced an amber codon between EcFtsZ and mNeonGreen for each variant to control the expression level of mNeonGreen . These plasmids , which expressed mNeonGreen-tagged mutant FtsZ and mutant FtsZ at a ratio of ~1:1 , were transformed into E . coli LY928-ftsz-avi cells ( Figure 7D , Materials and methods ) . The cellular proportions of mutant FtsZ ( mNeonGreen tagged and untagged ) increased to ~60% ( as shown in Figure 7D ) , while the total levels of mNeonGreen-tagged FtsZ and untagged FtsZ ( wild-type and mutant ) were expressed at similar levels as before ( Figure 7B ) . Fluorescence imaging demonstrated that midcell fluorescence was reduced to virtually undetectable levels for the laterally defective mutants , D304L and K121L ( Figure 7E ) . By contrast , Z-rings remained visible in cells expressing a high level of the laterally non-disruptive mutant R78L ( Figure 7E ) . Collectively , these results strongly suggest that lateral interactions are important for FtsZ protofilament assembly into the Z-ring . FtsZ and its eukaryotic counterpart tubulin share a similar overall structure and use similar longitudinal interfaces to form protofilaments . However , unlike tubulin , which exhibits strong lateral interactions to form multi-stranded microtubules , FtsZ polymerizes into single-stranded protofilaments ( Mukherjee and Lutkenhaus , 1994; Romberg et al . , 2001 ) , and undergoes GTP hydrolysis-driven treadmilling ( Bisson-Filho et al . , 2017; Chen and Erickson , 2005; Loose and Mitchison , 2014; Mukherjee and Lutkenhaus , 1994; Yang et al . , 2017 ) . These FtsZ protofilaments further coalesce and attach to the membrane at the division site through ZipA and FtsA ( Hale and de Boer , 1997; Pichoff and Lutkenhaus , 2002 ) , forming the Z-ring . The role of lateral interfaces between FtsZ protofilaments in Z-ring dynamics is fundamental to our understanding of Z-ring function in bacterial cytokinesis . FtsZ protofilaments associate laterally to form higher-order polymers in vitro ( Bramhill and Thompson , 1994; Erickson et al . , 1996; González et al . , 2003; Löwe and Amos , 1999; Löwe and Amos , 2000; Oliva et al . , 2003; Yu and Margolin , 1997 ) , and several studies have strongly suggested that lateral interfaces between protofilaments are important for FtsZ function ( Dajkovic et al . , 2008; Milam et al . , 2012 ) . In this study , we directly observed two different lateral interfaces of FtsZ protofilaments based on crystallographic analysis . We subsequently confirmed the presence of these lateral interfaces in living cells via in vivo photocrosslinking . Finally , we demonstrated that these weak , yet functionally important , lateral interfaces are involved in Z-ring assembly . Lateral interfaces between FtsZ protofilaments have been extensively probed for decades and several residues have been genetically implicated in lateral interactions ( Haeusser et al . , 2015; Jaiswal et al . , 2010; Koppelman et al . , 2004; Lu et al . , 2001; Márquez et al . , 2017; Moore et al . , 2017; Shin et al . , 2013; Stricker and Erickson , 2003 ) . However , some results from these studies were ambiguous and open to conflicting interpretations . Certain EcFtsZ mutants , such as E93R ( Jaiswal et al . , 2010 ) , L169R ( Haeusser et al . , 2015 ) , and D86K ( Lu et al . , 2001 ) , have elevated tendency to form protofilament bundles in vitro , but evidence of their function in vivo remains inconclusive ( Stricker and Erickson , 2003 ) . Other EcFtsZ surface residues , such as G124 and R174 , were identified as potential lateral residues since insertions of a fluorescent protein at these sites caused loss of function in vivo ( Koppelman et al . , 2004; Moore et al . , 2017 ) . However , insertion at R174 did not interfere with protofilament bundling in vitro , and insertion at G124 induced protein misfolding ( Moore et al . , 2017 ) , making it challenging to link loss of function to defects in lateral interactions . Our structures and random photocrosslinking results also failed to provide cross-verification for any of these residues . Our observation that lateral interactions are weak in nature may offer an explanation for the ambiguities from these studies . The association of FtsZ protofilaments due to additive effect of lateral interactions between two protofilaments means that single mutant studies can suffer from insensitivity in vivo and in vitro . For example , our in vivo complementation studies revealed that , among the ten rationally designed , potentially disruptive mutants , eight of them were able to complement . In addition , in vitro protofilament bundling relies heavily on bundling agents such as Ca2+ ( Löwe and Amos , 1999 ) , DEAE-dextran ( Erickson et al . , 1996 ) , or Ficoll 70 ( González et al . , 2003 ) . By contrast , our in vivo photocrosslinking study is able to unambiguously distinguish true lateral interactions from artifacts . Another alternative explanation for these ambiguous results is other cellular factors that regulate interactions between FtsZ protofilaments in vivo . ZipA and various Zap proteins ( ZapA , ZapC and ZapD ) have been reported to crosslink , or to promote the lateral interactions between FtsZ protofilaments ( Durand-Heredia et al . , 2011; Gueiros-Filho and Losick , 2002; Haeusser et al . , 2015; Hale et al . , 2000; 2011; Huang et al . , 2013 ) . Such enhancement of lateral association might enable ZipA/Zap to compensate for some intrinsic defects in lateral interactions between FtsZ protofilaments , thereby resulting in normal function for some of the mutants predicted to be disruptive . The residues involved in lateral association enhancement may be distinct from our findings . For example , the EcFtsZ L169R mutant can fully rescue the cell division defect of ΔzapAC cells ( Haeusser et al . , 2015 ) , but L169 is far apart from the lateral interfaces we identified . The inherently weak lateral interactions are unlikely to mediate formation of higher-order protofilament structures such as those observed in vitro . A recent electron cryotomography study showed a mean interprotofilament spacing of 6 . 8 nm , slightly too far apart to support tight interactions between FtsZ protofilaments ( Szwedziak et al . , 2014 ) . Moreover , membrane-targeting FtsZ ( either FtsZ and FtsA , or FtsZ alone with the addition of a membrane-targeting sequence ) is sufficient to reconstitute contractile Z-rings in liposomes ( Osawa et al . , 2008; Osawa and Erickson , 2013; Szwedziak et al . , 2014 ) , and to display treadmilling behavior and the reorganization of FtsZ protofilaments into dynamic vortices on supported membranes ( Loose and Mitchison , 2014; Ramirez et al . , 2016 ) . The antiparallel protofilament arrangement would preclude treadmilling of FtsZ . It is noteworthy that the cooperative assembly of single-stranded FtsZ protofilaments implies that the mechanism of FtsZ treadmilling is distinct from that of actin , for which lateral interactions are not directly involved . We propose that transient lateral interactions induce changes in treadmilling velocities of single protofilaments when they collide , rather than mediate the formation of a stable and static higher order architecture . Although the precise details of Z-ring dynamics remain to be determined , this study is a vital step toward understanding the architecture and assembly mechanism of the bacterial cell division machinery in living cells , and provides novel structural information to guide the development of novel antimicrobial compounds that specifically target the division machinery . The full-length ftsZ gene was amplified from M . tuberculosis genomic DNA and was subcloned into the pET15b plasmid vector . MtbFtsZ protein was overexpressed in BL21 ( DE3 ) /pLysS E . coli cells , cultured at 37°C in lysogeny broth ( LB ) medium and induced with 1 mM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) after OD600 reached 0 . 5 . The His-tagged MtbFtsZ protein was then purified with Cobalt affinity resin . After removal of the His tag by thrombin cleavage , the protein was subjected to size-exclusion chromatography performed with a Superdex 200 10/300 GL column ( GE Health Sciences ) that was pre-equilibrated with a buffer of 100 mM KCl , 0 . 1 mM EDTA , 20 mM Tris , pH 8 . 0 , and 10% glycerol . The protein was concentrated to 20 mg/mL ( as measured by ultraviolet absorbance ) , with 10 mM GTP added 30 min before crystallization . Well-diffracting crystals were grown by the sitting-drop vapor-diffusion method , in which 2 μL of the above MtbFtsZ-GTP solution were mixed with an equal volume of crystallization solution containing 1 M sodium citrate and 0 . 1 M imidizole , pH 8 . 0 . Crystals were cryo-protected from their mother liquid by adding 30% glycerol , and were frozen in liquid nitrogen . Diffraction data were collected at the Shanghai Synchrotron Radiation Facility BL19U beamline ( Shanghai , China ) . The data were indexed , integrated , and scaled using HKL-2000 ( Otwinowski and Minor , 1997 ) . Crystals are in space group P6522 and contain three GDP-MtbFtsZ subunits per asymmetric unit . The best crystal diffracted X-rays to 2 . 7 Å resolution , with unit-cell dimensions of a = 100 . 5 Å , c = 138 . 3 Å . Phases were determined by molecular replacement using PHASER ( McCoy et al . , 2005 ) with the MtbFtsZ monomer ( molecule A , PDB ID 1RQ7 ) ( Leung et al . , 2004 ) as a search model . Model adjustment was performed iteratively using Xtalview ( McRee , 1999 ) , and structure refinement was performed using REFMAC ( Collaborative Computational Project , Number 4 , 1994 ) . The models were refined with data to 2 . 7 Å resolution , maintaining a highly restrained stereochemistry . The final model contains an FtsZ molecule and a GTP molecule . All structural illustrations were prepared with PYMOL ( www . pymol . org ) . The complementation assay used here is based on the JKD7-1/pKD3 conditional null strain ( Dai and Lutkenhaus , 1991 ) and the pJSB100 complementation vector ( Stricker and Erickson , 2003 ) . JKD7-1 is an ftsZ-null strain that is maintained in the presence of the rescue plasmid pKD3 that contains a functional ftsZ allele . The pKD3 plasmid is temperature sensitive for its replication , such that it is lost in a majority of the transformed E . coli cells when cultured at 42°C . The pJSB100 plasmid , derived from the pBAD vector , was used to express the wild-type or mutant EcFtsZ protein at a moderate level upon induction by arabinose . When strains containing both the pKD3 and the pJSB100 plasmids are grown at 42°C in the presence of arabinose , pKD3 fails to replicate and thus the survival of the cells relies on the expression of a functional EcFtsZ variant from pJSB100 . The complementation assay was performed as follows . JKD7-1/pKD3 cells were transformed with pJSB2 ( carrying no ftsZ gene , as a negative control ) , pJSB100 ( carrying the wild type EcftsZ gene , as a positive control ) , or a particular pJSB100-EcftsZ variant . The transformed cells were cultured in the repression medium ( LB containing 34 μg/mL chloramphenicol , 100 μg/mL ampicillin , and 0 . 2% glucose ) overnight at 30°C , reaching an OD600 of 1 . 0–2 . 0 . Ten microliters of 10 , 000-fold dilutions of the overnight cultures were then plated either on induction plates containing 0 . 05% arabinose or on repression plates containing 0 . 2% glucose . Repression plates were then cultured at 30°C , and the induction plates were cultured at 42°C , to determine the number of colony forming units ( CFUs ) . CFU values on the induction plates were normalized to the CFU values from the repression plates . A mutant was considered to complement the ftsZ conditional-null strain when an induction plate produced at least 80% as many colonies as the repression plate . For each variant , the complementation assay was repeated three times . Liquid complementation assays were performed by culturing cells transformed with pJSB100-derived plasmids that express mutant EcFtsZ proteins at 30°C to an OD600 of 0 . 5 in repression medium . These cultures were then diluted 5 , 000 , 000-fold and cultured at 42°C for 24 hr in induction medium containing 0 . 05% arabinose . The successful complementation of ftsZ conditional-null cells by an EcFtsZ variant was defined by the ability for transformed cells to grow to an OD600 >0 . 5; failure of complementation was defined as lack of growth ( i . e . , no measurable turbidity after overnight growth ) . The unnatural amino acid ( pBpa ) incorporation system is based on a plasmid expressing orthogonal pBpa-tRNA synthetase/tRNApBpa pairs in E . coli ( Ryu and Schultz , 2006 ) . In generating a complementation system to screen for functional pBpa variants of FtsZ , we constructed the LY928-ΔftsZ ( pJSB100 ) conditional null strain , in which the optimized genes encoding the pBpa-tRNA synthetase and tRNApBpa ( Guo et al . , 2009 ) are integrated into the chromosome and a functional FtsZ protein is expressed from pJSB100 upon arabinose induction ( Stricker and Erickson , 2003 ) . For a pBpa variant of FtsZ that successfully rescued the growth of LY928-ΔftsZ ( pJSB100 ) in repression medium ( LB containing 50 μg/ml ampicillin and 0 . 2% glucose ) , the encoding plasmid was transformed into the LY928-ftsZ-avitag strain ( whose endogenous ftsZ gene was modified to encode FtsZ linked with an AviTag at the C-terminus ) . The transformed cells were then grown at 37°C to mid-log phase in repression medium supplemented with 100 μM pBpa . One milliliter was then transferred to a 1 . 5 mL Eppendorf tube , irradiated at room temperature with UV light ( 365 nm ) for 10 min using a Hoefer UVC 500 Crosslinker installed with 365 nm UV lamps ( Amersham Biosciences ) at a distance of 3 cm . Cells were subsequently harvested by centrifugation at 13 , 000 × g for 5 min , added into the loading buffer , and boiled . The cell lysate was then analyzed by tricine SDS-PAGE , and probed either by immunoblotting with FtsZ antibody or with streptavidin-alkaline phosphatase conjugate . Gel bands were scanned and processed using GIMP . A library of expression plasmids in which the amber codon was randomly substituted for any triplet nucleotide in the ftsZ gene was constructed using E . coli Top10 cells , based on a method modified from an earlier study ( Daggett et al . , 2009 ) . The plasmid library was used to transform LY928-ΔftsZ ( pJSB100 ) to select for variants that complement the ftsZ null phenotype . These complementing variants were subjected to in vivo photo-crosslinking analysis , and were sequenced to identify the site of the TAG codon replacement . The resulting library contains in-frame TAG mutations only in the N-terminal domain of FtsZ . Since the plasmid is leaky , and the in-frame TAG amber codon is read as a stop codon in E . coli Top10 cells , the generation of such a library would result in expression of truncated FtsZ proteins in cells . A likely explanation is that in-frame TAG mutation in the C-terminal domain would result in a truncated FtsZ with only the N-terminal domain , which is dominant negative . Plasmids constitutively expressing mutant FtsZ fused to mNeonGreen ( Shaner et al . , 2013 ) , with or without an amber codon inserted in between , were transformed into LY928 cells ( in which optimized genes encoding the pBpa-tRNA synthetase and tRNApBpa ( Guo et al . , 2009 ) were integrated into the chromosome ) , or LY928-ftsZ-avitag cells ( whose endogenous ftsZ gene was modified to encode FtsZ linked with an AviTag by a GSG linker at the C-terminus ) . The transformed cells were cultured at 37°C in LB ( containing 50 μg/mL ampicillin and 100 μM pBpa ) to mid-log phase . Cells were then loaded onto a glass dish ( NEST Biotechnology ) and covered with a cover glass . Images were acquired on an N-SIM imaging system ( Nikon ) at 30°C with a 100X/NA1 . 49 oil-immersion objective ( Nikon ) and 488 nm laser beam . The reconstructed images were further processed with NIS-Elements AR 4 . 20 . 00 ( Nikon ) and GIMP . For experiments in Figure 7C , we used plasmids with synthetic constitutive promoter PL3 ( selected from the Anderson promoter collection: parts . igem . org/Promoters/Catalog/Anderson ) to express FtsZ-mNeonGreen fusion protein . For experiments in Figure 7E , we used plasmids with the synthetic constitutive promoter P0 . 16 to express TAG inserted FtsZ-TAG-mNeonGreen . We used a counter-selective recombining technique based on lambda-Red recombination system to tag the ftsz gene in LY928 cells ( Lee et al . , 2009 ) . The expression levels of FtsZ were determined by Western-blot , and the proportions of mutant FtsZ were measured by analyzing the images with ImageJ gel analysis tool ( https://imagej . en . softonic . com/ ) . For Figure 1A and B , MtbFtsZ or EcFtsZ proteins ( 1 mg/mL ) were first incubated in MEMK6 . 5 buffer ( 100 mM morpholine ethane sulfonic acid , pH 6 . 5 adjusted with KOH , 1 mM EGTA , 5 mM Mg acetate ) with the addition of 0 . 6 mg/mL DEAE-Dextran , and in the presence of 2 mM GTP . For Figure 7A , wild-type and mutant EcFtsZ proteins ( 1 mg/mL ) were first incubated in MEMK6 . 5 buffer in the presence of 2 mM GTP . The reaction mixture was then incubated on ice for 5–10 min , then at 37°C for 5–10 min , before a 5 μL aliquot was placed on a glow-discharged carbon-coated copper grid and negatively stained with 2% aqueous uranyl acetate . The air-dried grids were subsequently examined with a HITACHI HT7700 transmission electron microscope operated at 80 kV , or with a FEI Tecnai-F20 transmission electron microscope operated at 200 kV . Images of FtsZ protein assemblies were acquired on a Gatan ORIUS CCD camera at a nominal magnification of 40 , 000X , or with a Gatan Ultra4000 CCD camera at a nominal magnification of 50 , 000X . We performed simulations based on the model of Z-ring formation described below , and calculated the percentage of laterally disruptive FtsZ subunits incorporated into the Z-ring . We assumed that in order to incorporate into the Z-ring through lateral bonds , a protofilament loses translational and rotational degrees of freedom and hence there is an entropic cost for immobilizing a protofilament . Since only wild-type FtsZ subunits contribute to lateral attachment , the balance between the energy of binding and the entropic cost results in an upper limit to the fraction of laterally disruptive subunits that a protofilament can tolerate and still incorporate into Z-ring . To simplify , we consider the dynamics of 200 protofilaments in the simulations , and all protofilaments are set to be 50 subunits long . We then use a variable threshold T , which is related to the critical fraction ( fc ) by: fc = T/50 . In each simulation , we set the overall proportion f of laterally disruptive subunits in a cell and generated a vector X = ( x1 , x2 , . . . , x200 ) , where xi represents the number of laterally disruptive subunits in the ith protofilament , and was selected based on a Binomial distribution with probability f ( Figure 7—figure supplement 1 ) . For a protofilament with more or less laterally disruptive subunits than the threshold T , we set the probability of Z-ring incorporation to 0 . 01 or 0 . 99 , respectively . We used a Boolean vector V = ( 0 , 1 , … , 1 ) to represent the states of protofilaments , where 1 or 0 indicate incorporation or not into the Z-ring , respectively . We then calculated the percentage of laterally disruptive subunits incorporated into the Z-ring as:X⋅V∑i=1200xi For each value of f and T , we performed 10 , 000 independent simulations .	New cells form when existing cells divide . When a cell divides it narrows at one point , which eventually allows it to split in two . This basic process of division happens in cells from all species , although they do not all use the same mechanisms to achieve it . In bacteria , a structure called the Z-ring guides where the cell narrows and divides . Although the importance of the Z-ring in bacterial cell division is clear , how it works was not known . A first step to understanding how the Z-ring works is to find out how it is made . The Z-ring consists of long ‘protofilaments’ made up of many copies of a protein called FtsZ . To find out how the protofilaments interact with each other to form the Z-rings , Guan , Yu , Yu , Liu , Li et al . studied the interactions between the FtsZ proteins in living cells . This revealed two key points of contact that allow two protofilaments to link together while aligned in opposite directions . Further experiments in living cells showed that disrupting either contact point prevents the cells from growing correctly and can cause cells to die . Guan et al . also show that these contacts are weak , so two protofilaments can only link together when many of their FtsZ proteins interact . Future research into how the Z-ring works can build upon these details of how the protofilaments interact . Because animal cells do not contain Z-rings , this could ultimately help researchers to design new antibiotics that can kill bacteria without affecting other cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2018	Lateral interactions between protofilaments of the bacterial tubulin homolog FtsZ are essential for cell division
RNA interference defends against viral infection in plant and animal cells . The nematode Caenorhabditis elegans and its natural pathogen , the positive-strand RNA virus Orsay , have recently emerged as a new animal model of host-virus interaction . Using a genome-wide association study in C . elegans wild populations and quantitative trait locus mapping , we identify a 159 base-pair deletion in the conserved drh-1 gene ( encoding a RIG-I-like helicase ) as a major determinant of viral sensitivity . We show that DRH-1 is required for the initiation of an antiviral RNAi pathway and the generation of virus-derived siRNAs ( viRNAs ) . In mammals , RIG-I-domain containing proteins trigger an interferon-based innate immunity pathway in response to RNA virus infection . Our work in C . elegans demonstrates that the RIG-I domain has an ancient role in viral recognition . We propose that RIG-I acts as modular viral recognition factor that couples viral recognition to different effector pathways including RNAi and interferon responses . The arms races between pathogens and their hosts have led to the evolution of sophisticated mechanisms to provide immunity against infection . Whilst adaptive immunity is specific to vertebrates , innate mechanisms are present in all multicellular organisms , allowing cells to recognize specific pathogens and instigate appropriate responses . RNA viruses are important pathogens of many multicellular organisms , which replicate without a DNA intermediate using RNA dependent RNA polymerase . Successful neutralization of invading RNA viruses by cells thus requires the viral genome to be recognized within the sea of endogenous RNA . The primary innate immune sensors for RNA viruses in mammals are RIG-I and its homolog MDA-5 ( Schlee , 2013 ) . Viral recognition by RIG-I and MDA-5 triggers activation of downstream signaling , mediated by the proteins’ N-terminal CARD domains , and results in the activation of the interferon pathway ( Yoneyama et al . , 2004 ) . Initial recognition of viral RNA is likely to be mediated by the DExD/H-box helicase domain and the C-terminal RIG-I domain . Though the precise ligands that activate RIG-I family proteins are not fully defined , MDA-5 appears to bind long dsRNA ( Gitlin et al . , 2006 ) , whilst RIG-I seems to recognize the 5′ end of double stranded RNA , but only if it has a 5′ triphosphate ( Pichlmair et al . , 2006 ) . As all known RNA polymerases leave a triphosphate at the 5′ end of newly synthesized RNA , the presence of a 5′ triphosphate is likely to be a signature of RNA virus replication and thus allow viral replication intermediates to be distinguished from endogenous mRNA , which will predominantly display a 5′ cap ( Rehwinkel and Reis e Sousa , 2010 ) . In plants and insects , interferon signaling is not involved in antiviral defense . Instead , the RNA interference ( RNAi ) pathway provides robust defense against RNA viral infection ( Ding and Voinnet , 2007; Ding , 2010 ) . The initial step in protection against positive strand RNA virus infection in plants and insects is detection and subsequent cleavage of the double-stranded replication intermediate by members of the Dicer family of endonucleases . Insects and plants possess dedicated Dicer enzymes responsible specifically for this antiviral response , Dicer-like 4 ( and to a lesser extent Dicer-like 1 ) in plants , and Dicer2 in insects ( Bouché et al . , 2006; Deleris et al . , 2006; Fusaro et al . , 2006; Galiana-Arnoux et al . , 2006; van Rij et al . , 2006; Diaz-Pendon et al . , 2007 ) . The small RNAs thus generated feed into the canonical RNAi machinery and can be used to silence the viral genome . In the nematode C . elegans , the RNAi pathway is best characterized as a response to artificial introduction of dsRNA by feeding or injection . However , exposing C . elegans to virus-derived dsRNA using transgenes or infection with mammalian viruses also triggers an RNAi response ( Lu et al . , 2005; Schott et al . , 2005; Wilkins et al . , 2005; Yigit et al . , 2006 ) . Additionally , we have shown previously that this response is also initiated upon infection with a positive strand RNA virus , named the Orsay virus , which infects C . elegans in the wild through horizontal transmission . Disruption of this pathway through mutation of the core components of the RNAi machinery results in greatly increased viral infection levels . Interestingly , the Orsay virus infects a wild isolate named JU1580 ( from which the Orsay virus was isolated ) to much higher levels than the N2 strain ( Félix et al . , 2011 ) . Despite these advances , our understanding of how RNAi-mediated antiviral defense in C . elegans is orchestrated is still limited . One major unsolved problem is how viral dsRNA within the cell is recognized in order to initiate the antiviral response . In contrast to the situation in both plants and insects , C . elegans only has one Dicer enzyme ( Knight and Bass , 2001; Yigit et al . , 2006 ) ; it is expected therefore that the activity of Dicer in different small RNA pathways will be controlled by different partner proteins within distinct complexes ( Yigit et al . , 2006; Thivierge et al . , 2012 ) . In this regard it is intriguing that C . elegans encodes three homologues of the mammalian viral recognition protein RIG-I: DRH-1 , DRH-2 and DRH-3 . The three RIG-I family members do not contain the CARD domains required for interferon induction; consistently they have yet to be implicated in signaling pathways . However , they do possess the RIG-I C-terminal domain and the helicase domains , potentially enabling them to recognize viral RNA . So far study of these genes has connected them to the RNA interference response . DRH-1 was initially characterized as a protein interacting with DCR-1 , RDE-1 and RDE-4 ( Tabara et al . , 2002; Duchaine et al . , 2006; Sijen et al . , 2007 ) . RNAi of drh-1 was not found to be required for RNAi per se but result in a defect in RNAi of a second gene ( Tabara et al . , 2002 ) , however it remains unclear whether this effect was solely due to knockdown of drh-1 . More recent studies did not observe defects in endogenous or exogenous small RNA pathways in drh-1 mutants ( Gu et al . , 2009; Lu et al . , 2009 ) . However , drh-1 mutants were found to be defective in the silencing of a flockhouse virus-derived replicon ( Lu et al . , 2009 ) . drh-1 and drh-2 are the result of a recent gene duplication event that occurred after the last common ancestor of C . elegans and its closest known sister clade including C . briggsae ( www . wormbase . org ) ( Stein et al . , 2003 ) , and drh-2 is the upstream gene in an operon containing drh-1 . drh-2 has lost some of its functional domains due to frame-shift mutations and its function remains unclear , although it has been suggested to act as a negative regulator of RNAi ( Lu et al . , 2009 ) . DRH-3 is required for endogenous small RNA pathways in the germline and efficient exogenous RNAi ( Gu et al . , 2009 ) , and is found in at least two distinct protein complexes including the ERI Complex ( ERIC ) ( Gu et al . , 2009; Thivierge et al . , 2012 ) . Thus , an intriguing possibility is that RIG-I family genes in C . elegans initiate RNAi rather than the interferon response . Here we discover a naturally occurring deletion in the gene drh-1 that is widespread in C . elegans populations despite predisposing individuals to viral sensitivity . We show that the increased viral sensitivity caused by this deletion results from a failure of the RNAi pathway . Mutations in drh-1 almost completely abolish the production of primary siRNAs , allowing us to place DRH-1 at the top of a hierarchical RNAi response . Our data supports a model whereby the conserved RNA virus recognition capability of drh-1 allows it to recruit the RNAi machinery to defend C . elegans from viral infection . To explore intraspecific variation in viral resistance in C . elegans , we assayed a worldwide set of 97 wild C . elegans isolates that had previously been genotyped ( Andersen et al . , 2012 ) . To assess viral sensitivity , we infected each isolate in triplicate and quantified the viral load after 7 days by qRT-PCR . Viral loads of the 97 isolates varied widely over five orders of magnitude ( Figure 1—figure supplement 1A ) . Genome-wide association for viral load revealed a single peak covering a 6 Mb region in the middle of chromosome IV ( Figure 1A ) . However , further mapping resolution is limited by the low natural recombination frequency in the species ( Cutter et al . , 2009; Andersen et al . , 2012 ) . 10 . 7554/eLife . 00994 . 003Figure 1 . A deletion polymorphism in drh-1 is a major determinant of Orsay virus sensitivity in wild isolates of C . elegans . ( A ) Genome-wide association analysis of Orsay virus sensitivity in 97 wild isolates of C . elegans . The mapped trait is the viral load of animals , measured by qRT-PCR on the Orsay virus RNA2 genome after 7 days of infection , using three independent infection experiments . The horizontal grey line is a Bonferroni-corrected threshold of significance at p=0 . 05 . Peaks reaching above this threshold are colored in red . ( B ) Fine mapping of the candidate region causative for virus hypersensitivity observed in JU1580 animals . The genotypes of chromosome IV and other chromosomes are represented for parental ( N2 and JU1580 ) and informative recombinant ( JU2196 and JU2197 ) strains . Regions in red or blue are identical to N2 or JU1580 , respectively . The inferred candidate region is delimited by dotted lines . Below each genotype are viral load measured by qRT-PCR of Orsay virus RNA2 , in two independent infections ( black and grey bars ) and normalized to JU1580 . ( C ) Diagram of the drh-1 locus . Positions of deletion alleles and a rescuing fosmid are indicated . ( D ) PCR analysis of niDf250 deletion in N2 and JU1580 strains . ( E ) drh-1 mRNA level in different strains ( as indicated ) , measured by RT-qPCR . ( F ) Diagram of C . elegans and human RIG-I like genes . DeXD = Pfam:DEAD , Hel = Pfam:Helicase_C , RIG-I = Pfam:RIG-I_C-RD , CRD = Pfam:CARD . ( G ) Viral load in different strains ( as indicated ) , measured by RT-qPCR of the Orsay virus RNA1 genome after 4 days of infection . JU1580 ( drh-1 rescue ) refers to JU1580 strains carrying three independent transgenic lines ( SX2375 , SX2376 , SX2377 ) . Transgenes include the fosmid WRM0640dC01 and a co-injection marker , they were integrated into the genome using X-rays . Error bars represent the standard error of the mean ( SEM ) of five biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 00310 . 7554/eLife . 00994 . 004Figure 1—figure supplement 1 . Variation in the ability of the Orsay virus to replicate in C . elegans . ( A ) For each of the 97 wild isolates listed on the horizontal axis , the mean of the logarithm and standard error of the RT-qPCR values on the viral RNA1 assayed 7 days post infection at 23°C are reported . For a given strain , the value of each replicate is represented as a dot . Isolates labeled in blue or red carry the niDf250 or the N2 allele of drh-1 , respectively . ( B ) To investigate whether the viral susceptibility of JU1580 was linked to drh-1 , we crossed N2 and JU1580 and allowed the F1 progeny to self-fertilize . Each line thus generated will carry a different combination of N2 and JU1580 SNPs allowing the separation of the drh-1 mutation from any unlinked additional differences in genetic background . After two generations of self-fertilization we then infected these lines and assayed both for sensitivity to infection and for the presence of the JU1580 drh-1 deletion . We saw good correlation between increased viral sensitivity and the drh-1 deletion ( p=2 . 9×10−8 , Wilcoxon test ) . ( C ) Distribution of JU1580 SNPs in recombinant lines from B . A pool of 20 sensitive recombinant lines was selected from a total of 110 independent F2 lines and subjected to high-throughput sequencing . Only average SNP frequencies between 20% and 70% were chosen to exclude false SNP calls . This showed that viral sensitivity was linked to chromosome IV . Red lines indicate an average SNP frequency of 50% . The blue line indicates position of the niDf250 deletion . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 00410 . 7554/eLife . 00994 . 005Figure 1—figure supplement 2 . Genotype and sensitivity to the Orsay virus of recombinants in the chromosome IV region . ( A ) Genotype at 17 loci ( 16 SNPs and the drh-1 allele ) along chromosome IV of the six introgressed and recombinant lines and their parents ( N2 and JU1580 ) . Chromosomal segments from N2 or JU1580 are represented in red or blue , respectively . The chromosomal positions of loci in the new candidate region are in bold italic . The viral load of each strain over that in JU1580 is represented below as in Figure 1B , for two replicate infections . ( B ) Summary of genetic differences between N2 and JU1580 in a 155 kb region on chromosome IV ( 6 , 567 , 528-6 , 676 , 736 ) based on resequencing of JU1580 . Only differences predicted to result in altered function were considered . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 005 We therefore focused on genetic variation in viral sensitivity between the N2 and JU1580 isolates . We first assayed viral sensitivity in 110 F2 recombinant families ( Figure 1—figure supplement 1B ) . Whole-genome sequencing of a pool of the 20 most sensitive families revealed linkage to chromosome IV ( Figure 1—figure supplement 1C ) . To narrow down the candidate region , we introgressed the center of chromosome IV ( IV:3 , 329 , 219 to IV:11 , 083 , 410 ) of JU1580 into N2 animals ( yielding strain JU2170 ) . As expected , JU2170 showed similar viral load to JU1580 ( Figure 1—figure supplement 2 ) . We then screened for recombinants in this region after crossing JU2170 with N2 . This allowed us to restrict the candidate region to a 155 kb interval carried by a virus-sensitive recombinant strain named JU2196 ( Figure 1B , Figure 1—figure supplement 2 ) . Genome sequencing of the JU1580 isolate and alignment to the N2 reference strain revealed 20 single nucleotide polymorphisms ( SNPs ) , including one non-synonymous SNP , and one 159 base indel in the 155 kb region ( Figure 1—figure supplement 2 ) . The 159 base deletion in the JU1580 genome , named niDf250 ( IV:6 , 607 , 635-6 , 607 , 793 ) , lies within the drh-1 gene locus ( Figure 1C ) . drh-1 is a homolog of the mammalian RIG-I family genes , whose products bind virus-derived RNAs , acting as pattern recognition receptors ( Parameswaran et al . , 2010; Yoneyama and Fujita , 2010 ) , and trigger antiviral innate immune responses in mammals ( Yoneyama et al . , 2004 ) . C . elegans expresses three RIG-I-like proteins: DRH-1 , DRH-2 and DRH-3 . The niDf250 deletion covers most of drh-1 exon 19 and part of exon 20 ( Figure 1C ) and was confirmed using genomic PCR ( Figure 1D ) . drh-1 mRNA levels are unchanged in JU1580 ( Figure 1E ) and the resulting transcript is predicted to encode a truncated protein of 987 amino acids , identical to the N2 form of DRH-1 up to amino acid 973 but with a novel C-terminus ( Figure 1F ) . Importantly , this truncates the RIG-I C-terminal domain ( amino acids 885–1014 ) , thought to be required for RNA recognition specificity ( Kowalinski et al . , 2011; Figure 1F ) . To establish a potential role for DRH-1 in Orsay antiviral resistance we tested whether a drh-1 deletion in the N2 background was sufficient to impart viral sensitivity , as assayed by viral load . Indeed , the drh-1 ( ok3495 ) mutant displayed an increased viral load compared to N2 animals , similar to that of JU1580 ( Figure 1G ) . Conversely , transgenic JU1580 animals carrying a fosmid containing the N2 allele of drh-1 were resistant to Orsay infection ( Figure 1G ) . Therefore , variation at the drh-1 locus explains the difference in viral load between N2 and JU1580 . An inactivating mutation in a pathogen-resistance gene could be expected to be a rare deleterious variant in natural populations , yet the wild niDf250 allele is found at an intermediate frequency at the global level , in 22/97 ( 23% ) of the tested wild isolates . The deletion is found in about one third of isolates from Europe and Africa ( 21/64 , 33% ) with a high incidence in France ( 14/30 , 47% ) , but is rarer ( 1/30 ) in those from the Americas and the Pacific regions ( Figure 2A ) . As expected , the presence of the deletion correlated strongly with viral load in the infection experiment ( Figure 1—figure supplement 1A; Wilcoxon test on viral load of isolates carrying each drh-1 allele , p=1 . 3×10−9 ) . Thus , surprisingly , the derived drh-1 allele has spread to intermediate frequency in natural populations , despite rendering the animals susceptible to viral infection . 10 . 7554/eLife . 00994 . 006Figure 2 . Geographic distribution and evolutionary genetic context of drh-1 alleles . ( A ) Geographic distribution of drh-1 alleles . The respective frequencies of the niDf250 and N2 alleles of drh-1 are represented for each world region in blue and red , respectively , based on genotyping of the 97 wild isolates . ( B ) Competition experiment between the N2 reference and the JU2196 introgression line . In the absence of the Orsay virus , the proportion of the N2 genotype remains close to 50% throughout the experiment ( 48 . 6 ± 5 . 3% ) . In the presence of the Orsay virus , the proportion of the N2 genotype increases and appears to stabilize around 90% after nine transfers ( 88 . 1 ± 4 . 6% ) . The presence of the virus has a significant effect ( linear model , p=1 . 6 × 10−4 ) . Error bars represents standard deviation . ( C ) Neighbor-network of the 97 isolates in the chromosome IV central region associated with Orsay virus sensitivity . Only one isolate per haplotype is represented; font size is relative to the number ( n ) of isolates sharing this haplotype . Haplotypes in blue or red carry the niDf250 allele of drh-1 , respectively . ( D ) Distribution of SNPs along chromosome IV between N2 and JU1580 , based on JU1580 whole-genome sequencing . ( E ) Molecular diversity ( left y axis scale ) is plotted along chromosome IV for isolates carrying the niDf250 or the N2 allele as blue or red lines , respectively . Linkage Disequilibrium D′ values ( right y axis scale ) between polymorphic RAD sites along chromosome IV and the niDf250 or N2 alleles of drh-1 are represented with blue or red circles , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 00610 . 7554/eLife . 00994 . 007Figure 2—figure supplement 1 . Infection by the Orsay virus has an effect on progeny production of drh-1 deleted strains and on longevity of the drh-1 ( ok3495 ) mutant . ( A ) Dynamics of progeny production of JU1580 in the absence or presence of the Orsay virus ( n = 40 animals ) . ( B ) Total progeny production in the same experiment . ( C ) Survival curves in the absence or presence of virus ( n = 130 animals ) . ( D–F ) Idem with the SX2377 strain ( JU1580 rescue ) . ( G–I ) Idem with the N2 strain . ( J–L ) Idem with the RB2519 strain ( the rde-1 ( ok3495 ) mutant ) . JU1580 animals and drh-1 ( ok3495 ) mutants show a significant delay in progeny production in the presence of the virus ( linear model , p=0 . 049 and p=7 . 5 × 10−4 , respectively ) , as well as a decrease in total progeny production ( Wilcoxon rank test , p=1 . 6 × 10−4 and p=1 . 2 × 10−4 , respectively ) . The N2 and SX2377 strains with the intact drh-1 gene do not show this viral sensitivity . Concerning longevity , the drh-1 ( ok3495 ) mutant has a significantly reduced lifespan in the presence of the virus ( logrank test , p=5 . 2 × 10−4 ) , but the JU1580 isolate does not ( p=0 . 85 ) . Strains only carrying deleted versions of drh-1 are represented in blue while strains carrying the N2 drh-1 are represented in red . **: p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 00710 . 7554/eLife . 00994 . 008Figure 2—figure supplement 2 . Chromosome IV haplotypes for the 97 isolates ( modified from Supplemental Figure 7 in Andersen et al . , 2012 ) . Each row represents one of the 96 isolates , ordered from the less divergent ( top ) to the most divergent ( bottom ) from N2 for the region associated with Orsay virus sensitivity ( IV:6 , 388 , 961 to IV:12 , 408 , 993; between black arrows ) . All clones that share a given region of a chromosome are shown with the same color in that region ( N2 is in red ) . Haplotypes unique to a single isolate are colored in gray . Isolates in blue or red carry the niDf250 or N2 allele of drh-1 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 008 A possible interpretation for the spread of the sensitive drh-1 allele might be that high Orsay viral load has no deleterious effect on fitness . However , in laboratory conditions , we found that viral infection leads to delayed and decreased total progeny of JU1580 and drh-1 ( ok3495 ) mutant animals relative to uninfected animals , whilst infection of N2 had no significant effect ( Figure 2—figure supplement 1 ) . Furthermore , we performed a competition experiment between N2 and JU2196 , which contains the drh-1 region from JU1580 introgressed into the N2 background ( Figure 1B ) . N2 rapidly outcompetes JU2196 in the presence of viral infection , but not its absence , confirming that the increased viral infection resulting from the drh-1 deletion is indeed detrimental for fitness ( Figure 2B ) . In the absence of viral infection , we could not detect in standard laboratory conditions over 10 generations of competition any positive or negative effect of the drh-1 deletion and the introgressed surrounding region . Thus the natural drh-1 deletion impairs fitness only in the presence of viral infection . To characterize the evolutionary history of the drh-1 region , we first focused on the 6 Mb region detected by genome-wide association . This region presents three main haplotypes among the 97 wild isolates ( Andersen et al . , 2012 ) : the N2 haplotype , the JU1580 haplotype and a distant haplotypic group ( including RW7000 and QX1211 ) , as well as a few recombinants ( Figure 2C , Figure 2—figure supplement 2 ) . The drh-1 ( niDf250 ) allele is exclusively found in isolates carrying the JU1580 haplotype in the 6 Mb region and in a few recombinants with either the N2 or distant haplotypes ( Figure 2C ) . The JU1580 and N2 haplotypes show fewer fixed differences in RAD polymorphic sites between them ( 24 SNPs ) than with the divergent haplotype group ( 63 and 57 SNPs , respectively ) . In addition , from our whole-genome JU1580 sequencing data , N2 and JU1580 display a very low level of molecular diversity in the central region of chromosome IV ( Figure 2D ) . Moreover , we observe a strong decrease in molecular diversity between IV:4529464 to IV:6662701 in isolates carrying the niDf250 allele ( Figure 2E , lines ) , but not in those carrying the drh-1 ( N2 ) allele . Thus , the divergence between the N2 and JU1580 haplotypes in this region , including the niDf250 deletion , appears recent relative to much of the species’ genetic diversity . Furthermore , the niDf250 allele is in high or even full linkage disequilibrium with a large region of chromosome IV ( Figure 2E , dots ) . This lack of diversity and the high linkage disequilibrium around niDf250 suggest a partial sweep of the haplotype linked to niDf250 . They also imply that the sensitive drh-1 allele may have spread by hitch-hiking with a favorable allele . Having established the major role of drh-1 allelic variation in natural variation of antiviral defense in C . elegans , we wished to understand the molecular mechanisms of DRH-1 action in the antiviral response . We previously showed that disruption of small RNA pathway genes such as rde-1 , which encodes an Argonaute protein essential for RNAi in response to exogenous dsRNA ( RDE-1 ) , renders N2 animals as sensitive to the Orsay virus as JU1580 ( Félix et al . , 2011 ) . DRH-1 interacts with the double-stranded RNA ( dsRNA ) binding protein RDE-4 and the dsRNA-specific endonuclease Dicer ( DCR-1 ) , both of which act upstream of RDE-1 in exogenous RNAi ( Tabara et al . , 1999; Parrish and Fire , 2001; Barber et al . , 2010 ) However , DRH-1 is dispensable for exogenous RNAi ( Gu et al . , 2009 ) . We therefore wondered if DRH-1 could act specifically to promote DCR-1 processing of long dsRNA that is produced during viral RNA replication . Indeed , we find that DCR-1 is required for viral resistance in the N2 strain ( Figure 3A ) . We therefore postulated that the antiviral response involves siRNAs processed by DCR-1 , which may initiate a cascade of events analogous to canonical C . elegans RNAi . 10 . 7554/eLife . 00994 . 009Figure 3 . DRH-1 is required for the Orsay antiviral response and primary viRNA generation . ( A ) qRT-PCR analysis of viral load after 4 days of infection with the Orsay virus . , dcr-1 mutants are sterile , data shown are homozygous mutant animals from heterozygous mothers . ( B ) Primary viRNA populations in strains as indicated . 5′ dependent small RNA sequencing captures only primary siRNAs with a 5′ monophosphate . Data are grouped as sense or antisense and according to length and the identity of the first nucleotide . From the same samples viral load was measured by qRT-PCR of the Orsay virus RNA1 genome after four days of infection ( heatmap , see also Figure 3A and Figure 3—figure supplement 1B ) . ( C ) Analysis of phasing of 23 nt primary viRNAs generated in infected N2 animals . The x axis shows the length of the overhang in nucleotides , either 5′ ( negative numbers ) or 3′ ( positive numbers ) , for each pair of sequences that map to overlapping regions on opposite strands . A value of 0 represents a pair of viRNAs with perfect complementarity that would form blunt ends . The y axis shows the number of times each particular overhang was observed relative to the number of times that such an overhang would be expected if overhangs were random . Green bar indicates the 2 nt 3′ overhang . ( D–F ) same as in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 00910 . 7554/eLife . 00994 . 010Figure 3—figure supplement 1 . Viral sensitivity in a number of mutants of small RNA pathway genes . ( A–C ) RT-qPCR analysis of viral levels after 4 days of infection with the Orsay virus . Different panels refer to different sets of experiments using different stocks of virus . Note that sago-2 appears to be highly sensitive to viral infection . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 01010 . 7554/eLife . 00994 . 011Figure 3—figure supplement 2 . Additional small RNA sequencing controls . ( A ) Virtually no small RNAs in libraries prepared from uninfected animals align to the viral genome . All reads aligning to RNA2 of the Orsay genome with up to one mismatch are shown . ( B and C ) 5´ dependent small RNA sequencing of infected drh-1; rde-4 double mutants ( compare to Figure 4K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 01110 . 7554/eLife . 00994 . 012Figure 3—figure supplement 3 . Relationship between antiviral small RNAs and predicted secondary structure within the viral genome . ( A ) Profile of the strength of predicted secondary structures for each 50 bp sliding window in 20 bp steps across the viral genome , as measured by the Z-score of the RNA-fold free energy compared to 100 random shuffles of the sequence . Thus negative values represent a stronger structure ( more negative free energy ) than random . The region with the strongest predicted structure is highlighted by a green bar . ( B ) Predicted minimum free energy structure of the region highlighted in ( A ) . ( C ) Distribution of read lengths for JU1580 small RNAs mapping to the viral genome anywhere within this region . ( D ) Detailed analysis of the reads mapping within the selected region indicating the number of reads that come from the predicted arms and loop of the hairpin shown in ( B ) . ( E ) Boxplot of read lengths for JU1580 small RNAs mapping either to regions with weak predicted secondary structures , or to regions with strong predicted secondary structures . The difference between unstructured or structured regions is statistically significant ( p=0 . 03 , Wilcoxon unpaired test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 012 In C . elegans , the RNAi pathway is divided into primary and secondary steps . Long dsRNA is processed by DCR-1 to generate a primary siRNA duplex ∼23 nucleotides ( nt ) in length with 5′ monophosphates and 2 nt 3′ overhangs . By examining libraries derived solely from small RNAs with 5′ monophosphates ( 5′ dependent libraries ) , we could interrogate the primary siRNA response specifically . In wild-type N2 animals , infection with virus leads to generation of predominantly 23 nt small RNAs with no first nucleotide bias , mapping both sense and antisense to the viral genome in equal proportions ( Figure 3B ) . These small RNAs are likely primary DCR-1 products generated from the double-stranded intermediate of viral replication , consistent with the small RNA response against single-strand RNA viruses in insects ( Flynt et al . , 2009 ) . Furthermore , overlapping RNAs mapping to complementary strands of the viral genome showed the 2 nt 3′ overhang characteristic of Dicer products ( p<10−9 , χ2 test against a uniform distribution of overhang length ) ( Figure 3C ) . The same viRNA pattern was observed in rde-1 mutants lacking the primary siRNA Argonaute protein RDE-1 ( Figure 3D ) . Thus rde-1 mutants are proficient in primary viRNA formation despite being sensitive to viral infection ( Figure 3—figure supplement 1 ) ( Félix et al . , 2011 ) . In contrast , JU1580 animals showed a markedly different pattern ( Figure 3E ) . First , a much higher proportion of small RNAs derived from the sense strand than the antisense strand ( 88% vs 49% in N2 ) . Second , the size distribution of viRNAs was flattened , with a greatly reduced proportion of 23 nt RNAs and an increased proportion of 15–20 nt RNAs . These shorter viRNAs were not present in libraries prepared from uninfected JU1580 or N2 controls ( Figure 3—figure supplement 2A ) and appeared to be derived from regions of strong secondary structure within the virus ( Figure 3—figure supplement 3 ) . The drh-1 mutant ( in an N2 strain background ) displayed an identical viRNA pattern to JU1580 ( Figure 3F ) . Thus JU1580 and drh-1 mutant strains are deficient in primary siRNA generation against the virus . Primary siRNAs act upstream of an amplification step by triggering the synthesis of secondary siRNAs antisense to targeted RNAs ( Sijen et al . , 2001 ) . Secondary siRNAs are synthesized by RNA-dependent RNA polymerases ( RdRPs ) , and bind to a number of secondary siRNA-specific Argonaute proteins to bring about target silencing ( Yigit et al . , 2006 ) . Secondary siRNAs have a modal length of 22 nt , are 5′ triphosphorylated , have a strong preference for a 5′ guanine ( G ) , and are referred to as 22G siRNAs ( 22Gs ) . Small RNAs can be enzymatically treated prior to adaptor ligation to allow sequencing ( 5′ independent ) of both primary and secondary siRNAs ( Pak and Fire , 2007; Sijen et al . , 2007 ) . N2 animals infected with Orsay virus showed a robust secondary 22G siRNA response , primarily antisense to the viral genome ( Figure 4A , Figure 4—figure supplement 1 ) . In contrast , rde-1 mutants lacked 22G siRNAs , consistent with a role for RDE-1 in initiating the secondary siRNA response ( Figure 4B , Figure 4—figure supplement 1 ) . In addition , mutants lacking DRH-3 or the RdRP RRF-1 and a strain deficient in 12 worm Argonaute proteins ( WAGO-1 through 12 ) that bind secondary siRNAs ( MAGO-12 ) ( Yigit et al . , 2006 ) lacked 22G viRNAs , but still produced primary siRNAs ( Figure 4F , L , M ) . The above have previously been implicated in secondary siRNA generation in other contexts ( Yigit et al . , 2006; Pak and Fire , 2007; Sijen et al . , 2007 ) and drh-3 , rrf-1 and MAGO-12 mutant strains are also sensitive to viral infection similar to rde-1 ( Figure 3—figure supplement 1 ) . Taken together these data show that a canonical secondary siRNA pathway is engaged to amplify the antiviral response . 10 . 7554/eLife . 00994 . 013Figure 4 . DRH-1 acts upstream of a 22G secondary siRNA pathway . ( A–M ) Primary and secondary viRNA populations in strains as indicated . 5′ independent small RNA sequencing captures 5′ primary siRNAs ( 5′ monophosphate ) and secondary siRNAs ( 5′ triphosphate ) . Data are grouped as sense or antisense and according to length and the identity of the first nucleotide . From the same samples viral load was measured by RT-qPCR of the Orsay virus RNA1 genome after 4 days of infection ( heatmap , see also Figure 3A and Figure 3—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 01310 . 7554/eLife . 00994 . 014Figure 4—figure supplement 1 . Distribution of viRNAs along the Orsay genome . ( A–E ) Small RNAs ( primary and secondary siRNAs from 5′ independent sequencing ) were mapped to Orsay RNA2 genomic sequence in strains indicated . JU1580 ( drh-1 rescue ) in panel ( E ) refers to SX2375 . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 01410 . 7554/eLife . 00994 . 015Figure 4—figure supplement 2 . Analysis of residual Dicer products in JU1580 mutants . ( A ) Number of 23 nucleotide long reads showing 3´ overlap as indicated on the x axis . ( B ) log2 observed/expected showing enrichment for a 2 nucleotide 3´ overlap . The p value is for a χ2 test against a uniform distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 01510 . 7554/eLife . 00994 . 016Figure 4—figure supplement 3 . Analysis of 22G-RNAs mapping to endogenous loci . ( A ) . Endogenous 22G secondary siRNAs mapping to genes in drh-1 and drh-3 mutants compared to wild-type ( N2 ) animals . All 22G siRNAs mapping to non-overlapping protein-coding genes for which at least 10 reads could be detected in at least one repeat of N2 sequencing are shown . Reads for each protein-coding gene were scaled so that each library was directly comparable by multiplying the number of reads by the ratio of the total size of the library to the size of the largest N2 library . 1 read was then added to allow the log2 of the number of reads to be taken . The density plot shows the distribution of differences of normalized reads averaged over 3 repeats of each of drh-1 mutants and N2 . The distribution of differences of drh-3 to N2 obtained using the same methods is shown for comparison . ( B ) 5´ RACE analysis of the Orsay viral genome . The left-hand panel shows the method used to asses whether the 5´ end of the virus carries a triphosphate or not . Primer pairs used for PCR are shown—F1 is a positive control annealing within the viral sequence and F2 is adaptor specific . The right-hand panel shows amplification with the primer pairs as indicated in the left-hand panel . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 01610 . 7554/eLife . 00994 . 017Figure 4—figure supplement 4 . drh-1 mutants are not hypersensitive to RNAi . The bar chart shows the number of L4 stage animals displaying either paralysis , twitching or no phenotype after feeding on RNAi E . coli for four days is shown for serial dilutions of unc-22 RNAi E . coli with E . coli expressing an empty vector control . No twitching was observed in any strain when the empty vector RNAi alone was used . We observed enhanced RNAi in eri-1 vs N2 ( p<1 x 10−4 , Fisher's exact test ) for every dilution but no difference N2 vs drh-1 ( p<0 . 1 , Fisher's exact test ) at any dilution . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 017 drh-1 mutants and JU1580 animals displayed the same profile of sense siRNAs in 5′ independent libraries as in 5′ dependent libraries ( Figures 3E–F and 4C , E ) . Mutants deficient in the adjacent gene drh-2 have a viRNA profile similar to N2 ( Figure 4I ) and are as resistant to viral infection as N2 ( Figure 3—figure supplement 1 ) , implying that this gene is not involved in the antiviral siRNA pathway . Moreover transgenic JU1580 animals carrying the drh-1 gene from N2 showed the same overall viRNA profile as N2 ( Figure 4D , Figure 4—figure supplement 1 ) , further supporting the conclusion that drh-1 deficiency is primarily responsible for the defective siRNA synthesis and virus sensitivity in JU1580 . Importantly however , there were residual antisense 22G siRNAs present in both JU1580 and drh-1 mutants . This suggests that the few DCR-1 products with the correct length in drh-1 mutants can still be used to generate secondary siRNAs as in N2 , implying that drh-1 is not essential for the secondary siRNA pathway . Consistent with this interpretation , the reduced 23 nucleotide long primary siRNA products in JU1580 still displayed a 2 nucleotide 3′ overhang characteristic of DCR-1 activity ( Figure 4—figure supplement 2 ) . Furthermore , residual 22G siRNAs are also present in mutants deficient in the DCR-1 accessory factor RDE-4 , which is required for efficient DCR-1 activity in the exogenous RNAi pathway ( Figure 4J; Tabara et al . , 1999; Parrish and Fire , 2001 ) . Additionally , drh-1 mutants displayed no difference in endogenous 22G siRNAs mapping antisense to protein-coding genes , whilst , in agreement with previous data ( Gu et al . , 2009 ) , drh-3 mutants showed markedly reduced levels of endogenous siRNAs ( Figure 4—figure supplement 3 ) . Together with our observation that drh-1 mutants are deficient in primary viRNA production , this suggests that DRH-1 acts early in the antiviral siRNA pathway and is not required for downstream amplification steps . These observations are in contrast to earlier work describing a role for DRH-1 downstream of secondary siRNA production in a flockhouse virus replicon model ( Lu et al . , 2009 ) . To further test this interpretation , we examined double mutant strains . The prominent 23 nt peak for sense and antisense viRNAs attributed to DCR-1 activity present in drh-3 and rde-1 single mutants was absent in both drh-3; drh-1 and drh-1; rde-1 double mutants ( Figure 4G , H ) . This is consistent with the idea that DRH-1 acts upstream of DRH-3 and RDE-1 in the antiviral siRNA pathway and acts in concert with DCR-1 . The rde-4; drh-1 double mutant showed a further reduction in primary 23 nucleotide long Dicer products compared to the drh-1 single mutant ( Figure 3—figure supplement 2 ) . Furthermore , residual antisense 22G siRNAs in drh-1 mutants ( Figure 4E ) were mostly absent in rde-4; drh-1 double mutants ( Figure 4K ) . These data support the conclusion that residual DCR-1 activity on viral dsRNA in drh-1 mutants is dependent on RDE-4 . Thus , the double mutant data confirms the position of DRH-1 as an upstream factor essential for the generation of robust levels of antiviral siRNA in response to infection . Overall our data support a key and unique role for the C . elegans RIG-I-like protein DRH-1 in primary siRNA synthesis by either guiding DCR-1 activity to the viral genome or assisting DCR-1 processing of the double-stranded viral RNA . We suggest that the physical interaction between DRH-1 and DCR-1 , and the potential for DRH-1 to recognize the viral genome as foreign , possibly through its well-conserved RIG-I domain , may enable DRH-1 to recruit DCR-1 to the double-stranded replicating viral genome and instigate a hierarchical antiviral siRNA response ( Figure 5 ) . Our data also show that the role of DRH-1 in viral recognition is distinct from that of its paralogs . Given the sequence and domain similarities between DRH-1 , DRH-2 and DRH-3 , it will be of interest to determine the mode of RNA recognition by DRH-2 and DRH-3 in the future . 10 . 7554/eLife . 00994 . 018Figure 5 . Model: DRH-1 triggers a hierarchical antiviral RNAi pathway . Upon infection of the N2 C . elegans strain by the Orsay virus , DRH-1 recruits DCR-1 and its partner RDE-4 to the viral dsRNA replication intermediate . DCR-1 cleaves the viral genome into 23 nt viRNA duplexes with a 2 nt 3′ overhang . Duplex viRNAs are incorporated into the Argonaute protein RDE-1 and one strand is lost to give rise to primary viral siRNAs ( primary viRNAs ) . Primary viRNAs and RDE-1 recruit an RdRP complex to the viral genome to synthesize secondary viral siRNAs , which act to silence viral transcripts or inhibit virus replication . The antiviral RNAi pathway is dependent on the SAGO-2 secondary Argonaute protein ( Figure 3—figure supplement 1C ) . The antiviral RNAi pathway has parallels to the exogenous RNAi pathway and the endogenous RNAi pathway thought to recognize aberrant endogenous transcripts ( Gu et al . , 2009 ) . A complex of DRH-1 , DCR-1 and RDE-4 has previously been observed in whole animal lysates ( Tabara et al . , 2002; Duchaine et al . , 2006; Thivierge et al . , 2012 ) . We refer to this complex as the Viral Recognition Complex ( ViRC ) . ERI , other ERI factors . ERIC , ERI Complex . DOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 018 Our findings imply that parallel RNAi pathways are involved in recognition of viral infection , aberrant endogenous transcripts and exogenous RNAi ( exo-RNAi ) in an experimental setting . It is interesting therefore that DRH-1 is found in a protein complex with DCR-1 and RDE-4 even in the absence of infection ( Thivierge et al . , 2012 ) . This might suggest that this complex limits the availability of DCR-1 for exo-RNAi . However , drh-1 and N2 worms respond equally to exo-RNAi , suggesting that this is not the case ( Figure 4—figure supplement 4 ) . The constitutive nature of the DCR/DRH-1 complex may allow cells to respond much more rapidly to the presence of viral replication intermediates . Recent observations from Rui Lu and colleagues are in agreement with our findings ( Guo et al . , 2013 ) . Antiviral small RNAs generated by Dicer are an evolutionarily conserved mechanism for fighting infection by positive-strand RNA viruses ( Aliyari and Ding , 2009 ) . However , there must be some mechanism to distinguish between the viral genome and cellular RNAs . Drosophila and plants have a dedicated Dicer enzyme responsible for viral dsRNA recognition . We suggest that in C . elegans DRH-1 may perform this function . The absence of a role for drh-1 in the endogenous small RNA pathway ( Figure 4—figure supplement 3; Gu et al . , 2009 ) supports the idea that it encodes a viral-specific recognition factor . In mammals , the C-terminal domain of the DRH-1 ortholog RIG-I is able to recognize the 5′ triphosphate on viral genomes in the context of the double-stranded replication intermediate ( Hornung et al . , 2006; Pichlmair et al . , 2006; Rehwinkel et al . , 2010 ) . The C-terminal domain is conserved in DRH-1 , thus suggesting that DRH-1 might use a recognition mechanism analogous to RIG-I to recruit DCR-1 to the Orsay virus RNA . In support of this , 5′ RACE experiments from total RNA from infected animals using Orsay virus specific primers fail to detect product unless the 5′ triphosphate is removed prior to adaptor ligation ( Figure 4—figure supplement 3 ) , suggesting that the majority of Orsay genomic RNA molecules are indeed phosphorylated at the 5′ end . It is interesting that the recognition function of DRH-1 may be conserved with mammalian RIG-I whilst the effector pathways are apparently distinct . It will therefore be intriguing to examine whether any part of the function of DRH-1 in antiviral RNAi might be conserved in mammals . As yet , although small RNA responses have been analyzed in mammalian cells infected with viruses ( Parameswaran et al . , 2010 ) it is not clear whether these have a significant role in the defense of viral infection or whether they display any similarities to antiviral RNAi pathway in plants or nematodes . Examining whether cells deficient in RIG-I show alterations in small RNA pathways upon viral infection may help to address this question . Equally , it will be interesting to identify whether DRH-1 has a signaling role in C . elegans upon viral infection . Such studies might indicate whether the gene expression or the RNAi function is the ancestral role of RIG-I-like genes . RIG-I and MDA-5 respond to viral infection with changes in cytoplasmic localization that result in the activation of the interferon response ( Nakhaei et al . , 2009 ) . In addition , RIG-I and MDA-5 are themselves interferon-induced genes . It will therefore be of great interest to explore the behavior of DRH-1 upon viral infection . A recent genome-wide analysis of gene expression upon Orsay virus infection in C . elegans did not detect any significant changes in drh-1 transcript levels ( Sarkies et al . , 2013 ) . Furthermore , a GFP-DRH-1 fusion protein did not show marked alterations in expression levels or subcellular localization upon infection with the Orsay virus ( our unpublished observations ) . It will be important to analyze the behavior of the endogenous DRH-1 protein upon infection in future studies . Our discovery of an inactivating deletion in drh-1 carried by many wild isolates of C . elegans is consistent with the rapid evolution under strong selection known to characterize proteins involved in immunity , including antiviral RNAi defense ( Obbard et al . , 2006; Vasseur et al . , 2011 ) . Yet it is puzzling that a derived allele with deleterious consequences in the presence of viral infection is found at intermediate frequency . One possible explanation might be the low natural occurrence of infecting viruses , meaning that the drh-1 deletion is effectively neutral . In support of this argument we have not been able to detect similar intestinal symptoms of viral infection in our extensive C . elegans sampling ( Félix and Duveau , 2012 ) , including in other years of sampling in the Orsay orchard . Alternatively the drh-1 deletion may have hitch-hiked to fixation with a closely linked beneficial mutation . Such a phenomenon is likely to be common in C . elegans natural populations due to the low effective outcrossing rate , which results in high linkage disequilibrium , especially when associated with a positive selective sweep ( Cutter et al . , 2009 ) . Indeed we find that the natural drh-1 deletion allele is in high linkage disequilibrium with the surrounding region of chromosome IV ( Figure 2E ) . A related possibility is that the drh-1 deletion itself might have a positive fitness effect in some natural conditions . Although we did not detect a fitness advantage for the region surrounding the drh-1 deletion under laboratory conditions , we cannot test all possible natural conditions , thus a beneficial effect for the drh-1 deletion or its surrounding region remains a possibility . Interestingly , RIG-I appears to have been lost in several clades , including chickens ( Zou et al . , 2009; Barber et al . , 2010; Table 1 ) , which might explain the increased sensitivity of chickens to avian influenza virus when compared to ducks ( Barber et al . , 2010 ) . 10 . 7554/eLife . 00994 . 019Table 1 . Evolution of Dicer and RIG-I family proteinsDOI: http://dx . doi . org/10 . 7554/eLife . 00994 . 019helicase + RIG-I structures in speciesDicerCnidariaNematostella vectensis21BilateriaProtostomiaPolyzoaPlatyzoaSchmidtea mediterranea01KryptochozoaMolluscaAplysia californica01AnnelidaPlatynereis dumerilii01EcdysozoaDrosophila melanogaster02Trichinella spiralis21Caenorhabditis elegans31DeuterostomiaBranchiostoma floridae21Meleagris gallopavo31Taeniopygia guttata31Gallus gallus21Homo sapiens31Presence of Dicer and RIG-I family proteins in selected animals . Data were obtained from Pfam ( version 26 . 0 ) ( Finn et al . , 2010 ) ( pfam . sanger . ac . uk ) . RIG-I family proteins were identified by having both helicase domains and the RIG-I C-terminal domain ( Pfam: PF11648 ) . Available sequence data is sparse for some clades and absence of data might not be sufficient evidence for absence of genes . In conclusion , we found that the RIG-I domain has an ancient role in antiviral immunity outside of mammals , yet that this broad conservation in animals is compatible with recurrent losses in insects , ducks , and down to some C . elegans isolates . Our results further indicate that the conserved biochemical activity of RIG-I is viral recognition , whereas downstream effector pathways , such as RNAi and interferon responses , may differ between C . elegans and mammals . C . elegans were grown under standard conditions at 20°C unless otherwise indicated . The wild-type strain was var . Bristol N2 ( Brenner , 1974 ) . All strains used are listed in Supplementary file 1A . Virus filtrate was prepared as described previously ( Félix et al . , 2011 ) . Non-synchronized animals were cultured in a 55-mm plate and collected in M9 just before starvation . RNA extraction and RT-qPCR were performed as described previously ( Félix et al . , 2011 ) . JU1580 animals were transformed as described ( Mello and Fire , 1995 ) with the fosmid WRM0640dC01 that contains the entire length of the drh-1 gene and its operon CEOP4647 . The injection mix contained 10 ng/µl of fosmid DNA , 5 ng/µl of the co-marker transgene myo-3::gfp::unc-54 , 85 ng/µl of 1 kb DNA ladder ( Invitrogen ) , 20 mM potassium phosphate pH 7 . 5 , and 3 mM potassium citrate pH 7 . 5 . The transgene was then integrated via X-ray irradiation as described ( Fire , 1986 ) . We controlled that the transgenic copy of drh-1 was transcriptionally active in transformed animals by qRT-PCR on a portion of the mRNA that is deleted in the wild JU1580 strain . Total RNA isolated from infected animals as described above was ligated to the 5′ adaptor from the Illumina Truseq small RNA kit either with or without prior treatment with 5′ polyphosphatase . Only monophosphorylated 5′ ends will be able to ligate to the adaptor , as for the small RNA sequencing . Ligated products were then reverse transcribed using a primer specific for RNA1 . The resulting cDNA was then analysed by standard Taq PCR and gel electrophoresis using either primers designed to amplify within the 5′ end of the viral genome as a positive control or primers to amplify from the adaptor into the 5′ end , thus determining whether the adaptor was able to ligate efficiently to the 5´ end ( Figure 4—figure supplement 3B ) . For each of the 97 isolates , 97 SNPs included in the 6 Mb central region of chromosome IV associated with Orsay virus sensitivity were extracted from RAD-sequencing data ( Andersen et al . , 2012 ) . DNAsp allowed us to classify all isolates in 28 different haplotypes for this region . The neighbor-net network ( Bryant and Moulton , 2004 ) was then drawn using the SplitTree software ( Librado and Rozas , 2009 ) . The average number of differences per polymorphic RAD site along chromosome IV was calculated using DNAsp ( Huson and Bryant , 2006 ) using a sliding window of 25 SNPs every 10 SNPs . The linkage disequilibrium ( D′ ) between drh-1 alleles and polymorphic RAD sites on chromosome IV was calculated by DNAsp . unc-22 and empty vector RNAi bacteria were grown for 6 hr at 37°C . unc-22 bacteria were then serially diluted with the empty vector bacteria and seeded onto NGM agar plates containing IPTG ( 1 mM ) and carbenicillin ( 25 μg/ml ) . After drying overnight , N2 , drh-1 or eri-1 worms were added and then grown at 20°C for 4 days . Each strain was tested in triplicate at each dilution and 15 animals selected at random from each plate were scored for either twitching or paralysis phenotypes .	Most organisms—from bacteria to mammals—have at least a rudimentary immune system that can detect and defend against pathogens , particularly viruses . This defense mechanism , which is known as the innate immune system , uses sensor proteins to recognize viral RNA , and then mobilizes other immune components to attack the invaders . The specific mechanisms used to destroy viruses differ between species . In mammals , a protein called RIG-1 binds to viral RNA and activates a signaling pathway that leads to the production of interferons: immune proteins named after their ability to ‘interfere’ with viral replication . Plants and insects do not use interferons , but instead use a mechanism called RNA interference , in which long double-stranded RNAs are cleaved into shorter fragments . The nematode worm C . elegans also deploys RNA interference against viruses but , in contrast to insects and plants , worms do not possess a specific set of RNA interference enzymes that participate solely in the antiviral response . They do , however , express a protein called DRH-1 that is related to the RIG-I protein found in mammals . To investigate whether DRH-1 contributes to innate immunity in C . elegans , Ashe et al . infected 97 strains of C . elegans from around the world with a virus , and showed that some strains were more sensitive to the virus than others , with certain strains showing complete resistance . By comparing a sensitive strain with a resistant one , Ashe et al . revealed that viral sensitivity was caused by a mutation in the gene encoding DRH-1 . Further experiments showed that DRH-1 is required for the first step in RNA interference . Ashe et al . have thus identified a conserved role for RIG-1 in initiating antiviral responses , and propose that the protein couples virus recognition to distinct defense mechanisms in different evolutionary groups .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2013	A deletion polymorphism in the Caenorhabditis elegans RIG-I homolog disables viral RNA dicing and antiviral immunity
Skeletal muscle maintenance depends on motor innervation at neuromuscular junctions ( NMJs ) . Multiple mechanisms contribute to NMJ repair and maintenance; however muscle stem cells ( satellite cells , SCs ) , are deemed to have little impact on these processes . Therefore , the applicability of SC studies to attenuate muscle loss due to NMJ deterioration as observed in neuromuscular diseases and aging is ambiguous . We employed mice with an inducible Cre , and conditionally expressed DTA to deplete or GFP to track SCs . We found SC depletion exacerbated muscle atrophy and type transitions connected to neuromuscular disruption . Also , elevated fibrosis and further declines in force generation were specific to SC depletion and neuromuscular disruption . Fate analysis revealed SC activity near regenerating NMJs . Moreover , SC depletion aggravated deficits in reinnervation and post-synaptic morphology at regenerating NMJs . Therefore , our results propose a mechanism whereby further NMJ and skeletal muscle decline ensues upon SC depletion and neuromuscular disruption . Skeletal muscle is composed of long multinucleated cells , muscle fibers ( myofibers ) , which function as primary effectors for force production and contribute to the regulation of whole body metabolism . Although primarily a post-mitotic tissue , adult skeletal muscle possesses a remarkable capacity for regeneration . This capacity is endowed by a population of resident stem cells , satellite cells ( SCs ) , identified by the expression of the paired box transcription factor Pax7 ( Tajbakhsh , 2009; Brack and Rando , 2012; Yin et al . , 2013 ) . In adults , SCs normally reside in a quiescent state at the interface between the myofiber and overlying basal lamina ( Yin et al . , 2013 ) . However , in response to degenerative stimuli , Pax7+ SCs activate and divide to produce myogenic progenitors for skeletal muscle regeneration ( Tajbakhsh , 2009; Yin et al . , 2013 ) . Since the initial identification of SCs over 50 years ago , many studies have examined roles for these cells in a plethora of distinct models of skeletal muscle injury , adaptability and disease ( Mauro , 1961; Relaix and Zammit , 2012 ) . Both depletion of Pax7+ SCs and targeted disruption of Pax7 have shown these cells to be an essential source of myonuclei for skeletal muscle regeneration ( Relaix and Zammit , 2012; Gunther et al . , 2013; von Maltzahn et al . , 2013; Lepper et al . , 2011; Murphy et al . , 2011 ) . Depletion studies have revealed roles for Pax7+ SCs in late stages of experimentally induced skeletal muscle hypertrophy ( Fry et al . , 2014 ) . Recently , SCs were also shown to function as a source of growth factors to facilitate bone fracture healing ( Abou-Khalil et al . , 2015 ) . Similar strategies have shown SCs contribute to myofibers and regulate aspects of skeletal muscle integrity during aging ( Fry et al . , 2015; Keefe et al . , 2015 ) . In adults , each myofiber is innervated by a single axon from a motor neuron ( Sanes and Lichtman , 1999 ) . Innervation occurs at a specialized site in the central region of myofibers , the neuromuscular junction ( NMJ ) ( Sanes and Lichtman , 1999; Wu et al . , 2010 ) . The NMJ , which occupies approximately 0 . 1% of the surface area of a myofiber , initiates action potential propagation required for excitation/contraction coupling to generate force for movement and maintain myofiber properties ( Sanes and Lichtman , 1999; Bassel-Duby and Olson , 2006; Wu et al . , 2010; Schiaffino and Reggiani , 2011 ) . Consistent with a vital role for NMJ integrity in skeletal muscle maintenance , neuromuscular disruptions elicit severe myofiber atrophy , and are frequently associated with skeletal muscle dysfunction observed in the context of neuromuscular diseases ( NMDs ) and aging ( Murray et al . , 2010; Gonzalez-Freire et al . , 2014; Moloney et al . , 2014 ) . The regeneration of NMJs in response to peripheral nerve lesions can occur , however the length and quality of recovery depends on the severity of injury ( Buti et al . , 1996; Williams et al . , 2009 ) . Remarkably , the initial reinnervation of synaptic basal lamina by motor axons can proceed in the absence of myofibers ( Sanes et al . , 1978; Sanes and Lichtman , 1999 ) . However , in the absence of myofibers and associated SCs , the continued maintenance of reinnervated NMJs on basal lamina ghosts eventually declines ( Sanes et al . , 1978; Sanes and Lichtman , 1999 ) . These observations indicate that myofiber derived factors or associated SCs may be required for the continued differentiation and maintenance of regenerated NMJs . Accordingly , myofiber components and derived factors have been identified to assist in the progressive differentiation of developing and regenerating NMJs ( Sanes and Lichtman , 1999; Fox et al . , 2007; Williams et al . , 2009 ) . In models of chronic denervation , where reinnervation is prevented , SCs activate and divide; however , the derived progenitors migrate into interstitial spaces , undergo defective differentiation or are lost via apoptosis ( Dedkov et al . , 2001; Borisov et al . , 2005; Bruusgaard and Gundersen , 2008 ) . Furthermore , little turnover of myonuclei and fusion of myogenic progenitors to chronically denervated parent myofibers have been observed ( Bruusgaard and Gundersen , 2008 ) . Collectively , these studies have suggested SCs have limited , if any , roles in the regeneration of NMJs upon neuromuscular disruptions ( Gundersen and Bruusgaard , 2008 ) . Through the use of targeted genetic strategies we sought to reexamine the fates and roles of SCs in a model of peripheral nerve injury that enables NMJ regeneration . In this study we find a limited proportion of SCs activate and divide during NMJ reestablishment . Remarkably , while SC depletion did not lead to additional loss of skeletal muscle mass , it was sufficient to reduce myofiber size , increase inter-myofiber connective tissue ( MCT ) accumulation , and aggravate myofiber type transitions connected to NMJ disruption . These phenotypes were associated with further declines in force generation capacity . Examination of fate revealed increased SC activity and fusion of indelibly labeled SC-derived progenitors to myofibers predominantly in the vicinity of regenerating NMJs . Consistent with a role for SCs in the regeneration of NMJs , we found that SC depletion led to deficits in NMJ reinnervation , reductions in post-synaptic morphology and loss of post-synaptic myonuclei . Collectively our findings reveal fates and roles for SCs in the regeneration of NMJs and regulation of skeletal muscle integrity upon neuromuscular disruption . To examine the roles of SCs in skeletal muscles upon neuromuscular disruption we generated Pax7CreER/+; Rosa26DTA/+ ( P7DTA ) and Pax7+/+; Rosa26DTA/+ ( Ctrl ) mice . These mice enable tamoxifen ( Tmx ) -mediated expression of diphtheria toxin-A ( DTA ) to deplete Pax7+ SCs to levels that prevent the regeneration of skeletal muscle ( Murphy et al . , 2011; Relaix and Zammit , 2012 ) . We employed 1-2 mm sciatic nerve transection ( SNT ) to disrupt lower-limb NMJs . This form of surgery leads to complete denervation of adult NMJs . Although delayed , reinnervation as assessed by immunofluorescence ( IF ) and physiological measures does occur 4–6 weeks after SNT ( Buti et al . , 1996; Williams et al . , 2009 ) . Consistent with previous reports , a modest albeit significant increase in Pax7+ SC number was observed 6 weeks after SNT surgery ( Figure 1 ) ( Snow , 1983; Viguie et al . , 1997 ) . After Tmx administration , extensive depletion of Pax7+ SCs occurred regardless of sham or SNT surgery ( Figure 1 ) . 10 . 7554/eLife . 09221 . 003Figure 1 . Depletion of Pax7+ SCs in P7DTA skeletal muscles . ( A ) Scheme demonstrating time of Tmx treatment , Sciatic nerve transection ( SNT ) surgery , and harvest of tissue . Representative images of TA transverse sections , stained with anti-Pax7 ( red ) , anti-Laminin ( white ) and DAPI ( blue ) . Red arrowheads indicate Pax7+ cells . ( B ) Quantification of Pax7+ satellite cell ( SC ) number from Ctrl and P7DTA TA muscles 6 weeks after sham or SNT surgery . N = 3 mice , 3 sections/mouse , 6 fields/ section . Scale bar = 50 μm . p < 0 . 05 compared to Ctrl , p < 0 . 05 compared to Ctrl sham , ANOVA/Bonferroni multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 003 We assessed various morphological parameters to determine the consequences of SC depletion on SNT-induced skeletal muscle atrophy . Gross examination of TA skeletal muscles 6 weeks after SNT surgery revealed overall reductions in girth , which were not noticeably different between P7DTA and Ctrl ( Figure 2A ) . While the mass of TA , extensor digitorum longus ( EDL ) and soleus muscles 6 weeks after SNT relative to contralateral sham controls remained low , SC depletion did not induce further loss of muscle mass ( Figure 2B–D ) . Examination of individual myofibers from EDL muscles that had been fixed prior to isolation from lower limbs revealed only modest reductions in myonuclear density after SNT and SC depletion ( Figure 2—figure supplement 1A and C ) . Based on these moderate losses , a small albeit significant level of myonuclear turnover was observed in P7DTA sham muscles . Therefore , the magnitude of myonuclear loss after SNT relative to sham in P7DTA skeletal muscles is small ( Figure 2—figure supplement 1C ) . Next we examined myofiber size based on Laminin IF analysis of transverse sections from TA muscles ( Figure 2E ) . Despite 6 weeks of recovery , Ctrl myofiber size after SNT remained ∼25% lower relative to contralateral sham , however , SC depletion led to further SNT-induced myofiber atrophy ( Figure 2F ) . Distribution analysis of myofiber size did not reveal any significant differences between Ctrl and P7DTA muscles after sham surgery , but a significant shift towards smaller sizes after SNT surgery was observed upon SC depletion ( Figure 2G ) . No change in myofiber numbers was observed after SNT and SC depletion ( Figure 2—figure supplement 1B ) . Therefore , SC depletion did not lead to a significant change of overall muscle morphology and mass , but aggravated SNT-induced myofiber atrophy . 10 . 7554/eLife . 09221 . 004Figure 2 . SC depletion exacerbates neuromuscular disruption induced myofiber atrophy . ( A ) Representative images of TA muscles . Scale bar = 5 mm . ( B–D ) Quantification of ( B ) TA ( C ) EDL and ( D ) Soleus ( SOL ) muscle wet weight after SNT as a percentage of contralateral sham . N = 6 for Ctrl and 8 for P7DTA . ( E ) Representative TA sections stained with anti-Laminin ( white ) and DAPI ( blue ) . Scale bar = 50 μm . ( F ) Quantification of TA myofiber size as a percentage of contralateral sham . ( G ) Histograms of TA myofiber size distributions . N = 4 mice , >1000 myofibers/mouse . p < 0 . 05 , t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 00410 . 7554/eLife . 09221 . 005Figure 2—figure supplement 1 . Retention of myofibers and modest loss of myonuclei 6 weeks after SNT . ( A ) Representative images of single isolated myofibers from EDL muscles ( fixed prior to isolation from lower limbs ) , stained with DAPI . ( B , C ) Quantification of EDL ( B ) myofiber number and ( C ) myonuclei number per 500 μm . N = 4 mice , 32 myofibers/mouse , p < 0 . 05 compared to Ctrl-sham , p < 0 . 05 compared to Ctrl-sham , Ctrl-SNT , and P7DTA-sham ANOVA/Bonferroni multiple comparisons test and t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 005 Previous studies on rodent models or human patients reveal that chronic denervation , aging of skeletal muscles or NMDs can induce an increase in MCT content , an indicator of fibrosis ( Peltonen et al . , 1982; Savolainen et al . , 1988; Goldspink et al . , 1994; Brack et al . , 2007 ) . Also , SC depletion has been shown to result in extracellular matrix accumulation in the context of skeletal muscle regeneration , functional overload-induced hypertrophy and aging ( Murphy et al . , 2011; Fry et al . , 2014 , 2015 ) . Therefore , to determine if elevated MCT is associated with reductions in myofiber size upon SC depletion and SNT , we performed hematoxylin and eosin ( H&E ) and Sirius Red staining for collagens ( Figure 3A , B ) . Surprisingly , although SNT surgery alone did not increase MCT content , when combined with SC depletion , a significant increase in fibrosis was observed ( Figure 3C ) . Therefore , the lack of change in the mass of SC-depleted skeletal muscles relative to Ctrl after SNT surgery was accompanied by both increased fibrosis and myofiber atrophy . 10 . 7554/eLife . 09221 . 006Figure 3 . SC depletion induces connective tissue accumulation in skeletal muscles after neuromuscular disruption . ( A ) Representative images of TA sections stained with H&E; scale bar = 100 μm , ( B ) Representative images of TA sections stained Sirius Red and pseudocolor images generated by VisioPharm software; numbers indicate myofiber connective tissue ( MCT ) ( red ) content in each representative image; scale bar = 50 μm . ( C ) Quantification of MCT content in TA muscles . N = 4 mice . p < 0 . 05 compared to Ctrl-sham , P7DTA-sham and Ctrl-SNT , ANOVA/Bonferroni multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 006 Most skeletal muscles are composed of heterogeneous mixtures of functionally distinct types of myofibers that differ in many parameters ( Schiaffino and Reggiani , 2011 ) . Based on the expression of skeletal muscle myosin heavy chain ( MyHC ) , different types of myofibers can be classified along a continuum: type I , IIA , IIX and IIB ( Schiaffino and Reggiani , 2011 ) . In this regard , type I fibers are characterized by slower contraction kinetics and lower force generation , whereas at the other end of the continuum type IIB fibers exhibit faster contraction kinetics and greater force generation ( Mantilla and Sieck , 2003 ) . Consistent with the vital influence of neural activity on myofiber type , myofibers innervated by the same motor neuron primarily express one MyHC isoform , and cross-innervation can induce the expression of isoforms indicative of the foreign nerve ( Murray et al . , 2010; Moloney et al . , 2014 ) . Neuromuscular disruption is frequently associated with abnormal profiles of MyHC isoform expression within a given skeletal muscle ( Klitgaard et al . , 1990; Krivickas et al . , 2002; Palencia et al . , 2005; Stevens et al . , 2008; Schiaffino and Reggiani , 2011; Rowan et al . , 2012 ) . A typical feature of NMD and aging skeletal muscle is an increased occurrence of hybrid myofibers whereby two or more MyHC isoforms are co-expressed ( Klitgaard et al . , 1990; Krivickas et al . , 2002; Palencia et al . , 2005; Stevens et al . , 2008; Rowan et al . , 2012 ) . To examine hybrid myofibers , we immuno-stained Ctrl and P7DTA inner TA/EDL muscle sections 6 weeks after sham or SNT surgery with antibodies that specifically detect MyHC IIX or all MyHCs except IIX , therefore hybrid IIX myofibers will be labeled with both . Very few MyHC IIX hybrid myofibers were found in adult inner TA/EDL muscles after sham surgery regardless of genotype ( Figure 4A , B ) . Elevations in the proportion of hybrid myofibers were observed in Ctrl inner TA/EDL muscles 6 weeks after SNT , a phenotype exacerbated by SC depletion ( Figure 4A , B ) . Assessment of inner TA/EDL muscles after SNT with antibodies specific for MyHC I , MyHC IIA and MyHC IIB revealed: i ) very little if any expression of MyHC I; ii ) inductions of MyHC IIA that were aggravated upon SC depletion; and iii ) loss of MyHC IIB ( Figure 4A and C–E ) . Together , these data indicate that the majority of hybrid myofibers observed after SNT are presumably type IIA/IIX . 10 . 7554/eLife . 09221 . 007Figure 4 . SC depletion aggravates myofiber type transitions connected to neuromuscular disruption . ( A ) Representative images of Ctrl and P7DTA inner TA/EDL muscle regions 6 weeks after sham and SNT surgery stained as indicated with anti-MyHC IIX , all MyHCs except IIX , MyHC IIA , MyHC IIB and MyHC I . Also depicted in Merge IIX+/IIX- , MyHC IIB/IIA and MyHC I labeled images are stains for anti-Laminin ( white ) and DAPI ( blue ) . ( B ) Quantification of type IIX pure ( green only ) and hybrid ( green and red , labeled with asterisks ) myofiber percentages . ( C–E ) Quantification of ( C ) Type IIB ( D ) Type IIA and ( E ) Type I fiber percentage . N = 4 mice , 3 sections/mouse , 3 fields/section . Scale bar = 50 μm . p < 0 . 05 compared to Ctrl-sham and P7DTA-sham , p < 0 . 05 compared to Ctrl sham , P7DTA-sham and Ctrl-SNT , ANOVA/Bonferroni multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 007 Modulations in factors that influence force transmission and excitation/contraction coupling , such as increased MCT and abnormalities in the expression of contractile elements can hinder intrinsic whole skeletal muscle contractile function ( Baldwin and Haddad , 2001; Kjaer , 2004 ) . Since we observed inductions in MCT and deviations in the expression of MyHC isoforms , major components of the contractile apparatus , after SC depletion and SNT , we measured ex vivo muscle contractility of Ctrl and P7DTA EDL muscles . Initially we examined the time taken by a given muscle to reach peak tension upon stimulation at 150 Hz , a frequency that elicited peak tetanic contractile force . A similar lengthening in time to peak tension ( TTP ) was observed in Ctrl and P7DTA EDL muscles after SNT ( Figure 5A , and Figure 5—figure supplement 1B ) . Although lengthening of TTP 6 weeks after SNT is consistent with shifts in MyHC expression with slower contractile character , SC depletion did not further influence this property . Next we examined the max tetanic force generated by Ctrl and P7DTA EDLs 6 weeks after SNT surgery upon stimulation at frequencies of increasing intensity . We found significant deficits in peak absolute and specific force generation in P7DTA EDL muscles in comparison to Ctrl after SNT at progressively higher frequencies ( Figure 5B , C , and Figure 5—figure supplement 1A ) . Therefore , the exacerbated deficits in skeletal muscle integrity such as myofiber atrophy , increased MCT content and abnormal profiles of MyHC expression upon SC depletion and SNT were accompanied by declines in whole skeletal muscle force generation . 10 . 7554/eLife . 09221 . 008Figure 5 . SC depletion leads to declines in force generation of skeletal muscles after neuromuscular disruption . ( A ) Average time to peak tension ( TTP ) during 150 Hz stimulation in EDL muscles . p < 0 . 05 compared to Ctrl and P7DTA sham . ANOVA/Bonferroni multiple comparisons test , N = 4–6 . ( B ) Absolute and ( C ) Specific force frequency curves for Ctrl and P7DTA EDL muscles 6 weeks after SNT surgery . p < 0 . 05 compared to Ctrl SNT at indicated frequency , t-tests , N = 4–6 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 00810 . 7554/eLife . 09221 . 009Figure 5—figure supplement 1 . Reduced contractile force and slowed force development in EDL muscles following SNT . ( A ) . Representative traces for specific force from in vitro muscle contraction measurements in EDL muscles stimulated at 150hz for 500ms . ( B ) . Same traces as ( A ) but normalized to the corresponding peak and expanded for the first 200ms of stimulation to show the delayed force development in muscles following SNT . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 009 Initially to assess SC fate we employed an in vivo BrdU incorporation assay to examine SNT-induced SC activation and division ( Chakkalakal et al . , 2012 ) . Consistent with limited SC activation , only a fraction ( ∼25% ) of the SC pool in TA muscles had incorporated BrdU after SNT ( Figure 6A , B ) . To further examine the fate of Pax7+ SCs in response to neuromuscular disruption , we generated a SC specific Pax7CreER/+; Rosa26mTmG/+ ( P7mTmG ) transgenic mouse line . The P7mTmG mouse ubiquitously expresses a loxP flanked membrane tomato red fluorescent reporter ( RFP ) that undergoes Tmx mediated excision to indelibly label Pax7+ SCs and derived cells with membrane GFP ( GFP ) , enabling lineage tracing of SCs in skeletal muscles and individual myofibers after SNT ( Murphy et al . , 2014 ) . To initially characterize the efficiency of the P7mTmG line , we assessed GFP label in SCs and myofibers 24 hr after the last Tmx administration . We found that 24 hr after the last Tmx administration , >90% of the SC pool was GFP+ , whereas myofibers were devoid of GFP label ( Figure 7—figure supplement 1 ) . Next , we examined SC derived GFP labeling after sham or SNT surgery . In comparison to sham , a marked increase in GFP + myofibers was observed in transverse sections from the mid belly of TA and EDL muscles after SNT ( Figure 7A , B ) . Although these data indicate a high proportion of myofibers within TA and EDL muscles have undergone a fusion event after SNT , this activity likely reflects limited fusion within a given myofiber consistent with modest losses of myonuclei within P7DTA myofibers ( Figure 2—figure supplement 1A , C ) . The occurrence of central nucleated myofibers ( CNF ) , an indicator of degenerative/regenerative events of myofibers , was minimal and cannot explain the observed induction of GFP + myofibers after SNT ( Figure 7—figure supplement 1C ) . 10 . 7554/eLife . 09221 . 010Figure 6 . Limited SC proliferation in skeletal muscles upon neuromuscular disruption . ( A ) Strategy to BrdU label SCs in adult mice 4 weeks after sham or SNT surgery and representative TA sections stained with anti-Pax7 ( red ) , anti-BrdU ( green ) and anti-Laminin ( white ) . Red arrowheads indicate Pax7+ cell; green arrowhead indicates BrdU + cell; yellow arrowhead indicates BrdU+/Pax7+ cell . ( B ) Quantification of BrdU+/Pax7+ percentage . N = 3 mice , 3 sections/mouse , 6 fields of view/section . p < 0 . 05 , t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 01010 . 7554/eLife . 09221 . 011Figure 7 . Neuromuscular disruption stimulates SC contribution in the vicinity of NMJs . ( A ) Scheme demonstrating time of tamoxifen treatment , SNT surgery , and harvest of tissue . Representative images to examine GFP label in myofibers and Pax7+ SCs of P7mTmG skeletal muscles 4 weeks after sham and SNT surgery . ( B ) Quantification of the percentage of GFP + myofibers from midbelly of EDL muscles 4 weeks after sham or SNT surgery . Scale bar = 50 μm . N = 3 mice , 3 sections/mouse , 6 fields/section . p < 0 . 05 , t-tests . ( C , D ) Representative images of single isolated P7mTmG EDL myofibers with no GFP ( RFP ) , GFP at ends ( End ) or GFP in middle portions where neuromuscular junctions ( NMJs ) are located ( Mid ) after ( C ) sham or ( D ) SNT surgery . Magnified inset images show SCs ( Pax7+ ) or NMJs ( Btx , AChRs ) . Scale bar for myofibers = 200 μm for inset = 25 μm . ( E ) Quantification of GFP + fiber percentage and distribution . Note a higher percentage of myofibers after SNT express GFP primarily in the Mid regions , the location of NMJs . N = 4 mice , 25 myofibers examined per mouse . p < 0 . 05 for all GFP + groups , *p < 0 . 05 for Mid GFP , t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 01110 . 7554/eLife . 09221 . 012Figure 7—figure supplement 1 . P7mTmG myofibers are not GFP + when examined immediately after Tmx administration . ( A ) Scheme demonstrating time of Tmx treatment , SNT surgery , and harvest of tissue . Representative images of adult P7mTmG TA muscle , taken 24 hr after Tmx treatment , sections stained for anti-GFP and anti-Pax7 . ( B ) Images of Ctrl and P7mTmG TA muscle sections , note the lack of GFP + myofibers . ( C ) Proportion of central nucleated myofibers ( CNF ) and GFP + CNFs after SNT , 3875 myofibers examined . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 012 Myofibers are long multinucleated cells , along which local events of turnover and gene activity can occur ( Pavlath et al . , 1989; Li et al . , 2011 ) . To determine if SNT leads to regional activity and fusion of indelibly labeled SCs and derived progenitors , we isolated EDL single myofibers by collagenase digestion and processed them for the detection of post-synaptic acetylcholine receptors ( AChRs ) with fluorochrome-conjugated α-bungarotoxin ( Btx ) and GFP ( Chakkalakal et al . , 2012 , 2014 ) . We scored the GFP labelling based on location: i ) ‘whole’ for GFP throughout the myofibers; ii ) ‘end’ for GFP at myofiber ends/tips; iii ) ‘middle’ ( Mid ) for central regions of myofibers where NMJs are located; and iv ) ‘RFP’ for no GFP labelling ( examples for RFP , End and Mid in Figure 7C , D ) . After sham surgery , only a small fraction of myofibers were GFP positive; none of these fibers had GFP throughout , and most labels were found either at the ends or middle regions , the latter being where NMJs are located ( Figure 7E ) . After SNT surgery a greater proportion of myofibers were GFP positive , indicating the activation of at least some SCs and fusion of SC derived progenitors to myofibers ( Figure 7E ) . In addition , the majority of the GFP label found in myofibers isolated from EDL muscles after SNT was located in the middle , within the vicinity of NMJs ( Figure 7E ) . Therefore , neuromuscular disruption led to regional SC derived contributions along the length of a myofiber in proximity to the NMJ . Due to the regionalized response of SCs along myofibers near the NMJ after SNT , we employed P7DTA mice to test the importance of this local SC activity to NMJ regeneration upon neuromuscular disruption . Confocal IF microscopy and 3-D image analysis with Amira were used to assess NMJ regeneration . NMJs were identified with post-synaptic ( AChR labeled with Btx ) and pre-synaptic ( SV2 , Syt-2 and neurofilament ) markers ( Figure 8A ) ( Williams et al . , 2009; Valdez et al . , 2010 ) . Initially , we examined the extent of reinnervation , defined as the coverage of post-synaptic regions by pre-synaptic markers . We considered a NMJ to be: i ) innervated , if the vast majority of post-synaptic regions are covered by pre-synaptic terminal markers; ii ) partially denervated , if > 5 μm length of an AChR enriched branch within the post-synaptic apparatus is not covered by pre-synaptic terminal markers while the other parts of the apparatus are; and iii ) fully denervated , if > 90% of the post-synaptic apparatus is devoid of pre-synaptic nerve terminal markers . Consistent with previous reports , a large increase in partially/fully denervated NMJs in Ctrl TA muscles was observed after SNT compared to sham surgery , suggesting incomplete regeneration of NMJs ( Figure 8B ) ( Williams et al . , 2009 ) . Examination of P7DTA NMJs revealed that SC depletion was associated with a significantly higher proportion of NMJs that remained partially/fully denervated 6 weeks after SNT surgery ( Figure 8B ) . 10 . 7554/eLife . 09221 . 013Figure 8 . Reductions in NMJ reinnervation , post-synaptic morphology , and post-synaptic myonuclei in SC depleted skeletal muscle . ( A ) Representative confocal IF images and 3-D Amira based reconstructions of Ctrl and P7DTA NMJs 6 weeks after sham or SNT surgery , stained for post-synaptic ( AChRs labeled with Btx , green ) , pre-synaptic markers ( SV2 , Syt-2 , neurofilament , red ) and myonuclei ( DAPI , blue ) . Post-synaptic myonuclei are indicated with asterisks . ( B ) Quantification of NMJ reinnervation: partially dennervated ( Part ) and fully denervated ( Full ) . ( C ) Quantification of degenerated NMJs based on post-synaptic morphology . ( D ) Percentage distribution of NMJs based on number of post-synaptic myonuclei . Scale bar = 10 μm . N = 4 mice , 20 NMJs/mouse . p < 0 . 05 compared to Ctrl and P7DTA sham , **p < 0 . 05 compared to Ctrl-sham , P7DTA-sham and Ctrl-SNT , ANOVA/Bonferroni multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 01310 . 7554/eLife . 09221 . 014Figure 8—figure supplement 1 . Loss of post-synaptic myonuclei is a feature of NMJ degeneration . ( A , B ) Quantification of Ctrl NMJ post-synaptic myonuclei: ( A ) Comparison between denervated and reinnervated NMJs; ( B ) Comparison between pretzel-like and plaque-like ( degenerated ) NMJs based on their post-synaptic AChR apparatus shape . DOI: http://dx . doi . org/10 . 7554/eLife . 09221 . 014 Post-natal NMJ maturation is associated with post-synaptic morphological transitions from a plaque-like to a pretzel-like shape due to the formation of post-synaptic membrane invaginations ( junctional folds ) ( Sanes and Lichtman , 1999; Wu et al . , 2010 ) . Aged and NMD-afflicted skeletal muscles are frequently associated with NMJ degeneration , manifested by loss of elaborate AChR-enriched branches and the characteristic pretzel-like morphology of mature adult NMJs ( Valdez et al . , 2010 , 2012 ) . Because post-synaptic morphology is linked to NMJ integrity and disease , we measured the occurrence of NMJ degeneration in Ctrl and P7DTA skeletal muscles after sham or SNT surgery . We considered an NMJ to be plaque-like , or degenerated , if the AChR-enriched area resembled a patch devoid of defined elaborate branches >5 μm in length . Very few degenerated NMJs were found in Ctrl or P7DTA TA muscles 6 weeks after sham surgery regardless of genotype ( Figure 8C ) . After SNT surgery there was a marked increase in the proportion of degenerated NMJs , a percentage that was substantially elevated by ∼ 25% with SC depletion ( Figure 8C ) . Enriched at most adult NMJs are clusters of post-synaptic myonuclei specialized for the expression of genes required for AChR consolidation and the differentiation of pre-synaptic nerve terminals ( Sanes and Lichtman , 1999; Hippenmeyer et al . , 2007; Zhang et al . , 2007; Mejat et al . , 2009 ) . Since SCs function as an essential source of myonuclei for skeletal muscle regeneration , local SC activity along denervated myofibers may be required for the maintenance of post-synaptic myonuclei ( Relaix and Zammit , 2012 ) . Therefore , we quantified the number of post-synaptic myonuclei ( >25% DAPI covered by the Btx + post-synaptic apparatus ) in TA muscles ( Grady et al . , 2005; Zhang et al . , 2007; Mejat et al . , 2009 ) . Assessment of Ctrl and P7DTA NMJs 6 weeks after sham surgery did not reveal any significant alteration in the distribution of NMJs based on post-synaptic myonuclei size ( Figure 8D ) . Also , no significant loss of post-synaptic myonuclei was found at Ctrl NMJs 6 weeks after SNT ( Figure 8D ) . However , partially/fully denervated or degenerated ( based on post-synaptic morphology , plaque-like ) Ctrl NMJs had significantly fewer post-synaptic myonuclei , indicating that loss of post-synaptic myonuclei is a feature of NMJ degeneration ( Figure 8—figure supplement 1 ) . Consistent with a higher proportion of degenerated NMJs , a significant shift towards smaller post-synaptic myonuclear size was observed at P7DTA NMJs 6 weeks after SNT ( Figure 8D ) . Collectively these data indicate that SC depletion , although not sufficient alone to trigger loss of NMJ integrity at homeostasis , did notably impair NMJ regeneration in response to neuromuscular disruption . In this report we interrogated the roles and fates of SCs in a model of neuromuscular regeneration . In doing so we found that upon NMJ disruption: i ) SC depletion exacerbates myofiber atrophy and type transitions; ii ) SC depletion leads to elevated fibrosis and declines in muscle force generating capacity; iii ) SC-derived contributions prevail in the vicinity of NMJs; and iv ) SC depletion leads to deficits in skeletal muscle reinnervation , reductions in post-synaptic morphology and loss of post-synaptic myonuclei . Through these findings we propose a cellular mechanism whereby SCs contribute to the regeneration of NMJs and skeletal muscle maintenance upon neuromuscular disruption . Based on the results presented here , an aspect of NMJ regeneration includes local SC activation , derived progenitor fusion and turnover of post-synaptic myonuclei . Although we did not observe any major alterations in whole muscle mass after SNT , SC depletion did lead to further declines in myofiber size and increased MCT accumulation . The connection between SC depletion and MCT accumulation or fibrosis has also been observed by other studies in the context of skeletal muscle regeneration , functional overload-induced hypertrophy and aging ( Murphy et al . , 2011; Fry et al . , 2014 , 2015 ) . A study by the Kardon group highlighted that MCT fibroblasts are dynamically regulated by SCs , and the positive feedback between SCs and fibroblasts ensures efficient and effective skeletal muscle regeneration ( Murphy et al . , 2011 ) . Results from the Peterson group revealed that SC depletion led to a fibrotic response in long-term overloaded muscles and elevated MCT accumulation in aged muscles , suggesting that SCs regulate the myofiber extracellular environment through inhibiting fibroblast function during skeletal muscle remodeling ( Fry et al . , 2014 , 2015; Lee et al . , 2015 ) . In addition , multiple populations of muscle interstitial cells ( MICs ) have been identified with fibro-adipogenic potential , such as Fibro-adiopogenic progenitors ( FAPs ) , PDGFRα+ MICs or ADAM12 + perivascular cells ( Joe et al . , 2010; Uezumi et al . , 2010; Dulauroy et al . , 2012; Malecova and Puri , 2012 ) . These cells may serve as functional niche components for SCs , and their cross-talk with SCs may regulate muscle regeneration and fibrosis associated with NMDs ( Malecova and Puri , 2012 ) . How the functional interactions between SCs , fibroblasts and MICs affect fibrosis either through cell contact or the release of soluble factors requires further study . Myofiber type transition and declines in force generation capacity are highly correlated with NMJ disruption ( Windisch et al . , 1998; Wu et al . , 2014 ) . Rodent models revealed that peripheral nerve lesions cause fast to slow myofiber transitions and a substantial increase in the proportion of type IIA myofibers ( Windisch et al . , 1998; Chakkalakal et al . , 2012 ) . A preferential loss of fast-type type IIB neuromuscular synapses has been reported in aged and NMD-afflicted muscles ( Frey et al . , 2000 ) . Consistent with previous studies , we showed an induction of type IIA and loss of type IIB myofibers after SNT . This SNT-induced increase of type IIA myofibers was aggravated upon SC depletion , in line with our later finding that more P7DTA NMJs remain denervated after SNT . Concomitantly , after SNT and SC depletion we observed a notable induction of hybrid myofibers , a typical feature of NMD-afflicted and aged skeletal muscle ( Klitgaard et al . , 1990; Krivickas et al . , 2002; Palencia et al . , 2005; Stevens et al . , 2008; Rowan et al . , 2012 ) . The mechanisms responsible for the occurrence of MyHC coexpression are still elusive . One possibility is induction of multiple MyHC isoforms following denervation and reinnervation by multiple motor neurons that specify distinct myofiber types ( Pette and Staron , 2000 ) . Our results also revealed that SC depletion led to declines in force generation capacity after neuromuscular disruption . Both accumulated MCT content and abnormal expression of contractile proteins could partially account for the loss of whole skeletal muscle force generation after SC depletion and SNT ( Baldwin and Haddad , 2001; Kjaer , 2004 ) . Also , impaired NMJ regeneration in response to neuromuscular disruption might contribute to the additional loss of force generating capacity in SC-depleted muscles compared to Ctrl after SNT . In support , declines in force generating capacity are a feature of partially denervated and aged skeletal muscles ( Urbanchek et al . , 2001; Kalliainen et al . , 2002 ) . Limited if any cellular turnover and no net loss of myonuclei has been reported in chronically denervated skeletal muscle ( Bruusgaard and Gundersen , 2008 ) . Therefore , it has been concluded that muscle mass loss in the context of denervation-induced atrophy primarily reflects imbalances between protein synthesis and degradation as opposed to cellular turnover ( Gundersen and Bruusgaard , 2008 ) . Our SNT surgery , however , leads to complete denervation and the eventual reinnervation of the vast majority of myofibers 4–6 weeks after surgery ( Buti et al . , 1996; Williams et al . , 2009 ) . Chronic denervation models do have greater levels of myofiber atrophy than observed with the SNT employed here ( Snow , 1983; Bruusgaard and Gundersen , 2008 ) . Although chronic denervation leads to SC activation , it is generally accepted that limited fusion of SC derived progenitors to atrophying parent myofibers occurs ( Borisov et al . , 2005; Bruusgaard and Gundersen , 2008; Gundersen and Bruusgaard , 2008 ) . Rather , SCs and derived progenitors tend to migrate into interstitial spaces where they undergo apoptosis , are believed to contribute to the formation of nascent myofibers with distinct basal lamina juxtaposed to parent atrophying myofibers , or contribute to other forms of failed myogenesis ( Borisov et al . , 2005; Bruusgaard and Gundersen , 2008 ) . It will be of interest to determine whether SCs are a source of nascent myofibers and failed myogenesis , and how this contributes to myofiber size changes during chronic denervation . Even though only a small percentage of SCs were found to activate after SNT , surprisingly we found that SC activation and SC-derived progenitor fusion primarily occurred in central portions of myofibers , in the vicinity of NMJs . Moreover , more denervated and degenerated ( based on post-synaptic morphology ) NMJs observed in SC depleted skeletal muscles after SNT were associated with significant declines in the number of post-synaptic myonuclei at NMJs . These phenotypes indicate the importance of regionalized SC activity for regenerating NMJs . Central nucleation was not a prominent feature of myofibers 4–6 weeks after SNT . Recently transient central nuclei were observed along the length of aging myofibers at random locations ( Li et al . , 2011 ) . Whether such transient degenerative regenerative events occur in the vicinity of NMJs in response to SNT as part of program of myogenic progenitor differentiation and fusion remains to be determined . The factors that control SC activity and derived progenitor fusion at central myofiber regions in the vicinity of NMJs during NMJ disruption are unknown . One possibility is the local expression of SC fate regulators that support progenitor activity and fusion from myofibers , the principal SC niche cell ( Bischoff , 1990; Yin et al . , 2013 ) . Alternatively , denervation may trigger the expression of factors that suppress SC activation and differentiation at extra-synaptic myofiber regions . Factors implicated in SC quiescence and activation include loss of Notch signaling and increased fibroblast growth factor ( FGF ) or hepatocyte growth factor ( HGF ) -induced receptor tyrosine kinase signaling ( Yin et al . , 2013 ) . TGFβ superfamily signaling , which is elevated in denervated skeletal muscle , is a well-established suppressor of myogenic differentiation ( Kollias and McDermott , 2008; Sartori et al . , 2014 ) . In addition , Jak/Stat and Wnt signaling are factors that regulate SC derived myogenic progenitor differentiation ( Yin et al . , 2013; Price et al . , 2014; Tierney et al . , 2014 ) . Whether any of these factors limit SC activation and differentiation in the vicinity of denervated NMJs will require molecular dissection of myofibers at distinct regions . Post-synaptic myonuclei are specialized for the expression of synapse-enriched genes required for the development , differentiation , consolidation and maintenance of both pre- and post-synaptic components ( Schaeffer et al . , 2001; Hippenmeyer et al . , 2007 ) . Consistent with vital roles , the etiology of some NMDs includes reductions in post-synaptic myonuclear number and integrity . Emery-Dreifuss muscular dystrophy is characterized by a loss of post-synaptic myonuclei together with deficits in skeletal muscle innervation , post-synaptic AChR morphology and the induction of gene expression programs consistent with denervation ( Mejat et al . , 2009 ) . A feature of Slow-channel syndrome , a congenital myasthenia disorder , includes apoptotic activity and the accumulation of DNA damage at post-synaptic myonuclei ( Zhu et al . , 2014 ) . Genetic studies have also shown the importance of post-synaptic myonuclei towards NMJ development . Mice null for Syne-1 , a nuclear anchoring protein , display loss of post-synaptic myonuclei together with gross deficits in the innervation of embryonic skeletal muscles ( Zhang et al . , 2007 ) . Therefore , reductions of post-synaptic myonuclei in SC-depleted skeletal muscles after SNT could manifest in the loss of gene expression programs required for the regeneration and maintenance of NMJs . Similar to what we observed with SC-depleted skeletal muscles after NMJ disruption , myofiber type transitions , decreases in NMJ regenerative capacity and force generation are also features of aged and NMD-afflicted skeletal muscles ( Frey et al . , 2000; Verdu et al . , 2000; Urbanchek et al . , 2001; Kalliainen et al . , 2002; Hegedus et al . , 2008; Rowan et al . , 2012; Kang and Lichtman , 2013 ) . Since loss of SC number and function is also associated with aging and NMDs , it will be of interest to determine in these contexts the interrelationship between reduced integrity of Pax7+ SCs impaired NMJ regeneration and correlated skeletal muscle dysfunction ( Pradat et al . , 2011; Chakkalakal et al . , 2012; Hayhurst et al . , 2012 ) . C57BL/6 , Pax7CreERT2 ( 017763 ) Rosa26mTmG ( 007576 ) and Rosa26DTA ( 009669 ) mice were obtained from Jackson Laboratories ( Bar Harbor , ME ) . Rosa26mTmG or Rosa26DTA mice were crossed with Pax7CreERT2 mice to generate Pax7CreER/+; Rosa26mTmG/+ ( P7mTmG ) or Pax7CreER/+; Rosa26DTA/+ ( P7DTA ) mice and control CreER negative ( Ctrl ) littermates . Transgenic mouse lines were used at 3–6 months of age . P7DTA and Ctrl mice were treated with ( 0 . 1 mg Tmx/g body weight ) for 5 consecutive days I . P . , 7 days after the first Tmx injection mice underwent SNT and sham surgeries . P7DTA and Ctrl mice were given additional Tmx 10 and 17 days after surgery . P7mTmG mice were similarly administered Tmx , however without additional Tmx injections after surgery . All animal procedures were conducted in accordance with institutional guidelines approved by the University Committee on Animal Recourses , University of Rochester Medical Center . Mice were anesthetized with intraperitoneal injections of ketamine ( 110 mg/kg ) and xylazine ( 10 mg/kg ) . The hindquarter was then carefully shaved and depilation completed with generic Nair hair removal cream prior to skin cleansing with gauze . The skin was incised 1 mm posterior and parallel to the femur , and the biceps femoris was bluntly split to expose the sciatic nerve . 1–2 mm sciatic nerve was then transected 5 mm proximal to its trifurcation , followed with realignment of the distal and proximal nerve ends and closure of the muscle with wound clips ( Autoclip , BD Clay Adams , Franklin Lakes , NJ ) . Mice were given analgesic ( 0 . 5–1 . 0 mg/kg buprenorphine ) and allowed to recover on a heating pad . Sham surgery was performed on the contralateral leg where procedures were performed without nerve transection . At the designated times left and right hind limb muscles were collected . The muscles used for histology were incubated at 4°C overnight in 30% sucrose solution and frozen in dry ice-cooled isopentane . Flash-frozen muscles were sectioned at 10 μm ( transverse ) or 30 μm ( longitudinal ) and stored at −80°C . Sections were fixed for 3 min in 4% paraformaldehyde ( PFA ) ( no PFA fixation for MyHC antibodies ) , and if needed , subjected to antigen retrieval: heating slides in citrate buffer ( 10 mM sodium citrate , pH 6 . 0 ) in a steamer ( Oster 6 . 1 quart , model 5712 , Racine , WI ) for 15 min followed by cooling at room temperature for 2 min ( Tang et al . , 2007 ) . Tissue sections were incubated with 0 . 2% Triton X-100 for 10 min , blocked in 10% normal goat serum ( NGS , Jackson ImmunoResearch , West Grove , PA ) 30 min at room temperature and stained with primary antibodies . If necessary ( when mouse primary antibodies were used ) , sections were blocked in 3% affinipure Fab fragment goat anti-mouse IgG ( H + L ) ( Jackson ImmunoResearch , West Grove , PA 115-007-003 ) with 2% NGS at room temperature for 1 hr . Primary antibodies were incubated at 4°C overnight or 2 hr at room temperature , and secondary antibodies were incubated for 1 hr at room temperature . DAPI ( Invitrogen , Carlsbad , CA ) staining was used to mark nuclei . All slides were mounted with Fluoromount-G ( SouthernBiotech , Birmingham , AL ) . For H & E staining , flash-frozen sections were fixed for 3 min in 4% PFA , stained with Mayers Hematoxylin and Alcoholic Eosin Y , dehydrated , equilibrated with xylene and mounted using Cytoseal 60 ( Richard-Allan Scientific , Kalamazoo , MI ) . For Sirius Red staining , a Picrosirius Red stain kit ( Polysciences , Warrington , PA ) was utilized . Briefly flash-frozen sections were fixed for 1 hr at 56°C in Bouin's fixative , washed in water , stained for 1 hr in Picrosirius Red , washed in 1 M HCl , dehydrated , equilibrated and mounted . Bright-field images were collected by a Zeiss Axioskop 40 microscope . Olympus VS110 virtual microscopy system was utilized for whole-slide scanning . Automatic quantification of MCT content was accomplished via VisioPharm software . Transverse sections and cells were imaged on a Zeiss Axio Observer A . 1 microscope . Longitudinal sections were stained with SV2 , Znp-1 , 2H3 , Btx and DAPI as described above and viewed with an Olympus Fluoview 1000 confocal microscope with 40X ( for quantification ) or 100X ( for representative pictures ) objectives at a 0 . 47 μm or 0 . 42 μm step size respectively . Amira software was used to analyze 3-D reconstructed NMJs for innervation analysis and to identify and enumerate post-synaptic myonuclei . Max-projection z-stack images of NMJs were generated with ImageJ software . The post-synaptic side was identified based on the entry of the terminal axon and as the concave side of the NMJ . For GFP localization , myofibers were purified by conventional collagenase digestion and trituration with fire polished glass pipets as previously described ( Zammit et al . , 2004 ) . Briefly , the EDL muscle was dissected , rinsed in Dulbecco's phosphate-buffered saline ( PBS ) , put into a 1 . 5 ml eppendorf tube containing 1 ml 0 . 1% type I collagenase ( Invitrogen ) and 0 . 1% type II collagenase ( Invitrogen ) in Dulbecco's modified Eagles medium ( DMEM , Sigma–Aldrich , St . Louis , MO ) , incubated in a shaker water bath at 37°C for 75 min and gently mixed by inversion periodically . Following digestion , the muscle was transferred to 100 mm × 15 mm plastic petri dishes containing 10 ml of plating media ( 10% horse serum in DMEM ) using fire-polished-tip Pasteur pipettes . Under a stereo dissecting microscope , single myofibers were released by gently triturating the EDL with a series of modified Pasteur pipettes that varied in tip diameter to accommodate the progressive decrease in muscle trunk size . Inseparable fibers and debris were removed . Purified single myofibers were fixed with 4% PFA for 3 min , washed with PBS and transferred to 5 ml polystyrene cell collection tubes for GFP , Pax7 and Btx IF . For assessing single myofiber size and myonuclear number , muscles were fixed in 4% PFA for 48 hr prior to dissection and NaOH mediated digestion ( Brack et al . , 2005 ) . Fixed muscles were incubated in 40% NaOH for 2 hr and agitated vigorously for 20 min . Released fibers were then washed in PBS and processed for DAPI staining . To assess cell proliferation C57BL/6 mice were fed BrdU ( Sigma–Aldrich ) ( 0 . 5 mg/ml supplemented with 5% sucrose ) in drinking water after denervation ( Chakkalakal et al . , 2012 ) . Muscles were collected 4 weeks after SNT and sectioned for Pax7 , BrdU and Laminin IF . Whole EDL muscle contractility and force generation were analyzed using an ASI muscle contraction system ( Aurora Scientific , Aurora , Canada ) as described previously ( Wei-Lapierre et al . , 2013 ) . Briefly , mice were anaesthetized and TA muscles removed . EDLs were then carefully isolated , adjusted to optimal length and stimulated at various frequencies to obtain absolute force values . To obtain specific force values , absolute force was normalized to pennation angle and cross-sectional area ( determined by EDL weight and length ) . Muscle force was recorded and analyzed using Dynamic Muscle Control , Clamp fit and Graph Pad Prism software . Pax7 ( mouse IgG1 , 1:100 , Developmental Studies Hybridoma Bank ( DSHB ) , Iowa City , IA ) , BrdU ( rat , 1:250 , Abcam ab6326 , Cambridge , UK ) , Laminin ( rat or rabbit , 1:1500 , Sigma–Aldrich L0663 or L9393 ) , GFP ( rabbit , 1:400 , Millipore AB3030P , Billerica , MA ) , F1 . 652 ( mouse IgG1 , 1:40 , DSHB ) , A4 . 840 ( mouse IgM , 1:40 , DSHB ) , NCL-MHC ( mouse IgG1 , 1:100 , DSHB ) , SC-71 ( mouse IgG1 , 1:40 , DSHB ) , 6H1 ( mouse IgM , 1:40 , DSHB ) , BF-35 ( mouse IgG1 , 1:40 , DSHB ) , BF-F3 ( mouse IgG1 , 1:40 , DSHB ) , SV2 ( synaptic vesicle protein-2 , mouse IgG1 , 1:100 , DSHB ) , Znp-1 ( synaptotagmin-2 , mouse IgG1 , 1:200 , DSHB ) and 2H3 ( neurofilament , mouse IgG1 , 1:200 , DSHB ) , AlexaFluor 488-conjugated α-Bungarotoxin ( 1:1000 , Life Technologies B-13422 , Grand Island , NY ) , AlexaFluor 647-conjugated α-Bungarotoxin ( 1:1000 , Life Technologies B-35450 ) , AlexaFluor 594-conjugated goat anti-mouse IgG ( 1:1500 , Life Technologies A-11032 ) , AlexaFluor 594-conjugated goat anti-mouse IgG1 ( 1:1500 , Life Technologies A-21125 ) , AlexaFluor 488-conjugated goat anti-mouse IgM ( 1:1500 , Life Technologies A-21042 ) , AlexaFluor 488-conjugated goat anti-rat IgG ( 1:1500 , Life Technologies A-11006 ) , AlexaFluor 647-conjugated goat anti-rabbit ( 1:1500 , Life Technologies A-21244 ) . Results are presented as mean +SEM . Statistical significance was determined by Student's t-tests for simple comparison , one-way ANOVA and Bonferroni multiple comparisons test for multiple comparisons with Graph Pad Prism software . p < 0 . 05 was considered as statistically significant .	New muscle fibers are made throughout our lives to replace those that have been damaged by normal wear and tear , and to meet new physical demands . These new muscle fibers develop from a pool of muscle stem cells . To create and maintain fully working muscles , nerve cells called motor neurons must also properly attach to the muscle fibers . These nerve cells transmit messages from the brain that tell the muscles what to do . If the muscle-nerve connections do not form correctly , or are severed , muscles can waste away . This may occur as part of a neuromuscular disease , and also happens to some extent as a normal part of aging . It was thought that muscle stem cells do not affect how the muscle-nerve connections form . By studying genetically engineered mice , Liu et al . now show that this is not the case . These mice had modifications to their muscle stem cells that allowed the number of these cells to be artificially reduced , and some cells also produced a fluorescent protein that allowed them to be tracked . Surgically severing some of the muscle-nerve connections in the mice triggered the rebuilding of the connections , but also weakened the muscles and caused some disease-related changes in the muscle tissue . During the healing process , the muscle stem cells are active near the regenerating connections . Reducing the number of muscle stem cells in the mice while these broken connections were healing further weakened the muscles . Closer inspection of the muscle-nerve connections also revealed poorer quality connections were formed in the stem-cell deficient mice . Further study of how stem cells help to form strong nerve-muscle connections may allow scientists to develop new treatments for age- or disease-related muscle loss .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "cell", "biology" ]	2015	Inducible depletion of adult skeletal muscle stem cells impairs the regeneration of neuromuscular junctions
Opioids , agonists of µ-opioid receptors ( µORs ) , are the strongest pain killers clinically available . Their action includes a strong central component , which also causes important adverse effects . However , µORs are also found on the peripheral endings of nociceptors and their activation there produces meaningful analgesia . The cellular mechanisms downstream of peripheral µORs are not well understood . Here , we show in neurons of murine dorsal root ganglia that pro-nociceptive TRPM3 channels , present in the peripheral parts of nociceptors , are strongly inhibited by µOR activation , much more than other TRP channels in the same compartment , like TRPV1 and TRPA1 . Inhibition of TRPM3 channels occurs via a short signaling cascade involving Gβγ proteins , which form a complex with TRPM3 . Accordingly , activation of peripheral µORs in vivo strongly attenuates TRPM3-dependent pain . Our data establish TRPM3 inhibition as important consequence of peripheral µOR activation indicating that pharmacologically antagonizing TRPM3 may be a useful analgesic strategy . Throughout the peripheral and central parts of the nociceptive system , µ-opioid receptors ( µORs ) are widely expressed and strongly control neuronal excitation ( Stein , 2016 ) . Agonists of µORs are the most potent analgesic drugs clinically available ( Pasternak and Pan , 2013 ) and are therefore often prescribed for the treatment of severe pain . These opioid substances are especially effective against acute pain states , such as post-operative pain , but they are also used , more controversially , for the treatment of longer lasting or chronic pain ( Rowbotham et al . , 2003; Chou et al . , 2015 ) . Much of the controversy around opioids arises because these substances cause important unwanted effects , such as addiction , tolerance ( Volkow and McLellan , 2016 ) , opioid-induced hyperalgesia ( Roeckel et al . , 2016 ) and , when overdosed , respiratory depression ( Pattinson , 2008 ) . Because of this unfavorable profile of unwanted effects , clinically used opioids are often implicated in fatal overdosing due to drug addiction or dosing accidents ( Compton et al . , 2016; Ray et al . , 2016 ) . While many actions of opioids are triggered by activation of µORs in the central nervous system , opioid receptors are also located on the peripheral nerve endings of nociceptor neurons ( Stein et al . , 1990a , 1990b; Stein , 2013 ) . Physiologically , in the skin , where many peripheral nociceptor nerve endings reside , opioid receptors are targeted by endogenous opioid substances , such as β-endorphin , released in the periphery from immune cells ( Stein et al . , 1990b ) or skin keratinocytes ( Ibrahim et al . , 2005; Fell et al . , 2014 ) . Activation of peripheral opioid receptors can provide clinically meaningful analgesia ( Farley , 2011; Stein and Machelska , 2011 ) . On the contrary , inhibiting peripheral µORs by antagonist application increases pain ( Jagla et al . , 2014 ) . Targeting peripheral µORs thus has been proposed as a strategy to provide analgesia with reduced adverse effects and an improved safety profile ( Stein et al . , 2003 ) . An alternative strategy , in which not the µORs themselves but downstream effectors of µOR signaling pathways are targeted , may also prove to be beneficial . However , such strategies have received less attention , partly because the downstream targets of peripheral µOR signaling are not well documented . At central synapses , several intracellular mechanisms leading to reduced neuronal excitation during µOR activation have been worked out in considerable detail . Activation of µORs causes inhibition of L-type voltage-gated Ca2+ channels ( Bourinet et al . , 1996; Heinke et al . , 2011 ) and activation of several types of K+ channels ( Law et al . , 2000; Marker et al . , 2005 ) . Importantly , also at several central synapses , µORs have been shown to interact with TRP channels , notably TRPV1 ( recently reviewed by Bao et al . , 2015 ) . The interaction of TRPV1 and µOR occurs at multiple anatomical locations in the central nervous system and its physiological consequences and mechanisms are highly diverse ( Maione et al . , 2009; Stein , 2016 ) . Proposed mechanisms range from influencing the cAMP-PKA pathway and the participation of several other kinases ( ERK , MAPK , JNK ) to β-arrestin2 and the PI3-kinase pathway ( Law et al . , 2000; Rowan et al . , 2014; Bao et al . , 2015 ) . In peripheral nerves and their endings , several , partially different , intracellular signaling events have been proposed to account for the effectiveness of peripherally restricted opioids , including inhibition of voltage-gated Na+ channels ( Gold and Levine , 1996 ) , inhibition of HCN channels ( Ingram and Williams , 1994 ) , activation of several classes of K+ channels ( Cunha et al . , 2010; Nockemann et al . , 2013; Baillie et al . , 2015 ) , and , importantly , again the inhibition of TRPV1 channels ( Vetter et al . , 2006; Endres-Becker et al . , 2007; Spahn et al . , 2013 ) However , despite the plethora of proposed targets , the magnitude and the interplay of these individual effects has not been determined . Furthermore , many of these ion channels are not inhibited by µOR activation per se , but rather their upregulation or sensitization by the cAMP/PKA pathway is blocked by µOR signaling . It is therefore unclear , how much each of these targets contributes to the overall effect of opioids on peripheral nerve endings , and how this contribution may vary during shifts from resting to inflamed states ( and back ) . Further , it is unclear , and perhaps even unlikely , that all important downstream targets of µORs have already been identified . Recently , TRPM3 channels ( Oberwinkler and Philipp , 2014 ) were described in primary nociceptive neurons ( Nealen et al . , 2003; Lechner et al . , 2009; Vriens et al . , 2011; Straub et al . , 2013a , 2013b; Held et al . , 2015; Usoskin et al . , 2015 ) and it was shown that these divalent-permeable cation channels ( Grimm et al . , 2003; Oberwinkler et al . , 2005; Wagner et al . , 2010 ) play an important role in noxious heat sensation since these channels are intrinsically thermosensitive ( Vriens et al . , 2011 ) . Activation of TRPM3 leads to release of pro-inflammatory CGRP from peripheral nociceptor nerve terminals ( Held et al . , 2015 ) , while TRPM3-deficient animals show severe defects in the development of inflammatory hyperalgesia ( Vriens et al . , 2011 ) . In this context , it is interesting that inhibitors of TRPM3 channels have been identified that exhibit strong anti-nociceptive properties ( Straub et al . , 2013b; Chen et al . , 2014; Suzuki et al . , 2016; Krügel et al . , 2017 ) . Hence , TRPM3 channels in peripheral nociceptors have pro-nociceptive and pro-inflammatory properties , making them an interesting target to study the mechanism of peripheral nociception and inflammation . Here , we demonstrate that TRPM3 channels in primary nociceptive neurons are rapidly and strongly inhibited by agonists of µORs such as morphine , through an intracellular signaling cascade relaying µORs to TRPM3 channels via Gβγ proteins . We show that local activation of peripheral µORs causes strong analgesia of TRPM3-dependent pain . Our results indicate that it may be worthwhile to further investigate TRPM3 channels as potential targets for the treatment of pain with reduced adverse effects in the central nervous system . TRPM3 channels possess pro-nociceptive properties in DRG neurons ( Vriens et al . , 2011; Straub et al . , 2013a , 2013b; Chen et al . , 2014; Held et al . , 2015 ) . We therefore investigated clinically used analgesic drugs for their effects on TRPM3 channels . To address this question , we first activated TRPM3 channels in isolated mouse DRG neurons by applying the known TRPM3 agonist pregnenolone sulfate ( PS ) ( Wagner et al . , 2008; Vriens et al . , 2011 ) . Under our recording conditions , the ensuing Ca2+ signals were almost completely dependent on the presence of TRPM3 proteins , as demonstrated in control experiments with neurons prepared from TRPM3-deficient ( TRPM3-knockout ) mice ( Figure 1—figure supplement 1 ) . Testing whether morphine , a traditional agonist of opioid receptors , or its synthetic , µOR-specific analog DAMGO ( Handa et al . , 1981 ) , influences TRPM3 channel activity , we found that both µOR agonists strongly reduced the Ca2+ signals induced by PS ( Figure 1a–c ) . The reduction of these Ca2+ signals was rapid and readily reversible within a time frame of 1–2 min ( Figure 1d ) . We ascertained that DAMGO acted via activating opioid receptors by co-applying the opioid receptor antagonist naloxone , which unspecifically blocks several opioid receptor subtypes ( Figure 1e–g ) . Naloxone alone did not affect TRPM3-dependent Ca2+ signals in DRG neurons ( Figure 1e , g ) , but , when co-applied with DAMGO , it prevented the action of DAMGO completely ( Figure 1f , g ) . In addition , we tested two further , structurally different , µOR agonists , herkinorin and loperamide . Both substances strongly inhibited PS-induced Ca2+ signals in isolated mouse DRG neurons ( Figure 1—figure supplement 2 ) . These data indicate that activation of µORs , independent of the chemical nature of the agonist , leads to inhibition of TRPM3 channels in DRG neurons . We successfully reconstituted the signaling pathway from µORs to TRPM3 channels in a heterologous overexpression system by co-expressing µORs and TRPM3 in HEK cells ( Figure 1h ) . Importantly , omitting µOR cDNA from the transfection abolished the effect of morphine ( Figure 1h ) and DAMGO ( as seen , for example , in Figure 6f ) , providing strong evidence that these agonists act through µORs and do not interfere with TRPM3 channel activity directly . Like many members of the TRP superfamily , TRPM3 channels are activated by several different agonists ( Oberwinkler and Philipp , 2014 ) . Taking advantage of the overexpression system , which avoids the complication of other nifedipine- or heat-sensitive conductances ( Fajardo et al . , 2008; Julius , 2013 ) , we demonstrated that TRPM3 channels activated by the application of nifedipine ( Wagner et al . , 2008; Drews et al . , 2014 ) or by heat applied in the absence of chemical agonists ( Vriens et al . , 2011 ) were also inhibited by µOR activation ( Figure 1i , j ) . Whole-cell patch-clamp electrophysiological recordings from HEK cells overexpressing TRPM3 and µORs demonstrated that currents through TRPM3 channels were inhibited by activating µORs rapidly in less than 5 s ( Figure 1k–m , Figure 1—figure supplement 3 ) . After measuring a dose-response curve by varying the concentration of DAMGO , we found that the IC50 values at +80 mV and −80 mV did not differ , providing an indication that inhibition of TRPM3 channels by µOR activation was voltage-independent ( Figure 1—figure supplement 4 ) . Similarly , performing patch-clamp experiments on isolated DRG neurons , we observed that the currents evoked by TRPM3 agonists were rapidly ( less than 5 s ) inhibited by µOR activation ( Figure 2a–c ) . Upon washout of DAMGO , the inhibition was at least partially reversible . Activation of opioid receptors causes many cellular responses in DRG neurons ( Law et al . , 2000 ) , including the activation of K+ channels ( Yoshimura and North , 1983; Ocaña et al . , 2004 ) . Importantly , stimulation of K+ channels might reduce Ca2+ signals in DRG neurons through hyperpolarization-induced closure of voltage-gated Ca2+ channels . Because K+ channels cannot directly be studied in Ca2+ imaging experiments , we continued to use the whole-cell patch-clamp technique on isolated DRG neurons . Aiming for small-diameter nociceptors , we selected cells that responded either to TRPV1- or to TRPM3-activating ( or to both types of ) agonists . In these cells , we could not detect outward currents upon application of the µOR agonist DAMGO ( Figure 2d–f ) . This indicates that the reduction of PS-induced currents by µOR activation is not due to the activation of outward currents . Consequently , the DAMGO-induced reduction of PS-evoked Ca2+ signals observed in Ca2+ imaging experiments ( e . g . in Figure 1a–g ) is unlikely to be caused by K+ channel-dependent hyperpolarization . Together , these data argue that TRPM3 channels in DRG neurons are inhibited by activating µORs . Our patch-clamp data of PS-sensitive DRG neurons indicated that the inhibition of TRPM3-dependent currents in DRG neurons was variable ( Figure 2c ) . We therefore measured a large number of DRG neurons in Ca2+ imaging experiments ( Figure 2g–i ) and found equally that not in all PS-sensitive neurons activation of µORs caused a reduction of Ca2+ signals . When we categorized these neurons , the DAMGO-reactive subset amounted to 91% ( 562 cells ) in a total of 614 PS-sensitive cells analyzed ( Figure 2i ) . We tested whether TRPM3 inhibition by µOR activation correlated with the presence of functional TRPV1 channels in these cells , the rational of this being that in adult mice , TRPV1 channels are restricted to peptidergic nociceptors ( Cavanaugh et al . , 2011 ) , in which µORs are also preferentially expressed ( Vetter et al . , 2006; Scherrer et al . , 2009 ) . In agreement with this , we found that most neurons ( 465 of 562 cells , 83% ) , which showed a reduction of TRPM3-dependent Ca2+ signals due to application of DAMGO , also showed capsaicin-induced Ca2+ signals , demonstrating the presence of functional TRPV1 channels . Conversely , only a minority of the cells without apparent effect of DAMGO on TRPM3-dependent Ca2+ signals ( 14 of 52 DAMGO-insensitive cells , i . e . 27% of the DAMGO-insensitive cells , corresponding to 2% of all PS-sensitive cells ) showed a response to capsaicin ( Figure 2i ) . These data are corroborated by the size distribution of both , DAMGO-sensitive and DAMGO-insensitive , subgroups of PS-sensitive cells ( Figure 2—figure supplement 1 ) . PS-sensitive cells of both subgroups showed the typical distribution of small-diameter C- and Aδ-type neurons ( Vriens et al . , 2011; Tan and McNaughton , 2016 ) , with only very few larger cells ( diameter >31 µm: 26 of 614 cells , equivalent to 4 . 2% ) . On the other hand , the group of neurons that responded neither to PS nor to capsaicin appeared to contain a somewhat higher number of larger cells ( diameter >31 µm: 146 of 1673 cells , equivalent to 8 . 9%; Figure 2—figure supplement 1c , d ) . Together , these data show that activation of µORs is capable of rapidly and strongly reducing currents through TRPM3 channels in a subset of DRG neurons , corresponding largely , but not exclusively , to putative peptidergic nociceptors . DRG neurons express a host of different G-protein-coupled receptors ( GPCRs ) ( Gold and Gebhart , 2010; Veldhuis et al . , 2015 ) , many of which have been implied in regulating the excitability of these neurons . µORs couple preferentially to members of the Gαi subfamily , i . e . Gi/o proteins ( Law et al . , 2000 ) . In general , GPCRs that couple to Gi/o proteins have been implied in dampening the activity of peripheral nociceptors . We therefore tested other GPCRs that couple to Gαi-containing G proteins and found that all of them inhibited TRPM3-dependent Ca2+ signals in DRG neurons , although the number of neurons responding to the specific agonists was variable ( Figure 3 ) . Specifically , we found that TRPM3 was inhibited by activating GABAB receptors , CB2 cannabinoid receptors , δ-opioid receptors , adrenoreceptors and somatostatin receptors ( Figure 3 ) . When we activated metabotropic glutamate type five receptors ( mGluR5 ) , which preferentially couple to Gαq proteins ( Bhave et al . , 2001 ) we observed the weakest inhibition of all agonists , which was only detectable in a relatively small number of neurons . In addition , the weak inhibition caused by DHPG did not recover upon washout ( Figure 3f ) . These data show that the activity of TRPM3 channels can be inhibited by a large variety of Gαi-coupled GPCRs . Apart from TRPM3 , other TRP channels are involved in temperature sensing and inflammatory processes in peripheral somatosensory neurons ( Julius , 2013 ) . We tested therefore whether other TRP channels prominently expressed in DRG neurons were inhibited by activation of µORs , by using the same experimental conditions as on TRPM3 channels . Ca2+ signals induced by the TRPV1 agonist capsaicin showed the same reduction in activity whether DAMGO or the solvent control ( DMSO ) was applied to the DRG neurons ( Figure 4a , b ) . Furthermore , when we used a different protocol and co-applied DAMGO together with longer applied capsaicin , there was a much weaker inhibition compared to the inhibition typically observed with TRPM3 ( Figure 4—figure supplement 1 ) . Notably , we did not observe any difference when we differentiated between PS-sensitive and PS-insensitive neurons ( Figure 4—figure supplement 1a , green vs red trace ) , although 91% of the PS-sensitive neurons are expected to possess functional µORs ( see Figure 2i ) . Additionally , the capsaicin-induced Ca2+ signals did not recover after DAMGO washout ( Figure 4—figure supplement 1a ) , again in marked contrast to the behavior of TRPM3 channels . We then investigated the sensitivity of TRPA1 channels on µOR activation in DRG neurons . Because TRPA1 channels are difficult to stimulate several times , we used the second type of protocol . Similarly to our results with TRPV1 , we observed that AITC-induced Ca2+ signals were not subject to inhibition by DAMGO , regardless of whether the neurons were PS-sensitive or not ( Figure 4c , d ) . When we overexpressed TRPV1 channels together with µORs in HEK cells , we found no evidence for any inhibitory action of µORs on TRPV1 channels in the time frame investigated ( 2 min; Figure 4e ) . Similarly , we investigated the coupling of µORs to TRPA1 and TRPM8 channels in overexpression systems , again without finding any indication of inhibitory action of µORs on these channels ( Figure 4f , g ) . In summary , from the group of TRP channels implicated in noxious thermoreception and inflammatory hyperalgesia , only TRPM3 channels displayed a substantial , rapid and reversible inhibition during µOR activation . Usually µORs couple to Gi/o proteins , but also other signaling pathways have been described for these receptors ( Law et al . , 2000 ) . We tested whether Gi/o proteins were involved in the functional coupling of µORs to TRPM3 channels by incubating cultured DRG neurons for 16–24 hr with pertussis toxin ( PTX ) , which selectively disrupts signaling via Gi/o proteins ( Mangmool and Kurose , 2011 ) . We found that this treatment strongly reduced the action of DAMGO on PS-evoked activity in DRG neurons ( Figure 5a , b ) . Moreover , in HEK cells overexpressing µORs and TRPM3 proteins , PTX treatment almost completely abrogated the DAMGO-induced inhibition of TRPM3 channels ( Figure 5c , d ) . These data strongly implicate the classical signaling pathway comprising G proteins containing Gαi/o subunits in the inhibition of TRPM3 . Activated Gαi proteins reduce the activity of adenylyl cyclases and may thereby lower the concentration of cytosolic cAMP . We tested whether this process is required for inhibiting TRPM3 channels . We found that application of IBMX ( to inhibit cAMP-degrading phosphodiesterases ) and forskolin ( to stimulate cAMP-producing adenylyl cyclases ) did not influence TRPM3 channel activity in DRG neurons nor their inhibition by µOR activation ( Figure 5e , f ) . We observed the same outcome in TRPM3- and µOR-overexpressing HEK cells after combined application of forskolin and IBMX ( Figure 5g , h ) . Changing concentrations of cAMP might influence the activity of protein kinase A ( PKA ) . However , when we dialyzed TRPM3- and µOR-expressing HEK cells with the unhydrolyzable ATP analog AMP-PNP through the pipette in patch-clamp experiments , effectively blocking the action of all kinases , including PKAs , we found no evidence that this treatment affected the inhibitory action of µOR activation on TRPM3 channels ( Figure 5i , k ) . On the other hand , when we removed Mg2+ ions from the intracellular solution , the inhibiting action of DAMGO on TRPM3 channel activity was lost after 5 min of internally dialyzing the cells via the patch pipette ( Figure 5j , k and Figure 5—figure supplement 1a ) , showing that under our recording conditions the exchange between pipette solution and the cytosol was sufficiently rapid . These data also support our earlier conclusion that µOR-mediated TRPM3 inhibition relies on G-protein signaling as G proteins need Mg2+ ions to bind GTP ( Gilman , 1987 ) . We made sure that the series resistance during the patch-clamp recordings was not significantly different between the experimental groups ( Figure 5—figure supplement 1b ) . Therefore , these experiments suggest that kinases are not involved in the signaling pathway of µORs to TRPM3 channels . This conclusion was further corroborated by experiments using kinase inhibitors ( H89 , a relatively non-specific kinase inhibitor ( Lochner and Moolman , 2006 ) ; KT5720 , inhibitor of PKA; BIM , inhibitor of PKC ) , none of which abolished the inhibition of TRPM3 channels by µOR activation ( Figure 6 ) . However , some of these pharmacological kinase inhibitors , in particular H89 and KT5720 , had unexpected and potentially unspecific effects on the measured Ca2+ signals and on TRPM3 activation ( Figure 6 and Figure 6—figure supplement 1 ) . We did not further investigate these effects . Taken together , these data ( Figures 5 and 6 , Figure 5—figure supplement 1 ) strongly indicate that TRPM3 inhibition after µOR activation is a G-protein-coupled process , but does not involve the second messenger cAMP or downstream kinases . Heterotrimeric G proteins consist of α and dimeric βγ subunits and both of them can act as intracellular messengers ( Wettschureck and Offermanns , 2005 ) . We overexpressed Gα or Gβγ subunits together with TRPM3 channels in HEK cells and assessed their activity by applying PS . Overexpression of each of the three Gαi proteins , either in wild-type form ( Figure 7a ) , or as constitutively active QtoL mutant ( Graziano and Gilman , 1989 ) of Gαi1 ( Figure 7b ) had no inhibitory effect on the activity of TRPM3 channels . We also tested YFP-tagged Gαi subunits ( Figure 7—figure supplement 1a ) , because these allowed to easily assess the expression of the proteins by fluorescence microscopy . These tagged Gαi subunits also failed to inhibit TRPM3 channels . Additionally , we assessed the expression of Gαi3 and the YFP-tagged Gαi subunits by Western blotting and found them to be well expressed in HEK cells ( Figure 8—figure supplement 3a ) . Since the proteins were detected in these blots at the location of their expected size and since we did not detect obvious signs of proteolysis , we concluded that Gαi subunits do not inhibit TRPM3 activity . Equally to the Gαi subunits , overexpression of Gαo1 ( Figure 7c ) or Gαo2 ( Figure 7d ) did not affect PS-induced TRPM3 channel activity . Finally , also overexpression of Gαq proteins , which recently have been shown to influence TRPM8 channels ( Zhang et al . , 2012 ) , did not reduce the TRPM3-evoked Ca2+ signals ( Figure 7—figure supplement 1b ) . However , when we overexpressed Gβ1 and Gγ2 proteins together with TRPM3 channels , TRPM3-dependent Ca2+ signals were strongly reduced ( Figure 7e ) . Importantly , the reduction of TRPM3 activity by Gβγ subunits was attenuated by the additional overexpression of ( myristoylated ) myr-βARKct ( Koch et al . , 1994; Rishal et al . , 2005 ) , a Gβγ-binding peptide , indicating that the inhibition of TRPM3 channels was caused specifically by the overexpressed Gβγ subunits in an unbound state ( Figure 7f , g ) . A suppression of TRPM3 activity by overexpressed Gβγ subunits could also be demonstrated in whole-cell patch-clamp recordings of transfected HEK cells ( Figure 7h , i ) . We next tried to increase the concentration of free Gβγ subunits with a different approach that does not necessitate transfections and overexpression of exogenous proteins . The peptide mSIRK , when applied extracellularly , enters the cells and induces the dissociation of heterotrimeric G proteins without inducing GDP/GTP exchange in the Gα subunits ( Goubaeva et al . , 2003 ) . Using this approach , we observed a significant reduction in TRPM3 activity in DRG neurons as well as in TRPM3 overexpressing HEK cells ( Figure 7—figure supplement 3 ) . The inactive analog mSIRK-L9A did not cause this effect . However , mSIRK also induced an increased Ca2+ concentration in the cells at baseline , which was especially prominent in DRG neurons . It is therefore unclear if the reduction in TRPM3 activity was caused by the increased concentration of free Gβγ subunits , or if it was caused more indirectly by the increased free Ca2+ concentration . When we tried to manipulate the concentration of free Gβγ subunits in the opposite direction by overexpression of Gβγ-binding peptides , myr-βARKct as before or myr-phosducin ( Schulz , 2001; Rishal et al . , 2005 ) , we found that the inhibitory signaling from µORs to TRPM3 channels was severely reduced ( Figure 7—figure supplement 2 ) . In aggregate , these data strongly suggest a model in which TRPM3 channels are inhibited by free Gβγ dimers , but not by Gαi/o subunits . Interestingly , in the experiments where non-active Gαi/o subunits were overexpressed ( Figure 7a–d , Figure 7—figure supplement 1a ) , we consistently observed an increase of the resting Ca2+ concentration . Possibly , this increase is due to an increased basal activity of TRPM3 channels caused by binding and thereby scavenging of free Gβγ subunits by the overexpressed Gα subunits . To test this speculation , further experimental work will be necessary . To examine whether TRPM3 proteins and Gβγ subunits form a protein complex , we immunoprecipitated TRPM3 proteins equipped with a C-terminal YFP tag with anti-GFP antibodies ( which also recognize YFP ) . In control experiments , we determined that fused tags ( YFP or myc ) on TRPM3 proteins do not interfere with the functional coupling of µORs to TRPM3 channels , irrespective of their N- or C-terminal location ( Figure 8—figure supplement 1 ) . After separating the precipitated proteins , we observed a single band in western blots at the molecular weight expected for myc-TRPM3-YFP using antibodies against GFP , indicating that our procedure indeed precipitated full-length TRPM3 proteins ( Figure 8a ) . When we probed blots with antibodies against Gβ subunits , we found evidence for Gβ co-precipitating with TRPM3 from HEK-TRPM3 cells . Using the same protocol on HEK cells without TRPM3 did not lead to precipitation of Gβ proteins ( Figure 8b , c , Figure 8—figure supplement 2 ) . We further tested the ability of the anti-Gβ antibodies to recognize Gβ proteins by overexpressing Gβ proteins with a FLAG tag and therefore increased molecular weight and successfully detected the predicted bands in Western blots ( Figure 8—figure supplement 3b ) . Finally , in order to ascertain the identity of Gβ proteins interacting with TRPM3 without the potentially confounding issue of antibody specificity , we identified in mass spectrometry experiments peptide fragments belonging unequivocally to Gβ1 proteins . These peptides originated from trypsin-digested SDS gels containing proteins co-immunoprecipitating with TRPM3 from HEK-TRPM3 cells ( Figure 9 ) . Although Gαi3 subunits were abundantly present in the cell lysates , we could not detect them on western blots of proteins co-precipitating with TRPM3 ( Figure 8d; see also Figure 8—figure supplement 3a for tests of the antibodies against Gαi3 ) . Because TRPM8 proteins have recently been shown to interact directly with Gαq proteins ( Zhang et al . , 2012 ) , we also tested for , but failed to detect Gαq proteins co-precipitating with TRPM3 ( Figure 8e ) . These data show that TRPM3 and Gβ proteins are present in the same protein complexes . This provides a mechanistic explanation for the inhibitory action of µORs exerted on TRPM3 channels . We injected either PS , which has been shown to evoke nocifensive behavior and thus pain in a strictly TRPM3-dependent manner ( Vriens et al . , 2011; Straub et al . , 2013a ) , or capsaicin to evoke TRPV1-dependent pain into the hind paws of mice and observed the duration of the ensuing nocifensive behavior . Co-injection of DAMGO to activate peripheral µORs ( Stein and Lang , 2009 ) strongly reduced the nocifensive behavior evoked by PS injection , but did not affect the TRPV1-dependent pain after capsaicin injection ( Figure 10a , b ) . Injecting higher concentrations of capsaicin into the hind paw produced considerably longer responses , showing that the lower capsaicin dose used in Figure 10b did not saturate the pain-evoked behavior under our conditions ( Figure 10c ) . Again , co-injecting DAMGO was ineffective in reducing the duration of the capsaicin-induced nocifensive behavior at these higher capsaicin doses ( Figure 10c ) . These data show that locally , in the peripheral skin of living mice , TRPM3 channels , much more than TRPV1 channels , are under rapid inhibitory control of µORs . These results add TRPM3 channels to the category of TRP channels expressed in primary somatosensory neurons that are under tight regulatory control of GPCRs and their associated signaling cascades ( Veldhuis et al . , 2015 ) . The only other intracellular regulatory mechanism affecting TRPM3 channels uncovered so far is the hydrolysis of PIP2 ( Badheka et al . , 2015; Tóth et al . , 2015; Uchida et al . , 2016 ) , which typically is brought about by activating Gαq-coupled receptors . In the panel of receptors that we tested here , we also included a Gαq-coupled receptor ( mGluR5 ) , for which we found that it induces only weak and inconsistent inhibition of TRPM3 that did not recover ( Figure 3f–i , Figure 3—figure supplement 1f ) . There are several reasons possible why mGluR5 seems not to couple strongly to TRPM3 , but it should be noted that DRG neurons express many more Gαq-coupled receptors , for which these results cannot necessarily be extrapolated . It is still largely unexplored and therefore entirely unclear whether PIP2 metabolism induced by Gαq-coupled receptor activation is physiologically relevant for the regulation of TRPM3 channels in primary nociceptor neurons . Mechanistically , the combined data from pharmacological experiments , overexpression studies and biochemical interaction analyses presented here strongly suggest a model in which TRPM3 channel complexes are inhibited by directly binding to Gβγ dimers , but not to Gα subunits . The key evidence for this assertion is that ( 1 ) overexpression of Gβγ , but not Gα proteins inhibits TRPM3 channels ( Figure 7 ) , that ( 2 ) TRPM3 proteins are found in the same protein complexes as Gβ proteins ( Figures 8 and 9 ) and that ( 3 ) alternative pathways involving cAMP and protein kinases could be ruled out ( Figure 5e–k and Figure 6 ) . Still other pathways that are entirely independent of heterotrimeric G proteins also cannot account for our findings , because sensitivity to PTX strongly implicates G proteins from the Gαi subfamily ( Figure 5a–d ) . Binding to Gβγ is the mechanism with which µORs affect many other ion channels , like voltage-gated Ca2+ and GIRK channels ( Bourinet et al . , 1996; Law et al . , 2000; Marker et al . , 2005; Heinke et al . , 2011 ) . In the family of TRP ion channels , however , direct regulation by Gβγ appears not to be common . One report linked TRPA1 channel activation to Gβγ dimers liberated after activation of the GPCR MrgprA3 ( Wilson et al . , 2011 ) . We tested therefore , whether µOR activation has an effect on TRPA1 channels in DRG neurons , but observed neither an inhibition nor a pronounced activation of these channels . This finding could indicate that µORs and TRPA1 channels are expressed in largely separate subpopulations of DRG neurons . However , we also failed to observe an effect of µOR activation in a heterologous overexpression system where µOR activation had a clear effect on TRPM3 ( Figure 4f ) , indicating that the regulation of TRPA1 channels by Gβγ proteins might not be direct and requires further investigation . The only other TRP channel that has been reported to be regulated by Gβγ dimers is TRPM1 ( Shen et al . , 2012 ) , the channel protein with the highest homology to TRPM3 . TRPM1 , however , is not known to be expressed in somatosensory neurons . The assertion that TRPM1 is inhibited by Gβγ dimers has been controversial since earlier work reported that purified Gβγ dimers are ineffective , while Gαo proteins inhibit heterologously expressed TRPM1 ( Koike et al . , 2010 ) . A recent publication apparently reconciles these findings by providing evidence that both G protein subunits bind to TRPM1 and inhibit the channel ( Xu et al . , 2016 ) . Still , it is an open question , which of these G protein entities is more important in regulating TRPM1 channels under physiological conditions . In addition to the proposed regulation of TRPM1 by Gαo proteins , TRPM8 has recently been shown to directly bind Gαq proteins and to be strongly inhibited by this event ( Zhang et al . , 2012 ) . It was therefore important to test whether TRPM3 binds to and whether it is regulated by Gα subunits . Our data indicate that neither is the case ( Figures 7 and 8 ) , reinforcing our model that the main signaling pathway from µOR to TRPM3 channels is via Gβγ proteins . The intracellular pathways by which peripheral µORs reduce the excitability of primary nociceptive neurons seem diverse . A PI3Kγ/NO pathway has been described for the stimulation of KATP channels , where PI3Kγ is activated by Gβγ subunits ( Cunha et al . , 2010 ) . Also , GIRK2 channels activated directly by Gβγ proteins have been implicated in peripheral opioid analgesia ( Nockemann et al . , 2013 ) . These channels are absent from mouse DRG neurons , but may play a role in human anti-nociception ( Nockemann et al . , 2013 ) . Other ion channels seem to be regulated by peripheral µORs via their effects on the cellular cAMP level , often through subsequently influencing PKA activity ( Ingram and Williams , 1994; Gold and Levine , 1996 ) . This implies that µOR activity only affects these targets when cellular cAMP levels are elevated , for instance during inflammatory conditions . Notably , TRPV1 channels are also regulated by µORs through the well-established cAMP/PKA pathway ( Vetter et al . , 2006; Endres-Becker et al . , 2007; Vetter et al . , 2008; Spahn et al . , 2013 ) . Since we were studying TRPV1 channels in DRG neurons under resting , non-inflamed conditions , these considerations might explain why we observed no or only very modest effects of µOR activation on TRPV1 channel activity ( Figure 4a , b , e; Figure 4—figure supplement 1 ) . An entirely different , β-arrestin2-dependent pathway leading from µOR stimulation to TRPV1 activation has also been proposed ( Rowan et al . , 2014 ) . Remarkably , and in stark contrast to the published regulation of TRPV1 by cAMP/PKA ( Huang et al . , 2006 ) , TRPM3 channels were largely unaffected by our attempts to manipulate intracellular cAMP levels ( Figure 5e–h ) . These results from our cellular studies are well matched by the outcome of the in vivo experiments . TRPM3-dependent pain ( evoked by the injection of PS into the hind-paw ) was strongly suppressed by concomitant injection of µOR agonists , while these substances did not significantly attenuate TRPV1-dependent pain due to the injection of capsaicin at any of the concentrations tested ( Figure 10 ) . Together , these data demonstrate that the strong and direct functional influence of activated µORs on TRPM3 channels also works in peripheral endings of primary nociceptive neurons in vivo . Our findings establish TRPM3 channels as privileged target of peripheral µORs and thus indicate that TRPM3 channels play an important role in the physiological control of nociceptor excitability . The presence of TRPM3 channels in small-diameter nociceptor neurons ( Vriens et al . , 2011; Usoskin et al . , 2015 ) , their ability to release inflammatory mediators like CGRP ( Held et al . , 2015 ) , some aspects of their pharmacology ( Straub et al . , 2013b; Chen et al . , 2014; Suzuki et al . , 2016 ) and the phenotype of TRPM3-deficient mice , which strongly implied TRPM3 channels in the sensation of noxious heat and in inflammatory heat hyperalgesia ( Vriens et al . , 2011 ) , have previously advanced the argument that inhibition of TRPM3 might be a viable strategy to combat pain , especially inflammatory hyperalgesia . The results presented in this study provide further , strong support for this contention , because they show that TRPM3 inhibition is an important aspect of the action spectrum of µORs . Potentially , the findings presented here therefore help to explain the clinical effectiveness of peripherally acting or peripherally restricted µOR agonists . However , peripherally restricted µOR agonists still can exhibit pronounced and dose-limiting adverse effects , such as tolerance and constipation ( Stein , 2016 ) . Pharmacologically inhibiting TRPM3 channels directly might therefore be a feasible alternative to the established administration of µOR agonists . Given the unremarkable phenotype of TRPM3-deficient mice , when not challenged with painful stimuli , ( Vriens et al . , 2011 ) , it is reasonable to hope that unwanted effects of TRPM3 inhibitors may be less limiting than those of µOR agonists . It will be important to elucidate in various painful conditions how strong and robust the relief is that can be obtained by antagonists of TRPM3 channels . This study provides a strong incentive for commencing such work . For cellular experiments , we used adult mice of a wide age range ( aged 12–77 weeks ) of both sexes . The animals were either C57BL/6 mice or TRPM3-deficient mice ( Vriens et al . , 2011 ) that were backcrossed for more than 10 generations to the C57BL/6 genetic background . The TRPM3-deficient mouse strain is a classical , unconditional knock-out strain , in which exon 19 has been substituted by a LacZ-neomycin cassette . Housing and killing of the animals were carried out with institutional approval and in compliance with the guidelines of the Regierungspräsidium Gießen ( AK-3–2014 ) . For behavioral experiments , only male C57BL/6 mice at an age of 7–9 weeks were used . These experiments were approved and carried out in compliance with institutional guidelines of the Max Planck Society and guidelines of the Landesamt für Verbraucherschutz und Lebensmittelsicherheit of Lower Saxony , Germany ( AZ 33 . 9-42502-04-14/1638 ) . Mice were killed by an overdose of isoflurane ( 5% , AbbVie , Wiesbaden , Germany ) and then decapitated . The spinal cord was exposed by a single dorsal-midline incision along the entire length of the mouse . The entire spinal cord was removed , washed and placed into ice-cold HBSS ( GIBCO , Thermo Fisher , Karlsruhe , Germany ) . The spine was bisected along the spinal canal in longitudinal direction , nerve trunks and connective tissue was removed . Dorsal root ganglia ( DRGs ) from all cervical , thoracic and lumbar segments were then harvested into ice-cold culture medium consisting of DMEM ( GIBCO ) supplemented with 10% fetal calf serum ( GIBCO ) , and 1% penicillin-streptomycin ( 100 U/ml and 100 µg/ml , GIBCO ) . Isolated ganglia were partially digested for 30 min in 1 . 8 U/ml liberase DH Research Grade ( Roche , Mannheim , Germany ) at 37°C . DRGs were then gently triturated with a 1000 µl pipette . The digestion was stopped by adding 10–12 ml culture medium and centrifugation of the dissociated DRGs for 5 min at 250 g; washing and centrifugation was typically repeated for a second time . The supernatant was discarded and the cells were suspended in culture medium . Subsequently , for Ca2+ imaging experiments one eighth of the cell suspension ( corresponding to 100 µl ) was plated onto the center of a glass coverslip pre-coated with laminin ( Sigma-Aldrich , Munich , Germany ) . The cells were left to adhere at 37°C in an incubator in a humidified atmosphere containing 5% CO2 . After 2 hr , 2 ml culture medium was added onto the coverslips . For electrophysiological experiments , 100 µl of the cell suspension was diluted with 2 ml culture medium and seeded as such in a laminin-coated plastic culture dish ( Falcon , VWR , Darmstadt , Germany ) . Cells were maintained in the incubator and all experiments were performed within 24–56 hr after plating the cells . Human embryonic kidney 293 ( HEK ) cells , HEK-TRPM3 cells , which stably express either myc-TRPM3α2 ( Frühwald et al . , 2012 ) or myc-TRPM3α2-YFP ( Oberwinkler et al . , 2005 ) and HEK cells stably expressing human TRPM8 ( Erler et al . , 2006 ) , kindly provided by Dr . U . Wissenbach ( Homburg , Germany ) , were cultured and handled as described previously ( Wagner et al . , 2008; Frühwald et al . , 2012; Drews et al . , 2014 ) . Alternatively , we used HEK cells transiently transfected with TRPM3α2 as described ( Wagner et al . , 2008 ) . Neither in this work , nor in our previous studies , did we observe differences in the TRPM3 channel properties ( apart from transfection efficiencies ) due to transfection methods ( whether transiently or stably ) or terminal protein fusion tags employed ( see also Figure 8—figure supplement 1 ) . TRPM3 proteins exist in many different isoforms , mainly due to alternative splicing ( Lee et al . , 2003; Oberwinkler et al . , 2005; Frühwald et al . , 2012; Oberwinkler and Philipp , 2014 ) . Here , we use the naming of the splice variants according to Oberwinkler and Philipp ( 2014 ) . Throughout this study , we refer to heterologously expressed TRPM3α2 and to TRPM3 channels endogenously expressed in DRG neurons ( which have not been characterized with respect to splice events ) as TRPM3 . HEK cells and derivative cell lines were grown in MEM ( GIBCO ) supplemented with 10% fetal calf serum . Geneticin ( 0 . 5 mg/ml , Sigma-Aldrich ) was added to the culture medium for stably transfected cells only . CHO-TRPA1 cells , i . e . chinese hamster ovary cells stably expressing mouse TRPA1 ( Story et al . , 2003 ) , kindly provided by Dr . A . Patapoutian ( San Diego , USA ) , were cultured in DMEM ( GIBCO ) supplemented with 10% fetal calf serum ( GIBCO ) , 1% penicillin-streptomycin ( GIBCO ) , 10 mM glutamax ( GIBCO ) and 10x non-essential amino acids ( GIBCO ) . As we used these cell lines merely and exclusively as containers for expression of ion channels and other signaling molecules , we did not routinely test the identity of these cell lines . We routinely and regularly tested , however , the expression of the stably transfected genes , both by western blotting and , more often , by functional tests ( Ca2+ imaging and/or electrophysiology ) to ascertain the presence of the stably transfected ion channels ( which the parental cell lines do not express ) . These tests also ensure that the stably transfected cell lines are not mis-identified . All cell lines were maintained at 37°C in a humidified atmosphere with 5% CO2 . Cells were passaged one to three times per week , care was taken to avoid passage numbers above 40 . For experiments , HEK and CHO cells ( and cells from derived lines ) were plated on coverslips coated with poly-L-lysine ( MW: 70 , 000; Sigma-Aldrich ) . Transient transfection of HEK cells ( or derived cells lines , see above ) and CHO-TRPA1 was achieved with PolyFect ( Qiagen , Hilden , Germany ) according to the manufacturer's instructions . Measurements were performed 24–72 hr after transfection . When cells were used for electrophysiological experiments , they were typically passaged 1 day before measurement to reduce their density . The following constructs and expression vectors were used . Human µOR in pcDNA3 . 1 was purchased from the cDNA Resource Center ( Bloomsburg , PA , USA ) . Starting from this vector , we generated the fusion construct human µOR-YFP ( in pcDNA3 ) with standard procedures and oligonucleotides obtained from Eurofins ( Ebersberg , Germany ) . Rat TRPV1-YFP in pcDNA3 ( Hellwig et al . , 2005 ) was obtained from Dr . T . Plant ( Marburg , Germany ) . TRPM3 proteins were expressed with the help of vectors containing wild-type TRPM3α2 , myc-TRPM3α2 , YFP-TRPM3α2 or TRPM3α2-YFP either in pcDNA3 or in pCAGGS , which contained an additional IRES-GFP sequence enabling easy identification of transfected cells . All TRPM3 expression vectors encoded for murine TRPM3 proteins . All wild-type Gαi cDNAs without tag ( human Gαi1 , Gαi2 and Gαi3 ) were obtained in pcDNA3 . 1 vectors from the cDNA Resource Center . Vectors ( either pcDNA3 or pCAGGS ) containing human Gαi1Q204L ( here named Gαi1QtoL ) , Gαo1 , rat Gαo1Q205L ( here named Gαo1QtoL ) , GαqQtoL and bovine Gγ2 ( which has the same amino acid sequence as human and mouse Gγ2 ) were kindly provided by Dr . M . X . Zhu ( Houston , USA ) . Rat Gαi1-YFP , Gαi2-YFP and Gαi3-YFP in pcDNA3 and human Gβ1 ( having the same amino acid sequence as mouse Gβ1 ) in pCMV were described previously ( Bünemann et al . , 2003; Frank et al . , 2005 ) . Human GαoB ( here named Gαo2 ) , GαoBQ205L ( here named Gαo2QtoL ) , FLAG-Gβ1 and GαqQ209L ( here named GαqQtoL ) were from the cDNA Resource Center ( all in pcDNA3 . 1 ) . Bovine βARKct ( Koch et al . , 1994 ) and mouse phosducin ( kindly provided by Dr . L . Hein , Freiburg , Germany ) were subcloned into the expression vector pCAGGS ( with the additional IRES-GFP cassette ) and N-terminal myristoylation tags were added ( Rishal et al . , 2005 ) with standard procedures . Specifically , we added a sequence encoding the first 15 amino acids of Src followed by the triplet GAT ( encoding for aspartate ) as a linker before the start codon of the original proteins . For control transfections we used empty pcDNA3 or pCAGGS-IRES-GFP vectors . To visually identify successfully transfected cells we co-transfected pcDNA3-IRES-GFP or ER-DsRed ( kindly provided by Dr . R . Jacob , Marburg , Germany ) when the other transfected plasmids did not express a fluorescent protein . Verification of DNA sequences was done by direct sequencing ( Seqlab , Göttingen , Germany ) . The concentrations indicated throughout this section are the final values after adjustment of pH . The standard extracellular solution contained ( in mM ) : 145–149 NaCl , 10 CsCl , 3 KCl , 2 CaCl2 , 2 MgCl2 , 10 HEPES , 3 D-glucose , 7 D-mannitol ( pH 7 . 2 ) . In some instances the D-mannitol was replaced by 7 mM D-glucose . The extracellular solution with an elevated concentration of K+ ( high-K+ solution ) contained ( in mM ) : 70–74 NaCl , 10 CsCl , 75 KCl , 2 CaCl2 , 2 MgCl2 , 10 HEPES , 10 D-glucose . These extracellular solutions were adjusted with NaOH to pH 7 . 2 . The osmolality was regulated to within 315–335 mOsm/kg by the addition of D-glucose or H2O . The monovalent-free extracellular solution contained ( in mM ) : 2 CaCl2 , 2 MgCl2 , 10 HEPES , 280 D-mannitol . NMDG was used here to adjust the pH to 7 . 4 , resulting in 4–5 mM NMDG in the solution . Osmolality was 312–316 mOsm/kg . Standard intracellular solution for patch-clamping contained ( in mM ) : 140–145 CsOH , 10 BAPTA , 50 CsCl , 80 aspartate , 4 Na2ATP , 3 MgCl2 , 10 HEPES . The ATP-free intracellular solution contained ( in mM ) : 140–145 CsOH , 10 BAPTA , 50 CsCl , 80 aspartate , 4 Li4AMP-PNP , 1 Na2GTP , 1 MgCl2 , 10 HEPES . The Mg2+-free intracellular solution contained in ( mM ) : 135 CsOH , 80 aspartate , 50 CsCl , 10 BAPTA , 10 HEPES , 5 Na2EDTA , 4 Na2ATP . The pH of all intracellular solutions was adjusted to 7 . 2 with CsOH and the osmolality of these solutions was in the range of 285–315 mOsm/kg . For the recordings shown in Figure 2d–f , we used solutions without Cs+ ions , in order to avoid inhibition of K+ channels . The Cs+-free extracellular solution contained ( in mM ) : 144 NaCl , 5 . 8 KCl , 0 . 9 MgCl2 , 1 . 3 CaCl2 , 0 . 7 NaH2PO4 , 5 . 6 D-glucose and 10 HEPES , with a pH of 7 . 4 ( adjusted with NaOH ) and an osmolality of 305 mOsm/kg . The Cs+-free intracellular solution was composed of ( in mM ) : 135 KCl , 3 . 5 MgCl2 , 2 . 4 CaCl2 , 5 EGTA ( resulting in a free Ca2+ concentration of 100 nM ) , 5 HEPES , 2 . 5 Na2ATP . The pH of this solution was adjusted to 7 . 3 with KOH , the osmolality was 285 mOsm/kg . Intracellular Ca2+ imaging was performed as described previously ( Drews et al . , 2014 ) . Generally , coverslips with the cells attached were incubated with 5 μM Fura2-AM ( 1 mM stock in DMSO , Biotrend , Cologne , Germany ) for 30 min in the respective culture medium . Loading and measurements took place at room temperature ( 22–25°C ) , except where indicated otherwise ( see Figure 1j ) . After loading , coverslips were transferred to a closed recording chamber ( Warner Instruments , Hamden , CT , USA ) and continuously perfused with standard extracellular solution . Alternatively ( for the experiments shown in Figure 7—figure supplement 3 ) , we imaged cells in an open chamber containing a small volume ( 300 µl ) of static ( not perfused ) solution containing the Gβγ-liberating peptide mSIRK or its inactive control mSIRK-L9A ( Goubaeva et al . , 2003 ) . At the indicated time point in the figure , an additional volume of 300 µL of the peptide containing extracellular solution , plus double-concentrated TRPM3-agonist was added manually to the pre-existing volume to ensure fast and efficient mixing . At the end of the experiments , further 600 µl of high-K+ solution was added ( raising the average K+ concentration to 39 mM ) in order to depolarize the cells and to identify neurons . During imaging , every 5 s a pair of images was taken at 510 nm wavelength with a Retiga-Exi ( Qimaging , Surrey , BC , Canada ) or a HQ2 camera ( Photometrics , Tucson , AZ , USA ) during alternating excitation at 340 and 380 nm wavelengths ( filters and dichroic mirrors from AHF , Tübingen , Germany ) using a motorized filter wheel ( Ludl , Hawthorne , NY , USA ) or a wavelength switcher ( DG4 , Sutter , Novato , CA , USA ) attached to Nikon ( Düsseldorf , Germany ) inverted microscopes equipped with 10x SFluor objectives ( N . A . 0 . 5 ) . From a fluorescence image with 380 nm excitation or a fluorescence image of the GFP/YFP fluorescence , several regions of interest representing each a single cell were selected manually ( see Figure 1—figure supplement 5 ) . Ratio images ( 340/380 nm ) were calculated with ImageJ ( Abràmoff et al . , 2004 ) using a modified version of the ‘ratio plus’ plug-in after background subtraction and thresholding to exclude pixels with low fluorescence intensity values . After imaging for approximately 5 min to establish baseline conditions , ligands were superfused onto the cells as indicated in figures using a gravity-driven perfusion system . For some experiments , cells were pretreated with pharmacological substances ( H89 , KT5720 , BIM , mSIRK or mSIRK-L9A , Figure 6 and Figure 7—figure supplement 3 ) , for 30 min by adding them to the culture medium during Fura2-AM loading . Pertussis toxin ( PTX , 100 ng / ml , List Biological Laboratories , Campbell , CA , USA ) was added to the culture medium in the incubator for 16–24 hr and washed off the extracellular medium before loading with Fura2-AM . For imaging DRG neurons , 20 μM verapamil was added to all extracellular solutions to block endogenous voltage-gated Ca2+ channels . However , to distinguish neuronal from non-neuronal cells in these cultures , the high-K+ solution was routinely used without verapamil to depolarize the cells at the end of the experiment . Whole-cell patch-clamping was performed with EPC10 amplifiers ( HEKA , Lambrecht/Pfalz , Germany ) as described previously ( Drews et al . , 2014 ) . Series resistances were compensated for 80% and all offset potentials were nullified before establishing the cell-attached configuration . All potential values are reported after being corrected for the calculated liquid junction potential ( 15 mV ) , except when using the Cs+-free solutions , where we assumed a negligible liquid junction potential . Voltage ramps ( −115 to +85 mV with a steepness of 1 mV/ms ) were applied with a frequency of 1 ramp per 1–2 s . For HEK and transfected HEK cells a holding potential of −15 mV was used between the ramps and the current amplitudes were analyzed at −80 and +80 mV offline . DRG neurons were voltage-clamped at holding potentials of −55 to −75 mV between the ramps and only the currents at −80 mV were analyzed . In an attempt to diminish endogenous currents during voltage-ramps , the standard extracellular recording solution was replaced by a monovalent-free extracellular solution ( Vriens et al . , 2011; Straub et al . , 2013a ) after establishing the whole-cell configuration ( Figure 2a–c ) . Alternatively , voltage ramps from −100 to −20 mV were recorded from DRG neurons in Cs+-free intra- and extracellular solution , and subsequently analyzed at −60 mV ( Figure 2d–f ) . From DRG neuron recordings used for Figure 2a–c , we only considered measurements , in which the application of TRPM3 agonists ( 50 µM PS + 50 µM nifedipine ) resulted in inward currents with an amplitude of more than 10 pA , indicating robust expression of TRPM3 channels . For Figure 2d–f , we used recordings from DRG neurons that showed currents > 10 pA to the application of the same TRPM3 agonists or to capsaicin , in an attempt to record from a broad population of small diameter neurons ( Vriens et al . , 2011; Tan and McNaughton , 2016 ) . Current densities were calculated offline with the help of the Igor software package ( version 5 . 05A , Wavemetrics , Lake Oswego , OR , USA ) . All chemical reagents were prepared as stock solutions in DMSO , with the exception of PTX ( pertussis toxin from Bordetella pertussis ) and morphine that were dissolved in H2O . Stocks were kept aliquoted and frozen at -20°C . Even when several compounds were applied simultaneously , the final DMSO concentration in the superfusing solution did not exceed 0 . 4% . The following substances were obtained from Sigma-Aldrich: AITC ( allyl-isothiocyanate ) , capsaicin , DAMGO ( [D-Ala2 , N-Me-Phe4 , Gly5-ol]-enkephalin acetate ) , menthol , morphine sulphate pentahydrate , IBMX ( 3-Isobutyl-1-methylxanthine ) , naloxone hydrochloride dihydrate , nifedipine , verapamil hydrochloride . The following substances were purchased from Biotrend: forskolin , herkinorin , loperamide hydrochloride , ( RS ) -baclofen , WIN 55 , 212-2 , somatostatin-14 , DHPG ( RS-3 , 5 dihydroxyphenylglycine ) , [D-Ala2] deltorphin II . L-noradrenaline was obtained from Alfa Aesar ( Thermo Fisher ) , BIM ( Bisindolylmaleimide IV ) was from Biomol ( Hamburg , Germany ) , H89 dihydrochloride hydrate from Biozol ( Eching , Germany ) , KT5720 from Santa Cruz ( Heidelberg , Germany ) and pregnenolone sulfate ( PS ) from Steraloids ( Newport , RI , USA ) . The cell-permeable peptides mSIRK and mSIRK-9LA were purchased from Calbiochem ( Merck-Millipore , Darmstadt , Germany ) . Ca2+ imaging time series of single cells were extracted from stacks of ratio images as averages over the entire cell area . The first 20 data points ( corresponding to the first 100 s of the experiment ) were averaged to form the baseline which , for quantitative analyses , was subtracted from all other values . For quantitative analysis and statistical testing , single cell time series traces were smoothed ( running window of five values ) and maximum , average or minimum values were obtained in a given time-window ( typically corresponding to the application of a substance ) from the smoothed single cell Ca2+ imaging traces . Inhibition due to a pharmacological manipulation was calculated for Ca2+ imaging ( and electrophysiological ) data by obtaining three ( baseline-subtracted ) values: One before ( Vbefore ) , one during ( Vduring ) and one after the application ( Vafter ) of the pharmacological substance . Inhibition ( in percent ) was calculated as: Inhibition = 100 * ( 1 – ( 2 * Vduring / ( Vbefore +Vafter ) ) . The averaging of the values before and after pharmacological interventions was done in an effort to correct for the pronounced and inevitable reduction in response during repeated or prolonged applications of agonists , such as PS or capsaicin . Please note , however , that this way of calculating inhibition ( which is a very common and standard way ) still is prone to artefacts and can report inaccurate values . For example , a cell that responded to the first application of an agonists ( Vbefore ) , but then simply ceased to respond to any further stimulation ( Vduring and Vafter would then be ‘0’ ) , would be reported as 100% inhibited . Please note also that the formula for calculating inhibition can report negative values , when a cell responds stronger during the application of the pharmacological intervention ( Vduring larger than the average of Vbefore and Vafter ) . Because DRG neurons are a highly diverse population and sometimes are also prone to spontaneous activity , the individual values reported in the ‘dot clouds’ often contain a minority of cells that displayed such negative inhibitory values . Another common source for artificial values is baseline shift , which we did not attempt to correct for . Isolated DRG cells were classified as neurons if they responded either to capsaicin or the solution with a high potassium concentration ( high-K+ solution ) , cells outside of this category were not considered further . For sorting individual DRG neurons into different categories ( e . g . PS-sensitive or capsaicin-sensitive ) , a threshold of 0 . 1 ratio units was set arbitrarily ( see Figure 1—figure supplement 1 for a justification of this value ) . As an exception , in Figure 7—figure supplement 3a , b , we have taken a threshold value of 0 . 05 , because the treatment with mSIRK has lowered the response amplitude to PS . For categorizing sensitivity to an inhibitory stimulus , a minimal inhibition of 7 . 5% of the agonist-induced response ( calculated as above ) was arbitrarily set as threshold . Traces of single cell time series corresponding to a category were averaged and are presented as mean ± SEM throughout this manuscript . Next to the representation of aggregated data , we also plotted the values from all individual neurons analyzed as ‘dot cloud’ . For each reported average trace , we also show responses from ten randomly chosen neurons in the figure supplements . Calculating the response values , categorizing , averaging and plotting the randomly chosen cells was done in R . For Ca2+ imaging experiments of ( transfected ) cell culture cells , we did not categorize individual cells as responders or non-responders , but took all cells according to transfection status or pharmacological treatment . Since these cells represent a much more homogenous group , we did not report single cell values , but only averaged responses . Cell diameters were estimated by drawing elliptical selections manually around single DRG neurons on ratio images during a strong stimulation ( in this case with high-K+ solution ) . Subsequently , Feret's diameter was measured with the built-in function of ImageJ . The distribution of single cell Feret's diameters was represented as histograms with a bin width of 2 µm . For Ca2+ imaging and electrophysiological experiments , we considered single individual cells as independent biological replicates . We additionally report the number of recordings for Ca2+ imaging experiments , which indicates the number of coverslips ( with the attached cells ) that were used in these experiments in sequential measurements . In electrophysiological experiments , we equally used the number of individual cells that were successfully recorded as the number of biological replicates . Each single , individual co-immunoprecipitation experiment was visualized on a single western blot , which was quantitatively analyzed and considered a biological replicate . Finally , in behavioral experiments , each individual mouse was recorded only once and considered a biological replicate . In all types of experiments , the number of biological replicates ( determined as described above and referred to as ‘n’ ) was used for statistical testing and to calculate SEM values . We did not perform any a-priori estimation of sample sizes . We followed , however , closely the standards in the field , and , accordingly , used 7–15 individual biological replicates for each data point in electrophysiological ( whole-cell patch-clamping ) , Western blotting and behavioral experiments . Such numbers for replicates mean that we were able to detect ( with a power of 80% and at a level of significance of p<0 . 05 ) differences between means that were in the order of 1–2 SD . For Ca2+ imaging experiments , we always performed at least two separate recordings . During each recording , we typically were able to record at least from 10 cells ( depending on parameters such as transfection efficiency and cell survival during pharmacological treatments ) , meaning that each trace in Ca2+ imaging recordings represents the mean of more than 20 cells . The precise number of cells is stated in the figures themselves or , alternatively , in the figure legends . Because of the small sample sizes in some experiments , and because we observed that many of the larger data sets did not conform to Gaussian normality ( as tested with Graphpad Prism , version 3 . 02 , Graphpad software , La Jolla , CA , USA ) , typically due to outliers ( which were never removed from any data set ) , we preferred non-parametric statistical tests . For comparing two groups , we used the Mann-Whitney test ( Figure 1c , Figure 1—figure supplement 2c , Figure 4b , Figure 5b , d , Figure 7g , i , Figure 8c , Figure 10a–c ) or the Wilcoxon signed rank test ( Figure 2b , e ) . For multiple comparisons , we used either the Kruskal-Wallis test ( Figure 5k , Figure 5—figure supplement 1b , Figure 7—figure supplement 2 , Figure 7—figure supplement 3a , c ) or the Friedman test ( Figure 1m ) . These two latter tests were followed by Dunn's multiple comparison test . All statistical testings were done with Graphpad Prism or R . In all cases , we accepted p-values smaller than 0 . 05 as statistically significant . In the figures , we use * to indicate p-values larger or equal than 0 . 01 and smaller than 0 . 05 , for p-values larger or equal than 0 . 001 and smaller than 0 . 01 and , accordingly , * for p-values smaller than 0 . 001 . Unless otherwise stated in the figure legend , bar graphs with error bars always represent the mean ± SEM . Fitting Hill functions ( with variable slope ) to dose-response curves was done with Graphpad Prism . For immunoprecipitation of TRPM3 , HEK cells stably expressing myc-TRPM3α2-YFP ( Oberwinkler et al . , 2005 ) were solubilized in a lysis buffer consisting of 25 mM Tris ( pH 8 ) , 0 . 5% ( w/v ) Triton X-100 , 0 . 5% ( w/v ) Na+-deoxycholate , 50 mM NaCl and a protease inhibitor mix as described ( Leitner et al . , 2016 ) . All reagents for immunoprecipitation and western blotting were obtained from Carl Roth ( Karlsruhe , Germany ) unless stated otherwise . Non-transfected HEK cells were used as controls . After removal of debris and nuclei by centrifugation for 15 min at 13 , 000 g and 4°C , a known fraction of the lysate ( typically 1% of the total lysate ) was separated for later use as input control and the remaining lysate was incubated with GFP-Trap agarose beads ( ChromoTek , Planegg-Martinsried , Germany ) at 4°C over night on a tube roller . Subsequently , beads were washed once with lysis buffer and four times with binding buffer ( Hu et al . , 2009 ) containing 20 mM HEPES ( pH 7 . 4 ) , 0 . 01% ( w/v ) CHAPS , 140 mM K+-aspartate , 5 mM MgCl2 , 10 mM EGTA and 0 . 04 mM dithiothreitol . Input and immunoprecipitated proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes ( GE Healthcare , Solingen , Germany ) . For protein visualization , we used fluorescence ( ODYSSEY Sa , LI-COR Biosciences , Bad Homburg , Germany ) or chemiluminescence ( ChemoCam Imager , Intas , Göttingen , Germany ) detection systems together with the following antibodies: anti-GFP ( Santa Cruz , sc-8334 , 1:500 ) , anti-Gβ ( Santa Cruz , sc-378 , 1:500 ) , anti-Gαi3 ( 1:1000; [Gohla et al . , 2007] ) , anti-Gαq/11/14 ( Santa Cruz , sc-365906 , 1:500 ) , anti-rabbit IgG-IRDye-800CW ( LI-COR , #926–32211 , 1:10 , 000 ) , anti-mouse IgG-HRP ( Santa Cruz , sc-2031 , 1:10 , 000 ) , anti-rabbit IgG-HRP ( Santa Cruz , sc-2030 , 1:10 , 000 ) . Immunoprecipitation experiments were quantified densitometrically using ImageJ ( Abràmoff et al . , 2004 ) , the density values for the immunoprecipitated Gβ proteins were subsequently normalized to the density values obtained for Gβ proteins from the total lysate . TRPM3 protein isolation was performed essentially as described previously ( Uchida et al . , 2016 ) . In brief , seven to eight 10 cm culture dishes with HEK cells stably expressing myc-TRPM3α2 ( Uchida et al . , 2016 ) were grown to ~80% confluence and used for a single immunoprecipitation probe . Cells were washed and collected with cold phosphate buffered saline ( PBS ) and resuspended in NCB buffer , containing ( in mM ) : 500 NaCl , 50 NaH2PO4 , 20 HEPES and 10% ( v/v ) glycerol ( pH 7 . 5 ) , with addition of 1 mM protease inhibitor phenyl-methyl-sulfonyl-fluorid and 5 mM β-mercaptoethanol . Next , the cells were lysed by freezing/thawing and centrifuged at 40 , 000 g for 2 . 5 hr . The pellet was resuspended in NCB buffer with the addition of a protease inhibitor mixture ( Roche ) , 0 . 1% ( w/v ) Nonidet P-40 ( Roche ) and 0 . 5% ( w/v ) dodecylmaltoside ( Calbiochem ) . The suspension was incubated overnight on a shaker with gentle agitation , and then centrifuged for 1 hr at 40 , 000 g . The supernatant was incubated with magnetic beads ( Pierce , Thermo Fisher ) conjugated with anti-myc antibodies ( Sigma-Aldrich ) . All steps of incubation were performed at 4°C . TRPM3 proteins were eluted from the beads with SDS-loading buffer by boiling . The eluted proteins then were separated by SDS-PAGE on 10% polyacrylamide gels and Tris-glycine-SDS buffer ( Bio-Rad , Hercules , CA , USA ) at a constant voltage of 180 V . Proteins were visualized by Coomassie blue staining . Besides the heavy and light chain antibody bands derived from immunoprecipitation , the gel showed a profound band at approx . 212 kDa , corresponding to the molecular weight of TRPM3 monomers . From each gel ( corresponding to one experiment ) , 12–14 bands were excised and digested with trypsin for mass spectrometry analysis according to a published protocol ( Shevchenko et al . , 2006 ) , with some modifications . The protein digests were analyzed by liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) using a nano flow liquid chromatography system ( Ultimate3000 , Thermo Fisher ) interfaced to a hybrid ion trap-orbitrap high resolution tandem mass spectrometer ( VelosPro , Thermo Fisher ) operated in data-dependent acquisition mode , as previously described ( Uchida et al . , 2016 ) . Data analysis were performed on ProteomeDiscoverer 2 . 1 ( Thermo Fisher ) using SequestHT ( 0 . 1% false discovery rate ) ( Tabb , 2015 ) and Percolator for peptide/protein identification and validation ( Käll et al . , 2007 ) . After acclimatization for at least 1 hr in transparent plastic boxes , the mice were injected subcutaneously with 10 µl of the respective experimental agent into the midplantar region of the left hind paw . The total duration of nocifensive behavior ( paw lifting , shaking and licking ) during a 20-min observation period was measured . For PS-induced pain mice received either 5 nmol PS together with 2 µg DAMGO , both dissolved in PBS or 5 nmol PS and the vehicle . For dissolving PS in PBS , the solution was heated to 30°C and sonicated for 20 min . For capsaicin-induced pain , mice were injected with 0 . 5 µg capsaicin ( dissolved in 3% ethanol/PBS ) together with 2 µg DAMGO or 0 . 5 µg capsaicin and vehicle ( PBS ) . Experiments were done with a minimum of three independent mouse litters . Whenever possible , littermates were used as controls ( vehicle injection ) on the same experimental day .	There are very few treatments available for people suffering from strong or long-lasting pain . Currently , substances called opioids – which include the well-known drug morphine – are the strongest painkillers . However , these drugs also cause harmful side effects , which makes them less useful . Like all drugs , opioids mediate their effects by interacting with molecules in the body . In the case of opioids , these interacting molecules belong to a group of receptor proteins called G-protein coupled receptors ( or GPCRs for short ) . These opioid receptors are widely distributed in the nerve cells and brain regions that detect and transmit pain signals . It was poorly understood how activation of opioid receptors reduces the activity of pain-sensing nerve cells , however several lines of evidence had suggested that a protein called TRPM3 might be involved . TRPM3 is a channel protein that allows sodium and calcium ions to enter into nerve cells by forming pores in cell membranes , and mice that lack this protein are less sensitive to certain kinds of pain . Dembla , Behrendt et al . now show that activating opioid receptors on nerve cells from mice , with morphine and a similar substance , rapidly reduces the flow of calcium ions through TRPM3 channels . Further experiments confirmed that activating opioid receptors in a mouse’s paw also reduced the pain caused when TRPM3 proteins are activated . GPCRs interact with a group of small proteins called G-proteins that , when activated by the receptor , split into two subunits . Based on studies with human kidney cells , Dembla , Behrendt et al . found the so-called G-beta-gamma subunit then carries the signal from the opioid receptor to TRPM3 . Two independent studies by Quallo et al . and Badheka , Yudin et al . also report similar findings . These new findings show that drugs already used in the treatment of pain can indirectly alter how TRPM3 works in a dramatic way . These results might help scientists to find drugs that work in a more direct way to dial down the activity of TRPM3 and to combat pain with fewer side effects . Though first it will be important to confirm these new findings in human nerve cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Anti-nociceptive action of peripheral mu-opioid receptors by G-beta-gamma protein-mediated inhibition of TRPM3 channels
After an initial stage of local analysis within the retina and early visual pathways , the human visual system creates a structured representation of the visual scene by co-selecting image elements that are part of behaviorally relevant objects . The mechanisms underlying this perceptual organization process are only partially understood . We here investigate the time-course of perceptual grouping of two-dimensional image-regions by measuring the reaction times of human participants and report that it is associated with the gradual spread of object-based attention . Attention spreads fastest over large and homogeneous areas and is slowed down at locations that require small-scale processing . We find that the time-course of the object-based selection process is well explained by a 'growth-cone' model , which selects surface elements in an incremental , scale-dependent manner . We discuss how the visual cortical hierarchy can implement this scale-dependent spread of object-based attention , leveraging the different receptive field sizes in distinct cortical areas . Neurophysiological studies over the past 40 years have revealed that the neuronal representation of an object in low-level areas of the visual cortex consists of a set of simple features such as colors and edge orientations . However , this is not how we perceive a visual scene . Our perception is much more structured , because our visual system groups the features into objects . Introspectively , this grouping process appears to be effortless because we hardly ever perceive features in isolation . Yet , the processes for perceptual organization are only partially understood . One influential theory suggested that perceptual grouping occurs instantaneously and in parallel if features are connected to each other , i . e . 'uniformly connected' ( Palmer and Rock , 1994 ) . Such a rapid grouping process would be in line with studies demonstrating that object recognition can be extremely fast and pre-attentive ( Thorpe et al . , 1996; Moore and Egeth , 1997; Treisman , 1985 ) . High-speed object recognition presumably relies on feedforward processing , leveraging the hierarchy of features represented in the visual cortex . Neurons in lower areas coding for elementary features rapidly propagate activity to shape selective neurons in higher visual areas ( Hung et al . , 2005; DiCarlo et al . , 2012 ) , grouping features into more complex constellations ( Roelfsema , 2006 ) so that perceptual grouping coincides with object recognition ( Biederman et al . , 1987; Nakayama et al . , 1989; Driver and Baylis , 1996; Pelli et al . , 2009 ) . However , there also exist conditions where perceptual grouping requires a slow and serial process ( Roelfsema , 2006 ) . Curve tracing is an example of a task that invokes such a serial , incremental grouping operation ( Jolicoeur et al . , 1986; Jolicoeur and Ingleton , 1991; Pringle and Egeth , 1988; Roelfsema and Houtkamp , 2011 ) when participants group contour elements of an elongated curve . In the example display of Figure 1a subjects judge if the two colored circles fall on the same curve . In this task , reaction times increase linearly with the length of the curve . Subjects first direct their attention to the red fixation point and attention then gradually spreads over the curve until it reaches the green circle ( Scholte et al . , 2001; Houtkamp et al . , 2003 ) . This process is implemented in the visual cortex as the propagation of enhanced neuronal activity over the curve’s representation ( Pooresmaeili and Roelfsema , 2014 ) ( Figure 1b ) . Curve-tracing is size invariant ( Jolicoeur and Ingleton , 1991 ) so that the reaction time of observers depends little on the viewing distance . This is remarkable , because the length of the curve in degrees of visual angle increases when subjects view the stimulus from nearby . However , now the distance between the curves also increases and this enhances tracing speed ( in degree/s ) compensates for the longer curves so that the total reaction time remains the same . Size invariance can be explained if perceptual grouping occurs at multiple levels of the visual cortical hierarchy ( Roelfsema and Houtkamp , 2011; Pooresmaeili and Roelfsema , 2014 ) . When curves are nearby , perceptual grouping requires the high spatial resolution provided by low-level areas where neurons have small receptive fields ( RFs ) and horizontal connections interconnect neurons with RFs that are nearby in visual space so that progress is slow . If curves are farther apart , however , neurons in higher areas can take over and their larger RFs could speed up the grouping process ( Pooresmaeili and Roelfsema , 2014 ) . Size invariance also occurs when subjects solve a maze , because paths are followed at a higher speed if the distance between the walls is larger ( Crowe et al . , 2000 ) . 10 . 7554/eLife . 14320 . 003Figure 1 . Mechanism of perceptual grouping . ( a ) Perceptual grouping of contour elements calls on a serial process as illustrated with a curve-tracing task . The actual stimulus contains one cue ( green , dashed circles show other possible cue locations ) and the participant indicates whether it falls on the same curve or on the other curve ( points labeled ‘d’ ) as the fixation point . Reactions times increase linearly with the distance between the fixation point and the second cue on the same object ( here 4 , 8 , 12 , or 16 degrees ) ( Jolicoeur et al . , 1986 ) . ( b ) Perceptual grouping corresponds to spreading object-based attention over the curve . Cortical neurons propagate an enhanced firing rate of cells over the representation of the relevant curve in the visual cortex ( Roelfsema , 2006 ) . ( c ) An example stimulus of a 2D shape for which we measure the time course of perceptual grouping . ( d ) We tested the hypothesis that grouping of 2D shapes also requires a serial grouping operation . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 003 Curve tracing and maze-solving might be special cases , however , and it is unclear whether perceptual grouping is serial in more typical visual scenes . We therefore investigated the time course of grouping for line drawings of relatively simple 2D shapes ( Figure 1c , d ) . Specifically , we addressed the following questions: ( 1 ) Does perceptual grouping of simple 2D stimuli rely on a serial , incremental process ? ( 2 ) What is the influence of image scale on the speed of grouping and the spread of object-based attention ? ( 3 ) Is there an effect of object-recognition on the speed of the grouping process ? The uniform connectedness hypothesis by Palmer and Rock , 1994 will serve as baseline . These authors suggested that image regions with homogenous surface properties are grouped instantaneously ( Palmer and Rock , 1994 ) so that the reaction time should not depend on the placement of cues within a white region enclosed by a black contour ( as in Figure 2a ) . The second model has been called 'pixel-by-pixel' in the context of curve-tracing ( McCormick and Jolicoeur , 1994; Jolicoeur and Ingleton , 1991 ) . Grouping is realized by spreading attention across pixels of the same color , at a fixed speed ( Figure 2b ) . Spreading starts at one of the cues and processing time is proportional to the length of the shortest path through the object . At a neuronal level , enhanced activity could spread among neurons tuned to the same color , at a single spatial scale ( Grossberg and Mingolla , 1985 ) . The third model is the growth-cone model ( Figure 2c ) , inspired by models for curve tracing ( Pooresmaeili and Roelfsema , 2014; McCormick and Jolicoeur , 1994; Jolicoeur et al . , 1986 ) and visual routines ( Ullman , 1984; Roelfsema et al . , 2000 ) . The growth-cone model predicts that grouping speed depends on the size of homogeneous image regions . Spreading is fast within large regions and slows down in narrow regions . In our implementation , we determined for every pixel the size of the largest circle centered at that pixel that did not touch the boundaries and assumed that speed was proportional to this size . This model is scale invariant because the total spreading time does not depend on variations in the overall size of the picture , caused e . g . by changes in viewing distance . In the visual cortex , this model could be implemented if neurons with various receptive field ( RF ) sizes tuned to the same feature spread the enhanced activity . In large , homogenous image regions , neurons with large RFs would make fast progress , but in narrow regions grouping speed decreases because neurons with small RFs take over . The fourth model is one where the grouping signal spreads inwards from the boundaries of the cued object so that the response time increases with the distance between the cue and the boundary ( Figure 2d ) ( Komatsu , 2006 ) . Such a 'filling-in' process has been proposed for brightness perception ( Paradiso and Nakayama , 1991 ) and texture segregation ( Lamme et al . , 1999 ) . The fifth and final model is a Euclidean model , which will serve as reference because it is simple . It holds that the reaction time depends on the distance between cues , irrespective of the shape of the objects ( Figure 2e ) . It differs from the pixel-by-pixel model where distance is measured as the shortest path within the object . Experiment 1 examined the time-course of perceptual grouping with wedge-shaped stimuli ( Figure 3a ) . Stimuli contained two adjacent wedges and a red fixation point that served as first cue . Twenty participants judged whether a second red cue ( one of the dots shown in blue/grey in Figure 3b per trial ) fell on the same wedge as the fixation point , by pressing a button . We measured eye-position to ensure that the participants maintained gaze on the fixation point . The screen coordinates of the cues were uninformative , because we mirrored and rotated the stimulus across trials . In Experiment 1A the second cue was at a fixed distance from the fixation point . Figure 3c illustrates the model predictions . The filling-in model predicts short RTs for cues near the edges and long RTs in the center . The uniform connectedness model , the Euclidean model , and the pixel-by-pixel model predict that RTs are constant , because all cues are at the same distance from fixation . In contrast , the growth-cone model predicts that RTs are short in the center and longer near the edges . 10 . 7554/eLife . 14320 . 005Figure 3 . Parsing a wedge-shaped object ( Experiment 1A ) . ( a ) Trial structure . After a fixation period of 500 ms , we presented two wedges and a cue . The subjects reported whether the cue fell on the same wedge as the fixation point . ( b ) The second cue could fall on the same ( blue dots ) or different wedge as fixation point ( grey dots ) . The cues in the actual experiment were red . ( c ) RT predictions of the filling-in model , Euclidean model , and growth-cone model . ( d ) Average RTs of 20 observers . Blue bars , RTs for the same object; grey bars , RTs for the different object . Error bars , s . e . m . across participants ( after correction for baseline differences in RT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 005 The average accuracy was 94% , without signs of a speed-accuracy trade-off ( correlation between RT and accuracy; r=0 . 17 , t ( 22 ) =0 . 80 , p>0 . 4 ) . Figure 3d shows the average RTs . If the two cues fell on the 'same' wedge , RTs were shortest in the center of the wedge and longer near the edges ( repeated-measures ANOVA , Greenhuise-Geisser corrected , F ( 4 . 5 , 85 . 7 ) =10 . 9 , p<0 . 001 ) . RTs in the 'different' condition were longer than in the 'same' condition ( F ( 1 , 19 ) =19 . 5 , p<0 . 001 ) . A regression analysis revealed that the growth-cone model accounted for 86% of the variance of the RTs , which was significantly better than models predicting that RT is constant ( p<0 . 001 bootstrap statistic , see Materials and methods ) . The Filling-in model predicted that RT is shortest near the edges , opposite to the RT data . Thus , these results support the growth-cone model and are incompatible with the other models described above . We considered the possibility that the longer RTs near the edges were a masking effect caused by the edges . We therefore carried out a control experiment in which participants viewed the same stimulus and performed a color discrimination task . Eleven participants reported the cue’s color , which could now be red or green . We did not observe a significant correlation between RT and the distance between cue and edge ( r=−0 . 2 , t ( 30 ) =−1 . 31 , p>0 . 1 ) . Thus , the edges did not act as a mask . Experiment 1B measured the influence of the distance between the fixation point and the cue on RT in thirty new participants ( Figure 4a ) . The average accuracy was 93% and we did not obtain evidence for a speed-accuracy trade off ( correlation between RT and accuracy , r=0 . 02; t ( 62 ) =0 . 14 , p>0 . 8 ) . The average RT was higher when the cue was farther from the fixation point and a further delay was observed near the edges ( warm colors in Figure 4b ) , in line with the predictions of the growth-cone model ( compare Figure 4b to the lower left panel of Figure 3c ) . We used a regression analysis to compare models ( Figure 4c ) . The filling-in model erroneously predicted that RTs are shortest near the edges . The Euclidean model ( and the pixel-by-pixel model , which made the same prediction in this experiment ) accounted for 49% of the variance but failed to explain the influence of edge vicinity . The growth-cone model did capture this effect and it accounted for 72% of the variance . The predictions of the growth-cone model were significantly better than those of the other models ( all ps<0 . 01 bootstrap statistic ) and the regression revealed that the average speed of perceptual grouping was 22 ms/growth cone ( 95% confidence interval 17–27 ms , bootstrap analysis ) . In other words , the RT increased by ~22 ms for every image patch ( with the size of one growth cone ) that was added to the perceptual group . Thus , grouping of surfaces invokes a serial perceptual grouping process that proceeds fast in homogeneous image regions and slows down near edges , in accordance with the growth-cone model . 10 . 7554/eLife . 14320 . 006Figure 4 . Time course of grouping a simple , wedge-shaped object ( Experiment 1B ) . ( a ) In Experiment 1B cue locations were at multiple eccentricities on the same ( blue ) or different object ( grey ) . ( b ) The pattern of RTs . Cold and warm colors show short and long RTs , respectively . ( c ) Regression analysis showing the fit of the models to the RT data . The growth-cone model explains 72% of the variance ( r=0 . 85 ) and is superior to the other models . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 006 We next measured the time-course of perceptual grouping with more complex images to address a number of questions . First , does the shape of the object determine growth cone size ? Does grouping speed decrease in narrow parts due to small growth cones and increase in broader parts ? Second , do internal features such as textures or color changes influence grouping speed ? Third , does object recognition affect perceptual grouping , because it can provide larger chunks to be grouped at once , as suggested by previous work ( Mahoney and Ullman , 1988; Peterson and Gibson , 1994; Vecera and Farah , 1997; Korjoukov et al . , 2012 ) ? We created four stimulus sets with 24 stimuli each; color pictures , detailed cartoons , cartoon outlines , and scrambled cartoons ( Figure 5—figure supplement 1 ) , based on 12 pictures with two animals and 12 pictures with two vehicles ( Korjoukov et al . , 2012 ) where animals and vehicles were easy to recognize . However , we made image recognition difficult for the scrambled cartoons by repositioning a few line segments but we left the region where the two objects intersected unchanged ( Figure 5—figure supplement 1 , lower ) . Object area , perimeter , and cue locations were similar across the stimulus sets . To measure grouping speed , we asked participants to report whether a cue appeared on the same object as the fixation point or on a different object ( Figure 5a ) . When the subjects had maintained gaze on the fixation point for 500 ms , we presented the cue for 1000 ms and then the two objects . The cue was at a distance of 4 . 7° ( 1/3 of the trials ) or 9 . 4° ( 2/3 of the trials ) from the fixation point and these distances were matched if cues fell on the other object . With 24 images per condition and 3 cue positions per object , we had 72 data points for the same object condition , and an equal number for the different object condition . We assigned a total of 88 participants to the four image categories ( between-subject condition ) , with 22 participants per condition . Sixty-four participants maintained gaze on the fixation point until the response , but we found that eye-movements had little influence on the pattern of RTs ( Materials and methods ) . 10 . 7554/eLife . 14320 . 007Figure 5 . Time course of parsing scrambled cartoons ( Experiment 2 ) ( the actual stimuli were white outlines on a black background ) . ( a ) After a fixation epoch of 500 ms , the subjects saw the cue for 1000 ms , and then also two objects . They reported whether the cue fell on the same object as the fixation point . ( b ) Left , Example stimulus . FP , fixation point . Numbers indicate the estimated number of growth cones between the cues and the FP . Right , RTs averaged across participants for the different cue locations . Error bars represent s . e . m . ( c-f ) Regression of the RT on the predictions of the filling-in ( c ) , Euclidean ( d ) , pixel-by-pixel ( e ) , and growth-cone model ( f ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 00710 . 7554/eLife . 14320 . 008Figure 5—figure supplement 1 . The full stimulus set for Experiment 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 008 Figure 5b shows an example scrambled cartoon picture and the RTs averaged across 22 participants . The RT was shortest for the nearest cue ( blue in Figure 5b , 627 ms ) and it increased by 24/30 ms ( green/red in Figure 5b ) for the more distant cues . We investigated how well the different grouping models account for the pattern of RTs across all scrambled cartoon pictures with a regression analysis ( Figure 5c–f ) . As in experiment 1 , the filling-in model made the wrong prediction that RT is shortest near boundaries ( r=−0 . 30 ) . The Euclidean model predicts that RT depends on the distance between fixation point and cue ( i . e . the eccentricity of the cue ) but this relationship was weak in the data ( r=0 . 29 ) . The predictions of the pixel-by-pixel model were better ( r=0 . 36 ) but those of the growth-cone model were best ( r=0 . 76 ) . The growth-cone model explained 58% of the variance , which was significantly better than the other three models ( Figure 6a ) ( bootstrap statistic: all ps<0 . 01 ) and the regression yielded an estimated time per growth cone shift of 24 ms ( 95% confidence interval; 18–29 ms ) , in accordance with the estimate of Experiment 1 . Thus , perceptual grouping of 2D image regions invokes a serial grouping process with a speed that depends on their scale: wide regions are grouped quickly and narrow regions more slowly . 10 . 7554/eLife . 14320 . 009Figure 6 . Model fits for the different types of images ( Experiment 2 ) . The bars show the variance in RT explained by the different models for ( a ) scrambled cartoons; ( b ) cartoon outlines; ( c ) detailed cartoons; and ( d ) color pictures . G , growth-cone; F , filling-in; E , Euclidean; P , pixel-by-pixel model . Asterisks ( *** ) represent p<0 . 01 ( bootstrap test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 009 We constructed the scrambled cartoons with the aim to make the objects difficult to recognize , as object-recognition can influence perceptual grouping . We did , however , also test the cartoon images in 22 subjects ( Figure 6b ) . The growth cone model explained 41% of the variance , which is more than any other model ( bootstrap: all ps<0 . 01 ) . However , the growth cone model was not better than the other models for the detailed cartoons with interior contours and for the color pictures ( Figure 6c , d ) . These results reveal that the growth-cone model provides an excellent account for the pattern of RTs for scrambled cartoons and cartoon outlines , but that the predictions for the detailed cartoons ( with internal contours ) and colored pictures are worse . It follows that the interior lines of the detailed cartoons and the colors and textures of the picture stimuli influence grouping speed . We only used the outlines of the shapes to estimate growth-cone size , which may explain why predictions worsened . Future grouping models might explain additional variance by taking interior features , texture gradients and color transitions into account , but these new models would require additional experimental and theoretical work . Previous studies suggested that grouping depends on the spread of object-based attention across the features that need to be bound in perception ( Rensink , 2000; Roelfsema , 2006; Roelfsema et al . , 2007 ) . So far , our experiments measured reaction times but we did not test the spread of 'attention' . Our third experiment measured the spread of object-based attention , capitalizing on the Egly cueing paradigm ( Egly et al . , 1994 ) to measure the speed . Egly et al . , 1994 used displays similar to the ones in Figure 7a . Their participants saw two bars , then one bar was cued at one of its ends and finally a target was presented to which the participants responded with a button press . If the cue was valid because the target appeared at the same location as the cue , the participants’ responses were faster than if it appeared at another location , because spatial attention was summoned by the cue . Their main finding , however , was that if the cue was invalid and the target appeared at the other end of the cued bar ( invalid same-object trials ) , the response was faster than if it appeared at the non-cued bar ( invalid different-object trials ) , even if the distance between the target and the invalid cue was constant . The reaction-time difference between invalid same- and different object trials is a measure for object-based attention , which is hypothesized to select the entire cued bar ( Egly et al . , 1994; Xu and Chun , 2007; Lamy and Egeth , 2002; Drummond & Shomstein 2010 ) . 10 . 7554/eLife . 14320 . 010Figure 7 . The time-course of spreading object-based attention ( Experiment 3; actual stimuli were white on a black background ) . ( a ) We presented two broad ( left ) or narrow bars ( right ) and one bar was cued at one of its ends for 100 ms . After a variable delay , we presented a target dot at the cued location ( valid trials , see inset ) or at one of two locations that were equidistant from the cue , on the same ( invalid same ) or on the other bar ( invalid different ) , i . e . only one target dot per trial . In catch trials ( not shown ) , the target did not appear . ( b ) Reaction times for the validly cued trials for the broad ( red ) and narrow bars ( blue ) did not differ ( t-test , p>0 . 1 ) . Reaction times for the validly ( black ) cued locations were faster than those for the invalidly ( grey ) cued locations ( * , p<0 . 01 in t-test ) . ( c ) The object-based advantage ( RTinvalid different – RTinvalid same ) as a function of cue-target onset asynchrony . The curves show the fit of Gaussian functions to the object-based advantage for the broad ( red ) and narrow bars ( blue ) , respectively . The red and blue horizontal bars on the x-axis indicate the 95% confidence interval of the peak of the Gaussian function as measured with a bootstrap method . Error bars represent s . e . m . of the data points . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 010 In our version of the experiment , eleven participants maintained gaze at a fixation point ( controlled with an eye-tracker ) and they saw two horizontal or vertical bars . We cued a corner of one of the bars ( Figure 7a ) and in 80% of trials the participants reported the appearance of a red target dot at one of three positions by pressing a button . The remaining 20% were catch trials that did not require a button press . In target trials , we presented the target dot at the cued location ( 60% , valid trials ) , at the opposite side of the same bar ( 20% , invalid same object ) , or in the other bar ( 20% , invalid different object ) . We varied the time between cue and target onset ( between 200 and 600 ms ) to measure the time-course of the object-based advantage ( invalid same trials vs . invalid different trials ) . To examine if the speed of object-based attention depends on the size of image regions , we presented either broad bars ( 3 . 9° wide , red panels in Figure 7a ) or narrow bars ( 2° , blue panels ) . The growth-cone model predicts that the spread of object-based attention is faster for broader bars because growth-cones are larger; we estimated that there were 3 . 3 and 7 . 2 growth-cones between cue and target for the broad and narrow bar , respectively ( see Materials and methods ) . The average accuracy across participants was 99% . On validly cued trials , bar width did not influence RT ( t ( 10 ) =−1 . 450 , p>0 . 1; Figure 7b ) . As expected , RTs were shorter on validly cued than invalidly cued trials ( t ( 10 ) =−3 . 468 , p<0 . 01 ) . We computed the object-based advantage for the broad and narrow bars and found that RTs were shorter with invalid cues on the same object than on a different object , an object-based advantage that is in accordance with previous studies ( Egly et al . , 1994; Xu and Chun , 2007; Lamy and Egeth , 2002; Drummond & Shomstein 2010 ) . The maximal object-based advantage occurred earlier for the broad bar than for the narrow bar ( Figure 7c ) . We fitted a Gaussian function to the object-based benefit and found that it peaked after 344 ms ( 95% confidence interval: 311–376 ms ) for the broad bar and after 463 ms ( 434–492 ms ) for the narrow bar ( p<0 . 001; bootstrap analysis ) . We estimated the speed of attention spreading by dividing the time difference of 119 ms by the difference in the estimated number of growth cones ( 7 . 2–3 . 3=3 . 9 ) and obtained a speed of 31 ms per growth cone , compatible with the speed estimates of Experiments 1 and 2 . The new growth-cone model holds that grouping speed depends on the size of homogenous image regions . The variation in the scale of processing could rely on multiple stages of the visual cortical hierarchy where neurons have different RF sizes ( Figure 8 ) . Near boundaries and in narrow parts of an object , the perceptual grouping process requires a high spatial resolution , which can be supported by small receptive fields in lower areas ( purple and blue in Figure 8 ) . In these areas the horizontal connections , responsible for the spread of enhanced activity , link neurons that represent nearby locations in visual space . If the homogeneous image regions are larger , the propagation may take place in higher areas with larger RFs so that the grouping speed is higher ( red in Figure 8 ) . This scale dependence of the spread of object-based attention is in accordance with neuroimaging studies revealing the stronger engagement of lower visual areas when the relevant spatial scale is finer ( Rijpkema et al . , 2008; Hopf et al . , 2006 ) . Future studies could aim to unravel the interactions between neurons at different levels of the visual cortical hierarchy during image parsing . 10 . 7554/eLife . 14320 . 011Figure 8 . Schematic representation of the contribution of different visual areas to perceptual grouping . Horizontal rows illustrate low- , mid- , and high-level visual areas with larger receptive fields in higher areas . The four columns illustrate different time steps during the grouping process; after the presentation of the cue , and after 1 , 3 and 7 growth-cone shifts . The labeling process begins at the cued location . Higher cortical areas with large receptive fields make great strides in the propagation of enhanced neuronal activity and this fast progress also impacts on lower areas through feedback connections ( downward pointing arrows ) . However , the higher visual areas cannot resolve fine-scale details and the grouping of narrower image regions therefore relies on the propagation of enhanced neuronal activity in lower visual areas with smaller receptive fields . Darker colors represent image regions that have been recently reached by the grouping process and lighter colors denote image regions that were labeled at an earlier point in time . White circles represent receptive fields that have not been reached by the grouping process . Note that the labeling process is serial and that the speed of grouping depends on the size of the receptive fields that contribute to the grouping process . DOI: http://dx . doi . org/10 . 7554/eLife . 14320 . 011 Although the growth-cone model capitalizes on the presence of receptive fields with different sizes at distinct levels of the visual cortical processing hierarchy , it does not require the receptive field size to be constant within a visual cortical area . Indeed , it is well known that the size of receptive fields increases with eccentricity . Suppose that the task is to group an elongated bar with a width of 2° ( as in Figure 7a ) . The propagation of enhanced neuronal activity would take place in area V1 for the more eccentric parts of the bar ( e . g . at 10° eccentricity ) ( Gattass et al . , 1981 ) but area V4 would make fastest progress where the bar is nearer to the fovea ( e . g . at 2° eccentricity ) ( Gattass et al . , 1988 ) . The horizontal distance between neurons with abutting , non-overlapping RFs is relatively constant across visual areas ( Harvey and Dumoulin , 2011 ) . Thus , if we assume that the propagation of enhanced neuronal activity through horizontal connections is also similar , it follows that the propagation speed ( in °/s or growth-cones/s ) should also be constant across eccentricities for a bar with a constant width . The growth-cone mechanism is therefore compatible with the fact that the receptive field size in visual cortical areas increases with eccentricity . The current implementation of the growth-cone model is based on the outlines of the shapes only and , accordingly , the interior features of detailed cartoons and color pictures decreased fit quality . Consider , for example , the picture with lemurs of Figure 6c . The interior contours in the tails form closed , disconnected compartments that cannot be grouped by the growth-cone model . The overall influence of the interior contours is difficult to predict , however . On the one hand , they might slow down parsing by presenting barriers . On the other hand , they might facilitate object recognition and thereby speed up the grouping . Hence , the interior contours , the influence of color and texture , as well as the influence of object-recognition go beyond the simple models considered by us so that variance had to remain unexplained . Importantly , perceptual grouping with the same color pictures also relies on a serial process ( Korjoukov et al . , 2012 ) . It would therefore be of interest to generalize the growth-cone model to account for perceptual grouping of objects with interior contours , colors , and textures . We expect that these generalizations necessitate a mechanism for object recognition . In the picture of the lemur ( Figure 6c , d ) , for example , we see that the left tail belongs to the right lemur , because we know what a lemur looks like . Segmentation based on the characteristic shape of objects is known as 'semantic segmentation' . Models of segmentation can take advantage of the recent breakthroughs in hierarchical convolutional neural networks for object recognition ( e . g . Krizhevsky et al . ( 2012 ) ; Zeiler and Fergus ( 2013 ) ; reviewed by LeCun et al . , 2015 ) . These networks consist of multiple hierarchically organized layers , which transform simple features at the lower layers , resembling lower visual cortical areas , to representations for shape recognition in higher layers , resembling the inferotemporal cortex ( Güçlü and van Gerven , 2015; Yamins and DiCarlo , 2016 ) . Recent studies have started to apply hierarchical convolutional networks to achieve semantic segmentation ( Hong et al . , 2015; Noh et al . , 2015 ) . The resulting networks can segment pictures of everyday visual scenes . They first use a convolutional network to determine the semantic categories present in the visual scene , which is followed by a 'deconvolutional network' that retrieves the lower-level features that are part of a particular semantic category . Hence , these models explain why object-recognition facilitates image parsing in human perception ( Vecera and Farah , 1997; Peterson et al . , 1991; Korjoukov et al . , 2012 ) . The deconvolutional network could be implemented in the visual cortex as a top-down influence from neurons in shape selective cortical areas to neurons in lower visual areas ( i . e . generalizing the feedback connections in Figure 8 ) ( Hochstein and Ahissar , 2002 ) . We expect that semantic segmentation networks may need to also include horizontal connections in order to account for the serial patterns of reaction times in grouping tasks with natural images ( as in Korjoukov et al . , 2012 ) . According to this view , the parsing of a natural scene would invoke an incremental grouping process that gradually adds simple features and more complex shape fragments to an evolving perceptual group ( Roelfsema , 2006; Roelfsema and Houtkamp , 2011 ) . We conjectured that grouping relies on the spread of object-based attention within homogenous image regions , and this spread should therefore depend on the size of these regions . Our third experiment confirmed this prediction by showing that the spread of object-based attention is slower for narrow regions . The last experiment yielded an estimate of the attentional propagation speed of ~30 ms per growth cone , which is similar to the estimates of 22 and 24 ms/growth cone of the other two experiments . In the cortex , a shift time of ~25 ms would correspond to the propagation of activity between neurons with abutting , non-overlapping RFs . In V1 , the distance between neurons with adjacent RFs is 2–4 mm , and it is only slightly larger in higher visual areas ( Hubel and Wiesel , 1974; Harvey and Dumoulin , 2011 ) . Our results therefore suggest that the propagation speed is 8–16 cm/s ( 2–4 mm/ 25 ms ) , which falls neatly in the range of previous neurophysiological measures for the horizontal propagation of neuronal activity in visual cortex ( Hirsch and Gilbert , 1991; Nauhaus et al . , 2009; Sato and Carandini , 2012 ) . Our results suggest that perceptual grouping of 2D image regions calls on a serial process that takes tens to hundreds of milliseconds for naturalistic pictures . Subjects apparently need this time to spread object-based attention within in the interior of image regions that need to be grouped in perception . The typical time between two saccadic eye movements is ~200–300 ms and it is therefore tempting to speculate that the duration of visual fixations is optimized to allot time for both object recognition and image parsing before the next saccade is made . The participants reported normal or corrected-to-normal acuity and were paid for their participation . The Ethics Committee at the University of Amsterdam approved the experiments . Informed consent was obtained before the start of the experiment . Eye position was sampled at 1000 Hz with an EyeLink eye-tracking system ( SR Research Ltd ) . The experiments were performed in a dimly lit room . We used a chinrest and the participants sat at a distance of 57 cm from a CRT screen . They practiced the task before data collection started .	When we look at an object , we perceive it as a whole . However , this is not how the brain processes objects . Instead , cells at early stages of the visual system respond selectively to single features of the object , such as edges . Moreover , each cell responds to its target feature in only a small region of space known as its receptive field . At higher levels of the visual system , cells respond to more complex features: angles rather than edges , for example . The receptive fields of the cells are also larger . For us to see an object , the brain must therefore 'stitch' together diverse features into a unified impression . This process is termed perceptual grouping . But how does it work ? Jeurissen et al . hypothesized that this process depends on the visual system’s attention spreading over a region in the image occupied by an object , and that the speed of the process will depend on the size of the receptive fields involved . If an image region is narrow , the visual system must recruit cells with small receptive fields to process the individual features . Grouping will therefore be slow . By contrast , if the object consists of large uniform areas lacking in detail , grouping should be fast . These assumptions give rise to a model called the “growth-conemodel” , which makes a number of specific predictions about reaction times during perceptual grouping . Jeurissen et al . tested the growth-cone model’s predictions by measuring the speed of perceptual grouping in 160 human volunteers . These volunteers looked at an image made up of two simple shapes , and reported whether two dots fell on the same or different shapes . The results supported the growth-cone model . People were able to group large and uniform areas quickly , but were slower for narrow areas . Grouping also took more time when the distance between the dots increased . Hence , perceptual grouping of everyday objects calls on a step-by-step process that resembles solving a small maze . The results also revealed that perceptual grouping of simple shapes relies on the spreading of visual attention over the relevant object . Furthermore , the data support the hypothesis that perceptual grouping makes use of the different sizes of receptive fields at various levels of the visual system . Further research will be needed to translate these findings to the more complex natural scenes we encounter in our daily lives .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Serial grouping of 2D-image regions with object-based attention in humans
As the biodiversity crisis continues , we must redouble efforts to understand and curb pressures pushing species closer to extinction . One major driver is the unsustainable trade of wildlife . Trade in internationally regulated species gains the most research attention , but this only accounts for a minority of traded species and we risk failing to appreciate the scale and impacts of unregulated legal trade . Despite being legal , trade puts pressure on wild species via direct collection , introduced pathogens , and invasive species . Smaller species-rich vertebrates , such as reptiles , fish , and amphibians , may be particularly vulnerable to trading because of gaps in regulations , small distributions , and demand of novel species . Here , we combine data from five sources: online web searches in six languages , Convention on International Trade in Endangered Species ( CITES ) trade database , Law Enforcement Management Information System ( LEMIS ) trade inventory , IUCN assessments , and a recent literature review , to characterise the global trade in amphibians , and also map use by purpose including meat , pets , medicinal , and for research . We show that 1215 species are being traded ( 17% of amphibian species ) , almost three times previous recorded numbers , 345 are threatened , and 100 Data Deficient or unassessed . Traded species origin hotspots include South America , China , and Central Africa; sources indicate 42% of amphibians are taken from the wild . Newly described species can be rapidly traded ( mean time lag of 6 . 5 years ) , including threatened and unassessed species . The scale and limited regulation of the amphibian trade , paired with the triptych of connected pressures ( collection , pathogens , invasive species ) , warrants a re-examination of the wildlife trade status quo , application of the precautionary principle in regard to wildlife trade , and a renewed push to achieve global biodiversity goals . At the close of a ‘decade of biodiversity’ , we have failed to meet any of the Aichi targets designed to safeguard biodiversity ( CBD , 2020 ) . One important driver of biodiversity loss is unsustainable wildlife exploitation ( IPBES , 2019 ) . Countering illegal wildlife trade is critical to limiting biodiversity loss; however , focusing solely on illegal wildlife trade can miss a potentially greater issue: that of legal wildlife trade . Gaps in trade regulations in terms of species covered by international regulation such as by the Convention on International Trade in Endangered Species ( CITES ) leave groups like amphibians and reptiles among the most frequently traded animals ( Herrel and van der Meijden , 2014 ) and largely outside the control of such conventions . Previous studies aiming to quantify global patterns of trade have relied upon accessible data ( such as CITES and IUCN data; i . e . , Scheffers et al . , 2019 ) ; relying on regulator data can miss critical legal un/under-regulated trade , as evidenced by analysis on reptiles which highlighted the proportion of species in trade fall outside the scope of CITES ( Marshall et al . , 2020 ) . Such analysis risks providing a false sense of assurance that we understand the dimensions of trade , while in reality the trade may be spanning far more species than those actively monitored ( Marshall et al . , 2020 ) . Marshall et al . , 2020 , highlighted the discrepancy in protection within the reptile trade , with only 8 . 3% under CITES regulations yet over 36% in trade and over 70% of individuals from some taxa ( e . g . , lizards ) harvested from the wild ( Marshall et al . , 2020; Uetz et al . , 2021 ) . Whilst trade of wild-collected individuals is not necessarily unsustainable , such a judgement should rely on data , as unregulated harvest from the wild , especially for rare or small-ranged species could potentially pose a significant risk to the continued survival of such populations ( Auliya et al . , 2016 ) . The need for a complete assessment of amphibian species in trade , their origins , and where native populations are at risk is emphasised by targeted studies revealing high rates ( 87% of individual Southeast Asian newts ) of wild collection ( Rowley et al . , 2016 ) . Given that species can be restricted to single drainage basins , unsustainable trade can represent a genuine risk to species future survival; limited trade assessments means that understanding when trade is or is not sustainable simply is not possible for many species , though recent studies show it can have an impact on population viability ( Morton et al . , 2021 ) . Despite experiencing similar pressures to reptiles and greater sensitivity to perturbations ( Stuart et al . , 2004 ) , amphibians are one of the least protected taxa under CITES regulation with only 2 . 4% of all known species listed ( second only to fish at 0 . 46%: http://www . fishbase . org/home . htm ) , despite showing faster population declines than any other vertebrate group ( Hoffmann et al . , 2010 ) . Often dubbed canaries in the coal-mine amphibians are sensitive to a myriad of anthropogenic stressors: pollution ( Blaustein et al . , 2003 ) , habitat loss ( Stuart et al . , 2004 ) , atmospheric changes ( Blaustein et al . , 2003 ) , introduced pathogens ( Lips , 2016 ) , invasive species ( Bellard et al . , 2016 ) , wildlife collection ( Phimmachak et al . , 2012 ) , and agricultural chemicals ( Trudeau et al . , 2020 ) ; such stressors are exacerbated by amphibians’ frequently small distributions and naturally fluctuating populations ( Nori et al . , 2018; Luo et al . , 2015; Hu et al . , 2012 ) . Amphibian trade is directly tied to the last three stressors . Trade can enable pathogen spread ( O'Hanlon et al . , 2018 ) , which has facilitated devastating amphibian species loss ( Scheele et al . , 2019; but see Lambert et al . , 2020 , for concerns over the number of species ) . Invasive amphibians ( often linked to trade; Lockwood et al . , 2019; Stringham and Lockwood , 2018 ) can be vectors for pathogen spread ( Bienentreu and Lesbarrères , 2020; Feldmeier et al . , 2016 ) , but also can compete with native species for resources such as space and prey ( Falaschi et al . , 2020 ) . Wild collection ( directly taking animals from the wild ) occurs at several scales: on local levels , humans collecting species for trade , consumption , and medicine ( Ribas and Poonlaphdecha , 2017; Van Vliet et al . , 2017; Onadeko et al . , 2011 ) , whereas more widely amphibian trade is augmented by demand for pharmaceutical products , pets , and even fashion ( Auliya et al . , 2016; Xiao et al . , 2011 ) . A recent literature assessment of amphibian pet trade found 443 traded species ( Mohanty and Measey , 2019 ) , but as we strive towards ever more complete and representative assessments of the amphibian trade , we must capture trade other than pets , as well as outside of literature ( that can often be skewed towards certain languages/regions; Konno et al . , 2020 ) . More standardised and comprehensive data are necessary to ensure that wildlife trade avoids harming species’ long-term survival prospects; the current lack of data and thus lack of transparency or access to baseline population data and compiled trade records frustrate trade mitigation efforts . Here , we aim to map amphibian species in trade , complementing previous regional efforts ( Yap et al . , 2015 ) , or those focusing on easily accessible data such as CITES ( CITES trade database; https://trade . cites . org ) and LEMIS ( United States Fish and Wildlife Service’s Law Enforcement Management Information System ) . We explore two major inventories of international trade , combining this with an automated web search of amphibian selling websites across six languages . We place these findings in the context of the findings of the Mohanty and Measey , 2019 , and species reported as traded within the IUCN Redlist species assessments . In addition , we examine the overlap between these five trade data sources and explore the different trade dimensions they represent , and how the trade may impact wild populations . We further explore where species origins and their threat status , thereby attempting to highlight trade vulnerability hotspots . This study builds towards a comprehensive assessment of amphibian trade , while attempting to highlight how many species are traded , the major drivers of trade , and where these species originate . Our online search efforts successfully examined a total of 139 amphibian selling websites and retrieved 2766 web pages to be searched ( mean of 19 . 91 ± 3 . 95 pages per website; range 1–302 ) . Our temporal online sample ( 2004–2019 ) added an additional 4568 pages , meaning our complete online species list is based on searches across 7334 pages in total . We detected 480 keywords ( i . e . , amphibian scientific names and synonyms ) that equated to 442 species in the 2020 snapshot , and 486 keywords that equated to 443 species in the temporal sample , resulting in a total of 575 species detected in the Online trade . Overall , the three data sources ( Online trade , LEMIS , and CITES trade database ) contained 909 species in total ( 11 . 06% of the 8212 total described amphibian species ) , of which LEMIS had the most ( 587 species , 31% unique ) , followed by Online trade ( 575 species , 30% unique ) then CITES ( 137 species , 4% unique ) . Most of this trade was commercial ( 99 . 6% ) with only 0 . 4% non-commercial . Unsurprisingly , anurans ( 729 species ) dominated the trade , followed by salamanders ( 162 species ) and caecilians ( 18 species ) . Based on these three trade inventories , a total of 157 species were threatened ( i . e . , listed as Vulnerable ( VU , EN , CR ) or worse on the IUCN Redlist ) , 27 Data Deficient , and 39 unassessed , and the remainder Least Concern ( Figure 1 ) . Whilst the majority of species in trade in CITES have a CITES appendix ( 95% ) , this is not the case for species detected via LEMIS ( 14% ) or online searches ( 16% ) . In terms of the degree of threat , 47% of species in trade via CITES are threatened according to the IUCN and 12% are unlisted by the IUCN , whereas this is lower for LEMIS ( 24%; 5% ) and Online ( 23%; 6% ) . However , due to the larger number of species traded , species detected via LEMIS and online searches account for a larger proportion of all threatened amphibian species . For example , 4% of Critically Endangered species and 5% of Endangered species were detected in trade via LEMIS , compared to 2% and 3% for CITES . In total , relying exclusively on CITES would suggest only 3% of threatened species are traded , whereas LEMIS and Online reveal 5% of threatened species traded , with most threatened species in trade not listed by CITES . Mapping reveals a global exploitation of amphibians . However , the number of species exploited in different regions varies dramatically ( Figure 2; Figure 2—figure supplement 1 ) . Both LEMIS and Online trade highlight high numbers of species across the tropics , especially in the Amazon . However , LEMIS highlights more traded species in Africa and Southeast Asia , and CITES misses almost all areas with only a fraction of species in the Amazon ( poison dart frogs ) covered ( Figure 2—figure supplement 2 and 3 ) . Particularly high proportions of species were in trade , not only in less diverse regions , but also across tropical Asian regions . In addition , particularly high percentages of species are in trade in South Cambodia and areas of Madagascar ( Figure 2—figure supplement 2 and 3 ) . Many traded species categorised as Vulnerable or worse originate from East and Southeast Asia , in addition to the Mediterranean and various parts of South America ( Figure 2—figure supplement 2 and 4 ) , whereas small-ranged species are in trade from across the tropics and various islands . At the national level , countries across the Middle East and Southeast Asia had more than half their species in trade classed as either threatened or Data Deficient/unassessed . South America , Madagascar , and the Caribbean have even higher percentages of threatened species in trade . South America and Southeast Asia have the highest numbers of species in trade without CITES regulations . The LEMIS trade inventory provides us with greater insights into the source of the amphibians being traded . Of the trade described in LEMIS 2000–2014 , and constituting/representing single individual animals , 99 . 9% is not from seizure and enters the USA ( 69 , 688 , 337/69 , 771 , 677 ) , and the vast majority is for commercial purposes ( 69 , 492 , 478/69 , 771 , 677; 99 . 6% ) . Of the 69 , 771 , 677 amphibians imported into the USA , recorded by LEMIS , 57 . 2% ( 39 , 921 , 289 ) are listed as captive sourced , leaving 42 . 3% ( 29 , 522 , 128 ) as originating from the wild ( the remaining 0 . 47% , 328 , 260 , classed as other or with an ambiguous source ) . The wild capture volumes and percentages vary among genera , from millions of individuals to fewer than 100 ( Figure 3—figure supplements 1–6 ) . The vast majority of imported genera are impacted by wild capture ( 254/259 ) with 141 genera exclusively wild-sourced; five genera are fully sourced from captive operations ( Peltophryne , Ranitomeya , Calyptocephalella , Cryptophyllobates , Samandrella; Figure 3—figure supplements 1–6 ) . On average 84 . 2% of each genera’s individuals come from the wild , and a per genera median of 100% is likely driven by the large number of genera exclusively taken from the wild but in much lower volumes ( e . g . , fewer than 100 individuals , or fewer than 10 individuals per year given the 2000–2014 timeframe; Figure 3—figure supplement 6 ) . Whilst the CITES trade has remained relatively consistent over time between 2000 and 2020 at around 50 species a year with a gradual increase of species , LEMIS shows an increase up to 2014 ( the limit of available data ) at 310 species ( Figure 3A ) . The Online trade shows much more interannual variation ( likely exaggerated by sampling effort fluctuations ) , increasing to 200 species in 2010 , decreasing up to 2014 at under 100 species , then increasing again up to over 200 species in 2019 . The number of pages scraped for online trade also followed this trend , peaking at over 1250 pages in 2014 , decreasing to under 200 in 2014 then increasing to over 1000 in 2018 ( Figure 3B ) . The residuals from a linear regression accounting for the number of pages searched suggests a steady increase in species ( Figure 3B ) . Thirty-eight species described since 1999 ( 1 . 38% of the 2747 amphibian species described after 1999; Figure 4A and B ) appeared in trade based on our three inventories ( and 41 with the addition of two further species described in 2018 and listed for sale in 2020; Altherr and Lameter , 2020 ) . Eight only appeared in the 2019 snapshot , so are discounted from time lag calculations , leaving 30 species with connected trade years and a mean time lag of 6 . 5±0 . 78 years between species description and appearance in the trade . Of the 38 species , 12 are Least Concern , 10 are unevaluated , three are Data Deficient , and 13 are threatened ( one of which is Critically Endangered ) . One species was in trade the year after it was described , but four were in trade in the second year , four in the third year , and seven within 4–5 years ( Figure 4C ) . We cannot differentiate instances of rapid exploitation after species description from instances of name updates pertaining to species already traded . Although it should be noted that even in these cases given the smaller population sizes and distributions of split species , they may be more vulnerable to population declines resulting from wild-harvest , as populations and ranges are likely to be smaller than currently known . Different language searches returned different species lists , with all languages containing species unique to that language . English and German detected the most species by far ( 293 , 289 ) , and each also contained the highest rates of unique species ( 81 , 97 ) . German produced a larger list of species , despite similar sampling efforts as Spanish , French , Japanese , and Portuguese ( Figure 5 ) . The top websites in terms of species were mostly commercial ( six out of the top ten ) , two of which prominently advertised wholesale options . The remaining four top websites ( including the top website with 278 species ) were hosting classified advertisements . To better capture all the species traded , we combined our contemporary analyses from the three data sources ( Online trade , LEMIS , and CITES trade database ) with the analyses from Mohanty and Measey , 2019 , and the IUCN Redlist assessments . Comparisons reveal that different sources detected different species in the trade , and no single source is sufficient to detect all species traded ( Figure 6 ) . Combining all sources yielded a total of 1500 amphibian species in trade before synonyms were removed , and 1215 once synonyms were removed , equivalent to 17% of amphibian species . The 1215 species included up to 413 species used for meat ( though a significant number were largely local consumption based on IUCN assessments ) , 805 species as pets ( though six are from separate lists: one from Germany; Altherr and Lameter , 2020; five from Asia; Choquette et al . , 2020 ) , 122 species used as medicine or in pharmacological research , and 664 species imported for research or breeding facilities ( including zoos and aquaria ) ; other purposes were also listed ( various fashion companies such as Prada and Gucci were listed as importers , and some amphibians are imported for bait ) but we have not listed these uses separately . In total over 930 species were used for other commercial purposes , and 1215 species in total when medicinal/pharmaceutical and research are included . In terms of status , 4% of species in trade are Critically Endangered ( 4% for pets , 4% for meat ) , 10% are Data Deficient or unassessed ( 9% pets , 11% meat , over 8% used in medicine or pharmacology ) . In total , 22% of species in trade are threatened ( i . e . , Vulnerable or worse , 28% when Near-Threatened are also considered ) , 25% for pets , 31% for meat , 39% for medical purposes and only 21% of those used for research . In terms of coverage of species for each type of trade by CITES ( 12% overall 151/1215 ) , this varied from 5% of species used for meat , to 16% of those used for pets or 18% for medicine , and 16% of those in research . Mapping out these patterns also revealed a variety of trends among different uses ( Figure 7 ) . In terms of commercial trade , pet trade dominated the global trade of amphibians and the pattern is most similar to the map of all trade with up to 51 species from any given area shown to be in trade for pets relative to the 71 from all trade . Trade for meat is more limited with only up to 26 species from any given area in trade , and up to eight species for medicine or pharmaceutical trade . Interestingly , research/zoos were associated with up to 57 species from any given area in trade and broadly mirroring the patterns seen in the pet trade . It should be noted that these may be underestimates , as the LEMIS does not specify exact purpose , and it must be inferred from the buyer and type of sale . Whilst the volumes likely differ substantially between animals traded for research relative to commercial sources , it highlights the numbers of species potentially vulnerable to at least low levels of international trade . Commercial trade of amphibians for meat is also shown to be from Asia using the United Nations Commodity Trade Statistics Database ( UN Comtrade: https://comtrade . un . org/data/ ) which shows that global export of frog legs is dominated by Indonesia ( at 8 , 005 , 997 kg in 2008–2009 alone ) , followed by China , Vietnam , and other Asian nations with the dominant markets in France , Belgium , and the USA , though these statistics are only available until 2010 and markets seem to be both growing and diversifying at that point , based on data available in the preceding years . Amphibian declines are often considered to provide an early warning of potential declines in other taxa as they are sensitive to pollution and habitat loss making their absence an early warning sign of habitat degradation; sensitivity to change combined with trade , and disease risk creates the perfect storm threatening future amphibian survival . Whilst regional and some global studies have explored the extent of pet trade ( Measey et al . , 2019 ) , or meat trade ( Carpenter et al . , 2014 ) , a well over double the known number of species are in trade relative to previous studies ( i . e . , Scheffers et al . , 2019 , 542 relative to 1215 ) , as well as a more representative understanding of what is currently in trade and how it has changed over the last two decades . The scope of the amphibian trade is larger than formerly realised with implications for the direct exploitation of these species , disease spread ( Schloegel et al . , 2009 ) , and the pool of potentially new invasive species ( Gippet and Bertelsmeier , 2021 ) . Each dataset we examined included unique species missing from the other datasets ( Figure 6 ) , illustrating the need to use multiple sources to characterise wildlife trade , and underscoring the need for a better system to centralise knowledge on what is being traded , and where animals are sourced . Concerns over the scale and scope of the trade are compounded by the lack of baseline population studies , frustrating efforts to truly understand sustainability of the trade , as understanding sustainable offtake is impossible without baseline population data . A recent meta-analysis on how trade impacts wild populations was unable to generate an estimate on amphibians because of a lack of standardised studies , but revealed abundance declines of 62% ( 95% CI 20–82% ) in traded populations of mammals , birds , and reptiles ( Morton et al . , 2021 ) . Amphibians in areas with high volumes of exports may be at particular risk given the high rates of wild capture . For example , meat trade is known to impact at least 40 species annually from Indonesia alone ( Gratwicke et al . , 2010 ) , with many coming directly from the wild , and even captive rearing facilities risk endangering wild species through pathogen exposure unless biosafety standards are improved . Understanding the impacts of harvest and trade on source populations requires a better understanding of what species are being traded , the volumes in trade and the status of the wild populations is critical for preventing negative impacts on source populations , especially given that the IUCN assessments can be decades old and not accurately reflect species’ current threat status ( Natusch and Lyons , 2012 ) . Furthermore , quantitative analysis of the volumes of species in trade often relies on import data ( e . g . , LEMIS ) and ignores mortality during transit and transport , which has been shown to be as high as 72% in some studies ( Ashley et al . , 2014 ) , with mortality in amphibians higher than all other groups ( 45% within 10 days of confiscation ) . Such statistics are alarming , and also highlight that the number of animals exported may be far higher than the anticipated demand to compensate for mortality before sale . Despite the impact of trade , the World Customs Organization still fails to list species data in exports – only basic data is needed to legally export most amphibians , providing no species-specific information to enable trade monitoring . With limited baselines on populations and disparate or inaccessible records of trade , we cannot hope to make effective management decisions or develop quotas and tools for sustainable use . A lack of systematic monitoring of global trade limits us to a basic understanding of traded species , origin , and impacts on native species . Monitoring deficiencies have been repeatedly highlighted over the past decade , but we still await the policy responses necessary to ensure the survival of vulnerable species ( Auliya et al . , 2016 ) . In fact , government funding for projects targeting basic monitoring initiatives has dwindled in recent years in favour of applied scientific applications , and ‘less charismatic’ species are most likely to be underfunded ( Bellon , 2019 ) and have lower investment in conservation ( Gerber , 2016 ) . We show 22% of the 1215 species in trade are threatened ( i . e . , IUCN Redlist status of Vulnerable or worse ) , and a further 8% remain unassessed or Data Deficient . One in ten traded species are already highly threatened ( 11% of species Endangered or Critically Endangered ) . The trade extends beyond captive-reared or ranched individuals , and is motivated in part by novelty and rarity ( as has been documented for the reptile trade previously; Marshall et al . , 2020; Lyons and Natusch , 2013 ) , potentially further illustrated by the appearance of 38 species described since 2000 in the trade . Whether these new species are the result of species splits or completely novel lineages being described , they highlight the knowledge gaps that need to be addressed before sustainability can be confidently assessed . However , Stringham et al . , 2021 , showed that new ( reptile ) species smuggled in Australia were well predicted by their existence in US markets , thereby suggesting a diminished role for novelty ( i . e . , recent description ) when compared to accessibility . Because of novelty dynamics in trade and changing taxonomy , CITES appears an inadequate tool to describe taxonomic or spatial trade patterns; CITES does not include 97 . 5% of amphibian species , and fails to provide any default ( or sufficiently rapid ) protection for newly described and potentially vulnerable species , and even scientific descriptions of species have been found to enable these newly described species to be targeted for trade ( Yang and Chan , 2015; Yeager et al . , 2020 ) . Tropical regions and islands , with high levels of endemism , still have a significant proportion ( often exceeding one-third or even half ) of species traded indicating the need to expand trade monitoring , and to prevent trade as a default until non-detriment findings can be assessed for any potential trade . Global monitoring continues to be inadequate; the lack of specificity hinders the utility of global data from the World Customs Organization ( Chan et al . , 2015 ) . Calls for improvements and increased specificity were made at the IUCN’s 5th World Conservation Congress ( WCC-2012-Res020 ) in 2012 . Changes remain elusive , with details on updates in the World Customs Organization , 2020 , edition failing to address animal trade ( World Customs Organization , 2020 ) . Thus , a decade has passed and reasonable actions for the conservation of biodiversity are still ignored in economically orientated databases . The dearth of reliable/accessible data ( both for baseline population and trade volumes ) undermines efforts to determine trade sustainability for the vast majority of non-CITES species ( i . e . , the vast majority of all amphibian species ) . The trade of Endangered and range-limited species , paired with the high rates of wild capture ( especially given that this is higher for pets than for other purposes ) , would suggest much of the trade could be unsustainable and damaging the future survival of species . To date , 94 cases out of the 159 extinct and potentially extinct species from the 2008 Global Amphibian Assessment are at least partially attributed to Batrachochytrium dendrobatidis ( Bd ) ( MacCulloch , 2008; Picco and Collins , 2008 ) , and suggestions that Bd is likely to be responsible for up to 500 species declines ( Scheele et al . , 2019; but see Lambert et al . , 2020 for discussion on the 500+ estimate ) . Furthermore Bd , Batrachochytrium salamandrivorans ( Bsal ) , Ranavirus and a range of other diseases , carried by amphibians and fish , can spread into naïve populations and move between aquatic taxa ( Bayley et al . , 2013; Mao et al . , 1999; Densmore and Green , 2007 ) . With millions of individuals exported annually ( peaking at around 5575K kg from Indonesia alone in a single year in the early 1990s , and fluctuating between 3600K and 5000K kg most years based on LEMIS ) , no systemic mechanism to ensure correct identity , and poor biosafety standards , water contamination resulting from continued unrestricted/uncontrolled trade is likely to lead to further disease spread , and population declines . Rates of Bd in live exports can be high ( over 60% of individuals ) , with studies linking the spread of Bd and Bsal to the trade of live animals in the pet trade ( Fitzpatrick et al . , 2018; Kriger and Hero , 2009; Yuan et al . , 2018 ) . As a consequence of this risk of disease , areas like the European Union have initiated the TRACES ( TRAde Control and Expert System ) programme to attempt to monitor what is imported and associated disease risk . Yet , such data is challenging to access and is unlikely to enable proactive monitoring for ecosystem health , despite the development of organisations such as the World Organisation for Animal Health ( OIE ) ( Martel et al . , 2020 ) . However , regional networks have been developed for specific cases such as Bd such as spatialepidemiology . net ( Aanensen , 2009 ) . The risk of both recognised and novel invasive pathogens should not be underestimated . Whilst we did not separately map it here , various amphibians are sold commercially as bait . Previous studies show that not only do the live animals kept in bait shops frequently carry fungal and viral pathogens , but they are also frequently released into the wild after use ( Picco and Collins , 2008 ) . Given that over 40% of individuals in this study are shown to come directly from the wild , the potential for spread of pathogens to spread to new areas must be addressed to avoid severely impacting native aquatic vertebrate communities ( Price et al . , 2017 ) . Many papers have highlighted the inadequacies of a CITES paper-based system for monitoring trade ( Berec et al . , 2018 ) . In the context of amphibians , the discrepancies on reporting ( such as species exported from the wild from countries to which they are not native; Auliya et al . , 2016 ) are well documented . Here again , we highlight that CITES fails to provide adequate safeguards both for species which are included , and more so for the 97 . 5% of amphibian species that are not . In recent years , millions of amphibians representing over 1200 species have been traded , with a considerable portion of individuals coming from the wild . The trade of range limited , Data Deficient , and newly described species with extremely limited data highlights how harm to species future survival prospects may be occurring out of sight . Inadequate biosafety standards , potential escape , and invasive species in combination with the direct exploitation threaten the future survival of species . The World Customs Organization must urgently address the lack of coding for these species , to enable steps towards sustainable trade . At present only LEMIS enables exact details of species imported and their origins and purchasers , and CITES and other UN conventions must interface better between environmental and economic conventions and targets . The lack of efficacy of coverage within CITES is also underscored by the EU Wildlife Trade Regulations , which build on the number of species under-regulation , but also highlights the need for a more comprehensive system globally . Whilst developing sustainable quotas for offtake are impossible for species with no data on range or populations , better means to monitor and control trade are necessary and could help form the baseline , especially given that over 40% of individuals come from the wild . The cost of enabling the status quo to continue is likely to guarantee the extinction of over-exploited rare , and range-restricted species , especially when the number of species traded annually may be increasing . The drive for rare species entering trade within a few years of description combined with access to more remote areas will expose areas with high endemism to potential exploitation from unsustainable and unmonitored trade , thus better monitoring and reporting standards are needed . Additionally , these naïve populations are vulnerable to pathogens and could potentially replicate the patterns of extinction so far seen in the Americas , and drive significant biodiversity loss . Further regulation , and better monitoring of both wild populations and species and individuals traded is urgently needed to slow the decline of populations and loss of species as a consequence of unsustainable , and largely unmonitored trade in wildlife . This would require databases to monitor international trade of individuals ( consistent with not only livestock , but all other commodities ) to provide accurate information on what species are being traded , their source , and at what volume . Consistent standards , such as those within LEMIS , provide a blueprint for what could become global wildlife trade databases . LEMIS serves as a framework for agencies wishing to monitor trade; we stress that the data should be fully open and accessible for review and not subject to slow freedom of information requests . For databases to be reliable , central authorities should be delegated at a national level for controlling and certifying traded wildlife , possibly with measures such as DNA barcoding to verify identity , then certify shipments , and be responsible for their export ( to prevent laundering ) . These two approaches would remedy the lack of data , and the potential for laundering , but to prevent trade being unsustainable a shift is needed so that proof of sustainability ( i . e . , through approved non-detriment findings ) is required before trade in a species is allowed . The precautionary principle should become standard practice to ensure that when trading occurs it is based upon a foundation of data to prevent over-exploitation of vulnerable populations; we cannot continue to trade species until we realise that species is already potentially endangered before taking action . We used Google and Bing search engines to discover contemporary websites selling amphibians . We targeted amphibian selling websites in English , French , German , Japanese , Portuguese , and Spanish , to cover the largest herpetofauna pet trade markets . We used appropriately localised versions of the search engines for each language we searched in ( Google: https://www . google . com/ , https://www . google . fr/ , https://www . google . de/ , https://www . google . jp/ , https://www . google . pt/ , https://www . google . es/; Bing: https://www . bing . com/ ? cc=en , https://www . bing . com/ ? cc=fr , https://www . bing . com/ ? cc=de , https://www . bing . com/ ? cc=jp , https://www . bing . com/ ? cc=pt , https://www . bing . com/ ? cc=es ) . Each localised search engine and language was searched with a Boolean search string: We completed the searches in a Firefox private window ( Firefox , 2019 ) , while signed out of search engine accounts to minimise the impact of previous search history . Our search terms may have missed specialist sellers , specialising in a single genus/species and advertising only with slang . We downloaded the first 10 pages of search results provided by each search engine ( 100 URL search results ) to produce a list of 200 URLs per language ( ~1200 URLs overall ) . We used assertthat v . 0 . 2 . 1 ( Wickham , 2019a ) , XML v . 3 . 99 . 0 . 3 ( Lang and The CRAN Team , 2018 ) and stringr v . 1 . 4 . 0 ( Wickham , 2019b ) to extract all URLs present ( Source code 1 ) . We filtered out URLs associated with internal search engine links , leaving us with a list of potential amphibian selling websites . We simplified the extracted URLs to their base URL ( so all URLs ended in . com , . org , . co . uk , etc . ) and removed duplicates . We reviewed each website with the goal of determining whether the site sells live amphibians , classifying the type of website ( classified ads , commercial , other ) , determining whether the site explicitly forbade automated data collection , identifying a page within the site to initiate data mining , identifying the most appropriate method of data collection , and identifying any ordering in amphibian listings ( the last review goal revealed that websites had a mix of ordering; thereby unlikely to bias results: 21 alphabetically , 17 by featured , 12 by date , 5 by price , 2 by popularity , and 30 whose ordering was unclear ) . If a website did not sell live amphibians , or explicitly forbade automated data collection , we excluded it . We randomly assigned all accepted websites with a unique ID for further sampling/analysis ( Source data 1 ) . The above sampling process was preregistered on 2020-08-29 ( osf . io/x5gse ) . On 2020-09-11 , we completed the preregistered sampling and review of 856 websites; we determined that 104 sites would be suitable for searching . However , this was considerably lower than the 151 websites used in previous work ( Marshall et al . , 2020 ) . Therefore , we completed a second search using a simpler search term ( ‘amphibians for sale’ , and translations ) taking the first five pages from both search engines . The new URLs located in the simpler search were reviewed bringing the total reviewed websites to 1069 and the suitable websites to 139 ( 906 excluded because they did not sell amphibians , 13 specifically stated no automated searching of the website , 6 were duplicates , and the remaining 5 had issues with access ) . We used five methods to collect data from websites , applied hierarchically to minimise server load and the number of irrelevant pages searched ( Source code 2 ) . We used species data from AmphibiaWeb as our taxonomic backbone ( AmphibiaWeb , 2020; https://amphibiaweb . org/amphib_names . txt; accessed 2020-08-29; 2 ) . We created a species list that included all current scientific names and all scientific synonyms . We excluded common names from the keyword list because we did not have common names for all languages nor species , and previous work has shown that common names provide only marginal gains in online data collection efforts ( Marshall et al . , 2020 ) . We also made no attempt to search for partial names or abbreviations ( e . g . , Duttaphrynus melanostictus listed as D . melanostictus or D melanostictus ) . Prior to the keyword search we undertook basic web page text cleaning . We removed all double spaces , special characters , numbers , and html elements , replacing them with single spaces . The basic cleaning meant that genus and species epithets would appear in the same format as the keyword list provided they occur next to each other on the web page . We used rvest v . 0 . 3 . 6 ( Wickham , 2019c ) , XML v . 3 . 99 . 0 . 3 ( Lang and The CRAN Team , 2018 ) , and xml2 v . 1 . 3 . 2 ( Wickham et al . , 2018 ) packages to clean and parse the html data . We used case-insensitive fixed string matching , with stringr v . 1 . 4 . 0 package ( Wickham , 2019b ) , to search all collected web pages for species names . We used fixed string matching because it has lower computation costs compared with collation matching . Fixed string matching is unable to distinguish between differently coded ligatures or diacritic marks , but our focus on scientific names avoided diacritical marks . Future search efforts using partial or approximate string matching could reveal species we missed if they had only listed with misspelt names or using abbreviations; however , such search efforts would require more computational time , a more thoroughly curated keyword library than what we had access to , and greater caution regarding false positives . Upon searching a web page for species names , we recorded whether a keyword ( species ) was present , what accepted species the detected species corresponded to , the page number of the web page , and the website ID ( Source code 4; Source data 3 and 4 ) . We combined final results from the online search with data from LEMIS and CITES ( Source code 5; retrieved via the R package lemis v . 1 . 1 . 0 ( Eskew et al . , 2019; Eskew et al . , 2020; Ross et al . , 2019 ) , and https://trade . cites . org/# , respectively ) . To understand the dimensions of trade , and how regions may be impacted with different types of trade , we included an additional two data sources ( the Mohanty and Measey data based on a collation of published literature , and the IUCN listings of species which state if the species is threatened by trade ) . We compiled all species on a spreadsheet with the listed purpose from each data source ( Source data 5 ) . All species for sale in online stores , we classified as ‘pet trade’ , whereas the Mohanty and Measey data we classified as ‘other’ and only used these in the total analysis . For IUCN data the entire list of species listed as ‘Use and Trade’ for food , medicine , or pets was downloaded . These listings were manually processed and those listing food , medicine , or pets listed , keywords ( ‘food’ , ‘pets’ , ‘medicine’ ) were used to make the process more efficient , but as ‘not’ was often included in these statements all listings were manually processed , so checking of all listings to verify status was essential . This was used to classify species by use as ‘food’ , ‘medicine’ , ‘pharmaceutical’ , ‘pet trade’ , or ‘other uses’ . Species for which no form of trade was listed ( e . g . , ‘there is no evidence of trade for this species’ ) were removed from the listings . For both CITES and LEMIS data , the purpose was collated from the commercially imported listings as well as the personal listings ( whilst other categories such as ‘research/zoo’ were listed directly based on subsets of scientific category data ) . CITES does not list the importer so only coarse categories listed were usable , whereas for LEMIS keywords could be used for both importers and exporters to determine the likely purpose of the item . Firstly , items were split into ‘live’ and ‘dead’ . Companies with dead items were likely to be sourcing items for either meat or pharmaceutical/medicine , whereas live imports could have a variety of purposes , we used a list of keywords associated with the importer and exporter ( Source data 2 ) to determine the category each imported item fell into . This still left many items unaccounted for , so as sellers were likely to specialise in one category items were then sorted by seller and other items from that seller listed with the same category . Where a conflict of different listings existed , these were compared to any dead specimens from the same seller , which would indicate that the items were likely to be meat ( or medicine/pharmaceuticals ) . Through this process most items could be sorted to one of the categories , and other suggestive keywords ( i . e . , ‘zoo…’ in listings not associated with an actual zoo were classed as pets ) , and then listings of species traded for each purpose collated in a spreadsheet based on all data sources . Individuals importing species , unless listed for research was also categorised as pets . Whilst there is a degree of uncertainty associated with some of these assigned purposes , it does show that species imported for meat may be a wider selection than realised , as well as those consumed more locally . This was then summed to list the different purposes each species was traded for using LEMIS , and combined with the categories in CITES as well as purposes listed by the IUCN Redlist assessments to produce a list of uses of each species in trade . For LEMIS summaries of wild capture and captive rearing ( Source code 6 and 7; Source data 7 ) , we filtered the data to only include items that represented single individuals: whole dead animal ( LEMIS code = BOD ) , live eggs ( EGL ) , dead specimen ( DEA ) , live specimen ( LIV ) , specimen ( SPE ) , whole skin ( SKI ) , entire animal trophy ( TRO ) , following the process described in Hierink et al . , 2020 , and Marshall et al . , 2020 . We define non-commercial trade as that termed by LEMIS as: Biomedical research ( M ) , Scientific ( S ) , and Reintroduction/introduction into the wild ( Y ) ; whereas captive origin covered Animals bred in captivity ( C and F ) , Commercially bred ( D ) , and Specimens originating from a ranching operation ( R ) ; and wild origin only included those listed as Specimens taken from the wild ( W ) . We included all amphibians in origin/purpose summaries , but we only included species detected in LEMIS in final species counts if the full species name listed in LEMIS could be matched to an AmphibiaWeb name or synonym . We relied on LEMIS listing of genus for genera summaries , excluding non-applicable terms ( e . g . , Non-CITES entry , Anura , Bufonidae , Tadpole ) . All mapping , bar Figure 2—figure supplement 2 and 1 ( which used on AmphibiaWeb ISOCC country data; Source code 8 ) , was completed in ArcMap 10 . 3 . Amphibian data range maps were downloaded from the IUCN ( iucnredlist . org ) and then species in trade , once corrected for synonyms joined to the shapefile using joins and relates . Individual species maps were then converted into rasters with a resolution of 1 km using the conversion tools . Mosaic to new raster was then used to quantify the species in trade both altogether , or based upon subsets of data such as endangerment , data source ( CITES: Source code 8 , LEMIS: Source data 7 , Online: Source data 3 and 4 ) or use ( pet , meat , research , medicinal/pharmaceutical ) to provide global maps depicting each type of pressure . We also explored temporal trends in CITES , LEMIS , and Online data , plotting changes over time and using a linear regression to account for search effort online ( i . e . , pages searched; Source code 9 ) . We also plotted the differences in species lists produced by different languages , and summarised the top 10 most-species rich ( by number of unique species ) websites’ purpose ( Source code 10 ) . To calculate the level of coverage on and trade on a national basis , the IUCN maps were intersected with each country to give a country list , and species lacking range maps were compiled to a national level using AmphibiaWeb data . Endangerment and CITES status for species in trade and not traded were associated with this data using the joins and relates function , and quantified using summary statistics before being rejoined to a global map to assay the level of coverage for species in trade at a national level . We retrieved all species years of description from the amphibian species of the world database ( accessed 2020-10-02; Frost , 2020 ) . We used rvest v . 0 . 3 . 6 ( Wickham , 2019c ) and xml2 v . 1 . 3 . 2 ( Yuan et al . , 2018 ) to call and retrieve the top search result from the database on a species-by-species basis ( each AmphibiaWeb species binomial being used a search term ) , saving the full character string detailing the species authority ( Source code 10 and 11 ) . We double-checked the retrieved species authority contained the required species binomial . In cases where species binomial was not included ( 174 ) , we used similiars v . 0 . 1 . 0 ( Sjoberg , 2020 , 2020 ) to detect minor spelling differences . Ultimately , we found 12 species with non-matching authorities and were detected in the trade; for these 12 species we manually found the appropriate authority . We used LEMIS , CITES ( Source data 9 ) , and the Online sampling to determine the earliest instance of a species appearing in the trade . We completed all keyword searches and data review in R v . 3 . 6 . 3 ( R Core Development Team , 2020 ) and R Studio v . 1 . 4 . 669 ( R Studio Team , 2020 ) . During data manipulation , we also made use of R packages: dplyr v . 1 . 0 . 2 ( Wickham et al . , 2020 ) and tidyr v . 1 . 1 . 2 ( Wickham and Henry , 2019 ) ; for data visualisation we used cowplot v . 1 . 1 . 0 ( Wilke , 2019 ) , ggplot2 v . 3 . 3 . 2 ( Wickham , 2016 ) , ggpubr v . 0 . 4 . 0 ( Kassambara , 2018 ) , ggtext v . 0 . 1 . 1 ( Wilke , 2020 ) , glue v . 1 . 4 . 2 ( Hester , 2020 ) , maps v . 3 . 3 . 0 ( Becker and Wilks , 2018 ) , scico v . 1 . 2 . 0 ( Pedersen and Crameri , 2018 ) , and UpSetR v . 1 . 4 . 0 ( Gehlenborg , 2019 ) . We added additional details to the upset plot using Affinity Designer v . 1 . 8 . 5 . 703 ( Serif , 2020 ) . We have made code used to search online , filter LEMIS data , generate Figures 1 and 3–5 , S4 , and elements of 6 , and retrieve species authorities available at Open Science Framework: https://osf . io/x5gse/ ? view_only=27109adbb3364dd2b9115752fd912b99 . Alongside the code , we have provided all datasheets listed as supplementary material .	In the last few decades , exotic pets have become much more common . In the UK in 2008 , reptiles and amphibians were more popular than dogs , with over eight million in captivity . But while almost all pet cats and dogs are born and bred in captivity , exotic pets are often taken from the wild , putting species and their habitats at risk . An international trade agreement called the Convention on International Trade in Endangered Species ( CITES ) strives to prevent unsustainable animal trade . But to get CITES protection , species depend on data showing that wildlife trade threatens their survival . In addition , their range countries need to first propose them to be listed . For most wild animal species , there are no data on population size or population decline . In the case of amphibians , CITES regulates the trade of just 2 . 5% of species . This leaves the rest with no protection from overarching international trade regulations . To protect these animals , researchers need to find out which species are in trade , where they are coming from , and how many are already threatened . To address this , Hughes , Marshall and Strine combined data from five sources , including official CITES trade records , recent research and an online search for amphibian sales in six languages . The data showed evidence of trade in at least 1 , 215 amphibian species , representing 17% of all amphibians . The figure is three times higher than previous estimates . Of the species in trade , more than one in five is vulnerable to extinction , endangered , or critically endangered . For a further 100 of the traded species , data on population were unavailable . Moreover , analysis of the origins of traded individuals showed that around 42% came from the wild . Tropical parts of the world had the highest number of species in trade , but the data showed exchanges happening across the globe . Unsustainable wildlife trade can have devastating consequences for wild animals . It has already driven at least 21 reptile species to extinction , and data of amphibian species are unknown . To prevent further species going extinct , legal wildlife trade should follow the precautionary principle when it comes to wildlife trade . Rather than allowing people to trade a species until CITES regulates it , a blanket ban should come into force for species that have not been assessed or are threatened . Trade would be able to resume for a species only when assessments show that it would not cause major population decline , or secure , captive breeding facilities can be guaranteed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2021	Gaps in global wildlife trade monitoring leave amphibians vulnerable
The intracellular iron transfer process is not well understood , and the identity of the iron transporter responsible for iron delivery to the secretory compartments remains elusive . In this study , we show Drosophila ZIP13 ( Slc39a13 ) , a presumed zinc importer , fulfills the iron effluxing role . Interfering with dZIP13 expression causes iron-rescuable iron absorption defect , simultaneous iron increase in the cytosol and decrease in the secretory compartments , failure of ferritin iron loading , and abnormal collagen secretion . dZIP13 expression in E . coli confers upon the host iron-dependent growth and iron resistance . Importantly , time-coursed transport assays using an iron isotope indicated a potent iron exporting activity of dZIP13 . The identification of dZIP13 as an iron transporter suggests that the spondylocheiro dysplastic form of Ehlers–Danlos syndrome , in which hZIP13 is defective , is likely due to a failure of iron delivery to the secretory compartments . Our results also broaden our knowledge of the scope of defects from iron dyshomeostasis . Iron is critical to the function of a variety of proteins , including hemoglobin and myoglobin , various iron–sulfur proteins , and the electron transport chain . In addition , iron is a necessary component in the secretory pathway . For example , lysyl hydroxylase ( LH , also referred to as PLOD ) and collagen prolyl 4-hydroxylases ( P4Hs ) are two post-translational modifying enzymes localized to the lumen of endoplasmic reticulum , which use Fe2+ as a cofactor ( Tuderman et al . , 1977; Pirskanen et al . , 1996 ) . These two enzymes are critical for the synthesis of collagen , a crucial part of the basement membrane , and are formed by a complicated process involving multiple co- or post-translational modifications ( Myllyharju and Kivirikko , 2004 ) . While the types of collagens and genes found in mammals are very complex , only one type of collagen IV , encoded by two genes Viking ( Vkg ) and Cg25C , is found in Drosophila , and it constitutes a major structural component of basement membranes in the developing fly ( Fessler and Fessler , 1989 ) . In adidition to lysyl hydroxylase and prolyl hydroxylase , ferritin is another iron-dependent protein residing in the secretory pathway of Drosophila . In constrast to mammalian ferritin , which is predominantly found in the cytosol , Drosophila ferritin binds iron in the early secretory compartments and is then secreted into the circulation system ( Mandilaras et al . , 2013; Tang and Zhou , 2013a ) . This process is central for systemic iron supply as well as tissue iron detoxification ( Tang and Zhou , 2013b ) . Therefore , in comparison to mammals , it is expected that Drosophila will carry a larger amount of iron through this pathway and abnormalities in this process will lead to serious iron deficiency . Despite considerate interest in iron homeostasis , the metabolic process of iron , particularly its intracellular trafficking , remains poorly characterized . One important question that remains unanswered is how cytoplasmic iron is transferred to the secretory pathway for the iron-dependent proteins found therein . During the process of using the fruit fly as a model to decipher the functions of zinc transporter dZIP13 , we unexpectedly discovered dZIP13 physiologically acts as an iron exporter . dZIP13 belongs to the metal transporter ZIP family ( zinc-regulated and iron-regulated transporter proteins or Slc39a ) ; members of this family was reported to generally promote zinc transport from the extracellular space or from intracellular vesicles to the cytoplasm ( Liuzzi et al . , 2006 ) . Our work thus identified the iron transporter required for iron loading in the secretory pathway , which is also the first time a ZIP member has been reported as an iron exporter . This finding implies that failure of iron delivery to the secretory compartments is probably the underlying cause for SCD-EDS ( the spondylocheiro dysplastic Ehlers–Danlos syndrome , OMIM #612350 ) , which is due to a mutation in hZIP13 ( Fukada et al . , 2008; Giunta et al . , 2008 ) . Because SCD-EDS displays none of the classical iron phenotypes , for example , anemia or iron accumulation toxicity , our results also suggest that iron dyshomeostasis is likely involved in a wider spectrum of biological abnormalities than previously thought . BLASTP searches using the amino acid sequences of mammalian ZIP family members revealed that the Drosophila genome encodes at least eight putative ZIP proteins ( Lye et al . , 2012; Qin et al . , 2013 ) . Among them , the protein encoded by CG7816 shares the highest overall homology with human ZIP13 ( 45% identity and 58% similarity ) ( Figure 1A ) , and was named dZIP13 accordingly . In the phylogenetic tree , dZIP13 clusters together with hZIP13 and several other members including catsup ( CG10449 ) and hZIP7 ( Figure 1B ) , all belonging to the LIV-1 subfamily of zinc transporters or LZT proteins ( Taylor and Nicholson , 2003 ) . Several typical features of ZIP family members ( Jeong and Eide , 2013 ) are found in dZIP13 , including eight transmembrane domains ( TM ) , particularly amphipathic TM4 and TM5 , and a predicted extracellular/luminal location of both the amino and carboxyl termini . Notably for dZIP13 and hZIP13 , there is only a single His residue in a generally histidine-rich region ( 2–14 His ) between TM3 and TM4 . The highly conserved potential metalloprotease His-Glu-X-X-His ( HEXXH , where X is any amino acid ) motif , located within TM5 ( Bin et al . , 2011 ) of LZT proteins , is also found in dZIP13 ( Figure 1A ) . 10 . 7554/eLife . 03191 . 003Figure 1 . Sequence analysis of Drosophila ZIP13 . ( A ) Alignment of Drosophila ZIP13 ( dZIP13 , the top ) , human ZIP13 ( hZIP13 , the middle ) , and human ZIP4 ( hZIP4 , the bottom ) proteins . Amino acid sequences for hZIP13 , hZIP4 , and dZIP13 ( CG7816 ) were obtained from GenBank and aligned by HMHMM software . Black and pink shadings indicate respectively identical and conservative amino acids . The eight putative transmembrane ( TM ) regions are underlined and denoted as ‘TM I’ through ‘TM VIII’ . ( B ) Phylogenetic tree analysis of human and putative Drosophila ZIP family members . The tree was generated using ClustalX version 1 . 81 and displayed with TreeView . Bootstrap probabilities for major clusters are shown by percentages . Accession numbers are listed for other Drosophila ZIPs used for the alignment . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 003 To analyze the functions of dZIP13 in vivo , transgenic lines of Drosophila dZIP13 RNAi and overexpression ( dZIP13-RNAi and dZIP13-OE ) were generated or obtained , and then tested for dZIP13 expression modulation . The dZIP13-RNAi and dZIP13-OE indeed efficiently altered expression of dZIP13 at both the mRNA and protein levels ( Figure 2—figure supplement1 and Figure 2—figure supplement2 ) . We initially anticipated dZIP13 as a zinc importer , similar to many other reported ZIP proteins . However , when dZIP13 was tissue specifically knocked down or over-expressed with NP3084-Gal4 , a driver specifically expressing the activator Gal4 and thus modulating dZIP13 in the midgut region , the iron levels of the whole body changed dramatically while the amount of zinc level stayed unaltered ( Figure 2A ) . Compared to the control , the iron levels in the dZIP13-RNAi fly dropped to about 50% of the normal iron content , while iron amount in overexpressing fly increased . This finding was confirmed with different RNAi lines ( data not shown ) , indicating it is not due to off-target effects . The iron effect of dZIP13 suggested to us dZIP13 might directly or indirectly affect dietary iron absorption in the gut . 10 . 7554/eLife . 03191 . 004Figure 2 . dZIP13-RNAi flies display iron-rescuable defects . ( A ) Body metal contents when dZIP13 expression was modulated . Shown are flies with modulated dZIP13 expression in the midgut ( NP3084 as the Gal 4 driver ) . A significant decrease in the whole body iron content , but not that of zinc or copper , was observed in dZIP13-RNAi flies , while dZIP13 overexpression led to an iron increase . Values represent three independent measurements and are normalized to the dry body weights; data are presented as means + SEM; n = 3 or 6 . p<0 . 05 , p<0 . 01; two-tailed Student's t test . ( B ) The eclosion rate of ubiquitously-RNA-interferenced-dZIP13 ( Da > dZIP13-RNAi ) larvae could be rescued by dietary iron supplementation . Da-Gal4 was crossed to wild-type or dZIP13-RNAi flies on juice-agar plates . Newly hatched progeny were transferred to normal food , or food supplemented with ZnCl2 , TEPN , FAC , or BPS . Percentages of flies that eclosed to adults were counted; n = 6 or 8 . ( C ) A control showing that the same amount of zinc , iron , or chelators supplemented in the food had no effect on the eclosion rate of wild-type Drosophila . ( D ) The shortened lifespan of Da > dZIP13-RNAi adults was partially rescued by dietary iron supplementation but not zinc . Percentages of flies that eclosed to adults were counted; n = 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 00410 . 7554/eLife . 03191 . 005Figure 2—figure supplement 1 . RT-PCR analysis of efficacy of dZIP13 knockdown or overexpression . RT-PCR analysis of dZIP13 mRNA abundance in third instar larvae . rp49 was used as the loading control . Da-GAL4 was used as the expression driver . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 00510 . 7554/eLife . 03191 . 006Figure 2—figure supplement 2 . Western blot analysis of efficacy of dZIP13 knockdown or overexpression . Western blot showing that the RNAi used in this study suppressed dZIP13 protein to a significantly reduced level . Tubulin was used as the loading control . Da-GAL4 was used as the expression driver . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 006 Under normal dietary conditions , ubiquitous RNAi of dZIP13 by daughterless ( Da-Gal4 ) produced developmental arrest in the pupal stage ( only about 10% eclosion rate ) ( Figure 2B ) . Amazingly , the eclosion defect resulting from dZIP13 knockdown could be rescued from ∼10% to ∼75% simply through dietary iron supplementation in the form of ferric ammonium citrate ( FAC ) ( Figure 2B ) . Addition of 0 . 1 mM iron-specific chelator bathophenanthrolinedisulfonic acid disodium ( BPS ) , on the other hand , exacerbated the phenotype as an even lower eclosion rate was observed ( Figure 2B ) , indicating abnormal iron absorption indeed plays a critical role in causing the eclosion defect in dZIP13-RNAi flies . Zinc , a substrate of many ZIP proteins , also to some extent exacerbate the eclosion defect . When zinc was added to the diet , almost no adult flies could eclose ( ∼0% eclosion rate ) ; however , addition of zinc chelator TPEN could ameliorate the phenotype slightly , although without statistical significance ( Figure 2B ) . There have been several reports showing that iron absorption can be competitively inhibited by additional zinc ( Rossander-Hulten et al . , 1991; Whittaker , 1998 ) . From that perspective it is possible that zinc addition to the food could exacerbate the iron deficiency phenotype of dZIP13-RNAi flies . Simultaneous addition of both zinc and iron could still rescue the eclosion defect very effectively ( ∼65% eclosion rate , statistically insignificant with iron only ) . The wild type flies' eclosion rate , in contrast , remained the same either in iron supplemented or deficient , zinc supplemented or deficient food ( Figure 2C ) . The dramatic rescue by iron instead of zinc suggests that lack of iron is the primary defect in dZIP13-RNAi flies , and the zinc effect is minimal and likely secondary . As a result of this unexpected finding , our original focus of dZIP13's function in zinc homeostasis was subsequently switched to investigating its role in iron homeostasis . Consistent with the results obtained in the eclosion experiments , the lifespan of dZIP13-RNAi flies was also prolonged by iron addition in the food ( Figure 2D ) : the flies raised on iron supplemented food have a prolonged median lifespan ( ∼25 days ) compared with files raised on normal food ( ∼4 days ) . Addition of zinc to the diet shortened their lifespan ( median lifespan of ∼1 day ) . No significant differences of lifespans were found between normal food and BPS or TPEN food . The activity of aconitase is often used as a molecular indicator for the availability of iron in the cell ( Haile et al . , 1992; Suzuki et al . , 2005 ) . There are two types of aconitase in cells , the cytosolic aconitase ( c-aconitase ) and the mitochondrial aconitase ( m-aconitase ) . Cytosolic aconitase needs to bind iron for its enzymatic activity and is an indicator of the cytosolic iron level ( Haile et al . , 1992; Tong and Rouault , 2006 ) . As shown in Figure 3 , the in-gel aconitase activity assay allowed clear separation of Drosophila m-aconitase and c-aconitase , providing a convenient method to evaluate the levels of iron in the two different subcellular compartments ( Tong and Rouault , 2006 ) . 10 . 7554/eLife . 03191 . 007Figure 3 . dZIP13 knockdown led to iron deficiency in the body but not the cytosol of gut cells . ( A ) Cytosolic aconitase activity was increased in the gut of NP3084>dZIP13-RNAi larvae and decreased in NP3084>dZIP13-OE larvae , suggesting respectively iron elevation and iron deficiency in the cytoplasm . Panel ( b ) was quantitative measurement of ( a ) . Results are presented as mean + SEM relative activity; n = 3 . p<0 . 05 , p<0 . 01 , p<0 . 001; two-tailed Student's t test . ( B ) Cytosolic aconitase activity was decreased in the whole body minus gut ( body parts other than the gut ) of NP3084>dZIP13-RNAi larvae and increased in NP3084>dZIP13-OE larvae . Panel ( b ) was quantitative measurement of ( a ) . Results are presented as mean + SEM relative activity; n = 3 . p<0 . 05 , p<0 . 01 , p<0 . 001; two-tailed Student's t test . ( C ) RT-PCR analysis of iron homeostasis genes of normal flies in response to iron changes ( feeding with iron or chelator ) . RNA was made from third instar larvae midguts . rp49 was used as the loading control . ( D ) RT-PCR analysis of iron homeostasis genes in the midgut of dZIP13 RNAi or OE third instar larvae . Expression is driven by the midgut driver NP3084 . ( E ) An analysis of metal contents in the gut when dZIP13 expression was modulated . Shown are metal levels from fly larvae with modulated dZIP13 expression in the midgut ( NP3084 as the Gal 4 driver ) . A significant decrease in the gut iron , but not zinc or copper , was observed in dZIP13-RNAi flies; dZIP13 overexpression led to an iron increase . Results are presented as mean + SEM relative activity; n = 3 . p<0 . 05 , p<0 . 01 , p<0 . 001; two-tailed Student's t test . ( F ) An analysis of metal contents in the whole-body-minus-gut parts when dZIP13 expression was modulated in the larval midgut ( NP3084 as the Gal 4 driver ) . A significant decrease in the whole-body-minus-gut iron , but not zinc or copper , was observed in dZIP13-RNAi flies; dZIP13 overexpression led to an iron increase . Results are presented as mean + SEM relative activity; n = 3 . p<0 . 05 , p<0 . 01 , p<0 . 001; two-tailed Student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 007 We examined how the gut itself would be affected when dZIP13 was specifically knocked down in the gut . As shown in Figure 3A , dZIP13-RNAi larvae exhibited ∼35% more c-aconitase activity in the gut as compared to the control larvae while dZIP13-OE larvae showed ∼20% c-aconitase activity reduction . In contrast to this finding , in the remaining body parts ( i . e . , the whole body minus the gut ) of dZIP13-RNAi larvae , c-aconitase activity was significantly reduced when compared to the control larvae , implying a general lack of iron in the body parts other than the gut . Conversely , c-aconitase activity was significantly elevated in dZIP13-OE , implying an iron elevation in the other body parts ( Figure 3B ) . Because the labile iron–sulfur cluster of aconitase is also subjected to conditions other than iron levels , such as oxidative stress , we further analyzed how genes involved in iron homeostasis might respond when dZIP13 expression is modulated . We reasoned that because these iron metabolism genes are sensitive to iron availability , their behaviors would also be a good indicator of cytosolic iron levels . Under iron deficiency , ferritin and iron uptake protein Malvolio ( Mvl ) is respectively down- and up-regulated , while under iron surplus , vice versa ( Figure 3C ) ( Missirlis et al . , 2007; Tang and Zhou , 2013b ) . Indeed , when dZIP13 was knocked-down , ferritin was significantly up-regulated while Mvl down-regulated , very much akin to the scenario when the larvae were fed with iron-supplemented diet ( Figure 3D , also see Figure 4E for ferritin protein levels ) . These results indicate dZIP13-RNAi larvae sensed a state of iron-replete condition in the cytosol of their gut cells , consistent with the above aconitase activity results . 10 . 7554/eLife . 03191 . 008Figure 4 . dZIP13 knockdown results in reduced ferritin iron loading in the gut . ( A ) Staining of ferric iron in the larval gut . The staining in the midgut constriction and ectopic ferric staining in the anterior midgut are noted separately by arrows ( blue ) and arrow heads ( green ) . The anterior midgut of NP3084>dZIP13-OE fly larvae deposited obviously a higher amount of iron than the control . Ferric iron significantly accumulated in the iron cell region of NP3084>dZIP13-OE while NP3084>dZIP13-RNAi adults showed almost no iron staining . Shown are representative images and in bright and dark fields . More images are shown in Figure 4—figure supplement 1 . ( B ) Staining of ferric iron in the anterior midgut of iron-fed larvae . The anterior midgut follows the preceding distinct proventriculus ( pv , red arrowheads ) . ( C ) Staining of ferric iron ( bound to ferritin ) on native PAGE . Same amounts of total protein extracts from control and NP3084>dZIP13-RNAi or NP3084>dZIP13-OE larval guts were loaded . The gel was directly stained with Prussian blue staining solution . For an intact gel image , see Figure 4—figure supplement 2 . Panel ( b ) was quantitative measurement of ( a ) . n = 3 . p<0 . 05 , *p<0 . 001; two-tailed Student's t test . ( D ) Ferritin expression in the gut . An obviously higher amount of ferritin was expressed in the gut of NP3084>dZIP13-RNAi larvae . The expression of ferritin was indicated with a protein trap line Fer1HCHG188 , which tags the endogenous Fer1HCH through an N-terminal GFP fusion . Shown are representative results and more images are shown in Figure 4—figure supplement 3 . ( E ) Western blot of ferritin of NP3084>dZIP13-RNAi and NP3084>dZIP13-OE larvae . Anti-ferritin light chain antibody was used . Tubulin was used as a loading control . For an intact gel image , see Figure 4—figure supplement 4 . Panel ( b ) was quantitative measurement of ( a ) . n = 3 . p<0 . 001; two-tailed Student's t test . ( F ) Eclosion rescue of ferritin-RNAi by dZIP13 . The eclosion rate of gut-specific ( NP3084 ) ferritin-RNAi flies was rescued from <5% to ∼30% by dZIP13 overexpression . Newly hatched progeny were transferred to normal food , and percentages of flies that eclosed to adults were counted . n = 6 . p<0 . 001; two-tailed Student's t test . ( G ) Rescue of iron deficiency of ferritin-RNAi by dZIP13 . The reduced body iron content in gut-specific ferritin-knockdown larvae was also partially rescued by dZIP13 overexpression . n = 6 . p<0 . 01; two-tailed Student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 00810 . 7554/eLife . 03191 . 009Figure 4—figure supplement 1 . Staining of ferric iron in the larval gut . The ferric staining in the midgut constriction and the ectopic ferric staining in the anterior midgut are noted separately by arrows ( blue ) and arrow heads ( green ) . The anterior midgut of NP3084>dZIP13-OE fly larvae deposited an obviously higher amount of iron than the control . Ferric iron accumulated prominently in the iron cell region of NP3084>dZIP13-OE while NP3084>dZIP13-RNAi larvae showed almost no iron staining . ( A ) , ( B ) and ( C ) are results from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 00910 . 7554/eLife . 03191 . 010Figure 4—figure supplement 2 . Staining of ferric iron ( bound to ferritin ) on native PAGE . Same amounts of total protein extracted from control and NP3084>dZIP13-RNAi or NP3084>dZIP13-OE larvae guts were loaded . The gel was stained with Prussian blue staining solution . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 01010 . 7554/eLife . 03191 . 011Figure 4—figure supplement 3 . A significantly higher amount of ferritin was detected in the gut of NP3084>dZIP13-RNAi fly larvae . The expression of ferritin was indicated with a protein trap line Fer1HCHG188 expressing an N-terminal GFP-tagged Fer1HCH protein . Shown are images from the bright and dark fields . ( A ) and ( B ) are results from two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 01110 . 7554/eLife . 03191 . 012Figure 4—figure supplement 4 . Western blot of ferritin of NP3084>dZIP13-RNAi and NP3084>dZIP13-OE larvae . Anti-ferritin light chain antibody was used . Tubulin was used as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 012 To directly quantify the total iron amount , we further performed ICP-MS ( inductively coupled plasma-mass spectrometry ) analysis of the gut and other body parts ( whole body-gut ) . Both the gut and the rest of the body as a whole exhibited iron reduction after dZIP13 knockdown and iron increase when dZIP13 was overexpressed ( Figure 3E and Figure 3F ) . Therefore , when dZIP13 was knocked down in the gut , the rest of the body experienced iron shortage . In the gut , however , although the total iron was lower , iron in the cytosol appeared to not be reduced , suggesting that dZIP13 is involved in iron extrusion from the gut to the body . Because ferritin is the major iron storage protein and is located in the secretory pathway in Drosophila , we speculate that when dZIP13 is down-regulated in the gut , iron may not be able to move from the cytosol to the secretory compartments , resulting in iron elevation in the cytosol , a feedback control of iron uptake and an overall reduction of iron in both the gut and the rest of the body . Because dZIP13 affects iron absorption as shown above , and it is known that the dietary iron absorption is mediated by ferritin in Drosophila ( Tang and Zhou , 2013b ) , we next investigated the influence of dZIP13 on ferritin iron assimilation . Ferritin , a heteropolymer composed of H and L subunits , can accommodate thousands of iron atoms in its protein shell in the ferric form . Ferritin is thought to be the major cytosolic iron-storage protein in mammalian organisms ( Andrews , 2005; Knovich et al . , 2009 ) , while in Drosophila , unlike mammals , ferritin sequences of both the H and L chains contain secretion signals , confining them to the secretory pathway and making them abundant in the hemolymph ( Nichol et al . , 2002 ) . A major function of Drosophila ferritin is absorption of dietary iron through iron loading and transporting via the secretory pathway supplying iron for the systemic use ( Tang and Zhou , 2013b ) . If dZIP13 is responsible for iron loading into the early secretory pathway , we expect it might affect ferritin iron assimilation . To explore this possibility , we stained the intestines from dZIP13-modulated fly larvae for ferric iron , which is mainly loaded in ferritin . In the iron cell region ( a cluster of cells in the middle midgut region where iron absorption occurs ) of dZIP13-RNAi larvae , we were unable to detect obvious iron signal , whereas a distinct blue iron-staining signal could be observed in the normal fly larvae ( Figure 4A , Figure 4—figure supplement1 ) , suggesting a lack of iron incorporation into ferritin after dZIP13 interference . Conversely , overexpression of dZIP13 led to a stronger staining in the iron cell region , and ectopically in the anterior midgut ( Figure 4A , Figure 4—figure supplement1 ) . This result is reminiscent of the observation that high iron levels , such as 5 mM FAC , can also induce ectopic ferric iron staining in the anterior midgut , suggesting an iron accumulation occurs in the secretory pathway of the gut cells when dZIP13 is overexpressed . Under high levels of dietary iron , iron was only faintly stained in the dZIP13-RNAi larvae , whereas staining was much stronger in dZIP13 overexpression larvae ( Figure 4B ) . This indicates that in the gut of dZIP13-RNAi larvae much less ferric ion is accumulated in ferritin . A native-PAGE staining for ferric iron further confirmed that the intestinal ferritin was indeed poorly loaded with iron in the dZIP13-RNAi larvae while the iron loading in the dZIP13-OE larvae was mildly elevated ( Figure 4C , Figure 4—figure supplement2 ) . To exclude the possibility that the poor iron loading observed in dZIP13-RNAi larvae was the result of poor ferritin expression , a protein trap line Fer1HCHG188 , which expresses an N-terminal GFP-tagged Fer1HCH , was used to indicate the ferritin expression pattern in the gut ( Missirlis et al . , 2007 ) . As shown in Figure 4D and reported previously ( Missirlis et al . , 2007 ) , in flies reared with standard food , GFP-tagged ferritin was most prominently expressed in the iron cell region . When dZIP13 expression was knocked down , expression of ferritin was ectopically and significantly induced in the anterior and posterior midguts , and was constitutively expressed in the iron cell region ( Figure 4D , Figure 4—figure supplement3 ) . This expression pattern of ferritin was repeated when flies were administered a diet containing 5 mM ferric ammonium citrate ( Missirlis et al . , 2007 ) , indicating iron is in excess in certain parts of the gut cells when dZIP13 is RNA interfered , consistent with the above studies in assessing cytosolic iron status of gut cells . Induction of ferritin expression in dZIP13-RNAi larvae was confirmed with a Western blot using a ferritin antibody against ferritin . A dramatic increase of ferritin level was seen in dZIP13-RNAi larvae , with a decrease observed in the dZIP13-OE larvae ( Figure 4E , Figure 4—figure supplement4 ) . An obvious genetic interaction of dZIP13 and ferritin was detected when dZIP13 was overexpressed in ferritin-RNAi flies . Midgut-specific ferritin RNAi leads to systemic iron deficiency . Ferritin RNAi with gut-specific driver ( NP3084 ) causes decreased total body iron contents , retarded growth , and death of most of the progeny at the larval or pupal stage ( Tang and Zhou , 2013b ) . Notably , the eclosion rate of these flies could be rescued from <5–30% through overexpressing dZIP13 ( Figure 4F ) . Consistently , the decreased total body iron content of midgut-specific ferritin RNAi larvae could also be partially rescued by dZIP13 overexpression ( Figure 4G ) . These results suggest that knockdown of dZIP13 would inhibit iron transport into the secretory pathway to be available to ferritin , reducing iron export from the gut for systemic use , while overexpressing dZIP13 would increase body iron amount by facilitating more iron transportation into the secretion pathway , making less iron available in the cytosol of the gut cells . Locations of hZIP13 have been reported in Golgi and some intracellular vesicles ( Fukada et al . , 2008; Jeong et al . , 2012 ) . To examine the intracellular position of dZIP13 , a C-terminal myc-tagged dZIP13 was introduced into human intestinal Caco2 cells . Immunofluorescence staining indicated it partially overlapped with ER and Golgi ( Figure 5A ) . A similarly tagged ( at the C-terminal ) dZIP13-EGFP was made and expressed in the flies and shown to be functional . This dZIP13-EGFP is also partially located to the Golgi apparatus ( data not shown ) . To check the intracellular positions of endogenous dZIP13 in the Drosophila gut , we generated an antibody against dZIP13 . The specificity of the antibody was confirmed by significantly reduced staining in dZIP13-RNAi larvae and loss of staining in dZIP13 mutant . Once more we observed an obvious overlapping of dZIP13 with the ER/Golgi apparatus in the Drosophila gut staining , along with some staining in additional intracellularlocations ( Figure 5B ) . Co-localization of dZIP13 with endosome markers , however , was poor ( Figure 5C ) . 10 . 7554/eLife . 03191 . 013Figure 5 . dZIP13 is located to ER/Golgi and involved in their iron regulation . ( A ) The localization of dZIP13 in Caco2 cells . dZIP13-myc was detected by myc antibody . This immunofluorescence staining showed that dZIP13 partially co-localizes with the ER/Golgi in Caco2 cells . ( B ) The localization of dZIP13 in Drosophila midgut epithelial cells . dZIP13 was detected directly by dZIP13 antibody . These images indicated dZIP13 partially co-localizes with ER/Golgi in Drosophila gut cells . ( C ) dZip13 does not co-localize well with endosome markers . The localizations of dZIP13 and endosomes in Drosophila midgut epithelial cells were shown . dZIP13 was detected directly with dZIP13 antibody . Lysosome-associated membrane protein 1 ( LAMP1 ) fused with GFP was used to indicate lysosomes and late endosomes; FYVE was used to mark the early endosomes . ( D ) Less iron in ER/Golgi from dZIP13-RNAi larvae and more in that of dZIP13-OE larvae . Copper was not affected while zinc contents were marginally different . n = 3 . p<0 . 05 , p<0 . 001; 2-tailed Student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 013 As previously discussed , the ferritin-loading experiments indicate iron homeostasis is affected in the secretory pathway after dZIP13 RNA interference . Because dZIP13 is expressed in the secretory pathway , we wondered whether iron levels are lower in these compartments when dZIP13 expression is inhibited . Metal content of ER/Golgi was tested and results showed that the iron level of these compartments was indeed reduced in dZIP13-RNAi and increased in dZIP13-OE larvae ( Figure 5D ) . Zinc level was marginally altered in these compartments , but negatively correlated to iron levels ( Figure 5D ) . We are not sure , however , if it is a secondary effect of disrupted iron levels or a result of some residual zinc importing activity of dZIP13 . No previous reports have shown a ZIP protein can be a metal exporter . The potential intracellular localization of dZIP13 makes direct assays of its metal transporting activity with intact cells difficult . We reasoned that if dZIP13 indeed functions as a membrane iron exporter , when expressed in E . coli , a heterologous platform lacking the sophisticated intracellular membrane system as in eukaryote cells , it might locate to the plasma membrane and facilitate iron cellular export directly outside of the cell . Interestingly , expression of dZIP13 in E . coli resulted in very poor growth under normal conditions . Strikingly , growth was partially restored in iron replete conditions whereas zinc had no effect ( Figure 6A ) , indicating the lack of growth of E . coli with dZIP13 expression was due to intracellular iron deficiency . dZIP13 expression also rendered E . coli more resistant to iron stresses compared with the control ( Figure 6A ) . Further analysis of cellular iron contents indicated that with iron-rich medium ( 2 mM iron ) , expression of dZIP13 reduced the iron level of the cells , while the change of zinc was insignificant ( Figure 6B ) . These results suggest that heterologous expression of dZIP13 in E . coli , a foreign and arguably a relatively clean system , can reduce intracellular iron availability to cells . 10 . 7554/eLife . 03191 . 014Figure 6 . Heterologous dZIP13 expression renders the growth of E . coli iron-dependent and iron-resistant . ( A ) E . coli expressing dZIP13 required iron addition to grow and was more resistant to iron excess . ( B ) Less iron content was detected in E . coli expressing dZIP13 while zinc and copper were not much affected . n = 3 . p<0 . 001; two-tailed Student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 014 Although previous data strongly suggest dZIP13 functions as an iron exporter , they have yet to provide direct proof . One experiment to show a protein as an iron exporter is via a radioactive Fe transport assay . Ferroportin , a plasma membrane iron exporter , for example , has been successfully shown to mediate iron export in Xenopus oocytes , by using Fe isotopes and measuring intracellular and extracellular radioactivity levels ( Donovan et al . , 2000 ) . However , as stated above , dZIP13 is located intracellularly and its expression would only directly alter redistribution of iron inside the cell rather than exporting iron out of the cell , as was seen with ferroportin in the oocyte . In order to show dZIP13 indeed acts as an iron exporter , we decided to measure iron effluxing activity of dZIP13 when expressed in E . coli , taking advantage of the fact that the only membrane an E . coli cell has is its plasma membrane . When dZIP13 is located on the plasma membrane , the radioisotope exporting activity can be directly measured by monitoring the radioactivity in the external buffer . A time-course study of iron transportation in E . coli expressing dZIP13 is shown in Figure 7A . The amount of iron exported from E . coli expressing dZIP13 increased roughly linearly within the tested time and was significantly higher in comparison to the control . Labeling with anti-dZIP13 indicates that dZIP13 was indeed located on the membrane of E . coli cells , while control with the empty vector was with minimal signal ( Figure 7B ) . These results demonstrate that dZIP13 in E . coli is capable of exporting iron out of the cell . 10 . 7554/eLife . 03191 . 015Figure 7 . Iron radioisotope transport indicates dZIP13 as an iron exporter . ( A ) A time-course of iron released from dZIP13-expressing E . coli cells . dZIP13 expression in E . coli significantly increased the iron efflux rate than the vector control . ( B ) Imunofluoresence of E . coli expressing dZIP13 indicated that at least some dZIP13 was located on the membrane of E . coli . ( C ) NADPH-cytochrome c reductase activity assay showed that the ER–Golgi preparation was indeed enriched with a large amount of ER/Golgi . ( D ) A time-course of iron uptake into the ER/Golgi sample . The rate of iron uptake was much higher in dZIP13-OE and lower in dZIP13-RNAi ER/Golgi samples . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 01510 . 7554/eLife . 03191 . 016Figure 7—figure supplement 1 . Western blot showing that the ER/Golgi samples purified contain both ER and Golgi . Golgi and ER were detected with anti-GM130 and anti-PDI antibodies respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 01610 . 7554/eLife . 03191 . 017Figure 7—figure supplement 2 . Purity of the ER/Golgi samples isolated from Drosophila . ( A ) Cytochrome c oxidase assay indicated that the purified samples still contained some residual mitochondrial parts . ( B ) β-N-Acetylglucosaminidase ( NAG ) activity assay showed that the samples were essentially free of lysomal contamination . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 017 We further isolated ER/Golgi from Drosophila larvae and tested their iron transporting activity . In this case , dZIP13 would transport iron into the ER/Golgi apparatus . By testing activities of enzymes for different compartments , we were able to determine that our extract preparations were reasonably pure except for some residual mitochondria ( Figure 7C , Figure 7—figure supplement1 and Figure 7—figure supplement2 ) . The results shown in Figure 7D indicate that the iron uptake of ER/Golgi isolated from dZIP13-OE Drosophila larvae was significantly increased whereas that from dZIP13-RNAi was decreased . Again , a linear correlation between time and radioactivity was observed . Collectively , these results provide strong direct evidences in support of our hypothesis that dZIP13 is responsible for iron transport from the cytoplasm into the secretory pathway . Bioinformatics analysis indicates that dZIP13 shares the highest homology with human ZIP13 ( Figure 1A , Figure 1B ) . Because ubiquitous overexpression of dZIP13 resulted in reduced body aconitase activity , we wanted to see whether overexpression of hZIP13 would cause a similar effect in the fly . Indeed , hZIP13 overexpression in Drosophila also led to a decrease of overall aconitase activity ( Figure 8A ) , suggesting hZIP13 similarly affected iron metabolism in Drosophila . Moreover , a survival assay showed that expression of hZIP13 rescued the eclosion rate of dZIP13-RNAi larvae from 40% to 75% ( Figure 8B ) , and doubled the lifespan of dZIP13-RNAi flies ( Figure 8C ) . 10 . 7554/eLife . 03191 . 018Figure 8 . dZIP13 is functionally analogous to hZIP13 . ( A ) Decreased aconitase activity was observed after either dZIP13 or hZIP13 overexpression , but not a closely related zinc transporter hZIP7 . n = 6 . p<0 . 01; two-tailed Student's t test . ( B ) The eclosion defect of dZIP13-RNAi flies could be significantly rescued by hZIP13 expression . Eclosion was rescued from ∼40% to ∼75% by hZIP13 . n = 6 . p<0 . 05 , *p<0 . 01; two-tailed Student's t test . ( C ) The shortened lifespan of dZIP13-RNAi was also rescued by hZIP13 . ( D ) dZIP13-RNAi flies exhibited iron-rescuable collagen defects . Shown are images of fat body cells in VkgG454/+ control larvae and dZIP13-RNAi larvae , cultured on normal food and food with 10 mM FAC . Green: Vkg-GFP , blue: DAPI . ( E ) The lysyl hydroxylation reaction . The reaction depends on the presence of ferrous iron . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 018 Another member of the ZIP family , hZIP7 , is closely related to ZIP13 ( Figure 1B ) and has been shown to be a zinc importer located on the ER/Golgi ( Taylor et al . , 2004; Huang et al . , 2005 ) . Proteins destined for secretion are at least transiently localized to the ER and could theoretically function during their temporary stay in this organelle . If ZIP13 had acted as a zinc importer as other ZIP members , we would expect the closely related hZIP7 might also exhibit effects similar to those seen with hZIP13 when expressed in the fly . However , despite similarity to hZIP13 , expression of hZIP7 produced neither whole body aconitase activity change , nor rescued the defects of dZIP13-RNAi flies ( Figure 8A–C ) . In fact , it appeared to have even slightly exacerbated the defects . hZIP13 mutations cause SCD-EDS , a disease with defective collagen hydroxylation leading to a decrease of collagen cross-linking and secretion . We asked whether dZIP13 knockdown could also affect collagen formation in flies . Type IV collagen , the main constituent of basement membrane , is encoded by two collagen genes Viking ( Vkg ) and Cg25C , which are expressed in Drosophila fat bodies . dPlod is expressed in the collagen-producing cells ( Bunt et al . , 2011 ) , and participates in the assembly and secretion of collagen . We used Vkg-GFP , a GFP protein-trap that identifies the localization of the endogenous Viking protein as a marker of the type IV collagen ( Morin et al . , 2001 ) . Ubiquitous knockdown of dZIP13 caused substantial retention of Vkg-GFP in the fat body cells , while iron supplementation rescued the Vkg-GFP retention ( Figure 8D ) . Because iron is an important cofactor/co-substrate of lysyl hydroxylase , a critical enzyme in collagen synthesis ( Figure 8E ) ( Murad et al . , 1985 ) , the observation of iron reduction in ER/Golgi when dZIP13 is inhibited ( Figure 5D ) suggests that the normal functions of lysyl hydroxylase and prolyl hydroxylase might be compromised under iron shortage ( see ‘Discussion’ for more ) . Taken together , our results indicate dZIP13 is analogous to hZIP13 , and is truly the Drosophila orthologue of hZIP13 . Their functions are fundamentally different from those of the other zinc importers including the closely related ZIP7 . We presented substantial evidence establishing that dZIP13 mediates iron export to the secretory pathway , identifying dZIP13 as the elusive iron transporter in the secretory pathway . Moreover , this is the first time to report that a member of the ZIP transporter family functions as an iron exporter . Previously , certain members of ZIPs , such as ZIP14 ( Liuzzi et al . , 2006 ) and ZIP8 ( Wang et al . , 2012 ) , have been reported to be capable of transporting iron . However , they are only complex , broad-scope metal-ion importers and their in vivo involvement in iron homeostasis is yet to be established . Our in vivo studies clearly demonstrated that expressional alteration of dZIP13 mainly affects iron instead of zinc metabolism . Modulating dZIP13 expression in the gut cell resulted in opposite iron changes in the cytosol and secretory organelles , and radioisotope iron transport assay further showed that dZIP13 mediates iron exporting , providing compelling evidence in supporting dZIP13 as an iron efflux pump . Future biochemistry/biophysical work is needed to decipher the exact mechanistic detail of the transportation process and evidences thus obtained may further substantiate the conclusion of dZIP13 as an iron exporter . The unanticipated finding of dZIP13 as an iron exporter was facilitated by the characteristic manner of dietary iron uptake in Drosophila gut , which is mediated by secreted ferritin and requires iron loading in the secretory pathway ( Tang and Zhou , 2013b ) . The subcellular distributions of ferritin are very different between mammals and insects . In mammals , ferritin is predominantly localized in the cytosol for storing cytosolic iron , and absorbed dietary iron is effluxed from enterocytes through ferroportin ( Vanoaica et al . , 2010 ) . However , in insects such as Drosophila , ferritin is mainly secreted , playing a central role in iron absorption by bringing iron out of enterocytes and into the hemolymph through the secretory pathway ( Tang and Zhou , 2013b ) . This feature of iron absorption demands higher levels of iron to be pumped into the secretory pathway for systemic use , and facilitated our characterization of dZIP13 as the iron pumper which fulfills this role . In addition to the iron transport function , we also occasionally observed a minor zinc connection for dZIP13 . It is possible that dZIP13 still retains some residual zinc-importing activity . However , in most cases , the zinc effect is physiologically marginal or insignificant . We would argue that the zinc dyshomeostasis is not the primary effect of dZIP13 function loss , because iron but not zinc levels were altered in the body of dZIP13-RNAi and dZIP13-OE larvae . In addition , iron but not zinc supplementation is able to rescue loss of dZIP13 . Finally , no genetic interaction was observed between dZIP13 and other ZIPs in the ER/Golgi . These results argue against the idea that zinc dyshomeostasis is the primary defect in the loss of dZIP13 function , and that iron dyshomeostasis is secondary to a zinc defect . Our other characterizations of dZIP13 , as reported in this work , are all consistent with this notion . Considering the unusual nature of dZIP13 as an iron exporter while other members of the family ( SLC39A ) so far identified all appear to be importers , it is noteworthy that zinc efflux protein ZnT5 ( SLC30A5 ) can mediate zinc transport in both directions ( Valentine et al . , 2007 ) . It would be interesting to know what sequences or structures endow dZIP13 with this unique property . At this stage we do not yet have clear answers to this question . Amino acid sequence comparisons of ZIP13 with other ZIP family members reveal some distinct differences . The most notabe difference is in transmembrane domain TM4 . The TM4 amino acid sequence for all ZIP13s are DNFTHG ( Figure 9 ) whereas other closely related ZIPs are HNFTDG , i . e . , the highly conserved region in TM4 has a D and H residue swap between ZIP13s and other ZIP members . Our preliminary evidences suggest this variation might be an important factor because when we switched this D and H dZIP13 was no longer functional . Additional experiments are clearly needed to further explore this promising lead . 10 . 7554/eLife . 03191 . 019Figure 9 . Two conserved amino acids D and H in the fourth transmembrane domain of other ZIPs are switched in ZIP13 . Shown here are the fourth transmembrane domain and adjacent sequences of ZIP13 and several other ZIPs . 1–3 are ZIP13 sequences from different organisms . 4–6 are several other representative ZIPs . Predicted TM4 segments are underlined . The highlighted two amino acids D and H are conserved in all other closely related ZIPs while their positions are switched in ZIP13s . Numbers in parenthesis indicate the starting/ending positions of the ZIPs shown . 1 . ZIP13 ( Danio rerio ) ; 2 . ZIP13 ( CG7816 ) ( Drosophila melanogaster ) ; 3 . ZIP13 ( Homo sapiens ) ; 4 . ZIP7 ( Homo sapiens ) ; 5 . ZIP8 ( Homo sapiens ) ; 6 . ZIP14 ( Homo sapiens ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 019 Mutations in LH family genes have been found in several human diseases including Bruck syndrome ( van der Slot et al . , 2003; Ha-Vinh et al . , 2004 ) ( OMIM #609220 ) and Ehlers–Danlos type VI ( Yeowell and Walker , 2000 ) ( OMIM #225400 ) . Recently mutations in ZIP13 , a member of the SLC39A/Zrt-Irt-like protein ( ZIP ) family , have been identified as the cause for the spondylocheiro dysplastic Ehlers–Danlos syndrome ( SCD-EDS ) , a form of EDS sharing some similar clinical presentations with EDS VI , which is caused by mutations in PLOD1 gene encoding lysyl hydroxylase ( LH1 ) ( Fukada et al . , 2008; Giunta et al . , 2008 ) . Because ZIP transporters were generally believed to import zinc into the cytosol from the extracellular milieu or organellar lumen ( Kambe et al . , 2006; Lichten and Cousins , 2009; Jeong and Eide , 2013 ) , the pathogenesis of SCD-EDS was difficult to understand and various hypotheses were proposed to explain how the zinc importer could affect iron homeostasis ( Fukada et al . , 2008; Giunta et al . , 2008; Jeong et al . , 2012 ) . The identification of dZIP13 as the iron exporter for the secretory compartments suggests a more direct interpretation of SCD-EDS: ZIP13 mutations impair iron transport to ER/Golgi , significantly attenuating the activities of iron-requiring enzymes such as lysyl hydroxylase , leading to defective collagen synthesis and accumulation in the secretory pathway . Several issues are worthy of clarification here . SCD-EDS was identified to be defective in collagen crosslinking from actions of lysyl and prolyl hydroxylases . However , lysyl and prolyl hydroxylase activities from the cellular extracts of patient cells were reported to be normal ( Giunta et al . , 2008 ) . This is likely the result from the enzymatic assays , which were performed in the presence of iron ( Murad et al . , 1985; Giunta et al . , 2008 ) . It has been shown that externally supplied iron is needed to do these assays ( Tuderman et al . , 1977 ) , suggesting iron is probably not tightly bound to these enzymes . However , the addition of iron would neutralize or mask the original iron-deficienct state of these enzymes in vivo ( in the ER/Golgi of SCD-EDS patient cells ) , and therefore produce ‘normal’ in vitro enzymatic activities despite defective in vivo hydroxylation . These seemingly contradictory observations , therefore , are very consistent with our conclusion that ZIP13 is an iron exporter that supplies iron to ER/Golgi . Jeong et al . ( 2012 ) used mammalian cells to measure metal uptake activity of ZIP13 . Because ZIP13 is normally an intracellular protein , the authors claimed that when overexpressed , some ZIP13 was mistargeted to the plasma membrane , making the uptake assay possible . In that experiment , zinc uptake was observed , but iron failed to compete with zinc in the importing assay . However , no exporting activity was examined in that experiment . As reported and discussed previously , sometimes we also observed slight changes in zinc levels , though we are not sure whether this was a result of direct dZIP13 transportation or due to secondary effects of iron dyshomeostasis . It is possible that dZIP13 may still retain some residual zinc importing activity . If this is the case , then the results from Jeong et al . would not be contradictory with our own findings . While Jeong et al . proposed zinc deficiency in the ER/Golgi , Giunta et al . ( 2008 ) , Fukada et al . ( 2008 ) , and Bin et al . ( 2011 ) proposed zinc accumulation in the ER/Golgi underlies SCD-EDS . According to this theory , zinc accumulation in ER/Golgi can compete with iron and then affect collagen hydroxylation . Indeed , zinc has been reported as an inhibitor of prolyl hydroxylase ( Tuderman et al . , 1977 ) . However in our hands , we saw only a marginal increase of zinc in the ER/Golgi when dZIP13 was knocked-down , but a much more dramatic change in the iron level . Because hZIP7 , a zinc importer , cannot rescue dZIP13-RNAi , we think when dZIP13 is knocked-down the primary defect is not zinc accumulation . We suggest iron dyshomeostasis instead of a zinc defect is probably the primary cause contributing to SCD-EDS . Although we cannot exclude zinc's effect , it appears that it is not the primary reason , or at most a subsidiary factor . Another issue worth pointing out is that none of the classical iron phenotypes are observed in SCD-EDS patients . Iron is involved in heme and Fe-S synthesis , and iron deficiency is typically observed as anemia . In SCD-EDS , it is possible that iron deficiency is localized to the secretory pathway and cytosolic and mitochondrial iron levels remain not much affected . Therefore , a ZIP13 mutation would result in a type of iron dyshomeostasis that would not present itself with the typical types of symptons ( e . g . , skin abnormality , bone malformation , and growth retardation ) observed with classical iron deficiency . We still do not know what other proteins in the secretory pathway require iron . In addition to the common EDS-like features , SCD-EDS patients do present other phenotypes such as generalized skeletal dysplasia involving the spine and striking clinical abnormalities of the hands ( Giunta et al . , 2008 ) . Are these additional characteristics also a result of collagen defects or due to abnormalities other than the collagen ? This question remains unanswered . Nevertheless , our results add another level of complexity in the regulation of iron homeostasis and some of the consequences that arise from its disruption . Constructs used for transgenic flies include pUAST-dZIP13 and pUAST-Golgi-mRFP . pUAST-dZIP13 was generated by PCR amplification of the coding region of CG7816 from Drosophila cDNA and cloning into pUAST using the following primers: pUAST-dZIP13 F: 5′-GGAATTCAGCCGAAAATGACCACGAACAG-3′ , pUAST-dZIP13 R: 5′-ATAAGAATGCGGCCGCCCTAGTGTTCGAATAGCATGGTCATC-3′; pUAST-Golgi-mRFP was generated by PCR amplification of monomeric RFP with rhomboid-1 ( Rho1 ) , a Golgi marker protein ( Lee et al . , 2001; Chen et al . , 2006 ) and cloning into pUAST . Construct used for transfecting human CHO cells was pIRESneo-dZIP13-myc , constructed in pIRESneo ( clontech ) by fusing myc in frame to the C terminal of dZIP13 using the following primers: pIRESneo-dZIP13 F: 5′-CTAGTGATATC AGCCGAAAATGACCACGAACAG-3′ , pIRESneo-dZIP13 R: 5′-CGGAATTCCTACAGGTCTTCTTCAGAGATCAGTTTCTGTTCGTGTTCGAATAGCATGGTCATC-3′ . Construct used in E . coli assay was pET28a-dZIP13 , which was generated by cloning dZIP13 into pET28a using the following primers: pET28a-dZIP13 F: 5′-ACG CATATG ACCACGAACAGCAGCTTCTTC-3′ , pET28a-dZIP13 R: 5′-TAC GAATTC CTAGTGTTCGAATAGCATGGTCATC-3′ , and then transfected into BL21 ( DE3 ) . All the constructs were verified by sequencing . Unless otherwise noted , flies were normally reared on standard cornmeal media at 25°C and third instar larvae were used . The concentrations of supplemented metals or metal chelators used were as follows: 5 mM ferric ammonium citrate ( FAC ) , 0 . 1 mM bathophenanthrolinedisulfonic acid disodium ( BPS ) ( Sigma , St . Louis , MO , USA ) , 25 μM N , N , N′ , N′-tetrakis ( 2-pyridylmethyl ) ethylenediamine ( TPEN ) ( Sigma , St . Louis , MO , USA ) , 2 mM ZnCl2 ( Beijing Yili Fine Chemicals Ltd . Co . , Beijing , China ) . Transgenic flies for each pUAST construct were generated in w1118 background by P-element-mediated transformation . Information about the flies used in this study is listed in Table 1 . 10 . 7554/eLife . 03191 . 020Table 1 . Drosophila used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 03191 . 020DrosophilaDescriptionsOriginDa-Gal4 ( #8641 ) Ubiquitous Gal4Bloomington Drosophila Stock CenterNP3084 ( #113094 ) Expresses Gal4 in salivary glands , gastric caecae , and whole midgut in third instar larvaeGenetic Resource Center at the Kyoto Institute of Technology ( DGRC ) Vkg-GFP ( #G00454 ) Carries a GFP fused to vikingFlytrapFer1HCHG188/TM3 ( #G00188 ) Carries a GFP fused to Fer1HCH ( ferritin 1heavy-chain homolog ) FlytrapdZIP13-RNAi ( #1364 ) CG7816 RNAi lineVienna Drosophila RNAi CenterdZIP13-OECG7816 over-expression lineThis studyFerHCH-RNAi ( #12925 ) Fer1HCH RNAi lineVienna Drosophila RNAi CenterhZIP7-OEHuman ZIP7 over-expression lineThis studyhZIP13-OEHuman ZIP13 over-expression lineThis studyGolgi markerCarries a RFP fused to Rho1This studyER marker ( #ZCL1503 ) Carries a GFP fused to PDI ( Morin et al . , 2001 ) Early endosome marker ( #39695 ) Carries a GFP fused to FYVEBloomington Drosophila Stock CenterLate endosome marker ( #42714 ) Carries a GFP fused to LAMPBloomington Drosophila Stock CenterdZIP13 mutant ( #18595 ) CG7816 mutant lineBloomington Drosophila Stock Center To examine the effects of metals or chelators on the eclosion of dZIP13-RNAi , Da-GAL4 homozygous flies were crossed to dZIP13-RNAi , and the progeny were reared on food containing different metals or metal chelators . The density of each vial was controlled to ∼100 progeny . The total number of emerging adults of each genotype was counted . Longevity assays was performed as described previously ( Xiao et al . , 2013 ) . 3-day-old adult females were collected . 20 flies were placed in a food vial and each vial was kept at 25°C with 60% humidity under a 12-hr light–dark cycle . Food vials were changed every 2 days , and dead flies were counted at that time . 10 parallel group tests were conducted for each genotype , and the experiments were repeated at least three times . Percentage increases in lifespan were based on comparing the median survivals to the controls . Flies of each genotype were reared on normal food from eggs until late third-instar larval stage . About 25 larvae or 110 guts or 50 whole body minus guts were collected , weighed , and sent for quantitative elemental analysis with inductively coupled plasma-mass spectrometry ( ICP-MS ) XII ( Thermo Electron Corp , Waltham , MA , USA ) by the Analysis Center of Tsinghua University ( Wang et al . , 2009; Xiao et al . , 2013 ) . For ER/Golgi metal content , Drosophila ER/Golgi was purified as previous described ( Graham , 2001 ) . After the protein concentration was determined , ∼1 . 4 mg protein was sent for ICP-MS . For E . coli metal content , ∼0 . 33 g ( dry weight ) E . coli was used . Protein was extracted with PBST ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 2 mM KH2PO4 , and 0 . 1% Triton X-100 , PH 7 . 4 ) from cells or tissue samples . Protein concentration was measured by the BCA kit ( Thermal ) . ∼60 μg protein for the adult gut extract , or ∼250 μg protein for the whole larvae extract was added to 700-μl reaction buffer ( 50 mM K2HPO4 , pH 7 . 4 , containing 30 µM citric acid ) . The increase of absorbance at 240 nm was monitored for 2 min as the relative aconitase activity . In-gel aconitase activity assays were performed essentially as described ( Tong and Rouault , 2006 ) . Gels consisted of a separating gel containing 8% acrylamide ( 132 mM Tris base , 132 mM borate , 3 . 6 mM citrate ) , and a stacking gel containing 4% acrylamide ( 67 mM Tris base , 67 mM borate , 3 . 6 mM cit-rate ) . The running buffer contained 25 mM Tris pH 8 . 3 , 192 mM glycine , and 3 . 6 mM citrate . Electrophoresis was carried out at 180 V at 4°C . Aconitase activities were assayed by incubating the gel in the dark at 37°C in 100 mM Tris ( pH 8 . 0 ) , 1 mM NADP , 2 . 5 mM cis-aconitic acid , 5 mM MgCl2 , 1 . 2 mM MTT , 0 . 3 mM phenazine methosulfate , and 5 U/ml isocitrate dehydrogenase . Total RNA was extracted with TRIzol reagent ( Invitrogen , Carlsbad , CA , USA ) . cDNA was reverse-transcribed from 1 μg total RNA with TransScript Reverse Transcriptase ( TransGen Biotech Co . , Beijing , China ) . Semiquantitative RT-PCR was performed using gene-specific primers to amplify partial regions of target genes . RNA isolation and reverse transcription were performed independently for three times , and no less than three PCR experiments were applied to each cDNA sample . Mouse polyclonal antibodies were raised against recombinant dZIP13-2 protein fragment ( MTEEKMAKEGYKDPADSKLLRSGSADEENPQPKCVEIANCLLRRHGGQLPEGETSESCGGACDIEDVGKVCFLREQEQKSKERKEQPKRSGFSRWDAARAQKEEERKESIKQLE ) . Briefly , the cDNA fragments encoding the cytosolic side of this protein ( dZIP13-2 ) were synthesized and cloned into pTwin1 ( NEB ) vector . The recombinant protein was expressed in E . coli and purified by chitin beads ( NEB ) , and injected into mice for antibody generation ( Institute of Genetics and Developmental Biology , Chinese Academy of Sciences , Beijing , China ) . The antibody was affinity purified and pre-absorbed before use . Anti-Fer2LCH was as described before ( Tang and Zhou , 2013b ) . Anti-tubulin rat monoclonal antibody ( ab6160 ) , anti-GM130 rabbit polyclonal antibody ( ab30637 ) , and anti-PDI mouse monoclonal antibody ( ab2792 ) were obtained from Abcam ( Cambridge , MA , USA ) . Secondary antibodies include HRP-conjugated goat anti-mouse IgG , goat anti-rabbit IgG and goat anti-rat IgG ( Zhongshan Goldenbridge Biotechnology , Beijing , China ) . For Western blot analysis , fly samples were homogenized in the buffer containing 1% Triton X-100 plus 10% proteinase inhibitor cocktail ( Sigma ) , centrifuged , separated on 10% SDS-PAGE , and transferred to nitrocellulose membranes ( Millipore , Watford , UK ) . Signals were developed with ECL detection kit ( Vigorous Biotechnology , Beijing , China ) . Caco-2 cells were maintained in MEM ( Invitrogen ) containing 10% fetal bovine serum ( FBS , Gibco BRL , Gaithersburg , MD , USA ) and MEM NEAA ( Gibco ) at 37°C . Cells were transfected with pIRESneo-dZIP13-myc and constructs of ER/Golgi markers using lipofectamine ( Invitrogen ) . After 24 hr , cells were fixed , taken pictures after staining for the myc antibody ( green ) and the Golgi marker ( anti- Giantin , 1:500 ) or ER marker ( anti-PDI , 1:500 ) . Anti-c-Myc rabbit polyclonal antibody ( 1:500 ) ( ab51156 ) , anti-Giantin rabbit polyclonal antibody ( ab80864 ) and anti-PDI mouse monoclonal antibody ( ab2792 ) were obtained from Abcam ( Cambridge , MA , USA ) . Secondary antibodies include cy3-conjugated goat anti-mouse and cy3-conjugated goat anti-rabbit IgG ( Zhongshan Goldenbridge Biotechnology , Beijing , China ) . For dZIP13 staining in E . coli , fixation and permeabilization of cells transformed with pET28a were performed as described previously ( Den Blaauwen et al . , 2003 ) . Poly-L-lysine-coated coverslips loaded with fixed cells were washed three times with PBS , and nonspecific binding sites were blocked for 1 hr in PBS containing 1% bovine serum albumin . Coverslips were incubated with anti-dZIP13 antibody ( 1:100 ) for 1 hr , washed three times with PBS , and incubated for an additional 1 hr with FITC- conjugated goat anti-mouse IgG ( 1:1000 ) . The coverslips were then washed three times with PBS , mounted onto glass slides , and taken pictures . For fly samples , tissues indicated were dissected , fixed , stained , and mounted following standard procedures ( Pastor-Pareja and Xu , 2011 ) . The following antibodies and dyes were used: mouse anti-dZIP13 ( 1:100 ) , goat anti-mouse IgG conjugated to FITC or cy3 ( 1:1000 , Zhongshan Goldenbridge Biotechnology ) and DAPI ( 1:5 , Beyotime C1005 ) . For DAPI staining , samples were incubated in 50 ng/ml DAPI for 10 min . Slides were mounted with 50% glycerol/PBS . Confocal images were taken with a Zeiss LSM710 Meta confocal microscope . Fluorescence of Fer1HCHG188 larvae was examined and recorded with a Nikon ECLIPES 80i microscope attached to a Nikon DXM1200F digital camera ( Nikon , Tokyo , Japan ) . Detection of ferric iron in the midgut or in-gel ferric analysis was performed as described previously ( Tang and Zhou , 2013b ) . 20 ml freshly inoculated E . coli was grown to a density of about OD600 = 0 . 05 in LB containing kanamycin . The cells were incubated with 10 μCi 55Fe ( PerkinElmer , NEZ04300 ) in the same medium containing 100 μM ascorbate ( Sigma ) for 1 . 5 hr . Then 0 . 5 mM IPTG was added into the medium and incubated for another 1 . 5 hr . The bacteria were collected at 8000×g for 5 min , washed with ice-cold PBS containing 100 μΜ EDTA to remove iron non-specifically bound to the cell surface , and twice with ice cold PBS . To measure 55Fe release , the bacterial cells were then incubated with 2 ml PBS at 37°C , 200 μl medium was sampled at each time point . Supernatants were collected after centrifugation and counted by liquid scintillation ( 1450 MicroBeta TriLux , PerkinElmer Life Sciences ) . % 55Fe release = ( cpm in supernatant ) / ( cpm in time 0 ) × 100% Drosophila ER/Golgi purification was performed as previous described ( Graham , 2001 ) . After determining the protein concentration , the ER/Golgi solution was diluted into 0 . 5 μg/μl protein in the presence of 0 . 25 M sucrose . To measure iron uptake , ER/Golgi samples were added with 50 μM ascorbate ( Sigma ) , 50 μM FeCl2 , and 10 μCi 55Fe and incubated at 37°C . 200 μl samples were collected by filters ( Millipore , VCWP01300 ) every 10 min , and washed three times with isotonic solution ( Sigma , I3533 ) . The filters were counted by liquid scintillation ( 1450 MicroBeta TriLux , PerkinElmer Life Sciences ) . The preparation was performed as previously described ( Miwa et al . , 2003 ) . Third instar larvae were collected and homogenized in a buffer containing 250 mM sucrose , 5 mM Tris–HCl , 2 mM EGTA , 1% ( wt/vol ) bovine serum albumin , pH 7 . 4 at 4°C . Protein concentration was measured by the BCA kit ( Thermal ) . Activities of NADPH-cytochrome c reductase , cytochrome c oxidase , and β-N-Acetylglucosaminidase were measured by the kit ( Sigma #CY0100 , # CYTOCOX1 , # CS0780 , St . Louis , MO , USA ) according to the manufacturer's instructions . Data were analyzed by Student's t-test between groups , and while multiple groups were compared ANOVA was used . Statistical results were presented as means ± SEM . Asterisks indicate critical levels of significance ( p<0 . 05 , p<0 . 01 , and *p<0 . 001 ) .	Iron is essential for life . Amongst its many important roles , iron is crucial for producing collagen—the protein that provides both strength and elasticity to bones , tendons , ligaments , and skin . Like many other proteins , collagens are produced inside the endoplasmic reticulum—an organelle inside the cell that is enclosed by a membrane that is similar to the plasma membrane that surrounds the cell itself . Two enzymes that are critical for producing collagen need to bind with iron in order to work correctly . To do this , iron in the cytoplasm of the cell has to cross the membrane that surrounds the endoplasmic reticulum . Small molecules are commonly transported across membranes by proteins called transporters , which tend to work on specific types of ions or molecules . However , researchers did not know the identity of the membrane transporter responsible for moving iron into the secretory pathway—including the endoplasmic reticulum—to bind with the enzymes that produce collagen . Xiao , Wan et al . have now investigated the function of the transporter ZIP13 in the fruit fly Drosophila . This transporter was thought to transport zinc across membranes and into the cytoplasm . Instead , Xiao , Wan et al . found that ZIP13 transports iron out of the cytoplasm and into the endoplasmic reticulum . Ehlers–Danlos syndrome is a condition that causes individuals to suffer from frequent joint dislocations , bone deformities , and fragile skin as a result of their body producing collagen incorrectly . One form of Ehlers–Danlos syndrome is caused by ZIP13 transporters working incorrectly . However , this was difficult to understand when it was thought that ZIP13 only transports zinc . The discovery that ZIP13 mostly transports iron rather than zinc can explain the link between this transporter and Ehlers–Danlos syndrome: if ZIP13 doesn't work , the collagen-building enzymes cannot get the iron they need to work properly . Disorders caused by iron deficiencies are normally identified by a few tell-tale symptoms , such as anemia , but these are not seen in Ehlers–Danlos syndrome . Xiao , Wan et al . suggest that iron transport problems could therefore be behind a wider range of diseases and disorders than is currently known .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	The metal transporter ZIP13 supplies iron into the secretory pathway in Drosophila melanogaster
Using cryo-electron microscopy ( cryo-EM ) , we determined the structure of the Escherichia coli 70S ribosome with a global resolution of 2 . 0 Å . The maps reveal unambiguous positioning of protein and RNA residues , their detailed chemical interactions , and chemical modifications . Notable features include the first examples of isopeptide and thioamide backbone substitutions in ribosomal proteins , the former likely conserved in all domains of life . The maps also reveal extensive solvation of the small ( 30S ) ribosomal subunit , and interactions with A-site and P-site tRNAs , mRNA , and the antibiotic paromomycin . The maps and models of the bacterial ribosome presented here now allow a deeper phylogenetic analysis of ribosomal components including structural conservation to the level of solvation . The high quality of the maps should enable future structural analyses of the chemical basis for translation and aid the development of robust tools for cryo-EM structure modeling and refinement . The ribosome performs the crucial task of translating the genetic code into proteins and varies in size from 2 . 3 MDa to over 4 MDa across the three domains of life ( Melnikov et al . , 2012 ) . Polypeptide synthesis occurs in the peptidyl transferase center ( PTC ) , where the ribosome acts primarily as an ‘entropic trap’ for peptide bond formation ( Rodnina , 2013 ) . To carry out the highly coordinated process of translation , the ribosome orchestrates the binding and readout of messenger RNA ( mRNA ) and transfer RNAs ( tRNAs ) , coupled with a multitude of interactions between the small and large ribosomal subunits and a host of translation factors . These molecular interactions are accompanied by a wide range of conformational dynamics that contribute to translation accuracy and speed ( Munro et al . , 2009; Javed and Orlova , 2019; Loveland et al . , 2020; Morse et al . , 2020 ) . Because of the ribosome’s essential role in supporting life , it is naturally the target of a plurality of antibiotics with diverse mechanisms of action ( Arenz and Wilson , 2016 ) . The ribosome also plays a unique role in our ability to study the vast array of RNA secondary and tertiary structural motifs found in nature , as well as RNA-protein interactions . Although X-ray crystallography has been central in revealing the molecular basis of many steps in translation , the resolution of available X-ray crystal structures of the ribosome in key functional states remains too low to provide accurate models of non-covalent bonding , that is , hydrogen bonding , van der Waals contacts , and ionic interactions . Furthermore , the development of small-molecule drugs such as antibiotics is hampered at the typical resolution of available X-ray crystal structures of the ribosome ( ~3 Å ) ( Arenz and Wilson , 2016; Yusupova and Yusupov , 2017 ) . Thus , understanding the molecular interactions in the ribosome in detail would provide a foundation for biochemical and biophysical approaches that probe ribosome function and aid antibiotic discovery . The ribosome has been an ideal target for cryo-EM since the early days of single-particle reconstruction methods , as its large size , many functional states , and multiple binding partners lead to conformational heterogeneity that makes it challenging for X-ray crystallography ( Frank , 2017 ) . Previous high-resolution structures of the bacterial ribosome include X-ray crystal structures of the Escherichia coli ribosome at 2 . 4 Å ( 2 . 1 Å by CC ½; Noeske et al . , 2015 ) and Thermus thermophilus ribosome at 2 . 3 Å ( Polikanov et al . , 2015 ) , and cryo-EM reconstructions at 2 . 1–2 . 3 Å ( Halfon et al . , 2019; Pichkur et al . , 2020; Stojković et al . , 2020 ) . While overall resolution is well-defined for crystallography , which measures information in Fourier space ( Karplus and Diederichs , 2015 ) , it is more difficult to assign a metric to the global resolution of cryo-EM structures . Fourier shell correlation ( FSC ) thresholds are widely used for this purpose ( Frank and Al-Ali , 1975; Harauz and van Heel , 1986 ) . However , the ‘gold-standard’ FSC ( GS-FSC ) most commonly reported , wherein random halves of the particles are refined independently and correlation of the two half-maps is calculated as a function of spatial frequency , is more precisely a measure of self-consistency of the data ( Henderson et al . , 2012; Rosenthal and Henderson , 2003; Subramaniam et al . , 2016 ) . Reporting of resolution is further complicated by variations in map refinement protocols and post-processing by the user , as well as a lack of strict standards for deposition of these data ( Subramaniam et al . , 2016 ) . A separate measure , the map-to-model FSC , compares the cryo-EM experimental map to the structural model derived from it ( DiMaio et al . , 2013; Liebschner et al . , 2019; Subramaniam et al . , 2016 ) . In this study , we use both metrics to evaluate our results but bring attention to the use of the map-to-model FSC criterion as it is less commonly used but more directly reports on the quality and utility of the atomic model . Here we determined the structure of the E . coli 70S ribosome to a global resolution of 2 . 0 Å , with higher resolution up to 1 . 8 Å in the best-resolved core regions of the 50S subunit . We highlight well-resolved features of the map with particular relevance to ribosomal function , including contacts to mRNA and tRNA substrates , a detailed description of the aminoglycoside antibiotic paromomycin bound in the mRNA decoding center , and interactions between the ribosomal subunits . We also describe solvation and ion positions , as well as features of post-transcriptional modifications and post-translational modifications seen here for the first time . Discovery of these chemical modifications , as well as a new RNA-interacting motif found in protein bS21 , provide the basis for addressing phylogenetic conservation of ribosomal protein structure and clues toward the role of a protein of unknown function . These results open new avenues for studies of the chemistry of translation and should aid future development of tools for refining structural models into cryo-EM maps . We determined the structure of the E . coli 70S ribosome in the classical ( non-rotated ) state with mRNA and tRNAs bound in the aminoacyl-tRNA and peptidyl-tRNA sites ( A site and P site , respectively ) . Partial density for exit-site ( E-site ) tRNA is also visible in the maps , particularly for the 3’-terminal C75 and A76 nucleotides . Our final maps were generated from two 70S ribosome complexes that were formed separately using P-site tRNAfMet that differed only by being charged with two different non-amino acid monomers ( see Materials and methods ) . In both complexes , we used the same mRNA and A-site Val-tRNAVal . Both complexes yielded structures in the same functional state , with similar occupancy of the tRNAs . As neither A-site nor P-site tRNA 3’-CCA-ends were resolved in the individual cryo-EM maps , we merged the two datasets for the final reconstructions ( Figure 1—figure supplement 1 ) . After Ewald sphere correction in RELION ( Zivanov et al . , 2018 ) , global resolution of the entire complex reached 1 . 98 Å resolution by GS-FSC , with local resolution reaching 1 . 8 Å ( Figure 1A , Figure 1—figure supplement 2 , Table 1 ) . The global resolution reached 2 . 04 Å using the map-to-model FSC criterion ( Figure 1—figure supplement 2 ) . We also used focused refinement of the large ( 50S ) and small ( 30S ) ribosomal subunits , and further focused refinements of smaller regions that are known to be conformationally flexible , to enhance their resolution ( Figure 1—figure supplement 3; von Loeffelholz et al . , 2017 ) . In particular , focused refinement of the 30S subunit improved its map quality substantially , along with its immediate contacts to the 50S subunit and the mRNA and tRNA anticodon stem-loops . Additional focused refinement of the central protuberance ( CP ) in the 50S subunit , the 30S head domain , and the 30S platform aided in model building and refinement . In the following descriptions , maps of specific ribosomal subunits or domains refer to the focused-refined maps . Details of the resolutions obtained are given in Figure 1—figure supplements 2–3 and Tables 2–3 . The high resolution indicated by the FSC curves is supported by several visual features observable in the maps , including holes in many aromatic rings and riboses , as well as ring puckers , the directionality of non-bridging phosphate oxygens in ribosomal RNA , and numerous well-resolved ions , water molecules , and small molecules ( Figure 1B–D , Figure 1—figure supplement 4 ) . The maps also reveal known post-transcriptional and post-translational modifications of ribosomal RNA and proteins in detail ( Figure 1—figure supplement 5 , Figure 1—figure supplement 6 ) , many of which are as previously described ( Fischer et al . , 2015; Noeske et al . , 2015; Polikanov et al . , 2015; Stojković et al . , 2020 ) . As seen in other ribosome structures , elements of the ribosome at the periphery are less ordered , including the uL1 arm , the GTPase activating center , bL12 proteins , and the central portion of the A-site finger in the 50S subunit ( 23S rRNA helix H38 ) , as well as the periphery of the 30S subunit head domain and spur ( 16S rRNA helix h6 ) . The resolution of the maps for the elbow and acceptor ends of the P-site and A-site tRNAs is also relatively low , likely as a result of poor accommodation of the unnatural substrates used . The 30S ribosomal subunit is highly dynamic in carrying out its role in the translation cycle ( Munro et al . , 2009 ) . Previously published structures demonstrate that even in defined conformational states of the ribosome , the 30S subunit exhibits more flexibility than the core of the 50S subunit . Furthermore , the mass of the 50S subunit dominates alignments in cryo-EM reconstructions of the 70S ribosome . Solvation of the 30S subunit has not been extensively modeled , again owing to the fact that it is generally more flexible and less well-resolved than the 50S subunit , even in available high-resolution structures ( Noeske et al . , 2015; Polikanov et al . , 2015 ) . Using focused-refined maps of the 30S subunit , we achieved the resolution necessary for in-depth chemical analysis of key contacts to mRNA and tRNAs , the 50S subunit , and to generate more complete models of 30S subunit components , Mg2+ ion positions , and solvation ( Figure 2—figure supplement 1 ) . The 30S ribosomal subunit controls interactions of tRNAs with mRNA and helps maintain the mRNA in the proper reading frame . The present structure reveals the interactions of the 30S subunit with mRNA and tRNA in the A and P sites including solvation . The tight binding of tRNA to the P site ( Lill et al . , 1986 ) is reflected in extensive direct contacts between the tRNA anticodon stem-loop ( ASL ) and both the 30S subunit head and platform domains ( Figure 2A–D ) . The C-terminal residues Lys129 and Arg130 of protein uS9 , which are important for translational fidelity ( Arora et al . , 2013 ) , form ionic and hydrogen bonding interactions with the nucleotides in the U-turn motif of the P-site ASL , and Arg130 stacks with the base of U33 ( Figure 2D ) . However , the C-terminal tails of ribosomal proteins uS13 and uS19 , which come from the 30S subunit head domain and are lysine-rich , are not visible in the map , suggesting they do not make specific contacts with the tRNA when the ribosome is in the unrotated state . Direct interactions between A-site tRNA and the 30S subunit are highly localized to the top of helices h44 and h18 in 16S rRNA and to helix H69 in 23S rRNA ( i . e . 16S rRNA nucleotides G530 , A1492 , and A1493 , and 23S rRNA nucleotide A1913 ) . By contrast , contacts between the A-site tRNA and 30S subunit head domain are entirely solvent mediated ( Figure 2E ) apart from nucleotide C1054 , possibly reflecting the weaker binding of tRNA to the A site ( Lill et al . , 1986 ) and the need for A-site tRNA conformational dynamics during mRNA decoding ( Rodnina et al . , 2017 ) . Protein bS21 , which resides near the path of mRNA on the 30S subunit platform ( Held et al . , 1973; Marzi et al . , 2007; Sashital et al . , 2014 ) , is essential in E . coli ( Bubunenko et al . , 2007; Goodall et al . , 2018 ) . Although partial structural models of bS21 have been determined ( Fischer et al . , 2015; Noeske et al . , 2015 ) , structural disorder in this region has precluded modeling of the C-terminus . In the present maps , low-pass filtering provides clear evidence for the conformation of the entire protein chain , including 13 amino acids at the C-terminus that extend to the base of the 30S subunit head domain ( Figure 3A ) . Most of the C-terminal residues are found in an alpha-helical conformation near the Shine-Dalgarno helix formed between the 3’ end of 16S rRNA and the mRNA ribosome binding site ( Shine and Dalgarno , 1975 ) , while the C-terminal arginine-leucine-tyrosine ( RLY ) motif makes close contacts with 16S rRNA helix h37 and nucleotide A1167 ( Figure 3B ) . The arginine and leucine residues pack in the minor groove of helix h37 , and the terminal tyrosine stacks on A1167 . Multiple sequence alignment of bS21 sequences from distinct bacterial phyla revealed that the RLY C-terminal motif is conserved in bS21 sequences of the Gammaproteobacteria phylum while in Betaproteobacteria , the sibling group of Gammaproteobacteria , bS21 sequences possess a lysine-leucine-tyrosine ( KLY ) C-terminal motif instead ( Figure 3C ) . Interestingly , such C-terminal extensions ( RLY or KLY ) are absent in other bacterial phyla ( Figure 3—figure supplement 1 ) . Recently , putative homologs of bS21 were identified in huge bacteriophages , which were shown to harbor genes encoding components of the translational machinery ( Al-Shayeb et al . , 2020 ) . Inspection of sequence alignments reveals that some phage S21 homologs also contain KLY-like motifs ( Figure 3C ) . The presence of phage S21 homologs containing KLY-like motifs is consistent with the host range of these phages ( Figure 3C , Supplementary file 1 ) . Other phages with predicted hosts lacking the C-terminal motif in bS21 encode S21 homologs that also lack the motif ( Supplementary file 1 ) . Although the C-terminal region of bS21 resides near the Shine-Dalgarno helix , the examination of ribosome binding site consensus sequences in these bacterial clades and the predicted ribosome binding sites in the associated phages do not reveal obvious similarities ( Supplementary file 1 ) . Ribosomal protein uS11 is a central component of the 30S subunit platform domain and assembles cooperatively with ribosomal proteins uS6 and uS18 ( Stern et al . , 1988 ) , preceding the binding of bS21 late in 30S subunit maturation ( Held et al . , 1973; Sashital et al . , 2014 ) . Protein uS11 makes intimate contact with 16S rRNA residues in a 3-helix junction that forms part of the 30S subunit E site and stabilizes the 16S rRNA that forms the platform component of the P site ( Stern et al . , 1988 ) . Remarkably , an inspection of the 30S subunit map uncovered a previously unmodeled isoaspartyl residue in protein uS11 , at the encoded residue N119 ( Figure 4; see Materials and methods ) , marking the first identified protein backbone modification in the ribosome . While it has been known that this modification can exist in uS11 ( David et al . , 1999 ) , its functional significance has remained unclear , and prior structures did not have the resolution to pinpoint its exact location . Conversion of asparagine to isoaspartate inserts an additional methylene group into the backbone ( the Cβ position ) and generates a methylcarboxylate side chain ( Reissner and Aswad , 2003 ) . In the present map , the shapes of the backbone density and proximal residues in the chain reveal the presence of the additional methylene , allowing it the flexibility to pack closely with the contacting rRNA nucleotides ( Figure 4A and B ) . A 30S-focused reconstruction using only early movie frames , in which damage to carboxylates would not be as severe ( Marques et al . , 2019 ) , shows improved density for the IAS sidechain , albeit at slightly lower resolution overall ( Figure 4—figure supplement 1 ) . Investigation of residues flanking the isoaspartate in uS11 reveals near-universal conservation in bacteria , chloroplasts , and mitochondria ( Figure 4C , Figure 4—figure supplement 2 ) , suggesting that the isoaspartate may contribute to 30S subunit assembly or stability . Consistent with this idea , the isoaspartate allows high shape complementarity including van der Waals contacts and hydrogen bonds between this region of uS11 and the 16S rRNA it contacts , which involves three consecutive purine-purine base pairs in bacteria ( Supplementary file 2 ) , and a change in rRNA helical direction that is capped by stacking of histidine 118 in uS11 on a conserved purine ( A718 in E . coli; Figure 4B , Supplementary file 2 ) . Strikingly , the sequence motif in bacterial uS11 is also conserved in a domain-specific manner in archaea and eukaryotes ( Figure 4C , Figure 4—figure supplement 1 ) , as are the rRNA residues near the predicted isoAsp ( Supplementary file 2 ) . Remodeling the isoAsp motifs in maps from recently-published cryo-EM reconstructions of an archaeal 30S ribosomal subunit complex at 2 . 8 Å resolution ( Nürenberg-Goloub et al . , 2020 ) and a yeast 80S ribosome complex at 2 . 6 Å resolution ( Tesina et al . , 2020 ) shows that the isoaspartate also seems to be present in these organisms based on the residue-level correlation between map and model ( Figure 4—figure supplement 2 ) . Taken together , the phylogenetic data and structural data indicate that the isoaspartate in uS11 is nearly universally conserved , highlighting its likely important role in ribosome assembly and function . The E . coli 30S ribosomal subunit has eleven post-transcriptionally modified nucleotides in 16S rRNA , all of which can be seen in the present maps or inferred from hydrogen bonding patterns in the cases of many pseudouridines . Interestingly , two methylated nucleotides–m7G527 and m62A1519–appear not to be fully modified , based on the density at ~2 . 1 Å resolution ( Figure 1—figure supplement 5 ) . In the map , m7G527 appears partially methylated , and m62A1519 lacks one of the two methyl groups . Loss of methylation at m7G527 , which is located near the mRNA decoding site , has been shown to confer low-level streptomycin resistance ( Okamoto et al . , 2007 ) , and possibly neomycin resistance in some cases ( Mikheil et al . , 2012 ) . The position of the methyl group is located in a pocket formed with ribosomal protein uS12 , adjacent to the post-translationally modified Asp89 , β-methylthio-Asp ( Anton et al . , 2008; Kowalak and Walsh , 1996 ) . The β-methylthio-Asp also has weak density for the β-methylthio group suggesting it is also hypomodified in the present structure ( Figure 1—figure supplement 6 ) . Notably , loss of m7G527 methylation is synergistic with mutations in uS12 that lead to high-level streptomycin resistance ( Benítez-Páez et al . , 2014; Okamoto et al . , 2007 ) . Loss of m7G527 methylation would remove a positive charge and open a cavity adjacent to uS12 , which may contribute to resistance by shifting the equilibrium of 30S subunit conformational states to an ‘open’ form that is thought to be hyperaccurate with respect to mRNA decoding ( Loveland et al . , 2020; Ogle et al . , 2002; Zaher and Green , 2010 ) . Within the 30S subunit platform near the P site , the two dimethylated adenosines–m62A1518 and m62A1519–have also been connected to antibiotic resistance . Although impacting the assembly of the 30S subunit ( Connolly et al . , 2008 ) and ribosome function ( Sharma and Anand , 2019 ) , loss of methylation of these nucleotides also leads to kasugamycin resistance ( Ochi et al . , 2009 ) . By contrast , bacteria lacking KsgA , the methyltransferase responsible for dimethylation of both nucleotides , become highly susceptible to other antibiotics including aminoglycosides and macrolides ( O’Farrell and Rife , 2012; Phunpruch et al . , 2013; Zou et al . , 2018 ) . In the present structure , m62A1519 is singly-methylated whereas m62A1518 is fully methylated ( Figure 1—figure supplement 5 ) . KsgA fully methylates both nucleotides in in vitro biochemical conditions ( O'Farrell et al . , 2012 ) , but the methylation status of fully-assembled 30S subunits in vivo has not been determined . The loss of a single methylation of m62A1519 , observed here for the first time , could be a mechanism for conferring low-level antibiotic resistance to some antibiotics without appreciably affecting assembly of the 30S subunit or leading to sensitivity to other classes of antibiotics , a hypothesis that could be tested in the future . Aminoglycoside antibiotics ( AGAs ) are a widely-used class of drugs targeting the mRNA decoding site ( A site ) of the ribosome , making them an important focus for continued development against antibiotic resistance ( Sati et al . , 2019 ) . Paromomycin , a 4 , 6-disubstituted 2-deoxystreptamine AGA ( Figure 5A ) , is one of the best-studied structurally . Structures include paromomycin bound to an oligoribonucleotide analog of the A site at 2 . 5 Å resolution ( Vicens and Westhof , 2001 ) , to the small subunit of the T . thermophilus ribosome at 2 . 5 Å resolution ( Kurata et al . , 2008 ) , and to the full 70S T . thermophilus ribosome at 2 . 8 Å resolution ( Selmer et al . , 2006 ) . While the overall conformation of rings I–III of paromomycin is modeled in largely the same way ( Figure 5B ) , the resolution of previous structures did not allow for unambiguous interpretation of ring IV , and only the oligoribonucleotide structural model of the decoding site includes some water molecules in the drug’s vicinity ( Vicens and Westhof , 2001 ) . Although ring IV remains the least ordered of the four rings in the present structure , the cryo-EM map of the focus-refined 30S subunit allows high-resolution modeling of the entire molecule and the surrounding solvation for the first time ( Figure 5C and D ) . The conformation of paromomycin in the oligonucleotide structure ( Vicens and Westhof , 2001 ) agrees most closely with the current structure , with ring IV adopting a chair conformation with the same axial and equatorial positioning of exocyclic functional groups . However , the N6’’’ group in the present structure points in the opposite direction and forms multiple contacts with the backbone phosphate groups of G1489 and U1490 ( Figure 5D ) . Tilting of ring IV in the present model also positions N2’’’ and O3’’’ to make contacts with the G1405 and A1406 phosphate groups , respectively . The paromomycin models in the previous structures of the 30S subunit and 70S ribosome differ further by modeling ring IV in the alternative chair conformation , which also breaks the contacts observed here . We do not see paromomycin bound in H69 of the 50S ribosomal subunit , a second known AGA binding site , consistent with prior work indicating that binding of aminoglycosides to H69 may be favored in intermediate states of ribosomal subunit rotation ( Wang et al . , 2012; Wasserman et al . , 2015 ) . The ribosome undergoes large conformational changes within and between the ribosomal subunits during translation , necessitating a complex set of interactions that maintain ribosome function . Contacts at the periphery of the subunit interface have been less resolved in many structures , likely due to motions within the ribosome populations . Additionally , some key regions involved in these contacts are too conformationally flexible to resolve in structures of the isolated subunits . In the present structure of the unrotated state of the ribosome , with tRNAs positioned in the A site and P site , improvement of maps of the individual ribosomal subunits and smaller domains within the subunits help to define these contacts more clearly . Helix H69 of the 50S subunit , which is mostly disordered in the isolated subunit , becomes better defined once the intact ribosome is formed . The 23S rRNA stem-loop closed by H69 is intimately connected to the 30S subunit at the end of 16S rRNA helix h44 near the mRNA decoding site and tRNA binding sites in the ribosome . During mRNA decoding , the RNA loop closing H69 rearranges to form specific interactions with the A-site tRNA ( Selmer et al . , 2006 ) . The stem of H69 also compresses as the 30S subunit rotates during mRNA and tRNA translocation , thereby maintaining contacts between the 30S and 50S subunits ( Dunkle et al . , 2011 ) . In the present reconstructions , helix H69 seems conformationally more aligned to the 30S subunit than the 50S subunit , as the cryo-EM density for H69 is much better defined in the map of the 30S subunit compared to the map of the 50S subunit . The loop comprising 23S rRNA nucleotides G713-A718 closing helix H34 forms an additional bridge between the 50S and 30S subunits and is also known to be dynamic in its position ( Dunkle et al . , 2011 ) . In the present reconstructions , this bridge is also better defined in the map of the 30S subunit compared to the 50S subunit , placing the more highly conserved arginine Arg88 in uS15 in direct contact with the RNA backbone of the H34 stem-loop , rather than the less conserved Arg89 ( Figure 6A ) . An additional conformationally dynamic contact between the 30S and 50S subunits involves the A-site finger ( ASF , helix H38 in 23S rRNA ) , which is known to modulate mRNA and tRNA translocation ( Komoda et al . , 2006 ) . In the present model , although the central helical region of the ASF is only visible in low-resolution maps , loop nucleotide C888 which stacks on uS13 residues Met81 and Arg82 in the 30S subunit head domain is clearly defined ( Figure 6B ) . Maps of the 30S subunit head domain and central protuberance of the 50S subunit also reveal clearer density defining the unrotated-state contacts between uS13 in the small subunit , uL5 in the large subunit , and bL31 ( bL31A in the present structure ) which spans the two ribosomal subunits . The core of the 50S subunit is the most rigid part of the ribosome , which has enabled it to be modeled to a higher resolution than the 30S subunit , historically and in the present structure . In the present 70S ribosome and 50S subunit reconstructions , which have global map-to-model resolutions of 2 . 04 Å and 1 . 90 Å , respectively , the resolution of the core of the 50S subunit reaches 1 . 8 Å ( Figure 1 , Figure 1—figure supplement 3A , Table 2 ) , revealing unprecedented structural details of 23S rRNA , ribosomal proteins , ions , and solvation ( Figure 1—figure supplement 4 , Figure 1—figure supplement 5 , Figure 7—figure supplement 2 ) . The resulting maps are also superior in the level of detail when compared to maps previously obtained by X-ray crystallography ( Noeske et al . , 2015; Polikanov et al . , 2015 ) , which aided in improving models of high-resolution chemical features like backbone dihedrals in much of the rRNA , non-canonical base pairs and triples , arginine side-chain rotamers , and glycines in conformationally constrained RNA-protein contacts . The density also enabled modeling of thousands of water molecules , dozens of magnesium ions , and polyamines ( Figure 1—figure supplement 4 , Figure 7—figure supplement 2 , Table 3 ) The present model now even allows for comparison of ribosome phylogenetic conservation to the level of solvent positioning . For example , water molecules and ions with conserved positions in the peptidyl transferase center ( PTC ) can be seen in comparisons of different bacterial and archaeal ribosome structures , even when solvation was not included in the deposited models ( Halfon et al . , 2019; Polikanov et al . , 2015; Schmeing et al . , 2005; Figure 7 ) . The central protuberance ( CP ) of the 50S subunit , which contacts the P-site tRNA and the head of the 30S subunit , is dynamic , however , is well-resolved with focused refinement , here reaching a resolution of 2 . 13 Å by GS-FSC and 2 . 26 Å in map-to-model FSC comparisons ( Figure 1—figure supplement 3 ) . The improved resolution of the CP aided in modeling ribosomal proteins uL5 and bL31A , as well as the CP contact to P-site tRNA . Despite the overall high resolution of the 50S subunit core , there are a number of regions where the RNA backbone density shows relatively poor connectivity ( Figure 1—figure supplement 7 ) . In some cases , linkages between the ribose and phosphate groups become weakly visible or broken , with the phosphate group having an overall more rounded appearance than in cases where density is strong throughout the backbone . In some cases , breaks in the ribose ring are observed . Our initial impression was that this may be indicative of damage from the electron beam . However , reconstructions using the first two or three frames ( corresponding to the first ~2–3 e-/Å2 in the exposure ) show similar patterns of weak or broken density in these regions ( Figure 1—figure supplement 7 ) . This suggests that RNA conformational flexibility rather than radiation damage may be responsible for the broken density . Details of post-transcriptional and post-translational modifications are also clear in the 50S subunit maps ( Figure 1—figure supplement 5 ) . Surprisingly , the post-translationally modified β-hydroxyarginine at position 81 in uL16 ( Ge et al . , 2012 ) is followed by unexplained density consistent with a thiopeptide bond between Met82 and Gly83 ( Figure 8A ) . Adjusting the contour level of the map shows map density for the modified atom similar to that of the sulfur in the adjacent methionine and nearby phosphorus atoms in the RNA backbone , in contrast to neighboring peptide oxygen atoms . In the other cryo-EM maps of the 50S subunit ( Pichkur et al . , 2020; Stojković et al . , 2020 ) , the density for the sulfur in the thioamide is not visible or barely visible ( Figure 8—figure supplement 1 ) . Notably , the mass for E . coli uL16 has been shown to be 15328 . 1 Da and drops to 15312 . 1 Da with loss of Arg81 hydroxylation ( Ge et al . , 2012 ) . However , this mass is still +30 . 9 more than the encoded sequence ( 15281 . 2 Da , Uniprot P0ADY7 ) . In E . coli uL16 is also N-terminally methylated ( Brosius and Chen , 1976 ) , leaving 16 mass units unaccounted for , consistent with the thiopeptide we observe in the cryo-EM map . We examined a high-resolution mass spectrometry bottom-up proteomics dataset ( Dai et al . , 2017 ) to find additional evidence supporting the interpretation of the cryo-EM map as a thiopeptide . Several uL16 peptides were found across multiple experiments that matched the expected mass shift closer to that of a thiopeptide’s O to S conversion ( +15 . 9772 Da ) rather than oxidation ( +15 . 9949 ) , a common modification with a similar mass shift ( Figure 8B ) . Fragmentation spectra localized the mass shift near the Met82-Gly83 bond , further supporting the presence of a thiopeptide ( Figure 8C ) . Taken together , the cryo-EM map of the 50S subunit and mass spectrometry data support the model of a thiopeptide between Met82 and Gly83 in E . coli uL16 . The enzymes that might be responsible for the insertion of the thioamide in uL16 remain to be identified . E . coli encodes the prototypical YcaO enzyme , which can form thiopeptides but for which no substrate is known ( Burkhart et al . , 2017 ) . A phylogenetic tree of YcaO family members shows a clear break separating YcaO proteins associated with secondary metabolism into a major branch ( Figure 8—figure supplement 2A ) . A sub-grouping in the other major branch includes YcaO family members within Gammaproteobacteria ( Figure 8—figure supplement 2A ) . The examination of genes in close proximity to YcaO across Gammaproteobacteria reveals three genes that form the focA-pfl operon involved in the anaerobic metabolism of E . coli ( Figure 8—figure supplement 2B; Sawers and Suppmann , 1992 ) . The combination of its unknown substrate in E . coli , the ability to catalyze thioamidiation in other species , and syntenic conservation in Gammaproteobacteria identify YcaO as a primary candidate for uL16 thioamidation . High-resolution cryo-EM maps are now on the cusp of matching or exceeding the quality of those generated by X-ray crystallography , opening the door to a deeper understanding of the chemistry governing structure-function relationships and uncovering new biological phenomena . Questions about the ribosome , which is composed of the two most abundant classes of biological macromolecules and essential for life , reach across a diverse range of inquiry . Structural information about ribosomal components can have implications ranging from fundamental chemistry to mechanisms underlying translation and evolutionary trends across domains of life . For example , in our cryo-EM reconstructions , we observed a surprising level of detail about modifications to nucleobases and proteins that could not be seen in prior X-ray crystallographic structures . The most unexpected of these is the presence of two previously unknown post-translational modifications in the backbones of ribosomal proteins , which would be otherwise difficult to confirm without highly targeted analytical chemical approaches . Precise information about the binding of antibiotics , protein-RNA contacts , and solvation are additional examples of what can be interrogated at this resolution . Beyond purely structural insights , these findings generate new questions about protein synthesis , ribosome assembly , and antibiotic action and resistance mechanisms , providing a foundation for future experiments . The remarkable finding of a thioamide modification in protein uL16 , only the second such example in a protein ( Mahanta et al . , 2019 ) , is a perfect example of the power of working at <2 Å resolution . The difference in bond length of a thiocarbonyl compared to a typical peptide carbonyl is ~0 . 4 Å , with otherwise unchanged geometry , and is too subtle to identify at a lower resolution ( Figure 8—figure supplement 1 ) . Moreover , the sulfur density is not as pronounced in maps at lower resolution . Analysis of previously published mass spectrometry data with sufficient mass accuracy to differentiate O to S modifications from a more common +O oxidation event ( Dai et al . , 2017; Figure 8 ) corroborates the finding . The possible role for the thiopeptide linkage in the E . coli ribosome , which is located near the PTC and involves contact between the thiocarbonyl sulfur atom and the hydroxyl group in the β-hydroxyarginine at position 81 in uL16 , remains to be shown . The mechanism by which its formation is catalyzed also remains an open question . One candidate enzyme for this purpose is E . coli protein YcaO , an enzyme known to carry out thioamidation and other amide transformations ( Burkhart et al . , 2017 ) . Although this enzyme has been annotated as possibly participating with RimO in modification of uS12 Asp89 ( Strader et al . , 2011 ) , genetic evidence for a specific YcaO function is lacking . For example , E . coli lacking YcaO are cold-sensitive and have phenotypes most similar in pattern to those observed with a knockout of UspG , universal stress protein 12 ( Nichols et al . , 2011 ) . Furthermore , knockout of YcaO has phenotypes uncorrelated with those of knockout of YcfD , the β-hydroxylase for Arg81 in uL16 adjacent to the thioamide ( Nichols et al . , 2011 ) . YcaO-like genes in Gammaproteobacteria genomes colocalize with the focA-pfl operon , a common set of genes involved in anaerobic and formate metabolism ( Figure 8—figure supplement 1; Sawers and Suppmann , 1992 ) . Since the ribosomes used here were obtained from aerobically grown cultures and the focA-Pfl operon is transcribed independently of the YcaO gene in E . coli ( Sawers , 2005 ) it is likely that the YcaO gene and the focA-Pfl operon encode proteins with unrelated functions . Interestingly , the clear phylogenetic separation between the YcaO gene in Gammaproteobacteria and the YcaO genes known to be involved in secondary metabolism in the phylogenetic tree suggests that , if YcaO is responsible for uL16 thioamidation , this modification may only be conserved in Gammaproteobacteria . The maps of the 30S subunit , resolved to a slightly lower resolution of ~2 . 0–2 . 1 Å ( Figure 1—figure supplement 3 , Table 2 ) , enabled the identification of the only known isopeptide bond in a ribosomal protein , an isoAsp at position 119 in uS11 . While isoaspartyl residues have been hypothesized to mainly be a form of protein damage requiring repair , previous work identified the existence of isoAsp in uS11 at near stoichiometric levels , suggesting it might be functionally important ( David et al . , 1999 ) . Certain hotspots in protein sequences are known to be especially prone to isoaspartate formation ( Reissner and Aswad , 2003 ) , including Asn-Gly , as encoded in nearly all bacterial uS11 sequences ( Figure 4C ) . However , the half-life of the rearrangement is on the timescale of days ( Robinson and Robinson , 2001; Stephenson and Clarke , 1989 ) . In archaea and eukaryotes , the formation of isoaspartate at this position would require dehydration of the encoded aspartate , which occurs even more slowly than deamidation of asparagine ( Stephenson and Clarke , 1989 ) . Importantly , the residue following the aspartate is nearly always serine in eukaryotes and is enriched for glycine , serine , and threonine in archaea ( Figure 4C , Figure 4—figure supplement 1 ) , consistent with the higher rates of dehydration that occur when aspartate is followed by glycine and serine in peptide models ( Stephenson and Clarke , 1989 ) . These results suggest that the isoAsp modification may be nearly universally conserved in all domains of life . Concordant with this hypothesis , isoAsp modeling provides a better fit to cryo-EM maps of uS11 in archaeal and eukaryotic ribosomes ( Figure 4—figure supplement 2 ) . Although it is possible that isoaspartate formation could be accelerated in specific structural contexts ( Reissner and Aswad , 2003 ) , it is not clear if the isoAsp modification in uS11 occurs spontaneously or requires an enzyme to catalyze the reaction . O-methyl-transferase enzymes have been identified that install a β-peptide in a lanthipeptide ( Acedo et al . , 2019 ) or serve a quality control function to remove spontaneously formed isoaspartates ( David et al . , 1999 ) . Deamidases that catalyze isoAsp formation from asparagine are not well described in the literature , although examples have been identified in viral pathogens , possibly repurposing host glutamine amidotransferases ( Zhao et al . , 2016 ) . Future work will be needed to identify the mechanisms by which the isoAsp in uS11 is generated in cells . Its biological significance , whether in the assembly of the small ribosomal subunit or other steps in translation , also remains to be defined . The resolution achieved here also has great potential for better informing structure-activity relationships in future antibiotic research , particularly because the ribosome is so commonly targeted . For example , we were able to identify hypomodified bases in 16S rRNA ( m7G527 and m62A1519 ) and possible hypomodification of Asp89 ( β-methylthio-Asp ) in uS12 ( Figure 1—figure supplement 5 , Figure 1—figure supplement 6 ) . These hypomodifications could confer resistance to kasugamycin and streptomycin antibiotics in some cases . Furthermore , we were also able to see more clearly the predominant position of paromomycin ring IV in the decoding site of the 30S subunit ( Figure 5 ) . The proposed primary role of ring IV has been to increase the positive charge of the drug to promote binding ( Hobbie et al . , 2006 ) , in line with its ambiguous modeling in previous structures ( Kurata et al . , 2008; Selmer et al . , 2006; Vicens and Westhof , 2001 ) . While ring IV’s features in the current map are weaker relative to those of rings I–III , we were able to identify interactions of ring IV with surrounding 16S rRNA nucleotides and ordered solvent molecules that were not previously modeled . Importantly , the observed interactions between the N6’’’ amino group and the phosphate backbone of nucleotides G1489-U1490 , in particular , are likely responsible for known susceptibility of PAR to N6’’’ modification ( Sati et al . , 2017 ) . While the same loss of interactions is expected for neomycin , which differs from paromomycin only by the presence of a 6’-hydroxy rather than a 6’-amine in ring I , the penalty of modifying N6’’’ in neomycin is likely compensated for by the extra positive charge and stronger hydrogen bonding observed with neomycin ring I ( Sati et al . , 2017 ) . The level of detail into modes of aminoglycoside binding that can now be obtained using cryo-EM thus should aid the use of chemical biology to advance AGA development . The cryo-EM maps of the 30S subunit also revealed new structural information about protein bS21 at a lower resolution , particularly at its C terminus . The location of bS21 near the ribosome binding site suggests it may play a role in translation initiation . The conservation of the RLY ( or KLY ) motif and its contacts to the 30S subunit head domain also suggests bS21 may have a role in modulating conformational dynamics of the head domain relative to the body and platform of the 30S subunit . Rearrangements of the 30S subunit head domain are seen in every stage of the translation cycle ( Javed and Orlova , 2019 ) . Although we could align putative S21 homologs from huge phages ( Al-Shayeb et al . , 2020 ) with specific bacterial clades , and show that many also possess KLY-like motifs , there were no clear relationships between the predicted consensus ribosome binding sites in these bacteria and these phages . It is possible that bS21 and the phage homologs interact with nearby mRNA sequences 5’ of the Shine-Dalgarno helix , affecting translation initiation in this way . Taken together , the structural and phylogenetic information on bS21 and the phage S21 homologs raise new questions about their role in translation and the phage life cycle , that is , whether they contribute to specialized translation and/or help phage evade bacterial defenses . The rotameric nature of nucleic acid backbones has historically been a challenge for modeling the sugar-phosphate conformation , in contrast to the generally well-ordered bases ( Murray et al . , 2003 ) . Ribose puckers , for example , are directly visualized only at better than ~2 Å resolution but significantly affect the remaining backbone dihedrals ( Richardson et al . , 2018 ) . Much work has been done to simplify the multidimensional problem of modeling RNA conformers given the scarcity of high-resolution RNA structures ( RNA Ontology Consortium et al . , 2008 ) . While some areas of the present structure show backbone details very clearly ( Figure 1C ) , some level of disorder is observed in the conformations of many other residues ( Figure 1—figure supplement 7 ) . Our initial impression was that this might be due to radiation damage . However , reconstructions with the first 2–3 frames of the exposure reveal similar breaks , and sometimes new breaks , in the EM density . Previous work has shown that at this dose , amino acid residues well known to be highly susceptible to radiation damage should be better preserved ( Hattne et al . , 2018 ) . Furthermore , it is known that nucleic acids tend to be more resilient to X-ray damage compared to the most beam-sensitive moieties in proteins ( Bury et al . , 2016 ) . Because the same general trends in specific radiation damage seem to hold for cryo-EM ( Hattne et al . , 2018 ) , and noting that the global resolutions of the low-dose 70S reconstructions remain resolved to ~2 . 1–2 . 2 Å , the persistence of the broken density at low doses is more likely to be a result of structural disorder in the backbone . Our observation that poorer connectivity in the RNA backbone seems to be more common in regions lacking close contacts to other regions of the structure tracks with this conclusion , while a minority of cases where only a single bond in the ribose appears to be broken are more of a puzzle . Closer investigation of these features may reveal more quantitative information about nucleotide rotamer preferences . With regard to resolution , in this work , we have supplemented the reporting of the ‘gold-standard’ FSC with map-to-model FSC curves for our maps and for comparisons to previous work . Although the map-to-model FSC metric has been described for some time , it is not routinely used in the ribosome field ( Halfon et al . , 2019; Loveland et al . , 2020; Nürenberg-Goloub et al . , 2020; Pichkur et al . , 2020; Stojković et al . , 2020; Tesina et al . , 2020 ) . Acknowledging that there is no substitute for visual inspection of the map to determine its quality , it is necessary to also consider which metrics are useful on the scale of the questions being answered . Sub-Ångstrom differences in resolution as reported by half-map FSCs have a significant bearing on chemical interactions at face value but may lack usefulness if map correlation with the final atomic model is not to a similar resolution . For example , maps from recent cryo-EM reconstructions of the bacterial 50S ribosomal subunit report resolutions of ~2 . 1 Å–2 . 3 Å but the deposited models reach global resolutions of ~2 . 3 Å–2 . 5 Å by the map-to-model FSC criterion ( Halfon et al . , 2019; Pichkur et al . , 2020; Stojković et al . , 2020; see Materials and methods; Figure 7—figure supplement 1 ) . Notably , for the 2 . 1 Å map of the E . coli 50S subunit based on half-map FSC values ( Pichkur et al . , 2020 ) , the map-to-model FSC fit of our 50S subunit model to that map has a higher resolution ( 2 . 07 Å ) , compared to the deposited model ( 2 . 29 Å , PDB entry 6xz7; Figure 7—figure supplement 1 ) . Thus , although the half-map FSC tells us something about the best model one might achieve , the map-to-model FSC captures new information that lies in how the model was generated and refined . Additionally , while the map-to-model FSC calculations carry intrinsic bias from the model's dependence on the map , model refinement procedures leverage well-defined chemical properties ( i . e . bond lengths , angles , dihedrals , and steric restraints ) that are entirely independent of the map and should ensure realism . This is perhaps a reason why the map-to-model FSC appears in recent work more focused on methods and tool development ( Nakane et al . , 2020; Terwilliger et al . , 2020a , Terwilliger et al . , 2020b ) . Aside from global high resolution , the conformational heterogeneity of the ribosome also calls attention to the tools used for working with complexes that display variable resolution . Methods for refining heterogeneous maps have proliferated , including multi-body refinement ( Nakane et al . , 2018 ) and 3D variability analysis ( Punjani and Fleet , 2020 ) among others , but ways to work with and create a model from many maps of the same complex have not yet been standardized and require substantial manual intervention . For example , we built and refined regions of the ribosome separately into focus-refined maps , using real-space refinement in Chimera ( Pettersen et al . , 2004 ) and Coot ( Casañal et al . , 2020 ) to ‘repair’ the breakpoints between model segments . The creation of composite maps from multiple refinements also suffers from imperfect stitching between refinements of distinct domains , and in our composite map , we observe that the highest resolution components degrade somewhat in the process . We also note that B-factor refinement in phenix . real_space_refinement is still under development , for example only allowing grouped B-factor refinement for nucleotides and amino acids . B-factor refinement in phenix . real_space_refine also results in unrealistic values for parts of the model , for example by exhibiting coupling to the global B factor applied to the map used in the refinement . Other strategies for deriving an analog to the B factor suitable for cryo-EM are still in development ( Zhang et al . , 2020 ) . Moreover , as it tends to be most convenient to develop tools with the highest resolution possible , it is common for new methods to utilize very high-resolution maps of apoferritin as a standard , which is highly symmetric and well-ordered . Specimens that do not share these characteristics may require new tools that move beyond these assumptions . Modeling tools for RNA also generally lag those for proteins . Here we were able to use the maps and the highly solvated nature of RNA secondary and tertiary structure to address the parameterization of solvent modeling in PHENIX ( phenix . douse ) ( Liebschner et al . , 2019 ) . For the present 70S map , the half-map FSC ≥0 . 97 up to 3 . 3 Å resolution , which is estimated to represent a theoretical correlation with a ‘perfect’ map up to 0 . 99 ( Terwilliger et al . , 2020a ) . Structural information with certainty up to so-called ‘near-atomic’ resolution has potential use in benchmarking newer tools and may specifically make our results valuable in addressing issues with focused or multi-body refinement . This structure also has potential use for aiding the future development of de novo RNA modeling tools , which are historically less developed compared to similar tools for proteins , and often rely on information generated from lower-resolution RNA structures ( Watkins and Das , 2019 ) . Finally , our micrographs uploaded to the Electron Microscopy Public Image Archive ( EMPIAR ) ( Iudin et al . , 2016 ) should serve as a resource for ribosome structural biologists and the wider cryo-EM community to build on the present results . E . coli 70S ribosome purification ( Travin et al . , 2019 ) and tRNA synthesis , purification , and charging ( Ad et al . , 2019 ) were performed as previously described . Briefly , 70S ribosomes were purified from E . coli MRE600 cells using sucrose gradients to isolate 30S and 50S ribosomal subunits , followed by subunit reassociation and a second round of sucrose gradient purification . Transfer RNAs were transcribed from PCR DNA templates using T7 RNA polymerase and purified by phenol-chloroform extraction , ethanol precipitation , and column desalting . Flexizyme ribozymes were used to charge the P-site tRNAfMet with either pentafluorobenzoic acid or malonate methyl ester and the A-site tRNAVal with valine ( Goto et al . , 2011 ) . Ribosome-mRNA-tRNA complexes were formed non-enzymatically by incubating 10 µM P-site tRNA , 10 µM mRNA , and 100 µM paromomycin with 1 µM ribosomes for 15 min at 37°C in buffer AC ( 20 mM Tris pH 7 . 5 , 100 mM NH4Cl , 15 MgCl2 , 0 . 5 mM EDTA , 2 mM DTT , 2 mM spermidine , 0 . 05 mM spermine ) . Then , 10 µM A-site tRNA was added and the sample was incubated for an additional 15 min at 37°C . Complexes were held at 4°C and diluted to 100 nM ribosome concentration in the same buffer immediately before grid preparation . The mRNA of sequence 5’-GUAUAAGGAGGUAAAAAUGGUAUAACUA-3’ was chemically synthesized ( IDT ) and was resuspended in water without further purification . The Shine-Dalgarno sequence is shown in bold , and the Met-Val codons are in italics . 300 mesh R1 . 2/1 . 3 UltraAuFoil grids from Quantifoil with an additional amorphous carbon support layer were glow discharged in a Pelco sputter coater . About 4 µL of each sample was deposited onto grids and incubated for 1 min , then washed in buffer AC with 20 mM NH4Cl rather than 100 mM NH4Cl . Grids were plunge-frozen in liquid ethane using a Vitrobot Mark IV with settings: 4°C , 100% humidity , blot force 6 , blot time 3 . Movies were collected on a 300-kV Titan Krios microscope with a GIF energy filter and Gatan K3 camera . Super-resolution pixel size was 0 . 355 Å , for a physical pixel size of 0 . 71 Å . SerialEM ( Schorb et al . , 2019 ) was used to correct astigmatism , perform coma-free alignment , and automate data collection . Movies were collected with the defocus range −0 . 6 to −1 . 5 µm and the total dose was 39 . 89 e-/Å2 split over 40 frames . One movie was collected for each hole , with image shift used to collect a series of 3 × 3 holes for faster data collection ( Cheng et al . , 2018 ) , and stage shift used to move to the center hole . Based on the 1 . 2/1 . 3 grid hole specification , this should correspond to a maximum image shift of ~1 . 8 μm , although the true image shift used was not measured . The beam size was chosen such that its diameter was slightly larger than that of the hole , that is , >1 . 2 μm , although we have observed variation in the actual hole size compared to the manufacturer specifications . Datasets of 70S ribosome complexes with the two differently charged P-site tRNAs were initially processed separately . Movies were motion-corrected with dose weighting and binned to the recorded physical pixel size ( 0 . 71 Å ) within RELION 3 . 0 ( Scheres , 2012 ) using MotionCor2 ( Zheng et al . , 2017 ) . CTF estimation was done with CTFFind4 ( Rohou and Grigorieff , 2015 ) , and micrographs with poor CTF fit as determined by visual inspection were rejected . Particles were auto-picked with RELION’s Laplacian-of-Gaussian method . The 2D classification of particles was performed in RELION , and 4× binned particles were used for all classification steps . Particles were separated into 3D classes in cryoSPARC heterogeneous refinement ( Punjani et al . , 2017 ) , using an initial model generated from PDB 1VY4 with A-site and P-site tRNAs ( Polikanov et al . , 2014 ) low-pass filtered to the default 20 Å resolution , and keeping particles that were sorted into well-resolved 70S ribosome classes . Particles were migrated back to RELION to generate an initial 3D-refined volume ( reference low-pass filtered to 60 Å ) on which to perform masked 3D classification without alignment to further sort particles based on A-site tRNA occupancy . CTF Refinement and Bayesian polishing were performed in RELION 3 . 1 before pooling the two datasets together , with nine optics groups defined based on the 3 × 3 groups for image shift-based data collection . The resulting 70S ribosome reconstruction was used as input for focused refinements of the 50S and 30S subunits . We used rigid-body docked coordinates for the 70S ribosome , individual ribosomal subunits or domains ( 30S subunit head , 50S subunit central protuberance ) to define the boundaries of the map regions to be used in the focused refinements . Focused refinement of the central protuberance was performed starting from the 50S subunit-focused refinement reconstruction , and head- and platform-focused refinements started from the 30S subunit focused refinement reconstruction . Ewald sphere correction , as implemented in RELION 3 . 1 with the single side-band correction ( Russo and Henderson , 2018; Zivanov et al . , 2018 ) , provided some additional improvements in resolution ( Tables 1–2 ) . In addition to using the 40-frame movies , we used the first three frames corresponding to a ~3 electron/Å2 dose to calculate 3D reconstructions , including focused refinements of the 30S and 50S subunits . The focused-refined map of the 30S subunit had a resolution of 2 . 45 Å by the map-to-model FSC metric . These maps were used to examine the density for the isoAsp in uS11 , which lacked clear density for the side chain in maps reconstructed from the full 40-frame movies . We also used the maps from the initial three frames to examine connectivity in ribose density , to determine if there is visual evidence for the impact of electron damage . The previous high-resolution structure of the E . coli 70S ribosome ( Noeske et al . , 2015 ) was used as a starting model . We used the ‘Fit to Map’ function in Chimera ( Pettersen et al . , 2004 ) to calibrate the magnification of the cryo-EM map of the 50S ribosomal subunit generated here to maximize correlation , resulting in a pixel size of 0 . 7118 Å rather than the recorded 0 . 71 Å . Focused-refined maps were transformed into the frame of reference of the 70S ribosome for modeling and refinement , using the ‘Fit to Map’ function in Chimera , and resampling the maps on the 70S ribosome grid . The 50S and 30S subunits were refined separately into their respective focused-refined maps using PHENIX real-space refinement ( RSR; Liebschner et al . , 2019 ) . Protein and rRNA chains were visually inspected in Coot ( Casañal et al . , 2020 ) and manually adjusted where residues did not fit well into the density , making use of B-factor blurred maps where needed to interpret regions of lower resolution . Focused-refined maps on smaller regions were used to make further manual adjustments to the model , alternating with PHENIX RSR . Some parts of the 50S subunit , including H69 , H34 , and the tip of the A-site finger , were modeled based on the 30S subunit focused-refined map . The A-site and P-site tRNAs were modeled as follows: anticodon stem-loops , 30S subunit focused-refined map; P-site tRNA body , 50S subunit focused-refined map , with a B factor of 20 Å2 applied; A-site tRNA body , 30S subunit focused-refined map and 50S subunit focused-refined map with B factors of 20 Å2 applied; tRNA-ACCA 3’ ends , 50S subunit focused-refined map with B factors of 20–30 Å2 applied . Alignments of uS15 were generated using BLAST ( Altschul et al . , 1997 ) with the E . coli sequence as reference . The model for bL31A ( E . coli gene rpmE ) was manually built into the CP and 30S subunit head domain focused-refined maps before refinement in PHENIX . A model for paromomycin was manually docked into the 30S subunit focused-refined map , followed by real-space refinement in Coot and PHENIX . Comparisons to prior paromomycin structural models ( PDB codes 1J7T , 2VQE , and 4V51; Kurata et al . , 2008; Selmer et al . , 2006; Vicens and Westhof , 2001 ) used least-squares superposition of paromomycin in Coot . Although ring IV is in different conformations in the various paromomycin models , the least-squares superposition is dominated by rings I–III , which are in nearly identical conformations across models . Ribosome solvation including water molecules , magnesium ions , and polyamines was modeled using a combination of PHENIX ( phenix . douse ) and manual inspection . The phenix . douse feature was run separately on individual focused-refined maps , and the resulting solvent models were combined into the final 30S and 50S subunit models . Due to the fact that the solvent conditions used here contained ammonium ions and no potassium , no effort was made to systematically identify monovalent ion positions . The numbers of various solvent molecules are given in . Along with the individual maps used for model building and refinement , we have also generated a composite map of the 70S ribosome from the focused-refined maps for deposition to the PDB and EMDB for ease of use ( however , experimental maps are recommended for the examination of high-resolution features ) . We made the composite map using the ‘Fit in Map’ and vop commands in Chimera . First , we aligned the unmasked focus-refined maps with the 70S ribosome map using the ‘Fit in Map’ tool . We then used the ‘vop resample’ command to transform these aligned maps to the 70S ribosome grid . After the resampling step , we recorded the map standard deviations as reported in the ‘Volume Mean , SD , RMS’ tool . Then , we added the maps sequentially using ‘vop add’ followed by rescaling the intermediate maps to the starting standard deviation using the ‘vop scale’ command . Initial real-space refinement of the 30S subunit against the focused-refined map using PHENIX resulted in a single chiral volume inversion involving the backbone of N119 in ribosomal protein uS11 , indicating that the L-amino acid was being forced into a D-amino acid chirality , as reported by phenix . real_space_refine . Of the 10 , 564 chiral centers in the 30S subunit model , the Cα of N119 had an energy residual nearly two orders of magnitude larger than the next highest deviation . Inspection of the map in this region revealed clear placement for carbonyl oxygens in the backbone , and extra density consistent with an inserted methylene group , as expected for isoAsp . The model of isoAsp at this position was refined into the cryo-EM map using PHENIX RSR , which resolved the stereochemical problem with the Cα chiral center . IsoAsp was also built and refined into models of archaeal and eukaryotic uS11 based on cryo-EM maps of an archaeal 30S ribosomal subunit complex ( PDB 6TMF; Nürenberg-Goloub et al . , 2020 ) and a yeast 80S ribosome complex ( PDB 6T4Q; Tesina et al . , 2020 ) . These models were refined using PHENIX RSR , and real-space correlations by residue calculated using phenix . model_map_cc . All archaeal genomes were downloaded from the NCBI genome database ( 2618 archaeal genomes , last accessed September 2018 ) . Due to the high number of bacterial genomes available in the NCBI genome database , only one bacterial genome per genus ( 2552 bacterial genomes ) was randomly chosen based on the taxonomy provided by the NCBI ( last accessed in December 2017 ) . The eukaryotic dataset comprises nuclear , mitochondrial , and chloroplast genomes of 10 organisms ( Homo sapiens , Drosophila melanogaster , Saccharomyces cerevisiae , Acanthamoeba castellanii , Arabidopsis thaliana , Chlamydomonas reinhardtii , Phaeodactylum tricornutum , Emiliania huxleyi , Paramecium aurelia , and Naegleria gruberi ) . Genome completeness and contamination were estimated based on the presence of single-copy genes ( SCGs ) as described in Anantharaman et al . , 2016 . Only genomes with completeness >70% and contamination <10% ( based on duplicated copies of the SCGs ) were kept and were further de-replicated using dRep at 95% average nucleotide identity ( version v2 . 0 . 5; Olm et al . , 2017 ) . The most complete genome per cluster was used in downstream analyses . Ribosomal uS11 genes were detected based on matches to the uS11 Pfam domain ( PF00411; Punta et al . , 2012 ) using hmmsearch with an E-value below 0 . 001 ( Eddy , 1998 ) . Amino acid sequences were aligned using the MAFFT software ( version v7 . 453; Katoh and Standley , 2016 ) . The alignment was further trimmed using Trimal ( version 1 . 4 . 22; --gappyout option; Capella-Gutiérrez et al . , 2009 ) . Tree reconstruction was performed using IQ-TREE ( version 1 . 6 . 12; Nguyen et al . , 2015 ) , using ModelFinder ( Kalyaanamoorthy et al . , 2017 ) to select the best model of evolution , and with 1000 ultrafast bootstrap ( Hoang et al . , 2018 ) . The tree was visualized with iTol ( version 4; Letunic and Bork , 2019 ) and logos were made using the weblogo server ( Crooks et al . , 2004 ) . 16S and 18S rRNA genes were identified from the prokaryotic and eukaryotic genomes using the method based on hidden Markov model ( HMM ) searches using the cmsearch program from the Infernal package ( Nawrocki et al . , 2009 ) and fully described in Brown et al . , 2015 . The sequences were aligned using the MAFFT software . S21 sequences were retrieved from the huge phage database described in Al-Shayeb et al . , 2020 . Cd-hit was run on the set of S21 sequences to reduce the redundancies ( Fu et al . , 2012; default parameters; version 4 . 8 . 1 ) . Non redundant sequences were used as a query against the database of prokaryotic genomes used for uS11 above using BLASTP ( version 2 . 10 . 0+; e-value 1e-20; Altschul et al . , 1997 ) . Alignment and tree reconstruction were performed as described for uS11 except that we did not perform the alignment trimming step . Similarly to uS11 , the YcaO sequences were identified in prokaryotic genomes based on its PFAM accession ( PF02624; Punta et al . , 2012 ) using hmmsearch with an E-value below 0 . 001 ( Eddy , 1998 ) . Amino acid sequences were aligned using the MAFFT software ( version v7 . 453; Katoh and Standley , 2016 ) . Alignment was further trimmed using Trimal ( version 1 . 4 . 22; --gappyout option; Capella-Gutiérrez et al . , 2009 ) . Tree reconstruction was performed using IQ-TREE ( version 1 . 6 . 12; Nguyen et al . , 2015 ) , using ModelFinder ( Kalyaanamoorthy et al . , 2017 ) to select the best model of evolution , and with 1000 ultrafast bootstraps ( Hoang et al . , 2018 ) . The tree was visualized with iTol ( version 4; Letunic and Bork , 2019 ) . The three genes downstream and upstream of each YcaO gene were identified and annotated using the PFAM ( Punta et al . , 2012 ) and the Kegg ( Kanehisa et al . , 2016 ) databases . Masks for each map were generated in two ways . First , to calculate the map-to-model FSC curves for comparisons of the present models with the cryo-EM maps generated here , we used masked maps generated by RELION during postprocessing ( Zivanov et al . , 2018 ) . The effective global resolution of a given map is given at the FSC cutoff of 0 . 5 in Table 2 and Figure 1—figure supplements 2–3 . Second , we used refined PDB coordinates for the 70S ribosome , individual ribosomal subunits or domains ( 30S subunit head , 50S subunit central protuberance ) for comparisons to the 70S ribosome map or focused-refined maps , and to previously published maps and structural models . Masks for each map were generated in Chimera ( Pettersen et al . , 2004 ) using the relevant PDB coordinates as follows . A 10 Å resolution map from the coordinates was calculated using molmap , and the surface defined at one standard deviation was used to mask the high-resolution map . For the present models and maps , the effective global resolution of a given map using this second approach was similar or slightly lower than that using the approach in RELION ( within a few hundredths of an Å ) . For the recent E . coli 50S subunit structure ( Stojković et al . , 2020 ) , we used Chimera to resize the deposited map ( emd_20353 ) to match the dimensions of the maps presented here . Briefly , our atomic coordinates for the 50S subunit were used with the ‘Fit to Map’ function and the voxel size of the deposited map was calibrated to maximize correlation . The resulting voxel size changed from 0 . 822 Å to 0 . 8275 Å in linear dimension . After rescaling the deposited map , we used phenix . model_map_cc to compare the map with rescaled atomic coordinates deposited in the PDB ( 6PJ6 ) or to the present 50S model , yielding a map-to-model FSC of 0 . 5 at ~2 . 5 Å . Similar comparisons of the structure of the Staphylococcus aureus 50S subunit to the deposited map ( PDB 6S0Z , emd_10077; Halfon et al . , 2019 ) yielded a map-to-model FSC of 0 . 5 at 2 . 43 Å , accounting for a change in voxel linear dimension from 1 . 067 Å to 1 . 052 Å . For comparisons to the map and model deposited by Pichkur et al . , 2020 ( EMD-10655 and PDB 6XZ7 ) , we removed tRNA bodies , the L1 arm , and the GTPase-associated-center coordinates from 6XZ7 since these regions are disordered or missing in the deposited 50S subunit reconstruction . Previously published E . coli tryptic peptide mass spectrometry ( MS/MS ) raw data was used for the analysis ( Dai et al . , 2017; MassIVE accession: MSV000081144 ) . Peptide searches were performed with MSFragger ( Kong et al . , 2017 ) using the default parameters for a closed search with the following exceptions: additional variable modifications were specified on residues R ( hydroxylation , Δ mass: 15 . 9949 ) and M ( thioamide , Δ mass: 15 . 9772 ) , maximum modifications per peptide set to four , and multiple modifications on a residue were allowed . Spectra were searched against a database of all E . coli proteins plus common contaminants concatenated to a decoy database with all original sequences reversed . Results were analyzed using TPP ( Deutsch et al . , 2015 ) and Skyline ( Pino et al . , 2020 ) . Cryo-EM maps were supersampled in Coot for smoothness . Figure panels showing structural models were prepared using Pymol ( Schrödinger ) and ChimeraX ( Goddard et al . , 2018 ) . Sequence logo figures were made with WebLogo 3 . 7 . 4 ( Crooks et al . , 2004 ) . Phylogenetic trees were visualized with iTol ( version 4; Letunic and Bork , 2019 ) and multiple alignments were visualized with geneious 9 . 0 . 5 ( https://www . geneious . com ) . Ribosome coordinates have been deposited in the Protein Data Bank ( entry 7K00 ) , maps in the EM Database ( entries EMD-22586 , EMD-22607 , EMD-22614 , EMD-22632 , EMD-22635 , EMD-22636 , and EMD-22637 for the 70S ribosome composite map , 70S ribosome , 50S subunit , 30S subunit , 30S subunit head , 30S subunit platform , and 50S subunit CP maps , respectively ) , and raw movies in EMPIAR ( entry EMPIAR-10509 ) .	Inside cells , proteins are produced by complex molecular machines called ribosomes . Techniques that allow scientists to visualize ribosomes at the atomic level , such as cryogenic electron microscopy ( cryo-EM ) , help shed light on the structure of these molecular machines , revealing details of how they build proteins . Understanding how ribosomes work has many benefits , including the development of new antibiotics that can kill bacteria without affecting animal cells . Watson et al . used cryo-EM techniques with increased resolution to examine the ribosomes of the bacterium Escherichia coli in a higher level of detail than has been seen before . The results revealed two chemical modifications in proteins that form the ribosome that had not been observed in ribosomes previously . Additionally , a protein segment with a previously undescribed structure was identified close to the site where the ribosome reads the genetic instructions needed to make proteins . Further genetic analyses suggested these structures are in many related species , and may play important roles in how the ribosome works . Watson et al . were also able to see how paromomycin , an antibiotic used to treat parasitic infections , is positioned in the ribosome . The antibiotic interacts with a site near where the genetic code is read out , which might explain why certain changes to the antibiotic can interfere with its potency . Finally , the new ribosome structure reveals thousands of water molecules and metal ions that help keep the ribosome together as it produces proteins . This study shows the value of advances in cryo-EM technology and illustrates the importance of applying these techniques to other cell components . The results also reveal details of the ribosome useful for further research into this essential molecular machine .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2020	Structure of the bacterial ribosome at 2 Å resolution
The Turing reaction-diffusion model explains how identical cells can self-organize to form spatial patterns . It has been suggested that extracellular signaling molecules with different diffusion coefficients underlie this model , but the contribution of cell-autonomous signaling components is largely unknown . We developed an automated mathematical analysis to derive a catalog of realistic Turing networks . This analysis reveals that in the presence of cell-autonomous factors , networks can form a pattern with equally diffusing signals and even for any combination of diffusion coefficients . We provide a software ( available at http://www . RDNets . com ) to explore these networks and to constrain topologies with qualitative and quantitative experimental data . We use the software to examine the self-organizing networks that control embryonic axis specification and digit patterning . Finally , we demonstrate how existing synthetic circuits can be extended with additional feedbacks to form Turing reaction-diffusion systems . Our study offers a new theoretical framework to understand multicellular pattern formation and enables the wide-spread use of mathematical biology to engineer synthetic patterning systems . How cells self-organize to form ordered structures is a central question in developmental biology ( Hiscock and Megason , 2015 ) , and identifying self-organizing mechanisms promises to provide new tools for synthetic biology and regenerative medicine ( Chen and Weiss , 2005; Guye and Weiss , 2008; Isalan et al . , 2008; Bansagi et al . , 2011; Chau et al . , 2012; Mishra et al . , 2014; Schaerli et al . , 2014; Wroblewska et al . , 2015 ) . More than six decades ago , Alan Turing proposed a theoretical model in which interactions between diffusible substances can break the initial symmetry of cell fields to form periodic patterns ( Turing , 1952 ) . Subsequent work from Gierer and Meinhardt postulated that such self-organizing processes require differential diffusivity between a short-range self-enhancing activator and a feedback-induced long-range inhibitor ( Gierer and Meinhardt , 1972 ) . Numerous studies have proposed models based on these concepts to explain pattern formation during development , including skin appendage specification ( Sick et al . , 2006; Harris et al . , 2005 ) , lung branching ( Menshykau et al . , 2012; Hagiwara et al . , 2015 ) , tooth development ( Salazar-Ciudad and Jernvall , 2010 ) , rugae formation ( Economou et al . , 2012 ) , and digit patterning ( Sheth et al . , 2012; Raspopovic et al . , 2014 ) . However , the evidence in support of specific activator-inhibitor pairs has been limited , and few studies have provided experimental support for the differential diffusivity of activators and inhibitors ( Kondo and Miura , 2010; Marcon and Sharpe , 2012; Müller et al . , 2012 ) . Pattern formation processes are regulated by the interactions between secreted signaling molecules and their receptors that activate complex cell-autonomous signaling events . However , since most reaction-diffusion models have been reduced to abstract networks of two diffusible reactants , the influence of immobile cell-autonomous factors on reaction-diffusion patterning is largely unknown . Previous theoretical studies on selected network topologies have challenged the differential diffusivity requirements and indicated that in the presence of an immobile substance , patterns can form for a wider range of reaction and diffusion parameters ( Othmer and Scriven , 1969; Pearson and Horsthemke , 1989; Pearson , 1992; Pearson and Bruno , 1992; Rauch and Millonas , 2004; Levine and Rappel , 2005; Miura et al . , 2009; Raspopovic et al . , 2014; Korvasova et al . , 2015 ) . These and other studies ( Meinhardt , 2004; Werner et al . , 2015 ) suggest that extending models beyond abstract two-node systems can reveal different pattern formation requirements and may uncover new biologically relevant network designs . However , due to the complex mathematical analysis required to identify and understand such systems , extending reaction-diffusion models to more realistic signaling networks has been challenging , and the main assumption in the field has remained that complex models should reduce to simple systems that require an effective differential diffusivity . Here , we developed the freely available and user-friendly software RDNets ( available at http://www . RDNets . com ) to perform a high-throughput mathematical analysis of complex reaction-diffusion networks with non-diffusible components . In comparison to previous numerical studies , this method guarantees completeness , reproducibility , and detailed mechanistic insights into the principles underlying pattern formation . We used RDNets to build a comprehensive catalog of minimal three-node and four-node reaction-diffusion networks that include interactions between diffusible signals and cell-autonomous factors . Our results show that reaction-diffusion systems have three types of requirements for the diffusible signals depending on the network topology: Type I networks require differential diffusivity , Type II networks allow equal diffusivities , and Type III networks allow for unconstrained diffusivity . Overall , 70% of the networks identified by our analysis are of Type II and Type III and thus do not require differential diffusivity to form a spatial pattern . This reveals that realistic reaction-diffusion systems are based on mechanisms that are fundamentally different from the concepts of short-range activation and long-range inhibition based on differential diffusivity ( Gierer and Meinhardt , 1972 ) that have been predominant in previous models of pattern formation . Our software can be used to explore these new networks and is a unique tool to understand in vivo reaction-diffusion systems and to engineer synthetic circuits with spatial patterning capabilities . We developed an automated linear stability analysis ( Murray , 2003 ) to derive the pattern forming conditions of networks with N nodes ( Figure 1a , Materials and methods ) . Linear stability analysis determines whether a system can form a pattern by testing i ) if the concentrations of the reactants are stable at steady state , and ii ) if diffusion-driven instabilities arise with small perturbations . Because of its mathematical complexity , this type of analysis has been the exclusive domain of mathematicians and systems biologists ( Koch and Meinhardt , 1994; Satnoianu et al . , 2000; Murray , 2003; Miura and Maini , 2004 ) , and its application beyond two-reactant models has required dedicated theoretical studies for selected networks ( Othmer and Scriven , 1971; White and Gilligan , 1998; Klika et al . , 2012; Korvasova et al . , 2015 ) . To generalize the analysis to networks with more than two nodes , we utilized a modern computer algebra system and developed the software pipeline RDNets that automates the algebraic calculations . Within this framework , secreted molecules like ligands and extracellular inhibitors are represented by diffusible nodes , and cell-autonomous components such as receptors and kinases are represented by non-diffusible nodes . Our software analyzes networks with k interactions between the nodes; these interactions are represented by first order kinetics rates , where a positive rate corresponds to an activation and a negative rate to an inhibition . 10 . 7554/eLife . 14022 . 003Figure 1 . High-throughput screen for reaction-diffusion patterning networks using RDNets . ( a ) Schematic representation of the software RDNets to identify pattern-forming networks . RDNets exploits a computer algebra system for high-throughput mathematical analysis of reaction-diffusion networks with N nodes and k edges . Diffusion and reaction constraints , including the number of diffusible ( blue ) and non-diffusible ( red ) nodes and quantitative parameters ( here: k2 , k8 ) , can be specified as inputs . Additionally , the phase of the resulting periodic pattern can be selected . A list of reaction-diffusion networks is given as output . ( b ) Bar charts summarizing the number of networks for the 2- , 3- , and 4-node signaling network cases . Resulting networks can be of three types: Type I requires differential diffusivity , Type II allows for equal diffusivity , and Type III is diffusivity-independent . Type II and Type III networks are more robust to parameter changes than Type I networks . ( c ) Simulations of the possible topologies associated with a given network show that the minimal three-node systems can form in-phase and out-of-phase periodic patterns depending on the network topology . See Appendix 6 for a full list of parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 00310 . 7554/eLife . 14022 . 004Figure 1—figure supplement 1 . Catalog of all 3-node networks with two diffusible nodes ( blue ) , one non-diffusible node ( red ) and six interactions . The relative robustness is shown below each network . Note the higher robustness of Type III networks . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 00410 . 7554/eLife . 14022 . 005Figure 1—figure supplement 2 . Comprehensive catalog of 4-node Type I reaction-diffusion networks with two diffusible ( blue ) and two non-diffusible ( red ) nodes representing the interaction between two signaling pathways . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 00510 . 7554/eLife . 14022 . 006Figure 1—figure supplement 3 . Comprehensive catalog of 4-node Type II reaction-diffusion networks with two diffusible ( blue ) and two non-diffusible ( red ) nodes representing the interaction between two signaling pathways . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 00610 . 7554/eLife . 14022 . 007Figure 1—figure supplement 4 . Comprehensive catalog of 4-node Type III reaction-diffusion networks with two diffusible ( blue ) and two non-diffusible ( red ) nodes representing the interaction between two signaling pathways . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 007 The software pipeline comprises six steps to identify patterning networks: Steps 4 and 5 represent the core part of the automated linear stability analysis and involve the majority of analytical computations . In Step 6 , our software screens the possible reaction-diffusion topologies associated with a network . A reaction-diffusion network of size k defines only a set of k regulatory links between nodes but does not make any assumption on whether these are activating or inhibiting interactions . In the following , we refer to the possible combination of activating and inhibiting interactions as 'network topologies' . We used our software RDNets to systematically explore the effect of cell-autonomous factors in reaction-diffusion models for the generation of self-organizing patterns . We studied two types of networks: a ) 3-node networks with two diffusible nodes and one non-diffusible node representing the interaction between two secreted molecules and one signaling pathway , and b ) 4-node networks with two diffusible nodes and two non-diffusible nodes representing the interaction between multiple ligands and signaling pathways . Table 1 shows the number of networks identified at each step of our automated mathematical analysis ( see Figure 1—figure supplements 1–4 for the complete catalog of the identified reaction-diffusion networks ) . Our analysis revealed that in the presence of cell-autonomous factors there are three types of networks with different constraints on the diffusible signals:Type I ( requires differential diffusivity ) :∃ ( di , dj ) ⊂D , dj≠dj∧∀ di∈D , di>0Type II ( allows for equal diffusivity ) :∀ ( di , dj ) ⊂D , dj=dj∧∀ di∈D , di>0Type III ( unconstrained diffusivity ) :∀ di∈D , di>0 where D is the list of diffusion coefficients that are non-zero . 10 . 7554/eLife . 14022 . 008Table 1 . From an initial number of possible networks ( Step 1 ) , RDNets progressively identifies reaction-diffusion networks that can form a pattern ( Step 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 008Steps3 nodes4 nodes# networks# topologies# networks# topologies1 . Minimal systems8453761144014643202 . Strongly connected48307222842923523 . Non-symmetrical251600597764164 . Stable2455632486405-6 . Reaction-diffusion218464512 We found that 70% of the identified networks with non-diffusible nodes are of Type II and Type III ( Figure 1b ) , showing that in the presence of cell-autonomous factors the differential diffusivity requirement is unexpectedly rare . Type III networks have never been characterized before and surprisingly have patterning conditions that are independent of specific diffusion rates . We found that Type III networks are not only numerous but also extremely robust to changes in parameter values compared to Type I and Type II networks ( Figure 1b , Materials and methods ) . Using numerical simulations , we systematically confirmed our mathematical analysis and determined that a network can form all possible combinations of in-phase or out-of-phase periodic patterns depending on the network topology ( Figure 1c , Appendix 1 ) . Together , our results show that realistic reaction-diffusion networks are intrinsically robust , do not require differential diffusivity , and have patterning capabilities identical to classical two-node reaction-diffusion models . Importantly , the novel class of Type III networks that we discovered suggests a new mechanism of pattern formation that is independent of short-range activation and long-range inhibition based on differential diffusivity . To obtain insight into the organizing principles underlying the three types of networks identified by our high-throughput analysis , we developed a novel graph-theoretical formalism to express the pattern forming conditions in terms of network feedbacks rather than reaction parameters ( see Materials and methods and Appendix 2 ) . This analysis determines which feedback cycles contribute to the stability and the instability conditions ( Figure 2a , b ) and defines the topological features that underlie Type I , Type II , and Type III networks . In agreement with previous studies ( Murray , 2003 ) , our analysis confirmed that two-node networks can only simultaneously satisfy the stability and instability conditions when the diffusion ratio d between the inhibitor and the activator is greater than one ( Figure 2 , left column ) . This observation has been linked with the widespread belief that reaction-diffusion systems require differential diffusivity to implement short-range auto-activation and long-range inhibition . Our analysis instead suggests that the differential diffusivity requirement arises from the opposite nature of the stability and instability conditions , which require that the destabilizing feedback must be both higher and lower than the stabilizing feedback . Since the diffusion term only appears in the destabilizing condition , it assumes the role of a unique pivot that can satisfy both conditions simultaneously when d > 1 . Our results indicate that the presence of non-diffusible nodes allows feedbacks that do not appear in the instability conditions to act as an additional pivot to satisfy both conditions simultaneously by increasing stability . This is the case for most Type II networks ( Figure 2 , middle column ) that contain additional negative feedbacks that allow for equal diffusivities ( Klika et al . , 2012; Korvasova et al . , 2015 ) . Importantly , our analysis also reveals that non-diffusible nodes can implement positive feedbacks that can drive the network unstable independently of stabilizing feedbacks and for any diffusion ratio d . This is the case for Type III networks ( Figure 2 , right column ) , where the stability and instability conditions are uncoupled and can be simultaneously satisfied for large parameter sets . This is possible because immobile factors can act as 'capacitors' that retain and amplify perturbations independently of the reactants’ diffusion coefficients ( see Appendix 3 for details ) . Such systems represent a fundamentally new pattern formation mechanism that has not been described previously . 10 . 7554/eLife . 14022 . 009Figure 2 . Analysis of the organizing principles underlying reaction-diffusion networks . ( a ) Schematic diagram of a 2-node network of Type I , a 3-node network of Type II , and a 3-node network of Type III . c1 to c4 indicate feedback cycles , red indicates overall inhibition and green overall activation , and d=dw/dv represents the diffusion ratio . The two-node network ( left column ) is a classical activator-inhibitor system , the other two networks are more realistic 3-node networks wired through a cell-autonomous factor u . ( b ) Linear stability analysis of the topologies shown in ( a ) reveals that pattern-forming conditions require a trade-off between stability and instability feedback cycles , which gives rise to the diffusion constraint . The blue volume highlights the parameter set that allows for pattern formation ( Turing space ) ; the three parameters c3 , c4 , and d vary independently along the axes . Intersecting the Turing space with a plane of equal diffusion coefficients d=1 shows that , in contrast to Type II and Type III networks , patterning in Type I networks is not possible with equal diffusivities . ( c ) 1D simulations show that the apparent longer inhibitor range ( blue arrows ) observed in the Type I network is also maintained in the Type II network even with d=1 and therefore does not result from differential diffusivity . The Type III network with d=0 . 1 surprisingly shows an apparent longer range for the activator v . 1D and 2D simulations show that Type II and Type III topologies form patterns similar to those generated by classical 2-node models . See Appendix 6 for a full list of parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 009 Together , our results show that models based on 'short-range auto-activation and long-range inhibition' implemented by differential diffusivity are only a special case of a general trade-off between stabilizing and destabilizing feedbacks required for pattern formation . The virtually indistinguishable simulations of Type I networks with differential diffusivity and Type II networks with equal diffusivities reveal that the final aspect of the periodic patterns does not reflect a difference in the range of activators and inhibitors but only a difference in their amplitude ( Figure 2c , see Appendix 3 for details ) . Indeed , in other Type II and Type III networks the relationship between the amplitude of activators and inhibitors can even be inverted , such that the perceived range of the activator appears larger than the perceived range of the inhibitor . Therefore , in contrast to previous studies ( Kondo and Miura , 2010 ) , we propose that long-range lateral inhibition is not required to limit the expansion of the activator ( Appendix 3 ) . To demonstrate the functionality and applicability of RDNets , we analyzed two known self-organizing developmental patterning networks , the Nodal/Lefty reaction-diffusion system and the BMP/Sox9/Wnt network . In the following , we show how quantitative and qualitative experimental data from these developmental systems can be used to constrain the high-throughput analysis and to characterize the possible underlying patterning topologies . It has been proposed that Nodal and Lefty implement an activator-inhibitor system that patterns the germ layers and the left-right axis in vertebrates ( Chen and Schier , 2001; Shiratori and Hamada , 2006; Shen , 2007; Meinhardt , 2009; Schier , 2009; Kondo and Miura , 2010; Rogers and Schier , 2011; Korvasova et al . , 2015 ) ( Figure 3a ) . In agreement with this hypothesis , the self-enhancing activator Nodal has been shown to diffuse 7 . 5 times slower than the feedback-induced inhibitor Lefty in living zebrafish embryos ( Müller et al . , 2012 ) . The Nodal/Lefty system has been modeled as a two-component activator-inhibitor system ( Nakamura et al . , 2006; Müller et al . , 2012 ) , but the influence of cell-autonomous factors including receptors and the well-characterized intracellular signal transduction cascade via phosphorylated Smad2/3 ( Schier , 2009 ) has not been studied . We used our software to screen for networks that extend the two-node Nodal/Lefty system with a non-diffusible node corresponding to active Nodal signaling ( Figure 3b ) . The screen was constrained with known qualitative regulatory interactions: a positive feedback loop between Nodal and its signaling , and a promotion of Lefty by Nodal signaling ( Figure 3b ) . Moreover , we constrained the two negative self-regulations on Nodal and Lefty , which represent their clearance from the diffusible pool , with the previously measured clearance rate constants ( Müller et al . , 2012 ) . Finally , we selected only reaction-diffusion networks that produced in-phase patterns of Nodal and Lefty , which recapitulate their overlapping expression domains ( Schier , 2009 ) . With these constraints , our mathematical analysis identified just two possible minimal networks: In one network Lefty inhibits Nodal signaling indirectly at the receptor level , and in the other network Lefty inhibits Nodal directly ( Figure 3b ) . These predictions are in agreement with the two possible mechanisms by which Lefty has been proposed to inhibit Nodal activity: by binding to the Nodal receptor or by directly sequestering Nodal ( Chen and Shen , 2004 ) . However , the role and significance of these two alternative mechanisms for Nodal/Lefty-mediated patterning has remained unclear ( Cheng et al . , 2004; Middleton et al . , 2013 ) . Our mathematical analysis predicts that the first mechanism ( Lefty blocks the receptor complex ) determines a Type II network , whereas the second mechanism ( Lefty blocks Nodal directly ) determines a Type III network . Importantly , both models suggest that the Nodal/Lefty system may form patterns without differential diffusivity of activator and inhibitor . Using the clearance rate constants of Nodal and Lefty as quantitative constraints , our mathematical analysis predicts a possible minimum diffusion ratio d = 0 . 55 for the Type II network , whereas the Type III network allows for any combination of diffusion coefficients ( Figure 3d ) . The robustness analysis of the networks shows that for unconstrained valued of d , the Type III network is more robust to parameter changes ( Figure 3c ) . However , when we fix the diffusion ratio to the experimentally quantified value ( Müller et al . , 2012 ) ( d = 7 . 5 ) , the Type II network becomes more robust than the Type III network ( Figure 3d ) . This shows that , while Nodal and Lefty do not necessarily need to have different diffusivities to form a pattern , the combination of differential diffusivity and clearance rate constants increases the robustness of the Type II system . 10 . 7554/eLife . 14022 . 010Figure 3 . Modeling of the Nodal/Lefty reaction-diffusion system with realistic signaling networks . ( a ) Schematic diagram of the Nodal/Lefty activator-inhibitor system . Nodal ( green ) is the self-enhancing activator that promotes the feedback inhibitor Lefty ( red ) . ( b ) Extension of the Nodal/Lefty system with an immobile cell/receptor-complex node ( blue ) to distinguish between two possible feedback modes . In both networks , the self-enhancing activation and the Nodal-induced Lefty expression occurs through a non-diffusible cell/receptor-complex represented by the activated signal transducer pSmad2/3 ( S , blue ) . In the Type II network , Lefty inhibits Nodal through the receptor node S , whereas in the Type III network , Lefty inhibits Nodal directly ( dashed lines ) . ( c , d ) The Type III network is more robust to parameter changes over a broader range of diffusivities ( bar chart on the left and bigger Turing space [blue volume] ) compared to the Type II network . However , constraining the two topologies with previously measured diffusion coefficients ( d=7 . 5 ) demonstrates that the Type II network is more robust for biologically relevant parameters ( bar chart on the right and bigger area of the green plane corresponding to d=7 . 5 within the Turing space [blue volume] ) . Experimental data for the previously measured clearance rate constants ( c1 , c2 ) of Nodal and Lefty predicts that the minimum allowed diffusion ratio for the Type II network is d=0 . 55 ( red dot ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 01010 . 7554/eLife . 14022 . 011Figure 3—figure supplement 1 . A possible evolutionary scenario for evolving the differential diffusivity of Nodal and Lefty . Network A ( Type III ) is more likely to evolve de novo with an initial equal diffusivity ( d=1 ) and for a wider range of diffusion ratios ( 100>d>0 ) . During evolution , if the negative feedback on Nodal signaling changes ( dashed line ) , network C ( Type II ) together with differential diffusivity can be selected to increase robustness and therefore fitness . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 011 As a second example , we used RDNets to analyze the BMP/Sox9/Wnt ( BSW ) self-organizing network that underlies digit patterning ( Sheth et al . , 2012; Raspopovic et al . , 2014 ) . The expression patterns and the signaling activity of the network components have been well-characterized showing that Sox9 forms periodic expression peaks that are out-of-phase of BMP expression and Wnt activity ( Figure 4a ) . A three-node reaction-diffusion network with two diffusible nodes for the secreted signals BMP and Wnt and a non-diffusible node for the transcription factor Sox9 has previously been derived based on the known regulatory interactions ( Figure 4b ) . It was shown that this network recapitulates the out-of-phase pattern between BMP/Wnt and Sox9 and forms a pattern with extremely low differential diffusivity requirements ( d = 1 . 25 ) . Our comprehensive mathematical analysis reveals that this three-node system is just another topology of the reaction-diffusion network that we analyzed for the extended Nodal/Lefty system; it is therefore a Type II network that can potentially form a pattern even when BMP and Wnt have equal diffusion coefficients . In previous studies , this observation was missed because the clearance rates of BMP and Wnt had been assumed to be identical ( Raspopovic et al . , 2014 ) . However , as we showed in the previous example for Nodal and Lefty , if BMP is cleared faster than Wnt , the diffusion ratio can be equal to or lower than one , d ≤ 1 . The three-node BSW model recapitulates the out-of-phase pattern between BMP/Wnt and Sox9 , but due to its high abstraction level it does not explain the opposite BMP expression and BMP activity patterns observed in the experimental data ( Figure 4a ) . We therefore used RDNets to screen for more complex models with five nodes that represent all components of the network: two diffusible nodes for BMP ( B ) and Wnt ( W ) and three non-diffusible nodes , one for the canonical BMP pathway through pSmad1/5/8 ( Sm ) , one for the intracellular Wnt signaling cascade ( β-catenin , β ) , and one for Sox9 ( S ) . We selected only networks that formed in-phase and out-of-phase patterns reflecting the experimental data ( Figure 4a ) . Previous studies ( Raspopovic et al . , 2014 ) showed that Sox9 is promoted by BMP signaling through pSmad1/5/8 and is inhibited by Wnt through β-catenin . Similar to the Nodal/Lefty example , we constrained the mathematical screen by incorporating these known regulatory interactions . Unexpectedly , the screen revealed that if β-catenin would directly inhibit Sox9 , no network could recapitulate the out-of-phase patterns between BMP expression and BMP signaling . By performing a more general screen that left this interaction unconstrained , we found that the opposite BMP expression and signaling patterns can be obtained when β-catenin indirectly inhibits Sox9 through pSmad/1/5/8 . RDNets also predicts that the most robust networks include the following additional interactions: i ) a negative feedback from Sox9 to Wnt , ii ) a negative feedback from pSmad1/5/8 to BMP , and iii ) either a positive feedback from β-catenin to Sox9 or a negative feedback from β-catenin on Wnt ( Figure 4b , gray arrows ) . Interestingly , the majority of networks identified by our screen was of Type III , suggesting that the proportion of Type III networks increases when more non-diffusible nodes are added . 10 . 7554/eLife . 14022 . 012Figure 4 . Modeling of mouse digit patterning with realistic signaling networks . ( a ) Experimental patterns of BMP ( green ) , pSmad1/5/8 ( purple ) , Sox9 ( blue ) , and β-catenin ( red ) in a mouse limb at stage E11 . 5 ( data reproduced from Raspopovic et al . , 2014 ) . ( b ) Extension of a previously proposed simple three-node network for digit patterning involving BMP , Sox9 , and Wnt to a more realistic five-node network incorporating known interactions ( black ) between Wnt ( W , red ) , BMP ( B , green ) , Smad1/5/8 ( Sm , pink ) , Sox9 ( S , blue ) , and β-catenin ( β , red ) ; interactions predicted by RDNets are shown in gray , and dashed lines correspond to alternative interactions that implement networks with similar robustness . The simulations of the new five-node network recapitulate the unintuitive out-of-phase pattern between BMP expression ( green ) and its own signaling through pSmad1/5/8 ( purple ) . The mathematical analysis predicts that these patterns can be formed when β-catenin inhibits Sox9 indirectly through pSmad1/5/8 . See Appendix 6 for a full list of parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 012 Although reaction-diffusion mechanisms have a simple network design , they exhibit unique self-organizing capabilities making them appealing for synthetic engineering ( Diambra et al . , 2015 ) . So far , the synthetic implementation of reaction-diffusion systems has been impeded by the small pattern-forming parameter space of simple two-node models , their requirement for differential diffusivity ( Carvalho et al . , 2014 ) , and a general gap between abstract models and real sender-receiver reaction-diffusion circuits ( Marcon and Sharpe , 2012; Barcena Menendez et al . , 2015 ) . RDNets provides a comprehensive catalog of reaction-diffusion networks that do not require differential diffusivity of the signaling molecules , which enables bioengineers to explore new mechanisms to form periodic spatial patterns in a robust manner . We demonstrate the utility of RDNets by proposing an extension to an existing synthetic circuit for cell-cell communication in yeast ( Chen and Weiss , 2005 ) . The original synthetic circuit introduced a diffusible plant hormone , cytokinin isopentenyladenine ( IP ) , and its receptor AtCRE1 into yeast ( Figure 5a ) . This circuit was used to implement a sender-receiver and a quorum sensing mechanism based on a positive feedback loop between IP-signaling and IP ( Figure 5a ) . We used RDNets to identify possible signaling networks that can extend this positive feedback with additional interactions to form a reaction-diffusion pattern . Since at least two diffusible nodes are required to form a pattern ( Murray , 2003 ) , we screened minimal 4-node networks that include the engineered positive feedback and candidate interactions with another diffusible node . In order to look for realistic and easily implementable signaling circuits , we explored only networks with interactions between diffusible nodes through non-diffusible factors representing intracellular signaling cascades . We also imposed self-regulations on diffusible nodes to be exclusively inhibitory , representing decay . With these constraints , our high-throughput analysis identified 16 minimal reaction-diffusion networks ( 5 Type I , 3 Type II , 8 Type III ) , of which the Type II and Type III networks were most robust to parameter changes ( Figure 5—figure supplement 1 ) . In the following , we demonstrate how the conditions derived by RDNets can be used to engineer the most simple and robust Type II network ( Figure 5a - right ) . In addition to the positive feedback loop , this network contains three additional negative feedbacks: two are self-regulations that correspond to decay , and one is a negative feedback between the newly introduced diffusible node and the non-diffusible node representing the receptor . This network suggests that a simple extension to the circuit developed in Chen and Weiss ( 2005 ) could be obtained by a ) destabilizing the signaling hormone and the receptor to increase their turn-over ( c1 , c2 ) , and b ) introducing another hormone that signals through the same receptor and implements a negative feedback loop to its own expression or activity ( c3 , Figure 5a - right ) . 10 . 7554/eLife . 14022 . 013Figure 5 . Combining signaling modules to form new synthetic reaction-diffusion networks . ( a ) Left: Schematic diagram of a four-node network to engineer a patterning system from an existing signaling module ( Chen and Weiss , 2005 ) that implements a positive feedback ( c4 , green ) . In the previously engineered synthetic network , the positive feedback highlighted by c4 was implemented by the hormone Cytokinin isopentenyladenine ( IP ) that activates the receptor AtCRE1 to induce the SSRE-promoter-driven expression of AtIPT4 , which catalyzes IP production . Right: A possible Type II reaction-diffusion network predicted by RDNets , in which the positive feedback module composed of w , u and z ( representing IP , receptors/transducers , and AtIPT4 shown in ( a ) ) is extended by a node v that activates u , which in turn inhibits v ( cycle c3 ) . The cycles c1 and c2 correspond to signal decay . ( b ) Stability and instability conditions of the predicted network . ( c ) Constraining RDNets with previous measurements of the positive feedback cycle c4 obtained by fitting experimental data ( graph on the left , Chen and Weiss , 2005 ) identifies exact parameter ranges for the new interactions in the synthetic reaction-diffusion network ( graph and formulae on the right ) . ( d ) 1D simulations show that different topologies of this synthetic network can be engineered to produce all possible in- and out-of-phase periodic patterns depending on the sign of the reaction rates shown above the graphs . ( e ) A 3D simulation of the synthetic patterning system forms tubular structures that could be exploited for tissue engineering . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 01310 . 7554/eLife . 14022 . 014Figure 5—figure supplement 1 . Catalog of possible synthetic networks that extend an existing feedback loop ( red arrows ) . Networks have 2 diffusible nodes ( blue ) , 2 non-diffusible nodes ( red ) and seven interactions . The relative robustness is shown below each network . Note the higher robustness of Type III networks . The two boxed circuits correspond to the networks presented in Figure 5 ( Type II ) and Appendix 4—figure 1 ( Type III ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14022 . 014 In addition to revealing possible topologies , our automated analysis provides the mathematical formulae of pattern forming conditions . This feature together with the specification of quantitative constraints can be used to calculate pattern forming parameter ranges . To determine the strength of the newly introduced negative feedbacks required for pattern formation , we constrained the positive cycle strength with the first-order kinetic rate quantified in Chen and Weiss ( 2005 ) by fitting measurements of the signaling activity of IP ( Figure 5c ) . Moreover , we assumed that the diffusion and decay rates are similar . With these constraints , RDNets determined that the newly introduced negative feedback has to be stronger than the positive feedback and decay rates ( Figure 5c - right ) . This could be implemented using the more responsive IP-signaling promoter ( TR-SSRE ) developed in Chen and Weiss ( 2005 ) to drive the expression of the inhibitor . This specific synthetic network represents only one possibility . We find other Type III networks to be even more robust to parameter changes , but they appear to require the design of more complex synthetic circuits ( Appendix 4 ) . Once a synthetic network is designed , RDNets can also be used to automatically derive kinetic models that can simulate the reaction-diffusion network ( Figure 5d , e , Appendix 5 ) . Numerical simulations can be used to investigate the qualitative aspect of the pattern and its spatial periodicity . In the long term , all these features open new avenues for designing synthetic reaction-diffusion circuits that could be coupled with gene expression to enable complex applications , such as fabrication of spatially patterned three-dimensional biomaterials and tissue engineering in mammalian cells ( Chen and Weiss , 2005; Carvalho et al . , 2014 ) . We developed the web-based software RDNets , which exploits a modern computer algebra system to identify new reaction-diffusion networks that reflect realistic signaling systems with diffusible and cell-autonomous factors . Our approach is a new example of high-throughput mathematical analysis , which has several benefits over previous numerical approaches ( Ma et al . , 2009 ) . First , RDNets can be run from most web browsers and does not demand large computational power . Second , our mathematical analysis yields closed-form solutions and is complete , in contrast to numerical simulations that can necessarily only sample from a small region of the entire parameter space . Third , RDNets derives the conditions for pattern formation and therefore provides mechanistic explanatory power to the users . In addition , it helps to identify reaction-diffusion topologies that are in agreement with qualitative and quantitative experimental constraints , which makes it an unprecedented tool for users that aim to study developmental patterning networks or to design reaction-diffusion synthetic circuits . Motivated by theoretical studies that showed that non-diffusible factors can influence pattern forming conditions , we used our software to systematically explore the effect of non-diffusible reactants in reaction-diffusion models . Our analytical approach is both comprehensive and informative and reveals that depending on the network topology , reaction-diffusion systems can belong to three classes: Type I systems that require differential diffusivity , Type II systems that can form patterns with equal diffusivity , and Type III systems that form patterns independent of specific diffusion rates . In particular , the novel class of Type III networks has not been described before and challenges models of short-range activation and long-range inhibition based on differential diffusivity that have dominated the field of developmental and theoretical biology for decades ( see Appendix 3 for details ) . We used RDNets to obtain new mechanistic insight into two developmental patterning systems . By using quantitative data to constrain possible patterning networks , we found a Type II and a Type III topology that extend the Nodal/Lefty activator-inhibitor system with realistic cell-autonomous signaling . In such extended networks , Nodal and Lefty do not necessarily need to have different diffusivities to form a pattern . However , our results suggest that the differential diffusivity can contribute to a more robust patterning system in Type II networks with indirect Nodal signaling inhibition . We propose that the high general robustness of the Type III network might have played a role for the evolution of the Nodal/Lefty reaction-diffusion system in the first place , and that the indirect Nodal signaling inhibition of Type II networks might have been fine-tuned during evolution ( Figure 3—figure supplement 1 ) . Similarly , we extended the three-node BSW digit patterning model with additional previously characterized cell-autonomous factors and constrained a five-node model with qualitative data . Our analysis identified realistic network topologies that accurately reflect the previously puzzling opposite pattern of BMP ligands and BMP signaling and predicts novel interactions between network components . Finally , we used RDNets to design a novel synthetic patterning circuit based on a previously engineered positive feedback module . Identifying a comprehensive catalog of gene networks that can perform a certain behavior has been shown to be a successful strategy to uncover the design space of stripe-forming networks ( Cotterell and Sharpe , 2010 ) , which can be directly useful to synthetic biology . In particular , this approach permitted a whole family of network mechanisms to be synthetically constructed in bacteria – all capable of forming a gene expression stripe in a bacterial lawn ( Schaerli et al . , 2014 ) . Similarly , our software provides a comprehensive catalog of reaction-diffusion networks and enables bioengineers to explore new mechanisms to form periodic spatial patterns in a robust manner . These networks explicitly include non-diffusible factors that mediate signaling and are easy to relate with sender-receiver synthetic toolkits ( Barcena Menendez et al . , 2015 ) . In addition , we found that the majority of realistic reaction-diffusion networks eliminate the differential diffusivity requirement that is difficult to implement synthetically ( Carvalho et al . , 2014; Barcena Menendez et al . , 2015 ) . The possibility to use qualitative and quantitative constraints to screen for reaction-diffusion networks makes RDNets a unique tool to customize patterning systems from initial starting networks . Moreover , the pattern-forming conditions derived by the software can be used to estimate parameter ranges and network robustness . Particularly promising is our finding that each network is associated with a set of topologies that exhaustively determine all the in-phase and out-of-phase relations between periodic patterns ( Figure 5d ) . It is therefore possible to design network topologies that promote the co-localized expression of any desired combination of factors . This will enable novel applications in tissue engineering , where the co-localized expression of differentiating factors can be used to induce specific tissues ( Kaplan et al . , 2005 ) . Coupled with the three-dimensional pattern-forming capabilities of reaction-diffusion mechanisms ( Figure 5e ) , this could be used to devise new strategies for engineering scaffolds or tissues with complex architecture . In summary , our analysis defines new concepts of reaction-diffusion-mediated patterning that are directly relevant for developmental and synthetic biology . We demonstrate three applications of our software RDNets to understand developmental mechanisms and to design synthetic patterning systems , but this approach can be extended to numerous other patterning processes ( Economou et al . , 2012; Menshykau et al . , 2012; Hagiwara et al . , 2015 ) . We therefore expect that RDNets will contribute to the wide-spread use of mathematical biology and that a similar approach could be applied to other dynamical processes such as oscillations and traveling waves ( Bement et al . , 2015 ) . We analyzed reaction-diffusion networks represented by a reaction matrix J and a diffusion matrix D of size NxN , where N is equal to the number of nodes . The matrix J corresponds to the Jacobian of the reaction-diffusion system and contains partial derivatives that describe the relative influence of one node on another . Elements of the reaction matrix represent the first order kinetics rates of the regulatory interactions in the network , where a positive rate corresponds to an activation and a negative rate to an inhibition . The matrix D contains the diffusion rates of the reactants along its principal diagonal and is zero otherwise . Our analysis aims to identify minimal reaction-diffusion networks , defined as the networks with the minimal number of edges k that can form a reaction-diffusion pattern . In the case of 2-node networks , it has been described that minimal reaction-diffusion networks must have 2x2=4 edges ( Murray , 2003 ) , and therefore only a completely connected network is allowed . This completely connected 2-node network allows for only two reaction-diffusion topologies: the 'activator-inhibitor system' that forms in-phase periodic patterns , and the 'substrate-depleted model' that forms out-of-phase periodic patterns . Our automated approach takes the following inputs through a graphical user interface: the number of network nodes N , constraints on J and D including reaction or diffusion rates set to zero , and the number of regulatory interactions k . This last parameter defines the number of edges that each network should have with an upper bound of NxN edges representing a completely connected network ( Figure 1a ) . This parameter also defines the number of possible networks that are analyzed by the software , which is calculated according to ( NxNk ) =NxN ! k ! ( NxN−k ) ! This number represents the possible subsets of size k that can be taken from J and corresponds to the number of possible networks of size k . An important part of the automated high-throughput mathematical analysis is the derivation of the characteristic polynomial , a mathematical expression that determines the stability of the reaction-diffusion system , which is calculated from the determinant of a matrix that combines J and D , the 'wave number' q , and the eigenvalue λ . For 3-node networks , the characteristic polynomial has the formλ3+λ2a1+λa2+a3=0 where λ is the eigenvalue associated with the reaction-diffusion system , and the coefficients a1 , a2 and a3 are polynomials formulated in terms of the elements of J , D and q ( see Appendix 1 ) . The eigenvalue λ determines the stability of the network: negative real solutions of λ represent a system that is stable around its steady state , while a positive real solution of λ represents an unstable system . The variable q that appears in the coefficients a1 , a2 and a3 is the wave number that is introduced by the linear stability analysis and is multiplied for D . For values q>0 , this parameter defines the periodicity of the reaction-diffusion pattern . Step 4 of our pipeline entails finding the ranges of the reaction parameters in J and diffusion parameters in D for each network , for which the solutions λ are all negative when q=0 . Similarly , Step 5 requires finding parameter intervals , for which at least one solution λ has a positive real part when q>0 . For characteristic polynomials of degree higher than 2 , this is usually done by using the Routh-Hurwitz stability criterion ( Murray , 2003 ) , a mathematical theorem that finds the necessary and sufficient condition for all negative roots in terms of the polynomial coefficients a1 , a2 . . . an . However , as the number of network nodes N increases , finding these parameter intervals becomes challenging and tedious because the coefficients a1 , a2 . . . an are also complex polynomials of high degree in q . We used a computer algebra system to automatically derive and analyze the Routh-Hurwitz criterion in terms of the coefficients a1 , a2 . . . an . Finally , Step 6 requires to evaluate which of the 2k possible topologies that exist for a given network are compatible with the pattern-forming conditions derived in Step 5 ( see Appendix 1 for details ) . The complete analysis of minimal networks is limited by the existence of analytical solutions . According to the Abel-Ruffini theorem , there is no general algebraic solution for systems with more than four nodes . However , in practice many five-node networks can be solved if the constraints specified in the input of RDNets lead to a simplification of the coefficients of the characteristic polynomial , as is the case for the five-node digit patterning network ( Figure 4 ) . Analytical approaches become also challenging when further diffusible nodes are added and when minimal models are extended with additional interactions . We analyzed the robustness of the networks by calculating the volume of the parameter space that respects the pattern-forming condition in relation to the unit length multidimensional space of all the possible parameter values . This robustness measure corresponds to the probability of randomly picking pattern-forming parameters . The pattern-forming parameter volume is calculated with a multiple integral of the pattern-forming conditions over all the parameters of the reaction-diffusion networks , in the form∭l1l2⋯∭dl1l2f ( k1 , …kNxN , d1…dN ) dk1…dkNxNdd1…ddN where k1 . . . kNxN are the reaction parameters and d1 . . . dn the diffusion parameters , f ( k1… kNxN , d1… dn ) are the pattern-forming conditions of the networks , and l1 , l2 and dl1 , dl2 are the limits of reaction and diffusion variables that are set respectively to ( -0 . 5 , 0 . 5 ) and ( 0 , 1 ) representing a multidimensional parameter space of unit side length . To investigate the topological basis of Type I , Type II , and Type III networks , we developed a new theoretical framework based on graph theory that can be used to rewrite the pattern-forming conditions in terms of network feedbacks rather than their reaction rates . Further details of this theory are provided in Appendix 2 . Our web-based software RDNets was written in Mathematica ( Wolfram Research Inc . , Champaign , Illinois ) and is available at http://www . RDNets . com . RDNets requires only the installation of the freely available Wolfram CDF player plugin ( http://www . wolfram . com/cdf-player/ ) . A simple graphical interface can be used to specify inputs and constraints and to run the high-throughput mathematical analysis . Constraints can be specified by clicking on the nodes or edges of the networks , or by providing specific values for the corresponding parameters ( see User Guide available at http://www . RDNets . com ) . These constraints are automatically translated into mathematical formulae that are coupled with the symbolic linear stability analysis performed by the computer algebra system . The graphical user interface can be used to explore and simulate the list of reaction-diffusion topologies given as output of the linear stability analysis . Additional constraints can be progressively added to the analysis to refresh the output and to narrow down the list of candidate topologies ( see User Guide available at http://www . RDNets . com ) .	Developing embryos initially consist of identical cells that specialize over time to create the different parts of the adult animal . More than sixty years ago , Alan Turing proposed that this spontaneous breaking of uniformity could be controlled by two molecules that interact with each other and move by diffusion at different rates between cells . In such “reaction-diffusion” systems , the interactions between the molecules cause repeating peaks in their concentrations in different locations , which could influence how different parts of the embryo develop . However , how these hypothetical molecules relate to the genes that control embryonic development has remained largely unknown . Marcon et al . have now developed a computational method to identify the conditions that enable periodic patterns to form spontaneously in realistic reaction-diffusion systems with mobile signaling molecules and immobile factors such as membrane-localized receptors . By computationally screening millions of biologically relevant networks , Marcon et al . found that a key requirement of classical Turing models – that the mobile signaling molecules must diffuse at different rates – does not need to be met for patterns to form . Instead , some networks can form patterns with signals that diffuse at equal rates , while others can form patterns with any combination of diffusion rates . The computational method developed by Marcon et al . can be used to interpret the mechanisms that allow patterns to form in biological systems , such as those that control embryonic development . It can also be used to develop synthetic networks that regulate genes for the formation of tissues in particular spatial patterns .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "developmental", "biology", "tools", "and", "resources" ]	2016	High-throughput mathematical analysis identifies Turing networks for patterning with equally diffusing signals
Despite anatomical similarities , there are differences in susceptibility to cardiovascular disease ( CVD ) between primates; humans are prone to myocardial ischemia , while chimpanzees are prone to myocardial fibrosis . Induced pluripotent stem cell-derived cardiomyocytes ( iPSC-CMs ) allow for direct inter-species comparisons of the gene regulatory response to CVD-relevant perturbations such as oxygen deprivation , a consequence of ischemia . To gain insight into the evolution of disease susceptibility , we characterized gene expression levels in iPSC-CMs in humans and chimpanzees , before and after hypoxia and re-oxygenation . The transcriptional response to hypoxia is generally conserved across species , yet we were able to identify hundreds of species-specific regulatory responses including in genes previously associated with CVD . The 1 , 920 genes that respond to hypoxia in both species are enriched for loss-of-function intolerant genes; but are depleted for expression quantitative trait loci and cardiovascular-related genes . Our results indicate that response to hypoxic stress is highly conserved in humans and chimpanzees . Understanding human susceptibility to disease , and the mechanisms that underlie disease susceptibility , are central goals of biomedical research . One common approach to investigate the regulatory mechanisms that underlie inter-individual disease susceptibility differences is to combine disease association studies with investigations of genetic variants that associate with molecular-level phenotypes using a quantitative trait locus ( QTL ) framework . However , one of the limitations of a QTL-based approach is that it is not clear what proportion of loci have actual functional consequences . A complementary approach to gaining insight into human susceptibility to disease is to investigate the genetic , molecular , and cellular differences between humans and our closest evolutionary relatives , the great apes . The challenge of a comparative approach is that it can be difficult to determine the specific basis of inter-species phenotypic differences , such as disease , because many observations are often correlated with each other across the species . The limitations and challenges of these approaches may be addressed by a combined analysis of comparative and QTL data , which can help us better understand the functional role of regulatory QTLs , by focusing on QTLs that impact genes whose regulation is either conserved or correlated with inter-species phenotypic differences . The framework for our comparative study begins with the observed inter-species difference in cardiovascular disease ( CVD ) . CVD is responsible for about a third of both human and captive chimpanzee deaths ( WHO; Varki et al . , 2009 ) . The anatomy of the healthy human and chimpanzee heart is similar; however CVD pathology differs . Chimpanzee disease is often associated with interstitial myocardial fibrosis , while human heart disease predominantly results from coronary artery artherosclerosis , leading to ischemic damage ( Lammey et al . , 2008; Varki et al . , 2009 ) . The interstitial fibrosis etiology prevalent in chimpanzees is also the most frequently diagnosed form of CVD in captive bonobos , gorillas and orangutans ( Lowenstine et al . , 2016 ) . It is unclear whether these apparent differences in susceptibility between species are due to genetic or environmental factors . A consequence of myocardial ischemia , the reduction of blood flow to the heart tissue , is oxygen deprivation . Oxygen sensing and response is an essential process across species . If the balance between anti-oxidants and pro-oxidants , such as reactive oxygen species ( ROS ) , is decoupled , redox signaling is disrupted , and oxidative stress ensues . The heart is the most oxygen-demanding tissue in the body ( Giordano , 2005 ) . Maintenance of oxygen homeostasis is essential for cardiac function as imbalance of ROS can result in myocardial infarction and heart failure . Indeed , 20–40 min of oxygen deprivation results in irreparable damage to the human heart ( Bretschneider et al . , 1975 ) . It is well appreciated that CVD is a complex disease with many contributing genetic and environmental factors . These effects are difficult to distinguish in in vivo studies within and between species because , in order to establish clear causal relationships and mechanism , directed perturbation is required . This is infeasible in humans and other apes due to practical and ethical considerations . More tractable model organisms such as mice are not optimal models of CVD given the differences in relative heart size and heart rate ( Doevendans et al . , 1998; Milani-Nejad and Janssen , 2014 ) . In addition to genome-level differences between humans and mice , the electrophysiology of mouse cardiomyocytes differs substantially from that of human cardiomyocytes ( Moretti et al . , 2013 ) . To understand intrinsic gene regulatory processes in cell types relevant to human disease , one might have to study human cells . Primary human cardiac cells are not easy to access , and have a limited lifespan in cell culture . The advent of induced pluripotent stem cell ( iPSC ) technology now allows us to access disease-relevant cell types across human individuals and other primates , control the extracellular environment of these cells in culture , and test the effects of perturbation . We have recently established a panel of human and chimpanzee iPSC lines ( Gallego Romero et al . , 2015 ) , and we have shown that cardiomyocytes derived from induced pluripotent stem cells ( iPSC-CMs ) can effectively model gene regulation in hearts from humans and chimpanzees ( Pavlovic et al . , 2018 ) . iPSC-CMs can be used to study CVD phenotypes including channelopathies such as long QT syndrome , and dilated cardiomyopathies such as Barth syndrome ( Tanaka et al . , 2015 ) . Cardiomyocytes make up 70–85% of the heart volume , 30–40% of the total cellular composition ( Pinto et al . , 2016; Zhou and Pu , 2016 ) , and are susceptible to ischemia following coronary artery occlusion . In order to gain insight into human gene regulatory adaptation in the heart , and the evolution of disease susceptibility and resistance , we developed a model of hypoxia and re-oxygenation in human and chimpanzee iPSC-CMs . This cell culture-based system enables an in-depth characterization of the inter-species response to , and recovery from , hypoxic stress . We can now determine intrinsic inter-species regulatory differences in response to a universal cellular stress , which could provide insight into the observed phenotypic differences in the manifestation of CVD between species . Hypoxia induces a transcriptional response following stabilization of the HIF transcription factors under conditions of low oxygen ( Samanta and Semenza , 2017 ) . We therefore determined both the global transcriptional response to hypoxia and re-oxygenation by RNA-seq , and the cellular response by measuring features of oxidative damage including lipid peroxidation and DNA damage , cytotoxicity , and cytokine release , in both species . While an iPSC-derived cardiomyocyte-based system has been previously used to study the effects of hypoxia in a single human and a single rhesus macaque individual , here we use a panel of human and chimpanzee individuals to identify a set of conserved and species-specific response genes ( Zhao et al . , 2018 ) . The identification of inter-species gene regulatory differences allowed us to develop hypotheses about molecular mechanisms that might explain phenotypic differences between species . We used RNA sequencing to characterize gene expression levels in all conditions , and study the regulatory response to hypoxia in the human and chimpanzee cardiomyocytes ( see Materials and methods ) . We processed the samples using a study design that was balanced with respect to a number of recorded potential technical confounders ( Supplementary file 1-Table S1 ) . Following sequencing of the RNA , we mapped reads to primate orthologous exons ( Figure 2—figure supplement 1 ) , and filtered lowly-expressed genes to yield a final data set with expression measurements for 11 , 974 genes ( see Materials and methods ) . Within each condition , inter-species correlation in read counts is somewhat lower than intra-species variation , as expected ( median Spearman’s correlation when comparing human samples = 0 . 97 , when comparing chimpanzee samples = 0 . 98 , and for human vs . chimpanzee samples = 0 . 92; Figure 2—figure supplement 2 ) . Using the RNA-seq data we confirmed that genes known to be expressed in cardiomyocytes are expressed in both our human and chimpanzee samples , including genes involved in cardiac structure , ion channels , and adrenoreceptors ( Figure 2—figure supplement 3 ) . As mentioned above , we included in our study differentiation replicates from a subset of samples ( see Supplementary file 1-Table S2 for details ) . We expect that gene expression data from pairs of replicates should be more similar to each other than to data from any other individual . We used this expected property of the data to account and correct the entire data set for unwanted technical variation ( see Materials and methods for more details ) . After accounting for unwanted variation , samples cluster by species and then by individual or condition ( Figure 2—figure supplement 4 ) . We note that one technical factor , the presence or absence of episomal reprogramming vectors ( three human samples tested positive; Figure 2—figure supplement 5 ) , remains a partial confounder with species . However , we confirmed that our conclusions are robust with respect to the inclusion of these three human samples ( Figure 2—figure supplement 6 and Supplementary file 1-Table S3 ) . To determine which genes respond to hypoxia , we analyzed the data from all four conditions using the framework of a linear model . The model included fixed effects for ‘species’ , ‘condition’ , a ‘species by condition interaction’ , a random effect for ‘individual’ , and four unwanted factors of variation as covariates ( see Materials and methods ) . For this analysis , we randomly selected one of each of the samples we had replicate data for . We were first interested in classifying genes into the following four categories within each species independently: genes that respond to hypoxia , genes that respond to short-term ( 6 hr ) re-oxygenation following hypoxia , genes that respond to long-term ( 24 hr ) re-oxygenation following hypoxia , and genes that differ between long-term re-oxygenation and baseline normoxia . Of 11 , 974 expressed genes , we identified ~4 , 000 genes that respond to hypoxia at 10% FDR in each species , and a slightly higher number of genes whose expression has changed upon re-oxygenation ( Figure 2; the results of all tests are in Supplementary file 1-Table S3A ) . We then focused on inter-species gene expression differences within each condition , independently , and found that roughly half of all expressed genes are differentially regulated between species , regardless of the condition ( at FDR of 10%; Supplementary file 1-Table S3A , Figure 2—figure supplement 7A ) . Using this approach we were also able to identify hundreds of genes that are differentially expressed between species exclusively in a single condition , for example following hypoxia ( Figure 2—figure supplement 7B ) . However , this approach does not provide strong evidence for true differences in the dynamic response to hypoxia between humans and chimpanzees , because of incomplete power to detect inter-species differentially expressed genes in any given condition . Thus , in order to determine the species-specificity of the global response to changing oxygen conditions we explicitly compared the effect size of expression change between pairs of conditions , for all genes , across species . Overall , there is a strong correlation in the global gene expression response to both hypoxia and re-oxygenation in humans and chimpanzees ( median Spearman correlation = 0 . 78; sign test p<10−4 for all comparisons; Figure 3 ) , suggesting that the general response to changes in oxygen level is conserved in the two species . Genes that respond to either hypoxia or re-oxygenation in both species include VEGFA ( a known hypoxia response gene ) , TRPV1 ( implicated in ischemia-reperfusion injury in the heart [Wang and Wang , 2005] ) , and DDX41 ( implicated in stress survival regulation [Shih and Lee , 2014] ) . The observation that the response to changes in oxygen level is generally conserved in the two species notwithstanding , we next focused on dynamic inter-species differences in our study . To do so , we used two approaches . First , we estimated a gene-specific interaction effect between species using the framework of the linear model described above . We identified 147 genes that responded to hypoxia in one species but showed little or no response in the other species , or that responded in both species but showed the opposite direction of effect ( at FDR of 10%; Figure 4; Supplementary file 2 ) . We did not find inter-species differences in the response to either the short or long re-oxygenation treatments . We did not find enrichment of particular pathways among the species-specific response genes , but this may not be surprising as stress response pathways are often regulated by a small number of key master regulator genes ( Haynes et al . , 2010; Li et al . , 2011; Natarajan et al . , 2013; Mahat et al . , 2016; Quirós et al . , 2017 ) . However , several of the genes with significant species by condition interactions have functions related to G-protein signaling , TGF-β signaling , and metabolism ( Figure 4—figure supplement 1 ) . Our most significant interaction corresponds to the RASD1 gene , which is up-regulated specifically in humans following hypoxia ( Figure 4 ) . This gene encodes a Ras GTPase , which activates G-protein signaling . Conversely , the LRRC25 gene responds to hypoxia specifically in chimpanzees , and has been found to inhibit NF-κβ signaling ( Feng et al . , 2017 ) ( Figure 4 ) . Directly modeling interaction effects with small numbers of samples is an underpowered approach . In order to side-step the challenge of incomplete power when performing multiple pairwise comparisons , we used a second approach; a joint Bayesian model , to classify genes based on their expression levels between conditions within each species during the course of the hypoxia-re-oxygenation experiment ( see Materials and methods ) . Four gene clusters were empirically determined to explain the predominant expression patterns in the data ( lowest BIC and AIC after testing 1–15 clusters; Figure 5—figure supplement 1 ) . Using this approach we categorized 9 , 414 genes as not responding to hypoxia in either humans or chimpanzees ( non-response genes ) , 1 , 920 genes that respond to hypoxia in both species ( conserved response genes ) , 430 genes that respond to hypoxia in chimpanzees only ( chimpanzee-specific response genes ) , and 199 genes that respond to hypoxia only in humans ( human-specific response genes; Figure 5 and Supplementary file 3 ) . It is notable that there is no prevalent pattern of genes responding specifically to re-oxygenation in either species , which suggests that the expression of most genes returns to baseline by the end of the experiment . We do not identify additional gene expression patterns even when we increase the number of clusters . To confirm that our approach identifies meaningful response genes , we considered the overlap of genes assigned to our four response categories with a set of genes previously identified to respond to hypoxia in human , and a more evolutionary distant primate , the rhesus macaque ( Zhao et al . , 2018 ) . As expected , genes previously found to respond to hypoxia are enriched among genes assigned to the ‘conserved response’ category in our study , and depleted among genes assigned to the ‘non-response’ category ( Chi-squared test; p<10−15 in both; Figure 6—figure supplement 1 ) . To explore properties of the response genes , we integrated our gene expression data with data from human chromatin immunoprecipitation followed by high through-put sequencing ( ChIP-seq ) experiments for three transcription factors that are known to bind to the genome in response to altered oxygen levels - HIF1α , HIF2α and FOXO3 ( Schödel et al . , 2011; Eijkelenboom et al . , 2013 ) . We arbitrarily designated genes as potentially regulated by the three transcription factors by identifying the closest orthologous gene to each human transcription factor-bound region . Consistent with previous literature ( Samanta and Semenza , 2017 ) , the 356 HIF1α-bound regions , and 301 HIF2α-bound regions are enriched near conserved response genes compared to non-response genes ( Chi-squared test; p<10−10 for both factors; Figure 6A ) . We thus asked whether differences in HIF binding could account for inter-species gene expression differences . Indeed , both HIF1α- and HIF2α-bound regions are enriched near conserved response genes compared to species-specific response genes ( p<0 . 01 for both factors; Figure 6A ) . In particular , chimpanzee-specific response genes are depleted for HIF binding ( p<0 . 05 for both factors; Figure 6A ) . The 934 FOXO3-bound regions are not enriched near conserved response genes compared to non-response genes , nor are conserved response genes enriched compared to species-specific response genes . We asked whether differences in sequence conservation at transcription factor-bound regions are associated with inter-species gene expression differences . To do so , we calculated the phyloP score at each bound region in close proximity to expressed genes . We found no difference in sequence conservation at HIF1α- , HIF2α- and FOXO3-bound regions when comparing conserved response genes to non-response genes , or conserved response genes to species-specific response genes ( Figure 6B ) . However , it should be noted that the numbers of species-specific response genes in proximity to binding sites for these transcription factors is small . In addition to transcription factor-mediated gene expression responses to stress , non-coding transcripts can also play a regulatory role in hypoxia , immune responses , and cardiac development and disease ( Scheuermann and Boyer , 2013; Choudhry et al . , 2014; Danko et al . , 2018 ) . To characterize the contribution of non-coding transcripts to the hypoxic response , we classified our 11 , 974 expressed genes as protein-coding , antisense , or long interspersed non-coding RNA ( lincRNA; see Materials and methods ) . The 205 antisense transcripts are enriched in conserved response genes compared to non-response genes ( 49 vs . 142; p=0 . 001; Figure 6C ) . However , species-specific response genes are no more likely to be antisense transcripts than conserved response genes . Conserved response genes are no more likely to be one of 59 annotated lincRNAs than non-response genes , and species-specific response genes are no more likely to be lincRNAs than conserved response genes . That said , considering the more stringent set of species-by-condition interaction genes , we found enrichment of lincRNAs compared to all expressed genes ( 4/147 interaction genes; Fisher’s exact test , p=0 . 007 ) . These lincRNAs are APTR , NEAT1 , RNF139-AS1 and LINC02615 . Finally , we wanted to determine whether response categories are enriched for particular pathways , using a background set of all expressed genes . In the non-response gene category , there is a significant enrichment in KEGG pathways related to the heart ( e . g . dilated cardiomyopathy , hypertrophic cardiomyopathy , arrhythymogenic right ventricular cardiomyopathy and adrenergic signaling in cardiomyocytes , 10% FDR; Figure 6—figure supplement 2 ) . In the conserved response category , various signaling pathways related to sensing the external environment , and responding to oxygen are significantly enriched including HIF1α , MAPK and FOXO1 . There are no significantly enriched pathways in the species-specific gene response categories . Given the apparent enrichment of cardiovascular genes in the non-response category , we explicitly tested the contribution of a set of cardiovascular-associated genes to the response to hypoxia ( see Materials and methods ) . Indeed , we found that there is a depletion of genes implicated in cardiovascular development and disease amongst the genes that respond to hypoxia in both species ( Chi-squared test; p=8 . 3×10−6; Figure 6D ) . It has been suggested that genetic variants that associate with gene expression levels ( eQTLs ) , may mediate disease phenotypes ( Emilsson et al . , 2008; Albert and Kruglyak , 2015; Yao et al . , 2015; GTEx Consortium et al . , 2018 ) . In order to test the contribution of eQTLs to the response to stress , a phenotype that is likely to provide insight into disease , we overlapped our four response gene categories with genes whose expression level is associated with genetic variants in human heart tissues ( eGenes; see Materials and methods ) . We observe a depletion of eGenes in the conserved response category , when compared to all expressed genes within each category , using data from the heart left ventricle ( Chi-squared test; p=0 . 01 ) , heart right atrial appendage ( p=1 . 4×10−5 ) , and iPSC-derived cardiomyocytes ( p=6 . 4×10−5; Figure 7A and Figure 7—figure supplement 1 ) . iPSC-derived cardiomyocytes consist mainly of ventricular-like cells . We therefore focused on eQTLs identified in the heart left ventricle . The depletion of eGenes corresponds to a difference in the contribution of eGenes to the non-response and conserved response categories ( p<10−15; Figure 7C ) . This observation is further supported by the fact that the absolute effect size of eQTLs , measured by allelic fold change , is significantly lower in the conserved response category compared to the non-response category ( p=0 . 001; Figure 7B ) . We confirmed that there is no correlation between the eQTL effect size , and gene expression level in either response category ( Figure 7—figure supplement 1B ) . The pattern of depletion of eGenes among conserved response genes is also observed across 12 other tested tissues; however the magnitudes of the effect differ between tissues ( Figure 7—figure supplement 1C ) . We found that human-specific response genes are also depleted of eGenes when we considered all eGenes identified in at least 1 of the 14 tested tissues ( p=7 . 1×10−5 ) , while chimpanzee-specific response genes are neither enriched nor depleted . Conversely , when we compare heart left ventricle eGenes to all expressed genes , we find that eGenes are depleted for conserved response genes ( p=0 . 02 ) . We next considered human gene expression response effect sizes , independent of response classification , in eGenes and non-eGenes . In accordance with the aforementioned findings , there is a lower absolute log fold change in expression in response to hypoxia in eGenes compared to non-eGenes ( p<0 . 002; Figure 7—figure supplement 2 ) . As we observed depletion of eQTLs found in healthy individuals among the conserved response genes in our study , we next considered eQTLs found among CVD patients . To do so , we investigated the contribution of eQTLs identified in left ventricle heart tissue from patients undergoing aortic valve replacement surgery pre- and post-cardioplegic arrest and ischemia , to our hypoxia response categories ( Stone et al . , 2019 ) . Again , we observed a depletion of eGenes in the conserved response category for both pre- and post-ischemia eGenes ( p=0 . 03 and p=0 . 006 respectively; Figure 7—figure supplement 3A ) . Interestingly , when we considered genes that are differentially expressed between pre- and post-ischemia samples , we observed the opposite pattern to that of the eGenes ( i . e . differentially expressed genes are enriched in the conserved response category , and depleted in the non-response category ) ; however the effect is not significant ( p=0 . 08 for non-response genes and p=0 . 09 for conserved response genes; Figure 7—figure supplement 3B; see Materials and methods ) . Differences in the gene expression response to hypoxia in eGenes compared to all expressed genes might imply differences in genetic tolerance to stress . We therefore overlapped our conserved response genes and non-response genes with genes associated with different levels of tolerance to mutation . When considering a set of 3 , 382 genes designated as loss-of-function intolerant in humans ( Lek et al . , 2016 ) , we found that there is an enrichment of these genes in the conserved response category ( p=0 . 02 , Figure 7D; see Materials and methods ) , and depletion in the non-response category ( p=0 . 005 ) . Similarly , the probabilities of loss-of-function intolerance ( pLI ) scores for all genes in the conserved response category are significantly higher than those of genes within the non-response category ( p=2×10−5 , Figure 7—figure supplement 4 ) . We then reasoned that 4 , 523 GWAS-associated genes are likely to be somewhat more tolerant to mutation than loss-of-function intolerant genes . Indeed , we found that there is no difference in the enrichment of these genes between the conserved response and non-response categories ( Figure 7D ) . We attempted to gain additional comparative insight in our system by characterizing cellular phenotypes that might relate to disease . First , we determined sensitivity to oxygen deprivation by measuring the level of cytotoxicity during the course of the experiment . A hallmark of cellular toxicity is the permeabilisation of the outer cellular membrane resulting in the release of intracellular components into the surrounding milieu . The activity of the lactate dehydrogenase ( LDH ) enzyme , which interconverts pyruvate and lactate , can be measured in the cell culture media as a proxy of this process . We observed a marginal yet significant increase in LDH activity following hypoxia in humans , and a significant increase following short-term re-oxygenation in both species ( Student’s t-test; p<0 . 05; Figure 8A ) . A significant increase is only observed in humans following hypoxia , and long-term re-oxygenation ( human Bmean = 0 . 61 , human Dmean = 4 . 82 , chimpanzee Bmean = 0 . 24 , chimpanzee Dmean = 2 . 49; Student’s t-test; p<0 . 05 for human A vs . B and B vs . D ) . Despite these apparent within-species differences in response to hypoxia , there is no significant difference in LDH activity between species within a condition . Second , we asked whether oxygen deprivation experienced by cardiomyocytes results in the secretion of cytokine signaling molecules , which could imply downstream consequences on other cell types in the heart such as cardiac fibroblasts . The TGFβ−1 cytokine mediates the development of fibrosis following stress in the heart ( Liu et al . , 2017 ) . We therefore measured secreted TGFβ−1 by ELISA for four individuals in each species . We found that TGFβ−1 release was significantly increased following re-oxygenation after hypoxic stress in both species ( p<7×10−3 ) ; however there is no difference between species under any condition ( Figure 8B ) . The results from these cellular assays support the gene expression data indicating a generally conserved response to hypoxia across species . Conserved hypoxic response genes correspond to signaling pathways related to oxidative stress and hypoxia including the FOXO1 and HIF1 signaling pathways . However , genes responding to hypoxia are significantly depleted for known cardiovascular-associated genes , suggesting that hypoxic stress response genes are expressed , and active in multiple tissues . This is supported by the fact that we observe that conserved response genes are depleted for eGenes identified in the heart as well as in other tissues . These results suggest that there is less tolerance for genetic variability that results in variation in the expression of genes that are necessary for eliciting a response to stress . Indeed , while eGenes are depleted in conserved response genes , genes identified to be intolerant to loss of function , are enriched in conserved response genes . GWAS-associated genes , likely to have individually small effects on a trait , and therefore more tolerance to mutation , are neither enriched nor depleted in conserved response genes . Cellular stress , including oxidative stress , can contribute to cellular damage and lead to disease pathology ( Giordano , 2005; Sack et al . , 2017 ) . A common notion is that genetic variants that modulate gene expression levels ( eQTLs ) are important in mediating disease phenotypes ( Emilsson et al . , 2008; Albert and Kruglyak , 2015; Yao et al . , 2015; GTEx Consortium et al . , 2018 ) . Thousands of genetic variants that associate with various phenotypes , including disease presentation , have been identified through genome-wide association studies ( GWAS ) . Given that most of these variants are located within non-coding regions of the genome , it is thought that integrating GWAS data with eQTL data will help to identify genes that are relevant to the phenotype of interest ( Hormozdiari et al . , 2016; Zhu et al . , 2016 ) . The observation that up to half of GWAS-identified variants are also eQTLs in at least one tissue ( Battle et al . , 2017 ) , provided some measure of support for this notion . However , there are several lines of evidence , which indicate that the relationship between eQTLs and complex disease may be relevant mainly to diseases that manifest late in life , and are therefore unlikely to have an effect on fitness: ( i ) Highly constrained genes with missense and protein-truncating variants , that might be expected to contribute to disease , are depleted for eQTLs but enriched for GWAS variants ( Lek et al . , 2016 ) . ( ii ) Most eQTLs are shared across a large number of tissues and are expected to have broad functional effects , which are therefore unlikely to be highly deleterious ( Battle et al . , 2017 ) . ( iii ) eQTLs identified in other primates are often also eQTLs in human ( Tung et al . , 2015; Jasinska et al . , 2017 ) suggesting that , across species , eGenes can tolerate the accumulation of associated mutations , which perturb their regulation . Together , these findings suggest that most eQTLs may be neutral . In this study , we performed perturbation experiments to determine the consequences on gene expression across species . Our results , which show a depletion of eQTLs in genes that respond to hypoxic stress in cardiomyocytes across species , suggest that eQTLs alone may not give immediate insight into stress phenotypes associated with cardiovascular disease . Genetic variants that modulate gene expression levels only in response to direct perturbation ( response QTLs ) are more likely to be informative in disease . Indeed , the association between GWAS variants is more pronounced in response QTLs than naïve QTLs ( Barreiro et al . , 2012; Alasoo et al . , 2018 ) . There are currently a limited number of data sets that allow for a systematic investigation of this association . However , using an available data set of heart tissue from CVD patients , we again observed a depletion of eQTLs in stress response genes . Our results suggest that eQTLs correspond to a set of genes that are largely distinct from genes , which respond to stress , or which are relevant to disease . Indeed , genes that are differentially expressed between pre- and post-ischemia samples show the opposite pattern to that of eQTLs in our data that is they are enriched in genes that respond to stress across species . While many eQTLs act in cis , it has been suggested that trans-eQTLs are more likely to associate with complex traits ( Westra et al . , 2013; Battle et al . , 2017 ) . However , we are currently underpowered to confidently identify these variants and determine their relevance to stress and disease . In addition to quantifying gene expression levels in response to hypoxia , we measured cellular stress phenotypes in our comparative cardiomyocyte system . The baseline level of DNA oxidation damage is similar in humans and chimpanzees , and increases following recovery from hypoxia in both species . The baseline level of lipid peroxidation is also similar in humans and chimpanzees , and shows a trend towards increased levels during the course of the experiment in both species; however this increase is only significant in chimpanzees . These findings are in line with a study of oxidative stress markers in blood from ten male humans and ten male chimpanzees , which showed that there is no significant difference in the levels of 8-OHdG between species , but there is significantly elevated 8-iso-PGF2α levels in chimpanzees compared to humans ( Videan et al . , 2009 ) . Cytoxicity and cytokine release also increase during the course of the experiment in both species . Another comparative study on the effects of hypoxia in human and rhesus macaque cardiomyocytes demonstrates that the secreted metabolome is highly correlated between species after 24 hr of hypoxia ( Zhao et al . , 2018 ) , suggesting an additional layer of regulatory conservation . Although the overall correlation in the response to hypoxia is high between humans and chimpanzees , a stringent interaction analysis identified 147 genes with species-specific expression in the hypoxic condition . The most significant species-specific response gene is RASD1 , which is similarly expressed in humans and chimpanzees in normoxic conditions but is significantly up-regulated after hypoxia only in humans . RASD1 was found to be up-regulated in samples from patients with ischemic disease compared to patients with dilated cardiomyopathy ( heart damage despite normal blood flow ) , and non-failing hearts ( Liu et al . , 2015 ) . These results suggest that aberrant RASD1 expression could be specifically involved in the response to oxygen deprivation , and the pathogenesis of ischemic heart disease . The RAI1-PEMT-RASD1 region is a replicated , genome-wide significant locus for coronary artery disease ( CAD ) ( McPherson and Tybjaerg-Hansen , 2016 ) . It is unclear what pathway , related to the CAD phenotype , is affected by the RASD1 locus ( Khera and Kathiresan , 2017 ) , and further experimentation specifically focused on this gene is beyond the scope of the current study . That said , several previous observations are also consistent with the notion of a relationship between RASD1 expression and the response to oxygen . The SNP in the RASD1-PEMT-RAI locus that is associated with CAD is not associated with various vascular-related traits ( Schunkert et al . , 2011 ) ; yet it is associated with ischemic stroke , with the same direction of effect ( Dichgans et al . , 2014 ) , suggesting a potential role for oxygen deprivation . It was also previously reported that many CAD case participants in GWAS studies have suffered from myocardial infarction , which results in myocardial ischemia and hypoxia , and that there is evidence for the SNP-GWAS association in the myocardial infarction sub-phenotype of CAD ( Schunkert et al . , 2011 ) . Moreover , the SNP is an eQTL for RASD1 expression in monocytes ( Emilsson et al . , 2008; Schunkert et al . , 2011 ) , suggesting that it has the potential to influence RASD1 expression in the right context . Finally , while RASD1 expression is induced following hypoxia in humans only , the RAI1 gene within this locus responds to hypoxia in both species , suggesting a link between hypoxia and this locus ( PEMT expression doesn’t significantly change in either species upon hypoxia ) . In addition to RASD1 , there are four other CAD-associated genes ( MRPS6 , SWAP70 , SNF8 and TRIB1 ) that respond to hypoxia in a species-specific manner . MRPS6 , another human-specific response gene , encodes a mitochondrial ribosomal protein likely important in the translation of mitochondrial mRNAs necessary for oxidative phosphorylation . RASD1 is an activator of G-protein signaling ( Cismowski et al . , 2000 ) , and is thought to contribute to the stress response ( Sato and Ishikawa , 2010 ) . Indeed , G-proteins are important sensors of the environment at the cell membrane and mediate a signaling cascade to initiate an intra-cellular response to external stimuli . RASD1 is one of several species-specific response genes related to G-protein signaling; other examples include ARRDC2 , RASL11B , ARL6 RASSF1 , GTPBP4 , RAB3A , RHOF , SYDE2 , and CNKSR1 . In fact , there are several genes that respond in a species-specific manner , which belong to similar pathways , or perform similar functions . For example , multiple genes ( ACVR2A , SNIP1 , JUNB , SMAD4 , TGFBR2 , SMAD6 , FGF9 ) are related to TGF-β signaling . TGF-β is induced following myocardial infarction , and mediates the development of fibrosis ( Bujak and Frangogiannis , 2007 ) . While we did not observe differences in the secretion of TGFβ−1 , it is tempting to speculate that differences in the expression of these genes under oxygen stress , could contribute to the fibrotic heart phenotype observed in chimpanzees . Several of these genes have been implicated in heart physiology and disease suggesting that they could be relevant to the phenotype ( Galvin et al . , 2000; Wang et al . , 2005; Alfonso-Jaume et al . , 2006; Tseng et al . , 2009; Itoh et al . , 2016; Lu et al . , 2016; Dogra et al . , 2017 ) . Many species-specific response genes are involved in post-transcriptional layers of the gene regulatory cascade including RNA modifications ( METTL14 ) , RNA folding ( DDX20 ) , splicing ( CCDC49 ) , nuclear-cytoplasmic transport ( NUP214 ) , and protein degradation ( CBLL1 , UBQLN4 , TRIM13 , DNAJB2 ) . This suggests that additional inter-species differences may emerge in processes downstream of transcription . For example , METTL14 deposits the N6-methyladenosine RNA modification , which has been implicated in the stabilization of mRNA molecules following hypoxia ( Fry et al . , 2017 ) . Intriguingly , differences in N6-methyladenosine levels have been reported between primates ( Ma et al . , 2017 ) . Alternative splicing is another mechanism for inter-species differences between humans and chimpanzees , and , interestingly , one gene that undergoes differential splicing between species is the GSTO2 gene , which is protective against oxidative stress ( Calarco et al . , 2007 ) . Genes that respond to oxygen deprivation in a species-specific manner could have phenotypic consequences as rapid changes in gene expression in response to stress can lead to evolutionary adaptation ( López-Maury et al . , 2008; de Nadal et al . , 2011 ) . This mechanism has been implicated in mediating inter-species differences in epithelial cancer incidence as coordinated gene expression differences between humans and chimpanzees have been observed in fibroblasts subjected to serum starvation ( Pizzollo et al . , 2018 ) . Conversely , it has been suggested that species-specific gene responses to stress across divergent yeast species may be adaptive , or , more likely , compensated by the response of related genes , or reflective of biological noise ( Tirosh et al . , 2011 ) . The latter could explain the overall inter-species similarity in the cellular and transcriptional response , and the fact that genes peripherally related to similar pathways show species-specific differences . We performed our experiments using an in vitro iPSC-CM model . To characterize the potential relevance of our observations to in vivo systems , we considered the results of Pavlovic et al . , who characterized regulatory differences between primary heart tissues and iPSC-CMs in humans and chimpanzees ( Pavlovic et al . , 2018 ) . Out of 2 , 459 genes that respond to hypoxia in one or both species based on our data , only 371 ( 16% ) were found to be differentially expressed between hearts and iPSC-CMs in Pavlovic et al . Similarly , 34 of the 147 species-by-condition interaction genes in our data ( 23% ) were found to be differentially expressed between hearts and iPSC-CMs . In addition , we find that there is no significant difference in the proportions of conserved response genes and non-response genes in the overlap with genes that are differentially expressed between heart tissue and iPSC-CMs . Similarly , there is no significant difference in the proportion of conserved response and species-specific response genes that overlap with genes differentially expressed between heart tissue and iPSC-CMs . Put together , this analysis suggests that there is no systematic bias in our cell culture system . It is important to note that following myocardial infarction and ischemia , highly metabolically active cardiomyocytes undergo rapid cell death thereby initiating a cascade of events including an inflammatory response by immune cells , cardiac fibroblast activation , and fibrosis ( Frangogiannis , 2014 ) . Our study was designed to measure the primary response to oxygen deprivation in the heart at the level of cardiomyocytes . Given the high mitochondrial content of cardiomyocytes , these cells are likely to be more susceptible to oxidative stress than other cell types . Indeed , cardiomyocytes are more sensitive to superoxide radicals than cardiac fibroblasts ( Li et al . , 1999 ) . However , our system was not able to capture secondary effects on extracellular matrix remodeling and fibroblast proliferation ( Ugolini et al . , 2017 ) , which may differ between species and also contribute to the different disease phenotypes . We attempted to measure secondary disease processes by assaying TGFβ−1 secretion by cardiomyocytes , but did not find a significant difference in the release of this factor by cardiomyocytes between species . In summary , to date there have been few well-powered studies investigating the evolution of the stress response in primates . Here we measured the genome-wide transcriptional response to a universal cellular stress , oxygen deprivation , across species in a CVD-relevant cell type . We find that the cellular and transcriptional response is largely similar across species; however there are hundreds of genes that respond in a species-specific manner . We used eight biological replicates ( individuals ) from human , and seven from chimpanzee . In addition , technical replicates ( independent cardiomyocyte differentiation and oxygen stress experiments ) from three human and three chimpanzee individuals were used to estimate unwanted factors of variation in the data . This number of biological and technical replicates is sufficient to be able to identify inter-species gene expression differences ( Gallego Romero et al . , 2015; Pavlovic et al . , 2018; Ward et al . , 2018 ) . All iPSC lines , from both species , were derived from fibroblasts using the same experimental design and reprogramming protocol as previously described ( Gallego Romero et al . , 2015 ) . Regardless , Gallego Romero et al . found the effects of different reprogramming protocols , population of origin , and originating cell types on inter-species DNA methylation and gene expression differences to be exceedingly small ( Gallego Romero et al . , 2015 ) . 13 iPSC lines have been described and characterized previously ( Gallego Romero et al . , 2015; Burrows et al . , 2016; Pavlovic et al . , 2018; Ward et al . , 2018 ) . Two additional iPSC lines are first described and characterized in this study ( H22422 and H25237 ) . All lines tested negative for mycoplasma contamination . Feeder-independent iPSCs were maintained at 70% confluence on Matrigel hESC-qualified Matrix ( 354277 , Corning , Bedford , MA , USA ) at a 1:100 dilution . Cells were cultured in Essential 8 Medium ( A1517001 , ThermoFisher Scientific , Waltham , MA , USA ) with Penicillin/Streptomycin ( 30002 Cl , Corning ) at 37°C , 5% CO2 and atmospheric O2 . Cells were passaged every 3–4 days with dissociation reagent ( 0 . 5 mM EDTA , 300 mM NaCl in PBS ) , and seeded with ROCK inhibitor Y-27632 ( ab12019 , Abcam , Cambridge , MA , USA ) . iPSC-CM differentiations were largely performed based on the protocol described by Burridge et al . ( 2014 ) . Importantly , the same differentiation protocol was used in both species . iPSCs cultured for 10–50 passages were seeded in 4 × 10 cm Matrigel-coated culture dishes until 70–100% confluent ( Days −4/–3 ) . The optimum cell density for efficient differentiation depended on the individual iPSC line . On Day 0 , 6 μM of the GSK3 inhibitor , CHIR99021 trihydrochloride ( 4953 , Tocris Bioscience , Bristol , UK ) was added to the cultures in 12 ml Cardiomyocyte Differentiation Media [500 mL RPMI1640 ( 15–040 CM ThermoFisher Scientific ) , 10 mL B-27 Minus Insulin ( A1895601 , ThermoFisher Scientific ) , 5 mL Glutamax ( 35050–061 , ThermoFisher Scientific ) , and 5 mL Penicillin/Streptomycin ) ] , and a 1:100 dilution of Matrigel . 24 hr later , on Day 1 , fresh Cardiomyocyte Differentiation Media , supplemented with 6 μM CHIR99021 was added to the cultures . Chimpanzee iPSCs , in general , were more sensitive to the addition of CHIR99021 hence the reduction from the optimal 12 μM CHIR99021 for 24 hr as described in Burridge et al . , to 6 μM for 48 hr . The GSK3 inhibitor was removed with the addition of fresh Cardiomyocyte Differentiation Media 24 hr later on Day 2 . After 24 hr , on Day 3 , 2 μM of the Wnt signaling inhibitor Wnt-C59 ( 5148 , Tocris Bioscience ) , diluted in Cardiomyocyte Differentiation Media , was added to the cultures . The media was replaced with 2 μM Wnt-C59 in Cardiomyocyte Differentiation Media 24 hr later , on Day 4 . Cardiomyocyte Differentiation Media was replaced on Days 5 , 7 , 10 and 12 . Spontaneously beating cells appear on Days 7–10 . To remove non-cardiomyocytes from the cultures , iPSC-CMs were purified by metabolic selection . 10 mL of glucose-free , lactate-containing media ( Purification Media ) [500 mL RPMI without glucose ( 11879 , ThermoFisher Scientific ) , 106 . 5 mg L-Ascorbic acid 2-phosphate sesquimagenesium salt ( sc228390 , Santa Cruz Biotechnology , Santa Cruz , CA , USA ) , 3 . 33 ml 75 mg/ml Human Recombinant Albumin ( A0237 , Sigma-Aldrich , St Louis , MO , USA ) , 2 . 5 mL 1 M lactate in 1 M HEPES ( L ( + ) Lactic acid sodium ( L7022 , Sigma-Aldrich ) ) , and 5 ml Penicillin/Streptomycin] was added on Day 14 . Purification Media was replaced on Days 16 and 18 . On Day 20 , iPSC-CMs were dissociated with 4 mL 0 . 05% Trypsin-EDTA solution ( 25–053 Cl , ThermoFisher Scientific ) for ~10 min , and quenched with double the volume of Cardiomyocyte Maintenance Media [500 mL DMEM without glucose ( A14430-01 , ThermoFisher Scientific ) , 50 mL FBS ( S1200-500 , Genemate ) , 990 mg Galactose ( G5388 , Sigma-Aldrich ) , 5 mL 100 mM sodium pyruvate ( 11360–070 , ThermoFisher Scientific ) , 2 . 5 mL 1 M HEPES ( SH3023701 , ThermoFisher Scientific ) , 5 mL Glutamax ( 35050–061 , ThermoFisher Scientific ) , 5 mL Penicillin/Streptomycin] . This change in carbohydrate source shifts metabolism away from glycolysis towards aerobic mitochondrial respiration , which is the predominant pathway used to generate energy in adult cardiomyocytes in vivo . A single cell suspension was generated by straining the cells through a 100 μm nylon mesh cell strainer three times , and once through a 40 μm mesh strainer . 1 . 5 million iPSC-CMs were plated per well of a Matrigel-coated 6-well plate in 3 mL Cardiomyocyte Maintenance Media . iPSC-CMs for each of the four conditions were plated on separate 6-well plates . iPSC-CMs were matured in culture for a further 10 days . Cardiomyocyte Maintenance Media was replaced on Days 23 , 25 , 27 , 28 and 30 . While cells are typically cultured in vitro at atmospheric oxygen levels ( 21% O2 ) , this oxygen level is not experienced by mammalian cells in vivo - arterial oxygen levels are ~13% , and levels drop to 5–10% within tissues ( Brahimi-Horn and Pouysségur , 2007; Carreau et al . , 2011; Jagannathan et al . , 2016 ) . We therefore chose to culture our cells at 10% O2 , which falls within this physiological range . On Day 25 , iPSC-CMs cultured at atmospheric oxygen levels , were transferred to an oxygen-controlled incubator ( HERAcell 150i CO2 incubator , ThermoFisher Scientific ) representing physiological oxygen levels ( 10% O2 ) . Oxygen levels are maintained through displacement of oxygen by nitrogen . To allow further maturation , and to synchronize iPSC-CM beating , iPSC-CMs were pulsed with the IonOptix C-Dish and C-Pace EP Culture Pacer from Day 27 until the end of the oxygen perturbation experiment . Cells were pulsed at a voltage of 6 . 6 V/cm , frequency of 1 Hz , and pulse frequency of 2 ms . iPSC-CMs were dissociated with 0 . 05% Trypsin-EDTA solution and quenched with four times the volume of Cardiomyocyte Maintenance Media . In order to obtain a single cell suspension , cells were strained twice through a 100 μm strainer , and once through a 40 μm strainer . 1 million cells were stained with Zombie Violet Fixable Viability Kit ( 423113 , BioLegend ) for 30 min at 4°C prior to fixation and permeabilization ( FOXP3/Transcription Factor Staining Buffer Set , 00–5523 , ThermoFisher Scientific ) for 30 min at 4°C . Cells were stained with 5 μl PE Mouse Anti-Cardiac Troponin T antibody ( 564767 , clone 13–11 , BD Biosciences , San Jose , CA , USA ) for 45 min at 4°C . Cells were washed three times in permeabilization buffer and re-suspended in autoMACS Running Buffer ( 130-091-221 , Miltenyi Biotec , Bergisch Gladbach , Germany ) . Several negative controls were used in each flow cytometry experiment: 1 ) iPSCs , which should not express TNNT2 , 2 ) an iPSC-CM sample that has not been labeled with viability stain or TNNT2 antibody , and 3 ) an iPSC-CM sample that is only labeled with the viability stain . 10 , 000 cells were captured and profiled on the BD LSRFortessa Cell Analyzer . Several gating steps were performed to determine the proportion of TNNT2-positive cells: 1 ) Cellular debris was removed by gating out cells with low granularity on FSC versus SSC density plots , 2 ) From this population , live cells were identified as the violet laser-excitable , Pacific Blue dye-negative population , 3 ) Two populations of TNNT2-positive cells were identified within the set of live cells: one conservative gate selected high-intensity TNNT2-positive cells , and a lenient gate included TNNT2-positive cells with a range of intensities . Both gates were created so as to exclude any cells that overlap the profiles of the negative control samples . iPSC-CM purity is reported as the proportion of TNNT2-positive live cells . Values from the lenient threshold are reported in the main text . Values for the conservative and lenient thresholds are reported in Figure 1—figure supplement 3 and Supplementary file 1-Table S2 . Importantly , there is no difference in purity between species whether a conservative or lenient threshold is used . Several pilot experiments were performed to determine the optimal oxygen conditions to initiate a gene expression response in iPSC-CMs . The expression of stress response genes is induced within 6 hr of hypoxia at 1% oxygen ( Figure 1—figure supplement 4B ) , and cell damage ( lactate dehydrogenase activity ) is evident after 24 hr ( Figure 1—figure supplement 4C ) . It is noteworthy that culturing the iPSC-CMs in media with a glucose carbohydrate source , instead of galactose , does not elicit a stress response following hypoxia ( Figure 1—figure supplement 4 ) . Given that we were interested in determining the early transcriptional response to hypoxia , prior to the induction of cell death , we chose the 6 hr time-point in our subsequent experiments . Oxygen stress experiments were conducted on Day 31 or 32 after the initiation of differentiation . At the start of the experiment ( total elapsed time = 0 hr ) , one plate ( A ) remained at 10% O2 , while three plates ( B , C and D ) were transferred to an oxygen-controlled cell culture incubator set at 1% O2 . After six hours , plates A and B were harvested , while plates C and D were transferred back to 10% O2 ( elapsed time = 6 hr ) . Plate C was harvested six hours after the end of the hypoxic incubation ( elapsed time = 12 hr ) . Plate D was harvested 24 hr after the end of the hypoxic incubation ( elapsed time = 30 hr ) . Ten batches of oxygen stress experiments were performed . For each of the four conditions , iPSC-CMs were harvested by manual scraping , flash-frozen as cell pellets , and stored at −80°C , together with the cell culture media from each sample , until further processing . Oxygen levels in the cell culture media of a representative iPSC-CM sample were measured during the course of the oxygen perturbation experiment , in each experimental batch . An oxygen sensitive sensor was applied to the inner wall of a well of a 6-well plate ( SP-PSt3-NAU-D5-YOP , PreSens Precision Sensing GmbH , Regensburg , Germany ) . Oxygen levels were measured non-invasively through the wall of the cell culture plate using a Polymer Optical Fiber ( NWDV29 , Coy , Grass Lake , MI , USA ) , and a Fiber Optic Oxygen Meter ( Fibox 3 Transmitter NWDV16 , Coy ) . 8-OHdG levels were measured by competitive enzyme-linked immunoassay using the OxiSelect Oxidative DNA Damage ELISA Kit ( STA-320 , Cell Biolabs Inc ) . Levels were measured in 50 μl of cell culture media , in duplicate , according to the manufacturer’s instructions . Samples were processed on three species-balanced 96-well plates . 8-OHdG was quantified relative to a standard curve using 4- and 5-parameter logistic models implemented in the drc package in R . Final 8-OHdG release is reported as four measurements either relative to the basline condition A or hypoxic condition B: A ( A-A ) , B ( B-A ) , C ( C-B ) and D ( D-B ) . For the three individuals with replicate experiments in each species , mean values from both experiments are reported . Secreted 8-iso-PGF2α was measured by competitive enzyme-linked immunoassay using the OxiSelect 8-iso-Prostaglandin F2α ELISA kit ( STA-337 , Cell Biolabs Inc , San Diego , CA , USA ) . Levels were measured in 55 μl of cell culture media , in duplicate , according to the manufacturer’s instructions . Samples were processed on three species-balanced 96-well plates . 8-iso-PGF2α was quantified relative to a standard curve using 4- and 5-parameter logistic models implemented in the drc package in R . Final 8-iso-PGF2α release is reported as four measurements either relative to the baseline , condition A , or the hypoxic condition that is A ( A-A ) , B ( B-A ) , C ( C-B ) and D ( D-B ) . For the three individuals with replicate experiments in each species , mean values from both experiments are reported . Lactate dehydrogenase ( LDH ) activity levels were measured by colourimetric determination of NAD reduction to NADH using the Lactate Dehydrogenase Activity Assay Kit ( MAK066 , Sigma-Aldrich ) . Samples were processed on four species-balanced 96-well plates , and each sample was assayed in triplicate . 5 μl of cell culture media was assayed as per the manufacturer’s instructions . LDH activity is reported as the difference in NADH levels measured at the start of the enzymatic reaction , and 25 min after the addition of the substrate . Enzyme activity is calculated relative to a linear standard curve . Final LDH activity is reported as four measurements either relative to the baseline , condition A , or the hypoxic condition that is A ( A-A ) , B ( B-A ) , C ( C-B ) and D ( D-B ) . For the three individuals with replicate experiments in each species , mean values from both experiments are reported . TGFβ−1 levels were measured by enzyme-linked immunoassay using the TGF beta 1 Human ELISA Kit ( ab100647 , Abcam ) . Levels were measured in 100 μl of cell culture media , in duplicate , according to the manufacturer’s instructions . Four representative individuals from each species were assayed on one 96-well plate . TGFβ−1 levels were quantified relative to a standard curve using 4- and 5-parameter logistic models implemented in the drc package in R . Final TGFβ−1 release is reported as four measurements: A ( A-A ) , B ( B-A ) , C ( C-B ) and D ( D-B ) . RNA was extracted from ~1 . 5 million cells from 84 iPSC-CM samples representing 15 individuals . Extractions were performed in ten species-balanced batches using the ZR-Duet DNA/RNA extraction kit ( D7001 , Zymo , Irvine , CA , USA ) . All four conditions from one human and one chimpanzee individual were extracted per batch ( one batch had three individuals ) . RNA concentration and quality was measured using the Agilent 2100 Bioanalyzer . RIN scores were greater than 7 . 5 for all samples ( human median: 9 . 1 , chimpanzee median: 9 . 2 ) . RNA-seq libraries were prepared from 250 ng of RNA in three species-balanced batches using the Illumina TruSeq RNA Sample Preparation Kit v2 ( RS-122–2001 and −2002 , Illumina ) . Libraries in each batch were multiplexed together to generate four pools for sequencing . Each pool was sequenced 50 base pairs , single-end on the HiSeq4000 according to the manufacturer’s instructions . Pools 1 , 3 , 4 were sequenced on three lanes ( 24 samples per pool ) , and Pool two was sequenced on two lanes ( 12 samples in the pool ) . RNA-seq data quality was determined by FastQC ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . Sequencing adapters were trimmed , and sequencing reads from each species aligned to their respective genome ( hg19 or panTro3 ) using TopHat2 ( version 2 . 0 . 11 ) ( Kim et al . , 2013 ) . The number of mapped sequencing reads is similar across species and conditions ( median human A: 53 , 547 , 009 , median human B: 40 , 872 , 328; median human C: 40 , 635 , 553 , median human D: 42 , 041 , 844; median chimpanzee A: 42 , 258 , 525; median chimpanzee B: 33 , 054 , 882; median chimpanzee C: 28 , 792 , 150 , median chimpanzee D: 36 , 903 , 485 ) . In order to have a comparable set of genes from which to identify gene expression differences , we quantified gene expression levels at orthologous meta-exons from 30 , 030 Ensembl genes from hg19 , panTro3 and rheMac3 ( Blekhman et al . , 2010 ) using featureCounts within subread ( version 1 . 4 . 6 ) ( Liao et al . , 2014 ) . In order to compare gene expression profiles equivalently across individuals regardless of sex , genes were filtered to only include those on the autosomes . Log2-transformed counts per million were calculated using edgeR ( Robinson et al . , 2010 ) . Lowly-expressed genes were filtered such that only genes with a mean log2 cpm >0 across samples were retained . Prior to differential expression analysis , unwanted factors of variation were estimated in the RNA-seq data using RUVSeq ( Risso et al . , 2014 ) . As we have two replicate samples from six of the individuals ( three in each species ) , the RUVs function for estimating the factors of unwanted variation using replicate samples was used . This approach takes advantage of the fact that replicate samples have constant covariates of interest . We tested different numbers of unwanted factors of variation ( k values ) until the data clustered best by our biological factors of interest that is species , individual and condition . Four factors of unwanted variation were thus selected . To assess data quality we performed Principal Component Analysis ( PCA ) on the RUVs-normalised log2 cpm expression values . We correlated known biological and technical factors with the first six PCs . The TMM-voom-limma pipeline was used to identify differentially expressed genes between species and conditions . Filtered read counts from a randomly selected replicate were taken forward in this analysis . Normalization factors were used to scale the raw library size to the effective library size of each sample using the trimmed mean of M-values ( TMM ) implemented in edgeR ( Robinson et al . , 2010 ) . The mean-variance relationship was removed using precision weights in voom ( Law et al . , 2014 ) . A linear model was fitted for expression values of each gene using limma ( Smyth , 2004 ) :Y∼β0+βspeciesXspecies+βBXB+βCXC+βDXD+βspecies , BXspecies , B+βspecies , CXspecies , C+βspecies , DXspecies , D+XRUV1+XRUV2+XRUV3+XRUV4+I+ϵwhere β0 is the mean expression level of gene g for chimpanzee cells grown under normoxic conditions ( A ) , βspecies is the fixed effect for species , βB is the effect for condition B , βC is the effect for condition C , βD is the effect for condition D , βspecies , B is the fixed interaction effect of condition B and species , βspecies , C is the fixed interaction effect of condition C and species , βspecies , D is the fixed interaction effect of condition D and species . The four unwanted factors of variation determined by RUVs are modeled as covariates XRUV1 , XRUV2 , XRUV3 , XRUV4 , and I is the random effect for individual , which was implemented using the limma function duplicateCorrelation . In order to obtain more precise gene-wise variability estimates , we used empirical Bayes moderation , which takes information across all genes into account . We used contrast tests in limma to identify genes that are differentially expressed between conditions within each species , genes that are differentially expressed between species at each condition , and species-by-condition interactions for conditions B , C , and D . We corrected for multiple testing at each gene using the Benjamini and Hochberg false discovery rate ( FDR ) ( Benjamini and Hochberg , 1995 ) . Genes with FDR-adjusted p values of < 0 . 1 are considered to be differentially expressed . In order to cluster genes by their gene expression trajectories during the course of the experiment , all data was jointly modeled using a Bayesian Hierarchical model across pairwise differential tests implemented in the Cormotif R package ( Wei et al . , 2015 ) . Cormotif fits correlation motifs to multiple pairs of tests to identify differential expression patterns . To identify gene expression trajectories ( correlation motifs ) , TMM-normalised cpm values for each gene were compared across three pairs of conditions within each species ( 6 pairs of tests in total; 1: human A vs . B , 2: human B vs . C , 3: human B vs . D , 4: chimpanzee A vs . B , 5: chimpanzee B vs . C , 6: chimpanzee B vs . D ) . In order to select the best model to fit the data , we varied the number of correlation motifs ( 1 through 15 ) . The best fit was determined using the Bayesian information criterion ( BIC ) and Akaike information criterion ( AIC ) . The BIC and AIC were minimized when four correlation motifs were modeled . Cormotif calculates the posterior probability of differential expression for each gene , in each of the six pairwise tests . These values are plotted in the heatmap in Figure 5A . We used a threshold posterior probability of 0 . 5 to classify genes into each of the four correlation motifs as suggested by the authors . Motif 1 ( non-response ) : p<0 . 5 in tests 1 , 2 , 3 , 4 , 5 , 6; motif 2 ( chimpanzee-specific response ) : p<0 . 5 in tests 1 , 2 , 3 and p>0 . 5 in tests 4 , 5 , 6; motif 3 ( conserved response ) : posterior probability >0 . 5 in tests 1 , 2 , 3 , 4 , 5 , 6; motif 4 ( human-specific response ) : p>0 . 5 in tests 1 , 2 , 3 and p<0 . 5 in tests 4 , 5 , 6 . The set of 187 genes that were previously found to respond to hypoxia in both human and rhesus macaque iPSC-derived cardiomyocytes were overlaid with our four Cormotif gene expression response categories ( Zhao et al . , 2018 ) . 164 of the human-rhesus conserved response genes are expressed in our data . We calculated the proportion of human-rhesus conserved response genes within each of our four response gene categories ( value one for each category ) , and the proportion of all genes that belong to each of the four response categories in our data ( value two for each category ) . We subtracted value twofrom value one to obtain an enrichment score per response category . This score is multiplied by 100 and plotted in Figure 6—figure supplement 1 . A chi-squared test is used to determine whether there is a significant enrichment or depletion of human-rhesus response genes in each of our four response gene categories compared to all expressed genes in that category . This approach is used in all subsequent analyses investigating the properties of the four response categories . Human transcription start sites ( TSS ) were downloaded from the UCSC genome bowser Table Browser ( http://genome . ucsc . edu/cgi-bin/hgTables ) using ‘txStart’ from Ensembl genes ( Karolchik , 2004 ) . Each gene was assigned a single TSS based on the 5’ most transcript of genes on the sense strand , and the 3’ most transcript from genes on the anti-sense strand . The list of TSS was filtered to include only those representative of orthologous genes used in the gene expression analysis . We obtained published human ChIP-seq data sets for HIF1α ( 356 sites ) , HIF2α ( 301 sites ) ( Schödel et al . , 2011 ) , and FOXO3 ( 934 sites ) ( Eijkelenboom et al . , 2013 ) . HIF1α and HIF2α binding was measured in a breast cancer cell line following HIF stabilization with dimethyloxaloylglycine . FOXO3 binding was measured in a colon cancer cell line with a Tamoxifen-inducible FOXO3A3-ER fusion protein . Binding locations for each transcription factor were converted to hg19 coordinates . Each transcription factor binding location was assigned its closest gene using bedtools ( Quinlan and Hall , 2010 ) . Genes were stratified into each of the four response gene categories , and the proportion of genes within each class calculated as described previously . The conservation status of each transcription factor binding location was determined using the phyloP score ( Pollard et al . , 2010 ) implemented on the Galaxy platform ( Afgan et al . , 2018 ) . The ‘gene_biotype’ annotation associated with each Ensembl gene ID was obtained through biomaRt . Biotypes include lincRNA , antisense RNA , transcribed unitary pseudogenes , transcribed unprocessed pseudogenes , processed transcripts , and protein coding genes . We focused on four gene types: protein coding genes , lincRNA and antisense transcripts given that the other categories have fewer than 50 instances in our data set . We determined the proportion of each gene type in our four response categories as previously described . The gene sets belonging to each of the four response categories were investigated for common pathway enrichment using the KEGG database ( Kanehisa et al . , 2017 ) , within the DAVID genomic annotation tool ( Huang et al . , 2009b; Huang et al . , 2009a ) . Enrichment was calculated relative to the set of all 11 , 974 expressed genes . Multiple testing was performed by the Benjamini-Hochberg method . Pathways enriched at 10% FDR are considered to be significant . A set of 5 , 010 genes implicated in cardiovascular development or disease ( BHF-UCL gene association file ) was obtained from the Cardiovascular Gene Ontology Annotation Initiative ( https://www . ebi . ac . uk/GOA/CVI ) . The 2 , 756 genes that are expressed in our study were overlapped with our four response categories . We determined the proportion of cardiovascular genes in our four response categories as previously described . For the overlap of our response categories with eGenes from healthy individuals , the list of eGenes in 14 GTEx tissues was downloaded from v7 in the GTEx portal ( www . gtexportal . org ) . eGenes were selected at 5% FDR in each tissue . eGenes from iPSC-derived cardiomyocytes were obtained from Banovich et al . ( 2018 ) ( 10% FDR ) . The eQTL effect sizes of eGenes are defined and reported as allelic fold change by the GTEx consortium . For the overlap of our response categories with eQTLs identified pre- and post-ischemia , an RNA-seq study of 114 patients undergoing aortic valve replacement surgery was used ( Stone et al . , 2019 ) . In this study , samples of heart left ventricle were obtained for each individual pre- and post- cardioplegic arrest/ischemia . We obtained lists corresponding to all eGenes ( 5% FDR ) identified in males and females combined within the pre-ischemic state ( 496 expressed in our data ) , all eGenes in males and females combined in the post-ischemic state ( 416 expressed ) , and differentially expressed genes ( 5% FDR ) between pre- and post-ischemia ( 6 , 571 in 46 females , and 6 , 572 in 68 males ) . Genes that are differentially expressed between pre- and post-ischemia in both males and females ( 5 , 115 ) were used to calculate enrichment in hypoxia response categories . We determined the proportion of eGenes in our four response categories as previously described . Existing gene lists , related to genetic tolerance , were overlapped with our hypoxia response genes . The following gene lists were obtained from the Macarthur lab ( https://github . com/macarthur-lab/gene_lists ) : genes nearest to GWAS peaks ( MacArthur et al . , 2017 ) , loss-of-function intolerant genes , and pLI scores ( Lek et al . , 2016 ) . We determined the proportion of gene in each gene set in our four response categories as previously described . All RNA-seq data have been deposited in the Gene Expression Omnibus ( www . ncbi . nlm . nih . gov/geo/ ) under accession number GSE117192 .	Understanding why some people get heart disease and others do not could help scientists find better ways to treat or prevent the condition . Genetics likely plays a role , and one way to identify genes that are important for heart health is to compare genes in humans and their closest evolutionary relatives , the chimpanzees . Though it is not exactly the same as seen in humans , chimpanzees do get heart disease . Differences in the genes involved in heart disease in humans and chimpanzees may help explain what leads to the disease in humans . Studying heart disease in chimpanzees and humans has been challenging because heart tissue from humans and chimpanzees is hard to come by . Yet scientists can now convert easy-to-access skin cells from humans and chimpanzees into heart cells and grow them under laboratory conditions . Ward and Gilad have used exactly this approach to see how human and chimpanzee cells respond when they are starved of oxygen , which simulates a heart attack . First , skin cells collected from eight humans and seven chimpanzees were coaxed into becoming heart cells and grown in the laboratory . Ward and Gilad then compared the activity levels of about 12 , 000 genes in these heart cells when their oxygen was limited . The responses were very similar , with 1 , 920 genes switching on or off in both species . But the activity of hundreds of other genes differed between humans and chimpanzees . For example , a gene called RASD1 , which is known to be important in human heart disease , became active in oxygen-starved human cells but not in chimpanzee cells . Genes that vary in their activity between healthy human individuals are thought to be important in disease . However , Ward and Gilad found that the activity of genes that switch on or off in both species after oxygen starvation did not vary a lot in a collection of heart samples from hundreds of individuals . These experiments may help scientists narrow down which genes are likely most important in heart disease . More studies are needed to understand what these genes do and how they contribute to heart disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2019	A generally conserved response to hypoxia in iPSC-derived cardiomyocytes from humans and chimpanzees
G protein gated inward rectifier K+ ( GIRK ) channels open and thereby silence cellular electrical activity when inhibitory G protein coupled receptors ( GPCRs ) are stimulated . Here we describe an assay to measure neuronal GIRK2 activity as a function of membrane-anchored G protein concentration . Using this assay we show that four Gβγ subunits bind cooperatively to open GIRK2 , and that intracellular Na+ – which enters neurons during action potentials – further amplifies opening mostly by increasing Gβγ affinity . A Na+ amplification function is characterized and used to estimate the concentration of Gβγ subunits that appear in the membrane of mouse dopamine neurons when GABAB receptors are stimulated . We conclude that GIRK2 , through its dual responsiveness to Gβγ and Na+ , mediates a form of neuronal inhibition that is amplifiable in the setting of excess electrical activity . Potassium channels oppose membrane electrical excitability by driving the membrane voltage towards the K+ reversal potential , near -90 mV in mammalian neurons . In the nervous system many inhibitory neurotransmitters act through G protein coupled receptors ( GPCRs ) , which regulate a G protein gated inward rectifier K+ ( GIRK ) channel ( Pfaffinger et al . , 1985; Lesage et al . , 1994; Lesage et al . , 1995; Wang et al . , 2014 ) . In this form of signaling G proteins are released by stimulated GPCRs and diffuse on the cytosolic surface of the membrane to a site on the K+ channel . A tight complex of the β and γ G protein subunits ( known as the 'Gβγ subunit' ) binds to GIRK , favors the open conformation , and drives the membrane potential towards the K+ reversal potential ( Pfaffinger et al . , 1985; Logothetis et al . , 1987; Reuveny et al . , 1994; Krapivinsky et al . , 1995 ) ( Figure 1A ) . 10 . 7554/eLife . 15751 . 003Figure 1 . Membrane anchored Gβγ binds to GIRK and activates the channel . ( A ) Inhibitory neurotransmitters activate Gi/o G protein coupled receptors ( GPCRs ) in neuron membranes . The GPCRs facilitate the exchange of GDP to GTP on the G protein hetero-trimer , releasing the Gαi/o subunit and Gβγ subunit . The membrane-anchored Gβγ subunit binds to and activates GIRK . ( B ) NTA lipid ( head group modified with a Ni2+ chelator NTA , DOGS-NTA ) is used to anchor non-lipid modified and His-tagged Gβγ ( sGβγ-His10 ) onto the lipid membranes . 2 μM of sGβγ-His10 was used to fully saturate all NTA lipid on the membrane . The sGβγ-His10 density on the membrane can be controlled by the NTA lipid mole fraction . 32 μM C8-PIP2 was included on the same side as sGβγ-His10 . ( C ) Membrane-bound sGβγ-His10 activates GIRK to different levels depending on the NTA lipid mole fraction . Lipid modified Gβγ is used to fully activate GIRK at the end of each experiment . GIRK currents corresponding to different NTA lipid mole fractions are normalized to the fully activated value . A detailed description of the experiment is shown in Figure 1—figure supplement 1 . Example current traces of activation by sGβγ-His10 and lipid modified Gβγ in liposomes are shown in Figure 1—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 00310 . 7554/eLife . 15751 . 004Figure 1—figure supplement 1 . Details of the planar bilayer experiment . To control Gβγ density in the membrane ( A ) GIRK2 channel proteoliposomes containing the corresponding mole fraction of DOGS-NTA-Ni2+ lipid were fused into planar lipid bilayers containing the same density of DOGS-NTA-Ni2+ lipid . ( B ) A high concentration KCl solution ( 1 M ) was applied at the membrane to facilitate complete fusion of proteoliposomes attached to the membrane . ( C ) 1 mM NiSO4 was applied at the membrane to ensure that all NTA groups were charged with Ni2+ . ( D ) sGβγ-His10 and C8-PIP2 were added to the top side of the membrane . GIRK2 channels with the cytoplasmic side facing the top side began to open . ( E ) A Na+ titration was then performed by stepwise addition of NaCl up to 32 mM final concentration . ( F ) After the Na+ titration , proteoliposomes of lipid modified Gβγ was fused to the membrane to maximize GIRK activation . ( G ) Maximum current of for each membrane is used for normalization of currents recorded in the same membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 00410 . 7554/eLife . 15751 . 005Figure 1—figure supplement 2 . Example traces of GIRK2 activation by sGβγ-His10 and lipid modified Gβγ in proteoliposomes . Solution contains 10 mM potassium phosphate at pH 8 . 2 with 150 mM KCl on both sides of the membrane . 2 nM NiSO4 , 2 mM MgCl2 and 32 μM C8-PIP2 were also included on the cytosolic side of the channel . Membrane voltage was held at -50 mV . ( A ) 0 . 03 mol fraction of Ni-NTA lipid was included in the lipid bilayer . 2 μM of sGβγ-His10 was applied to the membrane at the time indicated by the arrow . The activation by sGβγ-His10 generally takes a few seconds to a few tens of seconds to reach equilibrium . After reaching equilibrium , the current is stable for many minutes . ( B ) Application of Gβγ in proteoliposome vesicles activates GIRK with a slower apparent kinetics . ~700 mM KCl was included in the vesicles to facilitate fusion with the membrane . The decrease in current immediately after application of the high salt vesicles is due to the change in local electro-chemical gradient of K+ near the membrane . Mixing restores the ionic conditions . To ensure saturation of Gβγ binding on the channel , Gβγ vesicles were applied several times until the current ( after mixing ) no long increases . With good membranes , after saturation is reached , the current is stable for minutes . Voltage families were recorded during the breaks indicated by '/ /' which take around 1 min each . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 005 Extensive research on GIRK channels in both neurons and cardiac cells has identified three important regulators of GIRK channel gating: Gβγ subunits , the signaling lipid PIP2 and intracellular sodium ( Logothetis et al . , 1987; Wickman et al . , 1994; Huang et al . , 1998; Sui et al . , 1998; Ho and Murrell-Lagnado , 1999a; Petit-Jacques et al . , 1999 ) . Many aspects of how these ligands interact with the channel , whether they are absolutely required for channel opening , and how they interact with each other through their respective interactions with the channel have remained unclear . Some studies supported an absolute requirement for Gβγ subunits ( Logothetis et al . , 1987; Wickman et al . , 1994; Krapivinsky et al . , 1995 ) , while others concluded that Mg2+-ATP with Na+ ( Petit-Jacques et al . , 1999 ) or PIP2 ( Huang et al . , 1998 ) were by themselves sufficient to open GIRK channels in the absence of Gβγ subunits . Furthermore , attempts to determine Gβγ affinity for GIRK channels in cell membranes , assessed by applying detergent-solubilized Gβγ to membrane patches , yielded values ranging from 3 nM to 125 nM ( Wickman et al . , 1994; Krapivinsky et al . , 1995 ) . The problem with these studies is the membrane partition coefficient for detergent-solubilized Gβγ is not known and therefore the membrane concentration is not known . Isothermal titration calorimetry ( ITC ) measurements with a soluble form of Gβγ ( lipid anchor-removed ) and a soluble cytoplasmic domain of a GIRK channel ( removed from the transmembrane pore ) measured the affinity to be 250 μM ( Yokogawa et al . , 2011 ) . These experiments can only report binding affinity in the absence of energetic coupling to a gated pore , which was absent in the experiment . Given the difficulty in knowing accurately the composition and quantity of components in living cell membranes – the experimental system in which the majority of studies had been carried out – we developed a total reconstitution assay to investigate the regulation of neuronal GIRK2 channel gating ( Wang et al . , 2014 ) . Using planar lipid bilayers in which purified GIRK2 channels , G protein subunits and PIP2 were reconstituted , we found that Gβγ subunits and PIP2 are simultaneously required to open GIRK2 channels ( each alone is insufficient ) and that Na+ is not required for opening , but modulates GIRK2 channel opening . Because planar bilayers allow quantitative control of lipid concentrations , the reconstitution study also permitted a detailed characterization of channel opening as a function of the PIP2 concentration . While it is possible to specify lipid ( e . g . PIP2 ) concentrations in a planar bilayer membrane , it is not possible to specify protein concentrations by simply mixing components during the bilayer membrane synthesis . For this reason , the reconstitution study described above did not permit accurate control of membrane Gβγ concentration . In the current study we present a method to specify Gβγ subunit concentration in planar lipid membranes and use the method to determine the Gβγ-GIRK2 channel activity relationship . We then show that intracellular Na+ regulates GIRK2 channel gating mostly by increasing the GIRK2 affinity for Gβγ . Finally , we use the newly defined quantitative relationship between Gβγ , Na+ , and GIRK2 channel activity to estimate the membrane concentration of Gβγ subunits that appear in mouse dopamine neuron membranes upon stimulation of GABAB receptors . A method to control the concentration of G proteins on the surface of a lipid bilayer membrane is illustrated ( Figure 1B , Figure 1—figure supplement 1 ) . GIRK2 channels were reconstituted into planar lipid membranes formed with known mole fractions of Ni-NTA lipid , doped into otherwise biological phospholipids ( Nye and Groves , 2008; Knecht et al . , 2009; Platt et al . , 2010; Masek et al . , 2011 ) . Modified Gβγ subunits with a His-tag replacing the lipid anchor on the γ subunit were then added at known concentrations to the solution on one side of the membrane with the idea that these would anchor to the membrane via the Ni-NTA lipid ( Kubalek et al . , 1994; Schmitt et al . , 1994; Knecht et al . , 2009; Platt et al . , 2010 ) . All experiments were carried out in the presence of a fixed concentration of 32 μM C8-PIP2 to ensure high occupation of PIP2 sites on the channel: 32 μM C8-PIP2 , based on channel activity measurements , corresponds to 0 . 02 mol fraction ( 2% ) membrane PIP2 ( Wang et al . , 2014 ) . Example data using this assay are shown ( Figure 1C ) . In the absence of Ni-NTA lipid , addition of soluble Gβγ with a His-10 tag ( sGβγ-His10 ) to a solution concentration of 2 μM failed to activate the channel . The presence of channels in the membrane was subsequently confirmed by addition of a maximally effective ( but unknown ) concentration of lipid-anchored Gβγ through vesicle fusion with the membrane . In another experiment , when the same concentration of sGβγ-His10 was added to a membrane formed with 0 . 0019 mol fraction Ni-NTA ( 19 out of 10 , 000 lipid molecules in the membrane containing the Ni-NTA head group ) , channels were activated ( Figure 1C ) . Subsequent addition of excess lipid-anchored Gβγ to the same membrane showed that about 60% of the GIRK channels had been activated by the Ni-NTA lipid-anchored Gβγ . All further experiments were performed in the manner described , ending with saturation of the membrane with Gβγ to achieve maximal activation of the GIRK channels present . This normalization step enables comparison of currents measured in different membranes with different numbers of GIRK channels by placing them on a common scale ( normalized current ) . In the assay two equilibrium reactions occur , as depicted ( Figure 1B ) . First , sGβγ-His10 binds from solution to the Ni-NTA lipid , and second , the sGβγ-His10-Ni-NTA lipid complex binds to the channel . We are ultimately interested in the second reaction as this determines channel activation as a function of Gβγ concentration on the membrane ( Gβγ density in 2 dimensions ) . In Figure 2A the black symbols and curve show the normalized GIRK current level as a function of sGβγ-His10 solution concentration with a membrane containing Ni-NTA lipid at a mole fraction 0 . 0019 . Normalized currents under these conditions reach a maximum value around 0 . 6 ( 60% of current that is reached when the same membranes are saturated with lipid-anchored Gβγ ) . A maximum , saturated value below 1 . 0 can be explained if 2 μM sGβγ-His10 is sufficient to occupy all Ni-NTA lipid molecules in the membrane , but the concentration of Ni-NTA lipid in the membrane is too low to occupy all sites on the channel . This explanation is supported by the graph on the right ( Figure 2B ) in which normalized current is plotted as a function of Ni-NTA lipid mole fraction in the presence of 2 μM sGβγ-His10 ( i . e . a concentration that is sufficient to occupy all Ni-NTA lipid molecules ) . This graph is asymptotic to ~1 at higher values of Ni-NTA lipid mole fraction , and , as one would expect , 0 . 6 on the Y-axis corresponds to 0 . 0019 on the X-axis . A third graph ( Figure 2C ) of values from the X-axis in Figure 2B , plotted as a function of corresponding values ( dashed lines ) from the X-axis in Figure 2A , isolates the binding reaction of sGβγ-His10 to Ni-NTA lipid . The curve is a rectangular hyperbola ( binding isotherm ) with a Kd of 150 nM ( Figure 2C , black curve ) . A similar binding curve and affinity were determined for fluorescent sGβγ-His10 adsorption onto giant unilamellar vesicles ( GUVs ) containing Ni-NTA lipid at a mole fraction of 0 . 03 ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 15751 . 006Figure 2 . Calibration of sGβγ-His10 binding to NTA lipid . ( A ) Titration of GIRK activity by sGβγ-His10 ( black ) or sGβγ-His4 ( blue ) to lipid membranes containing a fixed 0 . 0019 mol fraction of NTA lipid ( n = 3–5 membranes , mean ± SEM ) . Solid lines are fits to the Hill equation: Current = Max × [Gβγ]n/ ( Kdn + [Gβγ]n ) with Kd = 200 ± 15 nM , n = 2 . 9 ± 0 . 4 ( black , sGβγ-His10 ) and Kd = 1 . 6 ± 0 . 1 μM , n = 1 . 7 ± 0 . 05 ( blue , sGβγ-His4 ) . Solutions contained 150 mM KCl on both sides of the membrane and 32 mM NaCl on the inside where sGβγ-His was applied . ( B ) Titration of GIRK activity as a function of NTA lipid mole fraction in the presence of 2 μM sGβγ-His10 in solution . The solid line corresponds to a model ( Figure 3—figure supplement 1 ) . ( C ) Mapping of the NTA lipid concentration in panel ( B ) to the solution sGβγ-His concentration in panel ( A ) through GIRK activity . The curves are rectangular hyperbolas ( Hill equation with n = 1 ) with Kd = 0 . 15 ± 0 . 03 μM ( black ) and Kd = 5 . 0 ± 1 . 0 μM ( blue ) . A similar affinity ( Kd = 0 . 09 ± 0 . 02 μM ) was obtained for fluorescently labeled sGβγ-His10 adsorption to giant unilamellar vesicles ( GUVs ) containing 0 . 03 mol fraction of NTA lipid ( Figure 2—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 00610 . 7554/eLife . 15751 . 007Figure 2—figure supplement 1 . Binding of fluorescently labeled sGβγ-His10 to GUVs containing 0 . 03 mol fraction of NTA lipid . ( A ) Typical confocal fluorescence images of the GUV equator planes . The corresponding concentration of sGβγ-His10 is indicated above each image . The first image from the right on the bottom row shows regions used for intensity quantification . The area between the yellow dashed concentric circles are used for calculating GUV fluorescence intensity , while the region inside the red dashed circle is treated as background . ( B ) Fitting of the dependency of GUV fluorescence intensity ( n = 5–7 GUVs , Mean ± SEM ) on sGβγ-His10-AF488 concentration to a simple 1:1 binding model . The apparent dissociation constant is around 90 nM . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 007 The blue data points and curve in Figure 2A show a similar set of experiments using sGβγ-His4 , that is , a soluble form of Gβγ with 4 instead of 10 histidine residues in its tag . The normalized current level is asymptotic to a value higher than 0 . 6 ( blue curve ) . The explanation for this becomes evident when the ( apparent ) Ni-NTA lipid mole fraction is plotted against the corresponding ( blue dashed lines ) sGβγ-His concentration ( Figure 2C , blue symbols and curve ) : in this binding isotherm the affinity is lower and the maximum apparent Ni-NTA lipid mole fraction is ~3 times higher . This result follows if sGβγ-His4 binds to a single Ni-NTA lipid and sGβγ-His10 binds to 3 Ni-NTA lipids . This stoichiometric difference is consistent with known structures of Ni-NTA-polyhistidine complexes , which show that a single Ni-NTA group is coordinated by 2 histidine residues separated by at least one histidine residue ( Knecht et al . , 2009 ) . Thus , sGβγ-His4 can only attach to a single Ni-NTA lipid while sGβγ-His10 can - and does - attach to three . All further experiments were carried out using sGβγ-His10 at 2 μM concentration to fully occupy the Ni-NTA lipid binding sites in the membrane and thus ensure that the concentration of Gβγ in the membrane would be controlled solely by the Ni-NTA lipid mole fraction . In other words , this approach isolates the reaction of interest – channel activity ( related to occupancy in a manner to be determined ) as a function of known membrane Gβγ concentration . The graphs report Gβγ concentration as Ni-NTA lipid mole fraction , while keeping in mind that the molar density of Gβγ in the membrane is one third that of the Ni-NTA lipid density . Having established an assay to control the concentration of Gβγ in the membrane , we measured the activity of GIRK2 as a function of membrane Gβγ concentration as well as solution Na+ concentration ( Figure 3A ) . At each Na+ concentration , normalized current increases as a steep sigmoidal function of membrane Gβγ concentration ( Figure 3B ) . The slope of these functions on a log-log plot at sufficiently low Gβγ concentrations ( achieved in these experiments for the Gβγ titrations at lower Na+ concentrations ) are consistent with four Gβγ subunits being required to open a GIRK2 channel ( see methods ) ( Figure 3C ) . A strong effect of Na+ on the functional relationship is clear and noteworthy because in cells Na+ is known to regulate GIRK currents , but by a mechanism that is unknown ( Sui et al . , 1996; Ho and Murrell-Lagnado , 1999b; Petit-Jacques et al . , 1999 ) . The titrations show that Gβγ activates the channel to a greater extent , especially at lower Gβγ concentrations , as Na+ is increased ( Figure 3B ) . To further understand how these two ligands interact with the channel to regulate its gating we constructed an equilibrium model . This model was guided by atomic structures , which show that a tetramer GIRK2 channel has 4 structurally identical Gβγ binding sites and 4 Na+ binding sites ( Whorton and MacKinnon , 2013 ) . The model contains 25 states , corresponding to an order-independent occupation number 0 to 4 for each ligand ( Figure 3D and Figure 3—figure supplement 1 ) . Fitted parameters in the model include a dissociation constant and cooperativity factor for each ligand , a cross cooperativity factor between Gβγ and Na+ and a parameter relating ligand occupancy to channel activity ( see Figure 3—figure supplement 1 ) . The data and modeling support the following conclusions . First , all four Gβγ binding sites must be occupied on the channel before it opens , consistent with the limiting slope analysis ( Figure 3C ) . Second , Gβγ binding is cooperative with a factor of 0 . 30 , which means the fourth Gβγ subunit binds with an affinity 37 times higher than the first . Attempts to fit the data imposing no cooperativity ( b = 1 ) yields higher residuals ( 0 . 126 compared to 0 . 064 when allowing cooperativity ) and fail to replicate the steep rise in channel activity as a function of Gβγ concentration ( Figure 3—figure supplement 2 ) . The strong cooperative binding of four Gβγ subunits accounts for the steep sigmoidal dependence of GIRK current on membrane Gβγ concentration ( Figure 3A , B ) . Third , Gβγ binds with a Na+ cross cooperativity factor ( η ) of 0 . 63 , which means Gβγ binds with 6-fold higher affinity when four Na+ sites are occupied compared to when the Na+ sites are not occupied . This effect of Na+ on Gβγ affinity accounts for channel opening at lower membrane Gβγ concentrations as Na+ concentration increases ( Figure 3B ) . 10 . 7554/eLife . 15751 . 008Figure 3 . GIRK activity as a function of Gβγ and Na+ concentration . 2 μM sGβγ-His10 was included in the solution on the intracellular side of GIRK . ( A ) Normalized GIRK current ( red spheres , mean ± SEM , n = 3–5 membranes ) is graphed as a function of Gβγ and Na+ concentration . Surface mesh shows predictions of a model for ligand activation ( Figure 3—figure supplement 1 ) . ( B ) Data points in ( A ) are graphed as a family of curves ( surface intersections ) corresponding to each Na+ concentration . ( C ) Log-log plot of normalized current against NTA lipid mole fraction . Data points corresponding to 0 . 0001 NTA lipid mole fraction were excluded because the current levels ( <0 . 02 normalized current ) were much smaller than background noise . Other data points are connected with solid lines . Dashed lines show the slope of the line connecting the first two graphed data points . ( D ) A schematic of the ligand activation model fit to the data . i and j are integers between 0 and 4 . The fitted parameters are: equilibrium dissociation constant for the first Na+ to bind in the absence of Gβγ , Kdn = 60 ± 20 mM , equilibrium dissociation constant for the first Gβγ in the absence of Na+ , Kdb = 0 . 019 ± 0 . 007 , cooperativity factor for each successive Gβγ binding b = 0 . 30 ± 0 . 06 , cross-cooperativity factor between Gβγ and Na+ binding η = 0 . 63 ± 0 . 04 and an activity term θ as described in Figure 3—figure supplement 1 . A comparison of fits to the data using cooperative and non cooperative models is shown in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 00810 . 7554/eLife . 15751 . 009Figure 3—figure supplement 1 . Modeling of Na+ and Gβγ binding equilibrium with the GIRK2 tetramer . GIRK binding with 0 to 4 Na+ and/or Gβγ is modeled with a system containing all 25 possible binding states . Symbols represent the following . C: GIRK tetramer with no ligand bound; C ( Na+ ) i: channel with i Na+ bound; C ( Gβγ ) j: channel with j Gβγ bound; Kdn: equilibrium dissociation constant of Na+ with ligand-free channel; Kdb: equilibrium dissociation constant of Gβγ with ligand-free channel; n: cooperativity factor for each successive Na+ binding; b: cooperativity factor for each successive Gβγ binding; η: cross cooperativity between Gβγ and Na+ binding . Experimentally measurable GIRK activity is the sum of the activities of all species described in this model: GIRK Activity = ∑i , j ( ci , j × θi , j ) where ci , j is the population of GIRK with i Na+ and j Gβγ bound and θi , j is the activity of the corresponding species . Since the Na+ titration to GIRK with saturated Gβγ follows a simple 1:1 binding isotherm ( rectangular hyperbola ) , we assumed the cooperativity factor n for Na+ binding to be 1 and that the channel activity increases proportionally with the number of occupied Na+ sites ( θi , j = θ0 , j + i × ( θ4 , j - θ0 , j ) / 4 ) . Reasonable fitting was achieved only when opening was associated with 4 Gβγ bound ( θi , j = 0 when j ≠ 4 ) . Under these assumptions , parameters for the best fit are: Kdn = 60 ± 20 mM , Kdb = 0 . 019 ± 0 . 007 , b = 0 . 30 ± 0 . 06 , η = 0 . 63 ± 0 . 04 , θ0 , 4 = 0 . 49 ± 0 . 04 and θ4 , 4 = 1 . 19 ± 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 00910 . 7554/eLife . 15751 . 010Figure 3—figure supplement 2 . Comparison of fitted model allowing or not allowing cooperativity in Gβγ binding . Solid lines represent surface intersections at various Na+ concentrations ( color coded as shown in the figure ) of the model allowing cooperativity of Gβγ binding ( the same fit as in Figure 3A and Figure 3—figure supplement 1 . ) . The scaled residual sum of squares is 0 . 064 . Dashed lines represent the same intersections of the model fit assuming that Gβγ binding is not cooperative ( b = 1 ) . Corresponding parameter values for this non cooperative case are: Kdn = 110 ± 60 mM , Kdb = 0 . 0016 ± 0 . 0003 , η = 0 . 62 ± 0 . 06 , θ0 , 4 = 0 . 49 ± 0 . 04 and θ4 , 4 = 1 . 22 ± 0 . 08 . The scaled residual sum of squares is 0 . 126 . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 010 The ability of Na+ to increase the affinity of Gβγ is demonstrable in another way , through a simple , intuitive analysis . The family of data points ( Figure 3B ) conform well to the Hill equation with a single global Hill coefficient ( n ≈ 3 ) but variable , Na+-dependent equilibrium constant for Gβγ binding ( Figure 4A ) . The Gβγ equilibrium constant decreases ( i . e . the affinity for Gβγ increases ) as Na+ concentration increases according to a rectangular hyperbola , with a ~6-fold difference between maximum and minimum values ( Figure 4B ) . The apparent equilibrium constant for the effect of Na+ on Gβγ activation is ~5 mM , which is very close to the physiological Na+ concentration in the cytoplasm of a resting neuron ( Rose and Ransom , 1997 ) . Thus , the GIRK2 channel’s response to Gβγ should be sensitive to changes in Na+ concentration right in the physiological range . 10 . 7554/eLife . 15751 . 011Figure 4 . Na+ concentration regulates Gβγ affinity . 2 μM sGβγ-His10 was included in the solution on the intracellular side of GIRK . ( A ) Normalized current values from Figure 3A are fit to the Hill equation: Current = Max × [NTA-lipid mole fraction]n / ( Kdn + [NTA-lipid mole fraction]n ) with a Hill coefficient ( n ) of 2 . 8 ± 0 . 2 for all curves and equilibrium dissociation constant ( Kd × 102 ) for Gβγ binding 0 . 6 ± 0 . 2 , 0 . 38 ± 0 . 06 , 0 . 27 ± 0 . 02 , 0 . 21 ± 0 . 03 and 0 . 16 ± 0 . 01 for Na+ concentrations 0 mM , 4 mM , 8 mM , 16 mM and 32 mM , respectively . ( B ) Gβγ Kd values from fits in ( A ) are plotted as function of Na+ concentration . The curve is the rectangular hyperbola Kd = Kdmax + ( Kdmin - Kdmax ) × [Na+] / ( Kd-Na+ + [Na+] ) , where Kd-Na+ = 5 . 1 ± 0 . 9 mM is the apparent Na+ dissociation constant as estimated through its effect on the affinity of Gβγ . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 011 The atomic structures of the GIRK2 channel and its complex with ligands offers clues to the mechanistic underpinnings of Gβγ and Na+ regulation of GIRK2 beyond the 4:1 stoichiometry of ligand binding ( Figure 5A ) . When four Gβγ subunits bind to the cytoplasmic domain of GIRK2 , which forms a ring made by the four channel subunits , they cause the ring to rotate as a rigid body with respect to the pore , which twists open the helical bundle that forms the pore’s gate ( Whorton et al . , 2013 ) . Because the rigid body rotation involves all four subunits at once , conformational changes induced by the binding of Gβγ to one site will favor binding at the neighboring sites ( i . e . positive cooperativity ) . A cartoon illustrating this concept depicts Gβγ binding more favorably to the channel’s open conformation ( Figure 5B ) . Because opening involves a concerted rotation of the subunits , all four Gβγ binding sites change to higher affinity at once , giving rise to strong positive cooperativity . 10 . 7554/eLife . 15751 . 012Figure 5 . Structural basis for cooperativity in Gβγ activation of GIRK . ( A ) Top view of the atomic structure of GIRK2 in complex with Gβγ , Na+ and PIP2 ( PDB ID: 4KFM ) . Four Gβγ , Na+ and PIP2 molcules bind to one GIRK2 homo-tetramer , associated with rotation of the cytoplasmic domain with respect to the transmembrane as the channel opens . ( B ) Gβγ binding favors the cytoplasmic-domain-rotated , open conformation of the channel . The rigid body rotation of the cytoplasmic domain is associated with increased affinity of four Gβγ binding sites simultaneously , giving rise to strong positive cooperativity . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 012 In addition to increasing Gβγ affinity , Na+ also increases the GIRK2 current when the Gβγ binding sites are fully occupied: at the highest ( 0 . 03 ) Ni-NTA lipid mole fraction Na+ increases current approximately 2 . 5-fold when Na+ is increased from 0 mM to 32 mM ( Figure 3A ) . This increase follows a rectangular hyperbola . The simplest physical explanation for this behavior , which is also consistent with the equilibrium model , is that Na+ stabilizes the open , conductive state of the channel in direct proportion to its occupancy on the channel . In other words , four Gβγ subunits bind to GIRK2 and permit opening to a probability that is higher in proportion to occupancy of the Na+ sites . Thus , by thermodynamic linkage , Na+ would increase the apparent affinity of Gβγ for GIRK2 and it would also increase the maximum level of current reached when four Gβγ subunits bind . This physical mechanism is consistent with the location of the Na+ binding sites at the interface between the cytoplasmic domains and the transmembrane pore , where opening is transduced through the binding of Gβγ ( Figure 5A ) ( Whorton et al . , 2013 ) . It is also consistent with the observations that Na+ in the absence of Gβγ does not open GIRK ( Figure 3A ) and in crystal structures Na+ binds to GIRK but does not cause a rotation of the cytoplasmic domain in the absence of Gβγ ( Whorton and MacKinnon , 2011 ) . Thus , Na+ facilitates Gβγ-mediated pore opening . How might neuronal electrical signaling be affected by the GIRK2 channel’s dual regulation by Gβγ and Na+ ? GIRK2 suppresses electrical activity in neurons when inhibitory neurotransmitters stimulate GPCRs on the cell surface , such as GABAB receptors , which release Gβγ on the intracellular membrane surface to open GIRK2 channels . At the same time , the level of GIRK2 channel opening – and therefore the level of neuronal inhibition – brought about by the released Gβγ potentially depends on the intracellular Na+ concentration . This conclusion derives from the family of Gβγ activation curves ( Figure 3B ) : at all Gβγ concentrations , the level of GIRK2 current is increased as Na+ is increased . We refer to this phenomenon as Na+-amplification of Gβγ-activated current . Na+-amplification is clearly not constant but is instead a function of the Gβγ concentration: at high Gβγ concentrations ( right side of graph ) amplification is 2 . 5-fold ( i . e . GIRK2 current increases 2 . 5-fold ) when Na+ is increased from 0 to 32 mM , while at lower Gβγ concentrations ( corresponding to the steep sigmoidal rise in current ) amplification approaches ten fold . The potential importance of Na+ amplification to neuronal electrical signaling lies in the fact that cytosolic Na+ increases with higher levels of electrical activity due to Na+ entry through both synaptic channels and voltage-dependent Na+ channels ( Lasser-Ross and Ross , 1992; Fleidervish et al . , 2010 , Rose and Konnerth , 2001 ) . Thus , Na+ amplification should , in principle , provide a mechanism for strengthening an inhibitory input to a more active neuron . How large is Na+-amplification in neurons ? As noted above , the magnitude of amplification depends on the concentrations of Gβγ generated inside a neuron when its GPCRs are stimulated . While the concentration of Gβγ inside cells is unknown , the data in this study provide an approach to estimate its value . The rationale is as follows: the curves in Figure 3B characterize the amplification as a function of Gβγ concentration , therefore we should be able to solve the inverse problem of deducing Gβγ concentration by measuring the Na+ amplification in a cell . Figure 6 shows this analysis applied to GIRK currents recorded in midbrain dopamine neurons when baclofen was used to stimulate GABAB receptors . A recording pipette was used to set the cytoplasmic Na+ concentration to either 0 or 27 mM . Baclofen-activated currents had the strongly inwardly-rectifying current-voltage relationship expected from GIRK channel activation ( Figure 6A ) . Baclofen-activated GIRK current was much smaller in neurons recorded with 0 mM internal Na+ compared with those recorded with 27 mM internal Na+ ( Figure 6B , C ) . Currents measured with 27 mM internal Na+ were amplified by an average of 8 fold compared to currents with 0 mM Na+ . From the Gβγ/Na+ titration data ( Figure 3B ) , 8-fold amplification corresponds to a Gβγ concentration of about 0 . 003 in NTA-lipid mole fraction units ( Figure 6D ) . In this concentration range the amplification curve becomes very steep , allowing relatively small cell-to-cell variations in Gβγ concentration to translate into larger differences in current response ( Figure 6E ) . This property offers an explanation for the large spread of current values measured upon baclofen activation in the presence of 27 mM Na+ . Most importantly , the estimated Gβγ concentration stimulated by baclofen in dopamine neurons ( Figure 6D , E ) is centered in the middle of the steep sigmoidal rising phase of the Gβγ-activation curves ( Figure 3B ) . In this regime even modest changes in intracellular Na+ concentration should amplify Gβγ-mediated inhibition of neuronal electrical activity . The intracellular Na+ concentration in neurons is subject to complex regulation by multiple channels and transporters and changes during neuronal activity ( Rose and Ransom , 1997 ) . Intracellular Na+ in dendrites can double during synaptic activity ( Rose and Konnerth , 2001 ) , and high local increases also occur in cell bodies and axons during action potential firing ( Lasser-Ross and Ross , 1992; Fleidervish et al . , 2010 ) . Thus , the Na+ amplification of GIRK currents likely occurs during normal physiological activity . Even stronger amplification is likely during epileptiform activity , when intracellular Na+ can likely reach 30 mM ( Raimondo et al . , 2015 ) . 10 . 7554/eLife . 15751 . 013Figure 6 . Estimation of Gβγ membrane density generated during GABAB receptor activation in mouse dopamine neurons . ( A ) Current-voltage relationship for current induced by 100 μM baclofen in a dissociated dopamine neuron from the mouse substantia nigra pars compacta recorded with an external solution containing 16 mM K+ and an internal solution containing 126 mM K+ and 27 mM Na+ . ( B ) Time-course of baclofen-induced current in two neurons recorded with an internal solution containing 0 mM Na+ ( top ) or 27 mM Na+ ( bottom ) . The same scale is used for both recordings . Current was measured as the average current between -142 mV and -147 mV evoked by voltage ramps ( 1 mV/ms ) from +8 to -147 mV delivered from a steady holding potential of -92 mV every 2 s . ( C ) Collected values for baclofen-induced GIRK current ( mean ± SEM ) in dopamine neurons equilibrated with 0 mM ( n = 10 ) and 27 mM ( n = 11 ) intracellular Na+ . ( D ) Data and curves from Figure 3B are used to estimate the concentration of Gβγ required to yield the 8-fold amplification of GIRK current observed in ( C ) . ( E ) A Na+ amplification curve is defined as the green curve ( 32 mM Na+ , which is near 27 mM ) divided by the black curve ( 0 mM Na+ ) in ( D ) . Amplification is a steep function of Gβγ concentration near the stimulated levels of Gβγ in dopamine neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 15751 . 013 Taking into account the surface area of a lipid head group and the stoichiometry of 3 Ni-NTA lipid molecules per sGβγ-His10 subunit , a mole fraction value 0 . 003 ( i . e . the concentration of Gβγ subunits estimated in dopamine neurons ) translates into approximately 1200 Gβγ subunits per μm2 of membrane . To place this 2-dimensional membrane density into more familiar concentration units we multiply the membrane surface area by the linear dimension of a Gβγ subunit ( about 70 Å ) to approximate a Gβγ concentration in the solution layer adjacent to the membrane equal to 280 μM . At this Gβγ concentration GIRK is between 10% and 80% activated , depending on the Na+ concentration ( Figure 3B ) . Thus , the apparent affinity of Gβγ for the GIRK channel is in this range . This estimate is close to the affinity reported using ITC to study the interaction of lipid anchor-removed Gβγ in solution with the soluble cytoplasmic domain of a GIRK channel ( 250 μM ) ( Yokogawa et al . , 2011 ) . A more careful comparison , however , reveals a fascinating difference . Removed from the pore , the cytoplasmic domain , even though it is a tetramer with four Gβγ binding sites like the full channel , binds to Gβγ according to a 1:1 binding isotherm . As we have shown here , the full GIRK2 channel by contrast exhibits strong cooperativity , the first Gβγ subunit binding with very low affinity ( equilibrium constant 0 . 019 mol fraction corresponding to 1 . 9 mM in the solution layer adjacent to the membrane ) and the fourth binding with higher affinity ( equilibrium constant ( 0 . 019 ) × ( 0 . 3 ) 3 mol fraction corresponding to 50 μM in the solution layer adjacent to the membrane ) . This cooperativity , which gives rise to the steep dependence of GIRK channel activity on Gβγ concentration , is completely lost when the transmembrane pore is removed . The mechanism proposed for coupling Gβγ binding to pore opening , illustrated in Figure 5 , offers an explanation: when the pore is removed , Gβγ binding free energy is no longer utilized to twist open the pore’s helical gate , and at the same time the rotational origin of cooperativity disappears . The ITC-determined affinity of 250 μM lies in between the affinities of the first ( 1 . 9 mM ) and fourth ( 50 μM ) Gβγ subunits to bind to the intact , cooperative system . In the context of other known protein complexes , the interaction of Gβγ with GIRK2 is weak , consistent with a short lifetime for the complex . For example , two proteins with a diffusion-limited association rate constant of , say , 107 M-1sec-1 , will remain in complex on average for less than 2 milliseconds if the equilibrium constant is 50 μM ( i . e . affinity of the fourth Gβγ subunit ) and the lifetime of an activated channel ( GIRK2 with 4 Gβγ subunits , any one of which can dissociate ) less than 0 . 5 milliseconds . Even if the association rate constant is smaller , the lifetime of an active channel will be brief compared to the duration of macroscopic GIRK current in a cell during GPCR stimulation ( ~1 s ) ( Ford et al . , 2009 ) . This means Gβγ apparently associates and dissociates many times on and off the channel during a period of stimulation . Because Gβγ binds to Gα ( GDP ) with greater than ten thousand times higher affinity ( Kd ~1 nM Sarvazyan et al . , 1998 ) than to the channel , whenever Gα ( GTP ) hydrolyses GTP to GDP it will rapidly bind free Gβγ and remove it from the channel by mass action . Thus , the very weak binding of Gβγ to the channel means that the duration of GIRK current activation during GPCR stimulation will be controlled by the lifetime of Gα ( GTP ) . The Gβγ concentrations reported here represent the thermodynamic activity concentrations in equilibrium with the GIRK channel . In living cells it is distinctly possible that GIRK and GPCRs/G proteins reside in specialized regions of the cell membrane . In this case the relevant density of Gβγ is the local density near GIRK channels , which would be much higher than that averaged over the entire membrane . Such specialized regions would promote locally high Gβγ densities , in line with the relatively low affinity for GIRK subunits that allows rapid control of free Gβγ by Gα . A method to control the concentration ( density ) of G protein subunits in lipid membranes has let us reach the following conclusions . ( 1 ) Four Gβγ subunits bind to the GIRK channel with high cooperativity to give rise to a steep dependence of channel activity on membrane Gβγ concentration . ( 2 ) Intracellular Na+ concentration increases Gβγ affinity with an apparent Kd-Na+ near the cytoplasmic Na+ concentration of a resting neuron . ( 3 ) Inhibitory GPCR stimulation generates membrane Gβγ concentrations corresponding to the steep regime of the Gβγ-activation curve . ( 4 ) Properties ( 1 ) – ( 3 ) give rise to Na+ amplification of Gβγ-activation . Such amplification provides a mechanism for strengthening GPCR inhibition when Na+ enters neurons during activity . ( 5 ) Gβγ binds to GIRK with low affinity . Rapid equilibrium between Gβγ and GIRK allows rapid signal termination when Gα hydrolyses GTP to GDP . Mouse GIRK2 ( residues 52–380 ) was expressed in Pichia pastoris and purified as previously described ( Whorton et al . , 2011 ) . High Five ( Life Technologies , Grand Island , NY ) insect cells were infected with baculovirus bearing Human G protein subunits β1 and γ2 . The G protein Gβγ subunit was then purified using an established protocol ( Whorton et al . , 2013; Wang et al . , 2014 ) . To produce non-lipid modified and His-tagged Gβγ , baculovirus bearing a mutant Gγ2 DNA with a C68S mutation and a 4- or 10-His tag connected to the C-terminus by a GSSG linker was generated . This mutant virus and the virus bearing β1 DNA were co-infected into High Five cells . The purification process of non-lipid modified and His-tagged Gβγ is essentially the same as non-lipid modified Gβγ ( Wang et al . , 2014 ) except that PreScission protease digestion was not necessary since no cleavable tag was used . GIRK2 proteoliposomes were reconstituted using a lipid mixture composed of 3:1 ( w:w ) 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine ( POPE ) : 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho- ( 1'-rac-glycerol ) ( POPG ) supplemented with 3% , 1 . 5% , 0 . 75% , 0 . 38% , 0 . 19% , 0 . 1% , 0 . 01% and 0% of DOGS-NTA-Ni2+ lipid ( Avanti , Alabaster , AL ) . 3:1 POPE:POPG was used for Gβγ reconstitution . The reconstitution protocol is essentially the same as previously described ( Wang et al . , 2014 ) . Purified sGβγ-His10 protein was exchanged into conjugation buffer ( 50 mM potassium phosphate pH 7 . 4 , 100 mM NaCl , 0 . 1 mM TCEP ) and diluted to ~1 mg/ml . 5-fold molar excess of Alexa-Fluor 488 maleimide was mixed with the protein . The mixture was rotated at 4°C overnight . Labeled protein was affinity purified using Ni2+-NTA ( Qiagen , Valencia , CA ) beads followed by size exclusion chromatography in a buffer containing 10 mM potassium phosphate pH 7 . 4 and 150 mM KCl . The labeling efficiency was approximately one dye per sGβγ-His10 protein . DOPE: POPC 1: 1 ( w: w ) lipid mixture supplemented with 3% DOGS-NTA-Ni2+ was used to produce GUVs . A few microliters of 1 mg/ml of the lipid mixture in chloroform were dried under vacuum on an electrically conductive Indium tin oxide coated glass slide ( Sigma , St . Louis , MO ) and electroformed ( Meleard et al . , 2009 ) in a buffer containing 5 mM sodium phosphate with 300 mM sucrose ( pH 7 . 4 ) . After electroformation , the solution containing GUVs was diluted 5 fold into a buffer containing 10 mM potassium phosphate pH 7 . 4 , 150 mM KCl and 2 nM NiSO4 . Fluorescently labeled sGβγ-His10 was then added to a final concentration of 2 μM , 1 μM , 0 . 5 μM , 0 . 25 μM , 0 . 125 μM and 63 nM . The equator plane of the GUVs was then imaged using a Leica DMI 6000 microscope controlled by Leica Application Suite X software ( Leica , Buffalo Grove , IL ) . An oil immersion 63x objective ( numerical aperture 1 . 40 ) was used . Fluorophore was excited with a white light laser positioned at 491 nm with a pinhole size of 1 airy unit , giving rise to a confocal plane thickness of about 360 nm . Emission light above 505 nm was detected with a Hyd detector in photon counting mode , which exhibited good linearity . A 3x ‘smart zoom’ was used . 1024 × 1024-pixel 8-bit depth images were recorded with 4x line averaging to increase signal to noise ratio . Each pixel corresponds to ~60 nm distance in x and y directions . Microscope and software settings were the same for all images acquired . The images were converted into tiff format using software Imaris . A Mathematica ( Wolfram Research , Champaign , IL ) script was used to quantify fluorescence from GUVs . A circle fitting algorithm was first performed to locate the GUV image . Fluorescence intensity in the center of the GUV was treated as background since no fluorophore should be present in this region ( Figure 2—figure supplement 1A ) . The intensity of a 40-pixel ( ~2 . 4 μm ) shell area around the fitted circle was integrated . This integrated value , divided by the circumference of the circle , is taken as the GUV fluorescence intensity . Images with saturated pixels in the quantification area were discarded . Measurements were replicated on 5–7 GUVs at each sGβγ-His10 concentration . Mean and SEM values were calculated . Data fitting was performed with Origin ( OriginLab , Northampton , MA ) . DOPE: POPC 1: 1 ( w: w ) lipid mixtures supplemented with 3% , 1 . 5% , 0 . 75% , 0 . 38% , 0 . 19% , 0 . 1% , 0 . 01% and 0% of DOGS-NTA-Ni2+ lipid were used to make planar bilayer lipid membranes across a ~100 micron diameter hole on a plastic transparency ( Montal and Mueller , 1972; Miller and Racker , 1976; Wang et al . , 2014 ) . Buffer contained 10 mM potassium phosphate pH 8 . 2 , 150 mM KCl on both sides of the membrane . 2 nM NiSO4 and 2 mM MgCl2 were also included in the top chamber . A detailed outline of the experimental procedure is illustrated in Figure 1—figure supplement 1 . To measure the GIRK activity at a specific density of Gβγ in the membrane , a lipid bilayer containing a certain mole fraction of DOGS-NTA-Ni2+ lipid was used for making the planar lipid bilayer . GIRK2 proteoliposomes with the same mole fraction of DOGS-NTA-Ni2+ were subsequently fused into the lipid bilayer ( Figure 1—figure supplement 1A ) . We use high KCl concentrations to facilitate vesicle fusion in our experiments . After the application of vesicles , 1 M KCl solution was applied at the membrane to complete the fusion process ( Figure 1—figure supplement 1B ) . Since reducing reagent DTT and divalent metal chelator EDTA were present in the GIRK2 proteoliposomes reconstitution to preserve channel activity , Ni2+ will not be present on the NTA lipid in these vesicles . To ensure that all NTA lipids are charged with Ni2+ , 1 mM NiSO4 solution was applied at the membrane ( Figure 1—figure supplement 1C ) . Then 2 μM sGβγ-His10 and 32 μM 1 , 2-dioctanoyl-sn-glycero-3-phospho- ( 1'-myo-inositol-4' , 5'-bisphosphate ) ( C8-PIP2 ) were added to one side of the membrane . Given the chamber volume and membrane area , the molar ratio of sGβγ-His10 to DOGS-NTA-Ni2+ always exceeded 10 , 000 , thus ensuring a constant concentration of sGβγ-His10 except in specific experiments to study channel activation as a function of sGβγ-His10 concentration at a fixed DOGS-NTA-Ni2+ lipid mole fraction ( e . g . Figure 2A ) . Note that the C8-PIP2 is maintained constant at 32 μM during all experiments . Only GIRK channels with their cytoplasmic surface facing the solution chamber containing C8-PIP2 can be opened ( Figure 1—figure supplement 1D ) ( Wang et al . , 2014 ) . A Na+ titration was then followed ( Figure 1—figure supplement 1E ) . Native ( lipid-modified ) Gβγ in the form of proteoliposomes was then fused to the membrane to saturate the Gβγ binding sites on GIRK , maximizing GIRK activation ( Figure 1—figure supplement 1F and G ) . The maximized current level was used to normalize recordings for each membrane . The titration of sGβγ-His10 and sGβγ-His4 to a membrane containing 0 . 0019 mol fraction ( Figure 2A ) of Ni-NTA lipid was performed by successive addition of the proteins to final concentrations of 0 , 0 . 13 , 0 . 25 , 0 . 5 , 1 . 0 , 2 . 0 μM for sGβγ-His10 and 0 , 0 . 25 , 1 . 0 , 2 . 0 , 4 . 0 and 8 . 0 μM for sGβγ-His4 . Native Gβγ was then used to saturate Gβγ binding . The analog signal was low-pass filtered at 1 kHz ( Bessel ) and digitized at 20 kHz with a Digidata 1322A or 1440A digitizer and recorded on a computer using the software suite pClamp ( Molecular Devices , Sunnyvale , CA ) . Data was fitted with Origin ( OriginLab , Northampton , MA ) . Measurements were replicated on 3–5 membranes and average and SEM values were calculated for each data point . Dissociated dopamine neurons from the substantia nigra pars compacta were prepared from 13- to 19-day-old mice ( Kimm and Bean , 2014 ) and studied with whole-cell patch clamp recording . GIRK current was evoked by application of 100 µM baclofen using an external solution containing 16 mM KCl , 139 mM NaCl , 1 . 5 mM CaCl2 , 1 mM MgCl2 , 13 mM glucose , 10 mM HEPES , pH adjusted to 7 . 4 with NaOH . The 0 Na+ internal solution contained 140 mM K-gluconate , 13 . 5 mM NMDG-Cl , 1 . 6 mM MgCl2 , 0 . 09 mM EGTA , 4 mM MgATP , 14 mM creatine phosphate ( Tris salt ) , 0 . 3 mM GTP ( Tris salt ) , 9 mM HEPES , pH 7 . 4 . The 27 mM Na+ internal solution contained 13 . 5 mM NaCl , 13 . 5 mM Na-gluconate , 126 mM K-gluconate , 13 . 5 mM NMDG-Cl , 1 . 6 mM MgCl2 , 0 . 09 mM EGTA , 4 mM MgATP , 14 mM creatine phosphate ( Tris salt ) , 0 . 3 mM GTP ( Tris salt ) , 9 mM HEPES , pH 7 . 4 . Pipette resistances were 2 . 5–3 . 5 MOhm . Whole-cell current was recorded during voltage ramps ( 1 mV/ms ) from +8 to -147 mV delivered from a steady holding potential of -80 mV every 2 s . Baclofen-induced current was measured after 9–11 min of cell dialysis to allow equilibration with the internal solution . Baclofen-induced current was measured by averaging the current between -142 to -147 mV over ramps delivered during a 10-sec application of baclofen , subtracting the current before application of baclofen . Baclofen was applied in <1 s by moving the cell between a pair of quartz fiber flow pipes ( 250 µm internal diameter , 350 µm external diameter ) glued onto an aluminum rod whose temperature was controlled by resistive heating elements and a feedback-controlled temperature controller ( TC-344B , Warner Instruments , Hamden , CT ) . Recordings were made at 37°C . For GIRK channels with 4 Gβγ binding sites , channel activity as a function of Gβγ concentration can be expressed as: ( 1 ) A=b6Kdb4θ0+4b6Kdb3mθ1+6b5Kdb2m2θ2+4b3Kdbm3θ3+m4θ4b6Kdb4+4b6Kdb3m+6b5Kdb2m2+4b3Kdbm3+m4 , where Kdb is the equilibrium dissociation constant for Gβγ binding to the first binding site , b is the cooperativity factor for each successive Gβγ binding , m is the Gβγ concentration and θj is the channel activity with j Gβγ bound . If 4 Gβγ subunits are required to open GIRK ( θj = 0 when j ≠ 4 , θ4 = 1 ) , then ( 2 ) Ln ( A ) =4 ln ( m ) −ln ( 1b6Kdb4+4b6Kdb3m+6b5Kdb2m2+4b3Kdbm3+m4 ) ≈ 4 ln ( m ) −4 ln ( 1b3/2Kdb ) when m≪s×Kdb . s is the smallest of 4–1 , ( b/6 ) 1/2 , 4–1/3 b and b3/2 . Equation ( 2 ) indicates that at low enough Gβγ concentrations a log-log plot of channel activity A will be a linear function with a slope of 4 . Moreover , if 3 bound Gβγ subunits are sufficient to activate the channel the slope will be less than 4 . Thus , a log-log plot will reveal the number of Gβγ subunits required to open the channel if the slope can be measured at sufficiently low concentrations of Gβγ .	Signals from outside of a cell can alter the activity inside the cell . This process often involves members of a large family of proteins called G protein-coupled receptors ( GPCRs ) that are found on the surface of many cells in the body . When these receptors are activated they release a G protein on the inside of the cell that then splits into two parts . One of these parts – called the Gβγ subunit – can directly bind to , and open , a protein channel called a GIRK channel in the cell membrane . Once opened , these channels allow potassium ions to flow into the cell . GIRK channels are involved in a number of processes in the body . For example , GIRK2 is a major type of GIRK channel found in nerve cells . When this channel is activated the flow of potassium ions into the cell inhibits the nerve cell’s activity and makes it less likely to send electrical impulses . However , it was not clear how many Gβγ subunits are required to activate a GIRK2 channel . Now , Wang et al . report that four Gβγ subunits must bind to a GIRK2 channel and then work together to open it . This means that a GIRK2 channel will switch between a closed and an open state whenever the density of Gβγ subunits released onto the cell membrane reaches a certain threshold . Wang et al . also found that a high concentration of sodium ions in the cell causes the Gβγ subunits to bind more strongly to the GIRK2 channel; this makes that channel more likely to open and inhibit the nerve cell’s activity . This action serves to dampen down the activity of the most active neurons , because highly active nerve cells contain more sodium . Also , in a related study , Touhara et al . – who include many of the same researchers – discovered that sodium ions affect GIRK4 channels from heart cells in a similar way . These findings shed new light on G protein signaling , but there is still more that is not yet completely understood . Wang et al . ’s findings suggest that the concentration of Gβγ subunits in certain nerve cells is much higher than previously expected , and further work is now needed to explore how this might be achieved .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "Methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Cooperative regulation by G proteins and Na+ of neuronal GIRK2 K+ channels
The Alphaproteobacteria is an extraordinarily diverse and ancient group of bacteria . Previous attempts to infer its deep phylogeny have been plagued with methodological artefacts . To overcome this , we analyzed a dataset of 200 single-copy and conserved genes and employed diverse strategies to reduce compositional artefacts . Such strategies include using novel dataset-specific profile mixture models and recoding schemes , and removing sites , genes and taxa that are compositionally biased . We show that the Rickettsiales and Holosporales ( both groups of intracellular parasites of eukaryotes ) are not sisters to each other , but instead , the Holosporales has a derived position within the Rhodospirillales . A synthesis of our results also leads to an updated proposal for the higher-level taxonomy of the Alphaproteobacteria . Our robust consensus phylogeny will serve as a framework for future studies that aim to place mitochondria , and novel environmental diversity , within the Alphaproteobacteria . The Alphaproteobacteria is an extraordinarily diverse and disparate group of bacteria and well-known to most biologists for also encompassing the mitochondrial lineage ( Williams et al . , 2007; Roger et al . , 2017 ) . The Alphaproteobacteria has massively diversified since its origin , giving rise to , for example , some of the most abundant ( e . g . Pelagibacter ubique ) and metabolically versatile ( e . g . Rhodobacter sphaeroides ) cells on Earth ( Giovannoni , 2017; Madigan et al . , 2009 ) . The basic structure of the tree of the Alphaproteobacteria has largely been inferred through the analyses of 16S rRNA genes and several conserved proteins ( Garrity et al . , 2005; Lee et al . , 2005; Rosenberg et al . , 2014; Fitzpatrick et al . , 2006; Williams et al . , 2007; Brindefalk et al . , 2011; Georgiades et al . , 2011; Thrash et al . , 2011; Luo , 2015 ) . Today , eight major orders are well recognized , namely the Caulobacterales , Rhizobiales , Rhodobacterales , Pelagibacterales , Sphingomonadales , Rhodospirillales , Holosporales and Rickettsiales ( the latter two formerly grouped into the Rickettsiales sensu lato ) , and their interrelationships have also recently become better understood ( Viklund et al . , 2012; Viklund et al . , 2013; Rodríguez-Ezpeleta and Embley , 2012; Wang and Wu , 2014 ) . These eight orders were grouped into two subclasses by Ferla et al . ( 2013 ) : the subclass Rickettsiidae comprising the order Rickettsiales and Pelagibacterales , and the subclass Caulobacteridae comprising all other orders . The great diversity of the Alphaproteobacteria itself presents a challenge to deciphering the deepest divergences within the group . Such diversity encompasses a broad spectrum of genome ( nucleotide ) and proteome ( amino acid ) compositions ( e . g . the A + T%-rich Pelagibacterales versus the G + C%-rich Acetobacteraceae ) and molecular evolutionary rates ( e . g . the fast-evolving Pelagibacteriales , Rickettsiales or Holosporales versus many slow-evolving species in the Rhodospirillales ) ( Ettema and Andersson , 2009 ) . This diversity may lead to pervasive artefacts when inferring the phylogeny of the Alphaproteobacteria , for example , long-branch attraction ( LBA ) between the Rickettsiales and Pelagibacterales , especially when including mitochondria ( Rodríguez-Ezpeleta and Embley , 2012; Viklund et al . , 2012; Viklund et al . , 2013; Luo , 2015 ) . Moreover , there are still important unknowns about the deep phylogeny of the Alphaproteobacteria ( Williams et al . , 2007; Ferla et al . , 2013 ) , for example , the divergence order among the Rhizobiales , Rhodobacterales and Caulobacterales ( Williams et al . , 2007 ) , the monophyly of the Pelagibacterales ( Viklund et al . , 2013 ) and the Rhodospirillales ( Ferla et al . , 2013 ) , and the precise placement of the Rickettsiales and its relationship to the Holosporales ( Wang and Wu , 2013; Martijn et al . , 2018 ) . Systematic errors stemming from using over-simplified evolutionary models ( which often do not fit complex data as well by , for example , not accounting for compositional heterogeneity across sites or branches ) are perhaps the major confounding and limiting factor to inferring deep evolutionary relationships; the number of taxa and genes ( or sites ) can also be important factors . Previous multi-gene tree studies of the Alphaproteobacteria were compromised by at least one of these problems , namely , simpler or less realistic evolutionary models ( because they were not available at the time; for example , Williams et al . , 2007 used the simple WAG+Γ4 model that cannot account for compositional heterogeneity across sites ) , poor or uneven taxon sampling ( because the focus was too narrow or few genomes were available; for example , Williams et al . , 2007 had very few rhodospirillaleans and no holosporaleans; Georgiades et al . , 2011 included only 42 alphaproteobacteria with only one pelagibacteralean ) or a small number of genes ( because the focus was mitochondria; for example , Rodríguez-Ezpeleta and Embley , 2012 used 24 genes; Wang and Wu , 2015 relied on 29 genes; Martijn et al . , 2018 also used 24 genes; or because only a small set of 28 compositionally homogeneous genes was used , for example , Luo , 2015 ) . The most recent study on the phylogeny of the Alphaproteobacteria , and mitochondria , attempted to counter systematic errors ( or phylogenetic artefacts ) by reducing amino acid compositional heterogeneity ( Martijn et al . , 2018 ) . Even though some deep relationships were not robustly resolved , these analyses suggested that the Pelagibacterales , Rickettsiales and Holosporales , which have compositionally biased genomes , are not each other’s closest relatives ( Martijn et al . , 2018 ) . A resolved and robust phylogeny of the Alphaproteobacteria is fundamental to addressing questions such as how streamlined bacteria , intracellular parasitic bacteria , or mitochondria evolved from their alphaproteobacterial ancestors . Therefore , a systematic study of the different biases affecting the phylogeny of the Alphaproteobacteria , and its underlying data , is much needed . Here , we revised the phylogeny of the Alphaproteobacteria by using a large dataset of 200 conserved single-copy genes and employing carefully designed strategies aimed at alleviating phylogenetic artefacts . We found that amino acid compositional heterogeneity , and more generally long-branch attraction , were major confounding factors in estimating phylogenies of the Alphaproteobacteria . In order to counter these biases , we used novel dataset-specific profile mixture models and recoding schemes ( both specifically designed to ameliorate compositional heterogeneity ) , and removed sites , genes and taxa that were compositionally biased . We also present three draft genomes for endosymbiotic alphaproteobacteria belonging to the Rickettsiales and Holosporales: ( 1 ) an undescribed midichloriacean endosymbiont of Peranema trichophorum , ( 2 ) an undescribed rickettsiacean endosymbiont of Stachyamoeba lipophora , and ( 3 ) the holosporalean ‘Candidatus Finniella inopinata’ , an endosymbiont of the rhizarian amoeboflagellate Viridiraptor invadens ( Hess et al . , 2016 ) . Our results provide the first strong evidence that the Holosporales is not closely related to the Rickettsiales and originated instead from within the Rhodospirillales . We incorporate these and other insights regarding the deep phylogeny of the Alphaproteobacteria into an updated taxonomy . We sequenced the genomes of the novel holosporalean ‘Candidatus Finniella inopinata’ , an endosymbiont of the rhizarian amoeboflagellate Viridiraptor invadens ( Hess et al . , 2016 ) , and two undescribed rickettsialeans , one associated with the heterolobosean amoeba Stachyamoeba lipophora and the other with the euglenoid flagellate Peranema trichophorum . The three genomes are small with a reduced gene number and high A + T% content , strongly suggesting an endosymbiotic lifestyle ( Table 1 ) . Comparisons of their rRNA genes show that these genomes are truly novel , being considerably divergent from other described alphaproteobacteria . As of February 2018 , the closest 16S rRNA gene to that of the Stachyamoeba-associated rickettsialean belongs to Rickettsia massiliae str . AZT80 , with only 88% identity . On the other hand , the closest 16S rRNA gene to that of the Peranema-associated rickettsialean belongs to an endosymbiont of Acanthamoeba sp . UWC8 , which is only 92% identical . Phylogenetic analysis of both the 16S rRNA gene and a dataset that comprises 200 single-copy conserved marker genes ( see below ) confirm that each species belongs to different families and orders within the Alphaproteobacteria ( Supplementary file 1 and Figure 2—figure supplement 1 ) . ‘Candidatus Finniella inopinata’ belongs to the recently described ‘Candidatus Paracaedibacteraceae’ in the Holosporales ( Hess et al . , 2016 ) , whereas the Stachyamoeba-associated rickettsialean belongs to the Rickettsiaceae , and the Peranema-associated rickettsialean belongs to the ‘Candidatus Midichloriaceae’ , in the Rickettsiales . The average-linkage clustering of amino acid compositions shows that the Rickettsiales , Pelagibacterales ( together with alphaproteobacterium HIMB59 ) and Holosporales are clearly distinct from other alphaproteobacteria . This indicates that these three taxa have divergent proteome amino acid compositions ( Figure 1A ) . These taxa also have the lowest GARP:FIMNKY amino acid ratios in all the Alphaproteobacteria ( Figure 1A; GARP amino acids are encoded by G + C%-rich codons , whereas FIMNKY amino acids are encoded by A + T%-rich codons . Proteomes that have low GARP:FIMNKY ratios are compositionally biased and therefore come from A + T%-rich genomes ) ; the Pelagibacterales ( including alphaproteobacterium HIMB59 ) being the most divergent , followed by the Rickettsiales and then the Holosporales . Such biased amino acid compositions appear to be the consequence of genome nucleotide compositions that are strongly biased toward high A + T%—a scatter plot of genome G + C% and proteome GARP:FIMNKY ratios shows a similar clustering of the Rickettsiales , Pelagibacterales ( including alphaproteobacterium HIMB59 ) and Holosporales ( Figure 1B ) . This compositional similarity in the proteomes of the Rickettsiales , Pelagibacterales ( plus alphaproteobacterium HIMB59 ) and Holosporales , which also turn out to be the longest-branched alphaproteobacterial groups in previously published phylogenies ( e . g . Wang and Wu , 2015 ) , could be the outcome of either a shared evolutionary history ( i . e . the groups are most closely related to one another ) , or alternatively , evolutionary convergence ( e . g . because of similar lifestyles or evolutionary trends toward small cell and genome sizes ) . As a first step to discriminate between these two alternatives , we used maximum likelihood to estimate a tree on a dataset that comprised 200 single-copy and rarely laterally transferred marker genes for the Alphaproteobacteria ( as determined by Phyla-AMPHORA; see Materials and methods for more details; Wang and Wu , 2013 ) under the site-heterogenous model LG+PMSF ( ES60 ) +F+R6 . The resulting tree united the Rickettsiales , Pelagibacterales ( with alphaproteobacterium HIMB59 at its base ) and Holosporales in a fully supported clade ( Figure 2A; see Figure 2—figure supplement 1 for labeled trees ) . The clustering of these three groups is suggestive of a phylogenetic artefact ( e . g . long-branch attraction or LBA ) ; indeed , such a pattern resembles the one seen in the tree of proteome amino acid compositions ( see Figure 1A ) . This is because the three groups have the longest branches in the Alphaproteobacteria tree and have compositionally biased and fast-evolving genomes ( see Figure 2 ) . If evolutionary convergence in amino acid compositions is confounding phylogenetic inference for the Alphaproteobacteria , methods aimed at reducing compositional heterogeneity might disrupt the clustering of the Rickettsiales , Pelagibacterales and Holosporales . To further test whether the clustering of the Rickettsiales , Pelagibacterales and Holosporales is real or artefactual , we used several different strategies to reduce the compositional heterogeneity of our dataset ( see Figure 2—figure supplement 2 for the diverse strategies employed ) . When removing the 50% most compositionally biased ( heterogeneous ) sites according to ɀ ( a novel metric that measures amino acid compositional disparity at a site; see Materials and methods ) , the clustering between the Rickettsiales , Pelagibacterales , alphaproteobacterium HIMB59 and Holosporales is disrupted ( Figure 2B; see also Figure 2—figure supplement 3 ) . The new more derived placements for the Pelagibacterales , alphaproteobacterium HIMB59 and Holosporales are well supported ( further described below ) , and support tends to increase as compositionally biased sites are removed ( Supplementary file 2A ) . Furthermore , when each of these long-branched and compositionally biased taxa is analyzed in isolation ( i . e . in the absence of the others ) , and compositional heterogeneity is further decreased , new phylogenetic patterns emerge that are incompatible , or in conflict , with their clustering ( Figure 2—figure supplement 4 and Figure 3—figure supplements 1–5 ) . Various strategies to reduce compositional heterogeneity , such as removing the most compositionally biased sites , recoding the data into reduced character-state alphabets , or using only the most compositionally homogeneous genes , converge to very similar phylogenetic patterns for the Alphaproteobacteria in which the clustering of the Rickettsiales , Pelagibacterales , alphaprotobacterium HIMB59 and Holosporales is disrupted; the Pelagibacterales , alphaproteobacterium HIMB59 and Holosporales have much more derived phylogenetic placements ( e . g . , Figure 3 , Figure 2—figure supplement 4 and Figure 3—figure supplements 1–5 ) . On the other hand , removing fast-evolving sites does not disrupt the clustering of these three long-branched groups ( Supplementary file 2B ) , suggesting that high evolutionary rates per site are not a major confounding factor when inferring the phylogeny of the Alphaproteobacteria . The Holosporales has traditionally been considered part of the Rickettsiales sensu lato because it appears as sister to the Rickettsiales in many trees ( e . g . Hess et al . , 2016; Montagna et al . , 2013; Santos and Massard , 2014 ) . It is exclusively composed of endosymbiotic bacteria living within diverse eukaryotes , and such a lifestyle is shared with all other members of the Rickettsiales ( with the possible exception of a recently reported ectosymbiotic rickettialean; see Castelli et al . , 2018 ) . When we decrease , and then account for , compositional heterogeneity , we recover tree topologies in which the Holosporales moves away from the Rickettsiales ( e . g . Figure 2B , Figure 2—figure supplement 4B and D ) . For example , the Holosporales becomes sister to all free-living alphaproteobacteria ( the Caulobacteridae ) when only the 40 most homogeneous genes are used ( Figure 2—figure supplement 4D ) or when 10% of the most compositionally biased sites are removed ( Supplementary file 2A ) . When compositional heterogeneity is further decreased by removing 50% of the most compositionally biased sites , the Holosporales becomes sister to the Rhodospirillales ( Figure 2B and Supplementary file 2A; and see also Figure 2—figure supplement 4B ) . Similarly , when the long-branched and compositionally biased Rickettsiales , Pelagibacterales , and alphaproteobacterium HIMB59 ( plus the extremely long-branched genera Holospora and ‘Candidatus Hepatobacter’ ) are removed , after compositional heterogeneity had been decreased through site removal , the Holosporales move to a much more derived position well within the Rhodospirillales ( Figure 3A , Figure 3—figure supplement 1B and C and Figure 3—figure supplement 6 ) . If the very compositionally biased and fast-evolving Holospora and ‘Candidatus Hepatobacter’ are left in , the Holosporales are pulled away from its derived position and the whole clade moves closer to the base of the tree ( Figure 3—figure supplement 7 ) . The same pattern in which the Holosporales is derived within the Rhodospirillales is seen when these same taxa are removed , and the data are then recoded into four- or six-character states ( Figure 3B , Figure 3—figure supplement 6 and Figure 3—figure supplement 8 ) . Specifically , the Holosporales now consistently branches as sister to a subgroup of rhodospirillaleans that includes , among others , the epibiotic predator Micavibrio aeruginosavorus and the purple nonsulfur bacterium Rhodocista centenaria ( the Azospirillaceae , see below ) ( Figure 3 ) . This new placement of the Holosporales has nearly full support under both maximum likelihood ( >95% UFBoot; see Figure 3 ) and Bayesian inference ( >0 . 95 posterior probability; see Figure 3—figure supplement 6 ) . Thus , three different analyses independently converge to the same pattern and support a derived origin of the Holosporales within the Rhodospirillales: ( 1 ) removal of compositionally biased sites ( Figure 3A ) , ( 2 ) data recoding into four-character states using the dataset-specific scheme S4 ( Figure 3B and Figure 3—figure supplement 7 ) , and ( 3 ) data recoding into six-character states using the dataset-specific scheme S6 ( Figure 3—figure supplement 8 ) ; each of these strategies had to be combined with the removal of the Pelagibacterales , alphaproteobacterium HIMB59 , and Rickettsiales to recover this phylogenetic position for the Holosporales . A fourth independent analysis further supports a derived placement of the Holosporales nested within the Rhodospirillales . Bayesian inference using the CAT-Poisson+Γ4 model , on a dataset whose compositional heterogeneity had been decreased by removing 50% of the most compositionally biased sites but for which no taxon had been removed , also recovered the Holosporales as sister to the Azospirillaceae ( see Figure 2—figure supplement 3 ) . The Rhodospirillales is an ancient and highly diversified group , but unfortunately this is rarely obvious from published phylogenies because most studies only include a few species for this order ( Williams et al . , 2007; Georgiades et al . , 2011; Ferla et al . , 2013 ) . We have included a total of 31 Rhodospirillales taxa to better cover its diversity . Such broad sampling reveals trees with five clear subgroups within the Rhodospirillales that are well-supported in most of our analyses ( e . g . Figures 2B and 3 ) . First is the Acetobacteraceae which comprises acetic acid ( e . g . Acetobacter oboediens ) , acidophilic ( e . g . Acidisphaera rubrifaciens ) , and photosynthetic ( bacteriochlorophyll-containing; for example , Rubritepida flocculans ) bacteria . The Acetobacteraceae is strongly supported and relatively divergent from all other families within the Rhodospirillales . Sister to the Acetobacteraceae is another subgroup that comprises many photosynthetic bacteria , including the type species for the Rhodospirillales , Rhodospirillum rubrum , as well as the magnetotactic bacterial genera Magnetospirillum , Magnetovibrio and Magnetospira ( Figure 3 ) . This subgroup best corresponds to the poorly defined and paraphyletic Rhodospirillaceae family . We amend the Rhodospirillaceae taxon and restrict it to the clade most closely related to the Acetobacteraceae . As described above , when artefacts are accounted for , the Holosporales most likely branches within the Rhodospirillales and therefore we suggest the Holosporales sensu Szokoli et al . ( 2016 ) be lowered in rank to the family Holosporaceae ( containing for example , Caedibacter sp . 37–49 and ‘Candidatus Paracaedibacter symbiosus’ ) , which is sister to the Azospirillaceae fam . nov . ( Figure 3 ) . The Azospirillaceae contains the purple bacterium Rhodocista centenaria and the epibiotic ( neither periplasmic nor intracellular ) predator Micavibrio aeruginosavorus , among others . The Holosporaceae and the Azospirillaceae clades appear to be sister to the Rhodovibriaceae fam . nov . ( Figure 3 ) , a well-supported group that comprises the purple nonsulfur bacterium Rhodovibrio salinarum , the aerobic heterotroph Kiloniella laminariae , and the marine bacterioplankter ‘Candidatus Puniceispirillum marinum’ ( or the SAR116 clade ) . Each of these subgroups and their interrelationships—with the exception of the Holosporaceae that branches within the Rhodospirillales only after compositional heterogeneity is countered—are strongly supported in nearly all of our analyses ( e . g . see Figures 2B and 3 ) . The Geminicocacceae is a recently proposed family within the Rhodospirillales ( Proença et al . , 2018 ) . It is currently represented by only two genera , Geminicoccus and Arboriscoccus ( Foesel et al . , 2007; Proença et al . , 2018 ) . In most of our trees , however , Tistrella mobilis is often sister to Geminicoccus roseus with full statistical support ( e . g . , Figures 2B and 3A , but see Figure 3—figure supplement 6 for an exception ) and we therefore consider it to be part of the Geminococcaceae . Interestingly , the Geminicoccaceae tends to have two alternative stable positions in our analyses , either as sister to all other families of the Rhodospirillales ( e . g . Figures 2A and 3A ) , or as sister to all other orders of the Caulobacteridae ( i . e . representing the most basal lineage of free-living alphaproteobacteria; Figure 2B and Figure 2—figure supplement 3B , Figure 3B and Figure 3—figure supplement 6 , or Figure 2—figure supplement 4B , Figure 3—figure supplement 1C , Figure 3—figure supplement 2B–D , Figure 3—figure supplement 3B and D , and Figure 3—figure supplement 5C ) . Our analyses designed to alleviate compositional heterogeneity , specifically site removal and recoding ( without taxon removal ) , favor the latter position for the Geminicoccaceae ( Figures 2B and 3B ) . Moreover , as compositionally biased sites are progressively removed , support for the affiliation of the Geminicoccaceae with the Rhodospirillales decreases , and after 50% of the sites have been removed , the Geminicoccaceae emerges as sister to all other free-living alphaproteobacteria with strong support ( >95% UFBoot; Supplementary file 2A ) . In further agreement with this trend , the much simpler model LG4X places the Geminicocacceae in a derived position as sister to the Acetobacteraceae ( Figure 2—figure supplement 5 ) , but as model complexity increases , and compositional heterogeneity is reduced , the Geminicoccaceae moves closer to the base of the Alphaproteobacteria ( Figures 2A and 3A ) . Such a placement suggests that the Geminicoccaceae may be a novel and independent order-level lineage in the Alphaproteobacteria . However , because of the uncertainty in our results we opt here for conservatively keeping the Geminicoccaceae as the sixth family of the Rhodospirillales ( Figure 3A ) . The clustering of the Pelagibacterales ( formerly the SAR11 clade ) with the Rickettsiales and Holosporales is more easily disrupted than that of the Holosporales , either when long-branched ( or compositionally biased ) taxon removal is performed to control for compositional attractions or not . The removal of compositionally biased sites ( from 30% on; 16 , 320 out of 54 , 400 sites; see Supplementary file 2A , Figure 2B , Figure 2—figure supplement 3B and Figure 3—figure supplement 4B ) , data recoding into four-character states ( Figure 3—figure supplement 4C ) , and a set of the most compositionally homogeneous genes ( Figure 3—figure supplement 4D ) , all support a derived placement of the Pelagibacterales as sister to the Rhodobacterales , Caulobacterales and Rhizobiales . Attempts to account for compositional heterogeneity both across sites ( e . g . Rodríguez-Ezpeleta and Embley , 2012; Viklund et al . , 2012; Viklund et al . , 2013; Martijn et al . , 2018 ) and taxa ( e . g . Luo et al . , 2013; Luo , 2015 ) tend to disrupt the potentially artefactual clustering of the Pelagibacterales and the Rickettsiales ( in contrast to the studies of for example , Williams et al . , 2007; Thrash et al . , 2011; Georgiades et al . , 2011 ) that did not account for compositional heterogeneity ) . The Caulobacterales is sister to the Rhizobiales , and the Rhodobacterales sister to both ( e . g . Figures 2B and 3 ) . This is consistent throughout most of our results and such interrelationships become very robustly supported as compositional heterogeneity is increasingly alleviated ( Supplementary file 2A ) . The placement of the Rickettsiales as sister to the Caulobacteridae ( i . e . all other alphaproteobacteria ) remains stable across different analyses ( see Supplementary file 2A , and also Figure 2B and Figure 3—figure supplement 2 ) ; this is also true when the other long-branched taxa , the Pelagibacterales , alphaproteobacterium HIMB59 and Holosporales , and even the Beta- Gammaproteobacteria outgroup , are removed ( see Figure 3—figure supplement 2 and Figure 3—figure supplement 3 ) . Yet , the interrelationships inside the Rickettsiales order remain uncertain; the ‘Candidatus Midichloriaceae’ becomes sister to the Anaplasmataceae when fast-evolving sites are removed ( Supplementary file 2B ) , but to the Rickettsiaceae when compositionally biased sites are removed ( Supplementary file 2A ) . The placement of alphaproteobacterium HIMB59 is uncertain ( e . g . see Figure 2 and Figure 2—figure supplement 3 , and Figure 2—figure supplement 4 and Figure 3—figure supplement 5; in contrast to Grote et al . , 2012 ) ; taxon-removal analyses suggest that alphaproteobacterium HIMB59 is sister to the Caulobacteridae ( Figure 3—figure supplement 5 ) , but the inclusion of any other long-branched group immediately destabilizes this position ( e . g . see Figure 2 and Figure 2—figure supplement 2 , and Figure 2—figure supplement 4 ) . This is consistent with previous reports that suggest that alphaproteobacterium HIMB59 is not closely related to the Pelagibacterales ( Viklund et al . , 2013; Martijn et al . , 2018 ) . We employed a combination of methods to decrease compositional heterogeneity in order to disrupt artefacts that arise when inferring the phylogeny of the Alphaproteobacteria . This is an example of the complex nature of the historical signal contained in modern genomes and the limitations of our current evolutionary models to capture these signals . A robust phylogeny of the Alphaproteobacteria is a precondition for placing the mitochondrial lineage . This is because including mitochondria certainly exacerbates the already strong biases in the data , and therefore represents additional sources of artefacts in phylogenetic inference ( as seen in Wang and Wu , 2015 ) where the Holosporales is attracted by both mitochondria and the Rickettsiales ) . The robust phylogenetic framework developed here will serve as a reference for future studies that aim to place mitochondria and novel not-yet-cultured environmental diversity within the Alphaproteobacteria . Rickettsidae emend . ( Alphaproteobacteria ) Rickettsia is the type genus of the subclass . The Rickettsidae subclass is here amended by redefining its circumscription so it remains monophyletic by excluding the Pelagibacterales order . The emended Rickettsidae subclass within the Alphaproteobacteria class is defined based on phylogenetic analyses of 200 genes which are predominantly single-copy and vertically inherited ( unlikely laterally transferred ) when compositional heterogeneity was decreased by site removal or recoding . Phylogenetic ( node-based ) definition: the least inclusive clade containing Anaplasma phagocytophilum HZ , Rickettsia typhi Wilmington , and ‘Candidatus Midichloria mitochondrii’ IricVA . The Rickettsidae does not include: Pelagibacter sp . HIMB058 , ‘Candidatus Pelagibacter sp . ’ IMCC9063 , alphaproteobacterium HIMB59 , Caedibacter sp . 37–49 , ‘Candidatus Nucleicultrix amoebiphila’ FS5 , ‘Candidatus Finniella lucida’ , Holospora obtusa F1 , Sneathiella glossodoripedis JCM 23214 , Sphingomonas wittichii , and Brevundimonas subvibrioides ATCC 15264 . Caulobacteridae emend . ( Alphaproteobacteria ) Caulobacter is the type genus of the subclass . The Caulobacteridae subclass is here amended by redefining its circumscription so it remains monophyletic by including the Pelagibacterales order . The emended Caulobacteridae subclass within the Alphaproteobacteria class is defined based on phylogenetic analyses of 200 genes which are predominantly single-copy and vertically inherited ( unlikely laterally transferred ) when compositional heterogeneity was decreased by site removal or recoding . Phylogenetic ( node-based ) definition: the least inclusive clade containing Pelagibacter sp . HIMB058 , ‘Candidatus Pelagibacter sp . ’ IMCC9063 , alphaproteobacterium HIMB59 , Caedibacter sp . 37–49 , ‘Candidatus Nucleicultrix amoebiphila’ FS5 , ‘Candidatus Finniella lucida’ , Holospora obtusa F1 , Sneathiella glossodoripedis JCM 23214 , Sphingomonas wittichii , and Brevundimonas subvibrioides ATCC 15264 . The Caulobacteridae does not include: Anaplasma phagocytophilum HZ , Rickettsia typhi Wilmington , and ‘Candidatus Midichloria mitochondrii’ IricVA . Azospirillaceae fam . nov . ( Rhodospirillales , Alphaproteobacteria ) Azospirillum is the type genus of the family . This new family within the Rhodospirillales order is defined based on phylogenetic analyses of 200 genes which are predominantly single-copy and vertically inherited ( unlikely laterally transferred ) . Phylogenetic ( node-based ) definition: the least inclusive clade containing Micavibrio aeruginoavorus ARL-13 , Rhodocista centenaria SW , and Inquilinus limosus DSM 16000 . The Azospirillaceae does not include: Rhodovibrio salinarum DSM 9154 , ‘Candidatus Puniceispirillum marinum’ IMCC 1322 , Rhodospirillum rubrum ATCC 11170 , Terasakiella pusilla DSM 6293 , Acidiphilium angustum ATCC 49957 , and Elioraea tepidiphila DSM 17972 . Rhodovibriaceae fam . nov . ( Rhodospirillales , Alphaproteobacteria ) Rhodovibrio is the type genus of the family . This new family within the Rhodospirillales order is defined based on phylogenetic analyses of 200 genes which are predominantly single-copy and vertically inherited ( unlikely laterally transferred ) . Phylogenetic ( node-based ) definition: the least inclusive clade containing Rhodovibrio salinarum DSM 9154 , Kiloniella laminariae DSM 19542 , Oceanibaculum indicum P24 , Thalassobaculum salexigens DSM 19539 and ‘Candidatus Puniceispirillum marinum’ IMCC 1322 . The Rhodovobriaceae does not include: Rhodospirillum rubrum ATCC 11170 , Terasakiella pusilla DSM 6293 , Rhodocista centenaria SW , Micavibrio aeruginoavorus ARL-13 , Acidiphilium angustum ATCC 49957 , and Elioraea tepidiphila DSM 17972 . Rhodospirillaceae emend . ( Rhodospirillales , Alphaproteobacteria ) Rhodospirillum is the type genus of the family . The Rhodospirillaceae family is here amended by redefining its circumscription so it remains monophyletic . The emended Rhodospirillaceae family within the Rhodospirillales order is defined based on phylogenetic analyses of 200 genes which are predominantly single-copy and vertically inherited ( unlikely laterally transferred ) . Phylogenetic ( node-based ) definition: the least inclusive clade containing Rhodospirillum rubrum ATCC 11170 , Roseospirillum parvum 930 l , Magnetospirillum magneticum AMB-1 and Terasakiella pusilla DSM 6293 . The Rhodospirillaceae does not include: Rhodocista centenaria SW , Micavibrio aeruginoavorus ARL-13 , ‘Candidatus Puniceispirillum marinum’ IMCC 1322 , Rhodovibrio salinarum DSM 9154 , Elioraea tepidiphila DSM 17972 , and Acidiphilium angustum ATCC 49957 . Holosporaceae ( Rhodospirillales , Alphaproteobacteria ) Holospora is the type genus of the family . The Holosporaceae family as defined here has the same taxon circumscription as the Holosporales order sensu Szokoli et al . , 2016 , but it is here lowered to the family level and placed within the Rhodospirillales order . The new family rank-level for this group is based on the phylogenetic analysis of 200 genes , which are predominantly single-copy and vertically inherited ( unlikely laterally transferred ) , when compositional heterogeneity was decreased by site removal or recoding ( and coupled to the removal of the long-branched taxa Pelagibacterales and Rickettsiales ) . The family contains three subfamilies ( lowered in rank from a former family level ) and one formally undescribed clade , namely , the Holosporodeae , and ‘Candidatus Paracaedibacteriodeae’ , ‘Candidatus Hepatincolodeae’ , and the Caedibacter-Nucleicultrix clade . Cultures of Viridiraptor invadens strain Virl02 , the host of ‘Candidatus Finniella inopinata’ , were grown on the filamentous green alga Zygnema pseudogedeanum strain CCAC 0199 as described in Hess and Melkonian ( 2013 ) . Once the algal food was depleted , Viridiraptor cells were harvested by filtration through a cell strainer ( mesh size 40 µm to remove algal cell walls ) and centrifugation ( ~1000 g for 15 min ) . For short-read sequencing , DNA extraction of total gDNA was carried out with the ZR Fungal/Bacterial DNA MicroPrep Kit ( Zymo Research ) using a BIO101/Savant FastPrep FP120 high-speed bead beater and 20 µL of proteinase K ( 20 mg/mL ) . A sequencing library was made using the NEBNext Ultra II DNA Library Prep Kit ( New England Biolabs ) . Paired-end DNA sequencing libraries were sequenced with an Illumina MiSeq instrument ( Dalhousie University; Canada ) . ( number of reads = 3 , 006 , 282 , read length = 150 bp ) . For long-read sequencing , DNA extraction was performed using a CTAB and phenol-chloroform method . Total gDNA was further cleaned through a QIAGEN Genomic-Tip 20/G . A sequencing library was made using the Nanopore Ligation Sequencing Kit 1D ( SQK-LSK108 ) . Sequencing was done on a portable MinION instrument ( Oxford Nanopore Technologies ) . ( total bases = 191 , 942 , 801 bp , number of reads = 73 , 926 , longest read = 32 , 236 bp , mean read length = 2 , 596 bp , mean read quality = 9 . 4 ) . Peranema trichophorum strain CCAP 1260/1B was obtained from the Culture Collection of Algae and Protozoa ( CCAP , Oban , Scotland ) and grown in liquid Knop media plus egg yolk crystals . Total gDNA was extracted following Lang and Burger ( 2007 ) . A paired-end sequencing library was made using a TruSeq DNA Library Prep Kit ( Illumina ) . DNA sequencing libraries were sequenced with an Illumina MiSeq instrument ( Genome Quebec Innovation Centre; Canada ) . ( number of reads = 4 , 157 , 475 , read length = 300 bp ) . Stachyamoeba lipophora strain ATCC 50324 cells feeding on Escherichia coli were harvested and then broken up with pestle and mortar in the presence of glass beads ( <450 µm diameter ) . Total gDNA was extracted using the QIAGEN Genomic G20 Kit . A paired-end sequencing library was made using a TruSeq DNA Library Prep Kit ( Illumina ) . DNA sequencing libraries were sequenced with an Illumina MiSeq instrument ( Genome Quebec Innovation Centre; Canada ) . ( number of reads = 35 , 605 , 415 , read length = 100 bp ) . Short sequencing reads produced in an Illumina MiSeq from Viridiraptor invadens , Peranema trichophorum , and Stachyamoeba lipophora were first assessed with FASTQC v0 . 11 . 6 and then , based on its reports , trimmed with Trimmomatic v0 . 32 ( Bolger et al . , 2014 ) using the options: HEADCROP:16 LEADING:30 TRAILING:30 MINLEN:36 . Illumina adapters were similarly removed with Trimmomatic v0 . 32 using the option ILLUMINACLIP . Long-sequencing reads produced in a Nanopore MinION instrument from Viridiraptor invadens were basecalled with Albacore v2 . 1 . 7 , adapters were removed with Porechop v0 . 2 . 3 , lambda phage reads were removed with NanoLyse v0 . 5 . 1 , quality filtering was done with NanoFilt v2 . 0 . 0 ( with the options ‘--headcrop 50 -q 8 l 1000’ ) , and identity filtering against the high-quality short Illumina reads was done with Filtlong v0 . 2 . 0 ( and the options ‘--keep_percent 90 --trim --split 500 --length_weight 10 min_length 1000’ ) . Statistics were calculated throughout the read processing workflow with NanoStat v0 . 8 . 1 and NanoPlot v1 . 9 . 1 . A hybrid co-assembly of both processed Illumina short reads and Nanopore long reads from Viridiraptor invadens was done with SPAdes v3 . 6 . 2 ( Bankevich et al . , 2012 ) . Assemblies of the Illumina short reads from Peranema trichophorum and Stachyamobea lipophora were separately done with SPAdes v3 . 6 . 2 ( Bankevich et al . , 2012 ) . The resulting assemblies for both Viridiraptor invadens and Peranema trichophorum were later separately processed with the Anvi’o v2 . 4 . 0 pipeline ( Eren et al . , 2015 ) and refined genome bins corresponding to ‘Candidatus Finniella inopinata’ and the Peranema-associated rickettsialean were isolated primarily based on tetranucleotide sequence composition and taxonomic affiliation of its contigs . A single contig corresponding to the genome of the Stachyamoeba-associated rickettsialean was obtained from its assembly and this was circularized by collapsing the overlapping ends of the contig . Gene prediction and genome annotation was carried out with Prokka v . 1 . 13 ( see Table 1 ) . The selection of 120 taxa was largely based on the phylogenetically diverse set of alphaproteobacteria determined by Wang and Wu ( 2015 ) . To this set of taxa , recently sequenced and divergent unaffiliated alphaproteobacteria were added , as well as those claimed to constitute novel order-level taxa . Some other groups , like the Pelagibacterales , Rhodospirillales and the Holosporales , were expanded to better represent their diversity . A set of four betaproteobacteria and four gammaproteobacterial were used as outgroup ( see Figure 2—figure supplement 6 for taxon names; see Supplementary file 2C for accession numbers ) . A set of 200 gene markers ( 54 , 400 sites; 9 . 03% missing data , see Figure 2—figure supplement 6 ) defined by Phyla-AMPHORA was used ( Wang and Wu , 2013 ) . The genes are single-copy and predominantly vertically inherited as assessed by congruence among them ( Wang and Wu , 2013 ) . In brief , Phyla-AMPHORA searches for each marker gene using a profile Hidden Markov Model ( HMM ) , then aligns the best hits to the profile HMM using hmmalign of the HMMER suite , and then trims the alignments using pre-computed quality scores ( the mask ) previously generated using the probabilistic masking program ZORRO ( Wu et al . , 2012; Wang and Wu , 2013 ) . Phylogenetic trees for each marker gene were inferred from the trimmed multiple alignments in IQ-TREE v1 . 5 . 5 ( Minh et al . , 2013; Nguyen et al . , 2015 ) and under the model LG4X + F model . Single-gene trees were examined individually to remove distant paralogues , contaminants or laterally transferred genes . All this was done before concatenating the single-gene alignments into a supermatrix with SequenceMatrix v 1 . 8 ( Vaidya et al . , 2011 ) . Another smaller dataset of 40 compositionally homogenous genes ( 5570 sites; 5 . 98% missing data ) was built by selecting the least compositionally heterogeneous genes from the larger 200 gene set according compositional homogeneity tests performed in P4 ( Foster , 2004 ) ; see Supplementary file 2D for a list of the 40 most compositionally homogenous genes ) . This was done as an alternative way to overcome the strong compositional heterogeneity observed in datasets for the Alphaproteobacteria with a broad selection of taxa . In brief , the P4 tests rely on simulations based on a provided tree ( here inferred for each gene under the model LG4X + F in IQ-TREE ) and a model ( LG + F + G4 available in P4 ) to obtain proper null distributions to which to compare the X2 statistic . Most standard tests for compositional homogeneity ( those that do not rely on simulate the data on a given tree ) ignore correlation due to phylogenetic relatedness , and can suffer from a high probability of false negatives ( Foster , 2004 ) . Variations of our full set were made to specifically assess the placement of each long-branched and compositionally biased group individually . In other words , each group with comparatively long branches ( the Rickettsiales , Pelagibacterales , Holosporales , and alphaproteobacterium HIMB59 ) was analyzed in isolation , that is , in the absence of other long-branched and compositionally biased taxa . This was done with the purpose of reducing the potential artefactual attraction among these groups . Taxon removal was done in addition to compositionally biased site removal and data recoding into reduced character-state alphabets ( for a summary of the different methodological strategies employed see Figure 2—figure supplement 2 ) . As an effort to reduce artefacts in phylogenetic inference from our dataset ( which might stem from extreme divergence in the evolution of the Alphaproteobacteria ) , we removed sites estimated to be highly compositionally heterogeneous or fast evolving . The compositional heterogeneity of a site was estimated by using a metric intended to measure the degree of disparity between the most %AT-rich taxa and all others . Taxa were ordered from lowest to highest proteome GARP:FIMNKY ratios; ‘GARP’ amino acids are encoded by %GC-rich codons , whereas ‘FIMNKY’ amino acids are encoded by %AT-rich codons . The resulting plot was visually inspected and a GARP:FIMNKY ratio cutoff of 1 . 06 ( which represented a discontinuity or gap in the distribution which separated the long-branched and compositionally biased taxa Pelagibacterales , Holosporales and Rickettsiales from all others ) was chosen to divide the dataset into low GARP:FMINKY ( or %AT-rich ) and higher GARP:FIMNKY ( or ‘GC-rich’ ) taxa ( Figure 2—figure supplement 7 ) . Next , we determined the degree of compositional bias per site ( ɀ ) for the frequencies of both FIMNKY and GARP amino acids between the %AT-rich and all other ( ‘GC-rich’ ) alphaproteobacteria . To calculate this metric for each site the following formula was used:ɀ= ( πFIMNKY%AT−rich−πFIMNKY%GC−rich ) + ( πGARP%GC−rich−πGARP%AT−rich ) where πFIMNKY and πGARP are the sum of the frequencies for FIMNKY and GARP amino acids at a site , respectively , for either ‘% AT-rich’ or ‘% GC-rich’ taxa . According to this metric , higher values measure a greater disparity between %AT-rich alphaproteobacteria and all others; a measure of compositional heterogeneity or bias per site . The most compositionally heterogeneous sites according to ɀ were progressively removed using the software SiteStripper ( Verbruggen , 2018 ) in increments of 10% . We also progressively removed the fastest evolving sites in increments of 10% . Conditional mean site rates were estimated under the LG+C60+F+R6 model in IQ-TREE v1 . 5 . 5 using the ‘-wsr’ flag ( Nguyen et al . , 2015 ) . Our datasets were recoded into four- and six-character state amino acid alphabets using dataset-specific recoding schemes aimed at minimizing compositional heterogeneity in the data ( Susko and Roger , 2007 ) . The program minmax-chisq , which implements the methods of Susko and Roger ( 2007 ) , was used to find the best recoding schemes—please see Figure 3 , Figure 2—figure supplement 4 and Figure 3—figure supplement 1–6 , and Figure 3—figure supplement 8 legends for the specific recoding schemes used for each dataset . The approach uses the chi-squared ( X2 ) statistic for a test of homogeneity of frequencies as a criterion function for determining the best recoding schemes . Let πi denote the frequency of bin i for the recoding scheme currently under consideration . For instance , suppose the amino acids were recoded into four bins: RNCM EHIPTWV ADQLKS GFY , then π4 would be the frequency with which the amino acids G , F or Y were observed . Let πis be the frequency of bin i for the sth taxa . Then the X2 statistic for the null hypothesis that the frequencies are constant , over taxa , against the unrestricted hypothesis ists= ∑is ( πis−πi ) 2/πi The X2 statistic provides a measure of how different the frequencies for the sth taxa are from the average frequencies . The maximum ts over s is taken as an overall measure of how heterogeneous the frequencies are for a given recoding scheme . The minmax-chisq program searches through recoding schemes , moving amino acids from one bin to another , to try to minimize the max⁡ts ( Susko and Roger , 2007 ) . The inference of phylogenies was primarily done under the maximum likelihood framework and using IQ-TREE v1 . 5 . 5 ( Minh et al . , 2013; Nguyen et al . , 2015 ) . ModelFinder in IQ-TREE v1 . 5 . 5 ( Kalyaanamoorthy et al . , 2017 ) was used to assess the best-fitting amino acid empirical matrix ( e . g . JTT , WAG , and LG ) , on a maximum-likelihood tree , to our full dataset of 120 taxa and 200 conserved single-copy marker genes ( see Supplementary file 2E and Supplementary file 2F ) . We first inferred guide trees ( for a PMSF analysis ) with a model that comprises the LG empirical matrix , with empirical frequencies estimated from the data ( F ) , six rates for the FreeRate model to account for rate heterogeneity across sites ( R6 ) , and a mixture model with 60 amino acid profiles ( C60 ) to account for compositional heterogeneity across sites—LG + C60+F + R6 . Because the computational power and time required to properly explore the whole tree space ( given such a big dataset and complex model ) was too high , constrained tree searches were employed to obtain these initial guide trees ( see Figure 2—figure supplement 6 for the constraint tree ) . Many shallow nodes were constrained if they received maximum UFBoot and SH-aLRT support in a LG + PMSF ( C60 ) +F + R6 analysis . All deep nodes , those relevant to the questions addressed here , were left unconstrained ( Figure 2—figure supplement 6 ) . The guide trees were then used together with a dataset-specific mixture model ES60 to estimate site-specific amino acid profiles , or a PMSF ( Posterior Mean Site Frequency Profiles ) model , that best account for compositional heterogeneity across sites ( Wang et al . , 2018 ) . The dataset-specific empirical mixture model ES60 also has 60 categories but , unlike the general C60 , was directly estimated from our large dataset of 200 genes and 120 alphaproteobacteria ( and outgroup ) using the methods described in Susko et al . ( 2018 ) ; ModelFinder ( Kalyaanamoorthy et al . , 2017 ) suggests that the LG + ES60+F + R6 model is the best-fitting model; the R6 model component , however , considerably increases computational burden; see Supplementary file 2F and Supplementary file 2G ) . Final trees were inferred using the LG + PMSF ( ES60 ) +F + R6 model and a fully unconstrained tree search . Those datasets that produced the most novel topologies under maximum likelihood were further analyzed under a Bayesian framework using PhyloBayes MPI v1 . 7 and the CAT-Poisson+Γ4 model ( Lartillot and Philippe , 2004; Lartillot et al . , 2009 ) . This model allows for a very large number of classes to account for compositional heterogeneity across sites and , unlike in the more complex CAT-GTR+Γ4 model , also allows for convergence to be more easily achieved between MCMC chains . PhyloBayes MCMC chains were run for at least 10 , 000 cycles until convergence between the chains was achieved and the largest discrepancy ( i . e . maxdiff parameter ) was ≤0 . 4 ( except for the untreated dataset analyzed in Figure 2—figure supplement 3A; see Supplementary file 2H for several summary statistics for each PhyloBayes MCMC chain , including discrepancy and effective sample size values ) . A consensus tree was generated from two PhyloBayes MCMC chains using a burn-in of 500 trees and sub-sampling every 10 trees . Phylogenetic analyses of recoded datasets into four-character state alphabets were analyzed using IQ-TREE v1 . 5 . 5 and the model GTR + ES60 S4+F + R6 . ES60S4 is an adaptation of the dataset-specific empirical mixture model ES60 to four-character states . It is obtained by adding the frequencies of the amino acids that belong to each bin in the dataset-specific four-character state scheme S4 ( see Data Recoding for details ) . Phylogenetics analyses of recoded datasets into six-character state alphabets were analyzed using PhyloBayes MPI v1 . 7 and the CAT-Poisson+Γ4 model . Maximum-likelihood analyses with a six-state recoding scheme could not be performed because IQ-TREE currently only supports amino acid datasets recoded into four-character states . The 16S rRNA genes of ‘Candidatus Finniella inopinata’ , and the presumed endosymbionts of Peranema trichophorum and Stachyamoeba lipophora were identified with RNAmmer 1 . 2 server and BLAST searches . A set of 16S rRNA genes for diverse rickettsialeans and holosporaleans , and other alphaproteobacteria as outgroup , were retrieved from NCBI GenBank . The selection was based on Hess and Melkonian ( 2013 ) , Szokoli et al . ( 2016 ) and Wang and Wu ( 2015 ) . Environmental sequences for uncultured and undescribed rickettsialeans were retrieved by keeping the 50 best hits resulting from a BLAST search of our three novel 16S rRNA genes against the NCBI GenBank non-redundant ( nr ) database . The sequences were aligned with the SILVA aligner SINA v1 . 2 . 11 and all-gap sites were later removed . Phylogenetic analyses on this alignment were performed on IQ-TREE v1 . 5 . 5 using the GTR + F + R8 model . A UPGMA ( average-linkage ) clustering of amino acid compositions based on the 200 gene set for the Alphaproteobacteria was built in MEGA 7 ( Kumar et al . , 2016 ) from a matrix of Euclidean distances between amino acid compositions of sequences exported from the phylogenetic software P4 ( Foster , 2004; http://p4 . nhm . ac . uk/index . html ) . Sequencing data were deposited in NCBI GenBank under the BioProject PRJNA501864 . The genomes of 'Candidatus Finniella inopinata' , endosymbiont of Peranema trichophorum strain CCAP 1260/1B and endosymbiont of Stachyamoeba lipophora strain ATCC 50324 were deposited in NCBI GenBank under the accessions GCA_004210305 . 1 , GCA_004210275 . 1 and GCA_003932735 . 1 . Raw sequencing reads were deposited on the NCBI SRA archive under the accessions SRR8145469 , SRR8145470 , SRR8156519 , SRR8156520 , SRR8156521 , SRR8156522 . Multi-gene datasets as well as phylogenetic trees inferred in this study were deposited at Mendeley Data under the DOI: 10 . 17632/75m68dxd83 . 2 .	The Alphaproteobacteria form one of the most abundant groups of bacteria on Earth , and one that is closely linked to all complex forms of life . Many bacteria within this class live inside the cells of other organisms . For example , mitochondria – the powerhouses of animal , plant and other eukaryotic cells – evolved from bacteria within this group . Other alphaproteobacteria act as parasites or beneficial symbionts within cells . The history of life on Earth can be thought of as a tree , with each branch representing the evolution of a new species from a common ancestor . But for many bacteria , the earliest stages of their evolutionary history are so tangled and complex that their origin remains largely unknown . For example , efforts to study the earliest history of the Alphaproteobacteria have been plagued with errors and artefacts . The extreme variation in the genetic sequences of different bacteria in the group make it particularly challenging to uncover relationships between the species . To overcome this problem , Muñoz-Gómez et al . focused on a set of 200 genes that occur in all alphaproteobacteria , and used a range of strategies to reduce potential errors in the data . The results propose a new general structure for the evolutionary tree of the Alphaproteobacteria . This shows that two groups of alphaproteobacteria that were thought to be closely related to each other – the parasites Rickettsiales and Holosporales – are unrelated . Instead , these groups evolved independently from different free-living alphaproteobacteria . The abundance and diversity of the Alphaproteobacteria means that the improved understanding of their evolutionary origins could influence the work of a wide range of scientists . Further research could help to shed light on how parasitic bacteria interact with the cells they invade; reveal how bacteria evolved certain abilities , such as the ability to photosynthesize; and uncover the precise origin of mitochondria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "microbiology", "and", "infectious", "disease" ]	2019	An updated phylogeny of the Alphaproteobacteria reveals that the parasitic Rickettsiales and Holosporales have independent origins
Temporal continuity of object identity is a feature of natural visual input and is potentially exploited – in an unsupervised manner – by the ventral visual stream to build the neural representation in inferior temporal ( IT ) cortex . Here , we investigated whether plasticity of individual IT neurons underlies human core object recognition behavioral changes induced with unsupervised visual experience . We built a single-neuron plasticity model combined with a previously established IT population-to-recognition-behavior-linking model to predict human learning effects . We found that our model , after constrained by neurophysiological data , largely predicted the mean direction , magnitude , and time course of human performance changes . We also found a previously unreported dependency of the observed human performance change on the initial task difficulty . This result adds support to the hypothesis that tolerant core object recognition in human and non-human primates is instructed – at least in part – by naturally occurring unsupervised temporal contiguity experience . Among visual areas , the inferior temporal ( IT ) cortex is thought to most directly underlie core visual object recognition in human and non-human primates ( Ito et al . , 1995; Rajalingham and DiCarlo , 2019 ) . For example , simple weighted sums of IT neuronal population activity can accurately explain and predict human and monkey core object recognition ( COR ) performance over dozens of such tasks ( Majaj et al . , 2015 ) . Moreover , direct suppression of IT activity disrupts COR behavior ( Afraz et al . , 2015; Rajalingham and DiCarlo , 2019 ) . These results were found in the face of significant variation in object latent variables including size , position , and pose , and the high performance of the simple IT readout ( weighted sum ) rests on the fact that many individual IT neurons show high tolerance to those variables ( DiCarlo et al . , 2012; Hung et al . , 2005; Li et al . , 2009 ) , reviewed by DiCarlo et al . , 2012 . But how does the ventral stream wire itself up to construct these highly tolerant IT neurons ? Simulated IT ‘neurons’ in the deep layers of artificial neural networks ( ANNs ) have such tolerance and provide quite accurate approximations of the adult ventral visual stream processing ( Khaligh-Razavi and Kriegeskorte , 2014; Rajalingham et al . , 2018; Yamins et al . , 2014 ) . However , those ANNs are produced by training with millions of supervised ( labeled ) training images , an experience regime that is almost surely not biologically plausible over evolution or postnatal development . That simple fact rejects all such ANNs as models of the construction of IT tolerance , regardless of whether or not the brain is executing some form of backpropagation-instructed plasticity ( Lillicrap et al . , 2020; Rumelhart et al . , 1986 ) . So the question remains open: how does the ventral stream wire itself up to construct a tolerant IT with minimal supervision ? The temporal stability of object identity under natural viewing ( i . e . , objects do not rapidly jump in and out of existence ) has been proposed as a key available source of unsupervised information that might be leveraged by the visual system to construct neural tolerance , even during adulthood ( Földiák , 1991; Hénaff et al . , 2019; Rolls and Stringer , 2006; Wallis et al . , 2009; Wallis et al . , 2009; Wiskott and Sejnowski , 2002 ) . Consistent with this view , psychophysical results from human subjects show that unsupervised exposure to unnatural temporal contiguity experience ( i . e . , laboratory situations in which object do jump in and out of existence ) reshapes position tolerance ( Cox et al . , 2005 ) , pose tolerance ( Wallis and Bülthoff , 2001 ) , and depth illumination tolerance ( Wallis et al . , 2009 ) as measured at the behavioral level . Similarly , neurophysiological data from adult macaque IT show that unsupervised exposure to unnatural temporal contiguity experience reshapes IT neuronal position and size tolerance ( Li et al . , 2009; Li and DiCarlo , 2008; Li and DiCarlo , 2010 ) , in a manner that is qualitatively consistent with the human behavioral data . Taken together , our working hypothesis is that the ventral visual stream is under continual reshaping pressure via unsupervised visual experience , that such experience is an important part of the construction of the tolerant representation that is ultimately exhibited at the top level of the ventral stream ( IT ) , that the IT population feeds downstream causal mechanistic chains to drive core object discrimination behavior , and that the performance on each such behavioral tasks is well approximated by linear readout of IT ( Hung et al . , 2005; Majaj et al . , 2015 ) . However , there is a key untested prediction in this working hypothesis: is the single neuronal plasticity in adult monkey IT quantitatively consistent with the adult human behavioral changes resulting from unsupervised temporal contiguity experience ? In this study , we chose to focus on testing that missing link as it was far from obvious that it would hold up . In particular , the prior IT neurophysiology work was with basic-level objects and produced seemingly large changes ( ~25% change in IT selectivity per hour of exposure in Li and DiCarlo , 2010 ) , and the prior human behavioral work was with subordinate-level objects and produced significant , but subtle , changes in behavior ( e . g . , ~3% performance change in Cox et al . , 2005 ) . Moreover , if we found that the link did not hold , it would call into question all of the elements of the overall working hypothesis ( especially IT’s relationship to COR behavior , and/or the importance of unsupervised plasticity to the IT representation ) . Thus , either result would be important . To test whether our working hypothesis is quantitatively accurate over the domain of unsupervised temporal contiguity-induced plasticity , we sought to build a model to predict the changes in human object discrimination performance that should result from temporally contiguity experience-driven changes in IT neuronal responses . This model has three components: ( 1 ) a generative IT model ( constrained by prior IT population response; Majaj et al . , 2015 ) that approximates the IT population representation space and can thus simulate the IT population response to any image of the objects ( within the space ) with variation in size; ( 2 ) an unsupervised plasticity rule ( constrained by prior IT neural plasticity data; Li and DiCarlo , 2010 ) to quantitatively describe and predict firing rate ( FR ) change of single IT neurons resulting from temporally contiguous pair of experienced images and can thus be used to update the simulated IT population representation; and ( 3 ) an IT-to-COR-behavior-linking model ( learned weighted sums , previously established by Majaj et al . , 2015 ) to predict behavioral discrimination performance from the state of the IT ( simulated ) population both before and after each epoch of unsupervised experience . To overcome the limitation of non-overlapping tasks in previous psychophysics and neurophysiology studies and to extend prior psychophysical work , we carried out new human behavioral experiments . Specifically , we measured the progression of changes in size-specific human object discrimination performance that resulted from unsupervised temporal contiguity experience using the same exposure paradigm as the prior monkey neurophysiology work ( Li and DiCarlo , 2010 ) . We did not use the exact same images as prior work , but we expected the model to still make accurate predictions of all behavioral changes . We made these behavioral measurements for a wide range of object discrimination tasks , ranging from subordinate ( specifically different face objects ) to basic level . Because humans make sensory-independent mistakes due to inattentional state , these sensory-independent random choices ( referred to as lapse rate ) set a ceiling in the measurable human behavioral performance ( Prins , 2012; Wichmann and Hill , 2001 ) . When tasks are in the saturated regime , it is hard to detect any learning effect as any changes in sensory representation would be hidden by the behavioral ceiling ( see later ) . Therefore , we focused our psychophysical study in the mid-range of task difficulty where learning effects can be measured . However , this meant that the task difficulty in human psychophysics could not be in the basic object regime where the neural data were collected . Thus , to make behavioral predictions from the neural data , we took advantage of the overall model to build this bridge: we first tuned the unsupervised plasticity rule by neural data with basic-level object images ( Li and DiCarlo , 2010 ) ; we then used a generative IT model – capable of simulating the response of each artificial IT neuron for a wide range of image discriminability levels – to make quantitative predictions of behavioral change in the regime where the human behavioral learning effects can be readily measured . Indeed , our behavioral tests revealed a strong dependency of learning effect on the initial task difficulty , with initially hard ( d’ < 0 . 5 ) and initially easy ( d’ > 2 . 5 ) COR tasks showing smaller measured learning effects than COR tasks of intermediate initial difficulty . We found that our overall model was quite accurate in its predictions of the direction , magnitude , and time course of the changes in measured human size tolerance in the regime where behavioral effects were readily measured for all of the tested unsupervised experience manipulations . The overall model also predicted how the behavioral effect size depended on the initial d’ once we assume behavioral lapses ( Prins , 2012 ) in the model at approximately the same level as those inferred in our subject pool . We note that , because of the ( expected ) inability to observed behavioral changes for tasks with initial high d’ , this study cannot confirm or refute the hypothesized linkage between IT neural effects and behavioral effects in that particular regime . Taken together , this result shows that at least three separate types of studies ( human unsupervised learning , IT unsupervised plasticity , and IT-to-COR-behavior testing ) are all quantitatively consistent with each other . As such , this result adds support to the overall working hypothesis: that tolerant COR is instructed – at least in part – by naturally occurring unsupervised temporal contiguity experience that gradually reshapes the non-linear image processing of the ventral visual stream without the need for millions of explicit supervisory labels ( Krizhevsky et al . , 2017; LeCun et al . , 1989; Riesenhuber and Poggio , 1999 ) and reviewed by LeCun et al . , 2015 . The basic experimental strategy is that , after testing initial object discrimination performance on a set of discrimination tasks ( ‘Test phase , ’ Figure 1A ) , we provide an epoch of unsupervised visual experience ( ‘Exposure phase , ’ Figure 1A ) that is expected to result in IT plasticity ( based on the results of Li and DiCarlo , 2010 ) . At the end of the exposure epoch , we remeasure discrimination performance ( Test phase ) , then provide the next epoch of unsupervised experience ( Exposure phase ) , etc . ( see Figure 1A ) . This strategy allowed us to evaluate the accumulation of positive or negative behavioral changes ( a . k . a . ‘learning’ ) resulting from four unsupervised experience epochs ( 400 exposure ‘trials’ each ) over approximately 1 . 5–2 hr . We include control discrimination tasks to subtract out any general learning effects . Specifically , we evaluated changes in discrimination performance ( relative to initial performance ) of each of a set of size-specific object discrimination tasks . A total of 174 human subjects on Amazon Mechanical Turk ( see Materials and methods and Kar et al . , 2019; Majaj et al . , 2015; Rajalingham et al . , 2018 ) participated in this experiment . To measure object discrimination performance in each subject , we used a set of two-way alternative forced choice ( 2AFC ) sub-tasks ( size-specific object discrimination tasks; see Materials and methods ) . These sub-tasks were randomly interleaved ( trial by trial ) in each test phase , and the key test conditions used in the analyses ( brackets indicated with d’s in Figure 1B ) were embedded within a balanced set of six sub-tasks and cover trials ( see Figure 1B and Materials and methods ) . Our first experiments used pairs of faces as the objects to discriminate , and we targeted our exposure manipulations at the big size ( 2× the baseline size; see Materials and methods and Figure 1; later , we targeted other pairs of objects and other sizes ) . Specifically , we used eight face objects from a previous study ( Majaj et al . , 2015 ) . We chose these face objects at this size because , prior to unsupervised exposure , they had intermediate discriminability ( mean d’ = 2 . 0 ± 0 . 1 for big size , frontal view , n = 28 pairs of faces ) , thus allowing us the possibility to measure both positive and negative changes in discrimination performance . For each subject , two target faces ( manipulated during exposure ) and two control faces ( not shown during exposure ) were randomly chosen from these eight faces . Subjects were instructed to identify the single foreground face in a briefly presented test image ( 100 ms ) by choosing among two alternative choice faces immediately presented after the test image , one of which is always correct ( i . e . , 50% chance rate ) . The test image contained one foreground object with variation in view ( position , size , pose ) , overlaid on a random background ( see Materials and methods for test image generation ) . The choice images were always baseline views ( i . e . , size of ~2° , canonical pose ) without background . Similar to prior work testing the effects of unsupervised exposure on single-site IT recordings ( Li and DiCarlo , 2010 ) , each experiment consisted of two phases ( Figure 1A ) : test phases to intermittently measure the size-specific object discrimination performance ( d’ ) for the target face pair and control face pair ( three d’ measured in each group of subjects , see Figure 1B bottom ) ; and exposure phases to provide unsupervised visual experience ( pairs of images with different sizes in close temporal proximity; Figure 1A ) that – based on prior work – was expected to improve or decrease the discrimination performance on the exposed objects . The purpose of the exposure phase was to deploy unsupervised visual experience manipulations to target a particular object pair ( two ‘target’ objects ) at particular views ( e . g . , sizes ) of those target objects . For each exposure event , two images , each containing a different size object ( frontal; no background ) , were presented consecutively ( 100 ms each ) ( see Materials and methods for details ) . In non-swapped exposure events , both images contained the same object ( expected to ‘build’ size tolerance under the temporal contiguity hypothesis ) . In swapped exposure events , each images contained a different target object ( expected to ‘break’ size tolerance under the temporal contiguity hypothesis ) . The conceptual predictions of the underlying IT neural population target object manifolds ( DiCarlo and Cox , 2007 ) are that non-swapped exposure events will straighten the manifold of each target object by associating size exemplars of the same object ( as in the natural world ) , and that swapped exposure events will bend and decrease the separation between the two manifolds by incorrectly associating size exemplars of different objects ( Figure 1C ) . This logic and experimental setup are adopted entirely from prior work ( Li and DiCarlo , 2008; Li and DiCarlo , 2010 ) . In our studies here , we specifically focused on manipulating the size tolerance in the medium size ( ×1 of baseline view; ~2° ) to big size ( ×2 of baseline view; ~4° ) regime . Thus , the images shown during the exposure phase ( indicated by * in Figure 1B ) were always medium- and big-size , frontal view of the target objects . We conducted three types of unsupervised exposure experiments ( u ) : swapped ( u1 ) , non-swapped ( u2 ) and non-swapped , followed by swapped ( u3 ) . In experiment u1 ( swapped exposure events ) , we found that discrimination of the target face pair viewed at big size decreased with increasing numbers of exposure events ( Figure 2A; top rows; red solid line; n = 102 subjects ) . We found little to no change in the performance for the non-exposed ( small size ) versions of those same faces ( black dashed line; mean initial d´ is 1 . 2 ± 0 . 1 ) or for non-exposed control faces ( also tested at big size , black solid line ) . Lower panels in Figure 2A show the learning effect defined by subtracting changes in control face discrimination performance ( to remove general learning effects over the experience epochs , which turned out to be small; see Figure 2A , upper panel ) . In sum , we demonstrated an unsupervised , object-selective , size-selective temporal contiguity-induced learning effect that was qualitatively consistent with prior work in ‘breaking’ tolerance ( Cox et al . , 2005; Wallis and Bülthoff , 2001 ) and measured the accumulation of that learning over increasing amounts of unsupervised exposure . In experiment u2 ( non-swapped exposure events ) , we found that discrimination of the target face pair viewed at big size increased with increasing numbers of exposure events ( Figure 2B; top rows; blue solid line; n = 36 subjects ) . As in experiment u1 , we found little to no change in performance for the non-exposed ( small size ) versions of those same faces or for non-exposed control faces ( also tested at big size , black solid line ) . This shows that , as predicted by the temporal contiguity hypothesis , unsupervised experience can build size tolerance at the behavioral level . Interestingly , after ~800 exposure events , the exposure-induced learning effects appeared to plateau in both ‘breaking’ tolerance conditions ( experiment u1 , Figure 2A ) and ‘building’ tolerance conditions ( experiment u2 , Figure 2B ) , suggesting a limit in the measurable behavioral effects ( see Discussion ) . To test whether this unsupervised learning effect is reversible , we measured human performance in a combined design ( experiment u3 ) by first providing exposure epochs that should ‘build’ tolerance , followed by exposure epochs that should ‘break’ tolerance ( n = 37 subjects ) . Consistent with the results of experiments u1 and u2 , we found that size tolerance first increased with non-swapped ( ‘build’ ) exposures and then decreased with swapped ( ‘break’ ) exposures ( Figure 2C ) , and that the effect did not spill over to the control objects . In sum , these results confirmed that the effect of unsupervised visual experience was specific ( to manipulated object and sizes ) and strong even in adults . Furthermore , the measured human learning effect trajectories with different unsupervised visual exposure conditions ( u1 , u2 , u3 ) were taken as behavioral effects that must – without any parameter tuning – be quantitatively predicted by our working hypothesis ( that links IT neural responses to COR behavior; see Introduction ) . We next describe how we built an overall computational model to formally instantiate that working hypothesis to make those predictions . To generate predictions of human behavior performance , we need to measure or otherwise estimate individual IT neural responses to the same images used in the human psychophysical testing ( above ) for a sufficiently large set of IT neurons ( a . k . a . IT population responses ) . Because each of the objects we used in human psychophysics had been previously tested in neural recording experiments from monkey IT , we did not collect new IT population responses ( very time consuming ) , but we decided instead to make suitably accurate predictions of the initial population pattern of IT response for test images of those objects ( i . e . , the IT response patterns prior to any simulated unsupervised plasticity effects ) . To do this , we built a generative model of the IT population based on the previously recorded IT population response to those objects . The output of this model is the FR of a simulated IT population to one presentation of a newly rendered test image ( generated from the 64 base objects used in the previous study ) . With this model , we could simulate the initial IT population responses to any image rendered from the psychophysically tested objects ( approximately ) without recording more neurons in behaving animals . This generative IT model captures the IT neuronal representation space with a multi-dimensional Gaussian ( MDG ) model , assuming the distribution of IT population responses is Gaussian-like for each object ( see Materials and methods for Gaussian validation ) ( Figure 3A ) . Because the MDG preserves the covariance matrix of IT responses to 64 objects , any random draw from this MDG gives rise to an object response preference profile ( one response level for each of 64 objects ) of a simulated IT neural site . To simulate the variance in object size , for each simulated site , we randomly chose one size-tuning kernel from a pool of size-tuning curves that we had obtained by fitting curves to real IT responses across changes in presented object size ( n = 168 recording sites; data from Majaj et al . , 2015 ) . This process is repeated independently for each simulated site . Motivated by prior work ( Li et al . , 2009 ) , we assumed separability of object representation and size tuning , and simulated the response to any of the 64 objects . To check if the simulation is statistically accurate in terms of the layout of images in IT population representation space , we compared the representation similarity matrix ( RSM; correlation between neuronal population responses to different images ) of different draws of a simulated IT with the RSM measured from the actual IT neural data ( Figure 3B ) . One typical example of that is shown in Figure 3C , revealing high correlation of the two RSMs ( r = 0 . 93 ± 0 . 01 ) . While this does not guarantee that any such simulated IT population is fully identical to an IT population that might exist in an actual monkey or human IT , our goal was simply to get the simulated IT population response distribution in the proper range ( up to second-order statistics ) . To make predictions about how IT neural changes will result in behavioral changes , we first needed a model to establish the linkage between IT population response and core object discrimination behavior prior to any experience-induced effects . We have previously found that simple weighted linear sums of IT neural responses accurately predict the performance ( d’ ) of human object discrimination for new images of those same objects ( here termed the IT-to-COR-behavior-linking model ) ( Majaj et al . , 2015 ) . That model has only two free hyperparameters: the number of neural sites and the number of labeled ( a . k . a . ‘training’ ) images used to set the weights of the decoder for each object discrimination . Once those two hyperparameters are locked , it has been empirically demonstrated that the performance for any object discrimination task on new images is accurately predicted by its trained decoder ( Majaj et al . , 2015 ) . To test whether the simulated IT population activity from the generative IT model ( above ) could quantitatively reproduce those prior results and to lock these two hyperparameters , we compared the predicted performance ( for any given object recognition task ) based on the simulated IT population ( Figure 3D; red solid line ) with the predicted performance based on the previously recorded IT neural population ( black solid line ) . We did this as a function of number of recording sites for a set of object recognition tasks . Figure 3D illustrates two example tasks ( error bar is standard error across 40 random subsamples of recording sites ) . As expected , we found that the model predictions overlapped with decoded performance of real IT neural sites , indicating that our generative IT model has captured the relevant components of the IT population response . We next set out to choose the two free hyperparameters ( number of sites and number of training examples ) . The crossing point with human performance in Figure 3D reflects how many neural sites are necessary to reach human performance level for a given number of training samples . Unlike the real IT neural data ( n = 168 recording sites ) that required extrapolation to estimate the number of sites matching human absolute performance ( Majaj et al . , 2015 ) , we simulated up to 1000 IT sites with the generative model to cover the range of neural sites necessary to reach human performance . Consistent with Majaj et al . , 2015 , we found that the number of simulated IT sites required to match human was similar across different tasks ( 260 ± 23 ) IT sites given 20 training images ( tested over 24 object discrimination tasks: low variation eight-way tests: eight basic level , eight car identification , and eight face identification tasks; previously used in Majaj et al . , 2015 ) . Specifically , we here used 260 sites with 20 training samples for all tasks , and the match between the decoded simulated IT performance and human performance over all discrimination tasks was r = 0 . 83 ± 0 . 05 ( n = 24 tasks ) , similar to previously reported match between decoded neural IT performance and human for the same tasks ( r = 0 . 868 from Majaj et al . , 2015 ) . Note that other specific combinations of the number of IT sites and the number of training examples are also suitable ( Figure 3F ) , and we explore this later . In sum , by setting the two decoder hyperparameters to match initial human performance , we established a fixed linear decoder rule that could be applied to our simulated IT population to quantitatively predict the expected performance of the subject ( i . e . , the owner of that IT population ) for any object discrimination task . The consequence is that , because the linkage model between the IT population and behavior is now fixed in the model , any changes in the model IT population are automatically mapped to predicted changes ( if any ) in behavioral performance . From here on , we locked down the generative IT model and the decoders that matched human initial performance ( before learning ) , and combine both of these models later to make predictions of direction and magnitude of behavioral performance change ( if any ) that should result from any given change in the IT population driven by unsupervised plasticity ( Figure 2 ) . To model the IT neural population response changes that result from the unsupervised visual experience provided to the human subjects , we developed an unsupervised IT plasticity rule guided by previous studies of IT plasticity effects in the rhesus monkey that used the same paradigm of unsupervised visual experience that we provided here to our human subjects ( Li et al . , 2009; Li and DiCarlo , 2008; Li and DiCarlo , 2010 ) . In particular , we set out to build an unsupervised IT plasticity rule that could predict the ( mean ) response change that occurs in each and every IT neuron as a result of each presented pair of temporally contiguous visual images . We assumed that the same model would also apply to human ‘IT’ without any parameter modifications ( see Discussion ) . Those prior monkey studies revealed that exposure to altered ( ‘swapped’ ) visual statistics typically disrupts the size tolerance of single IT neurons , while exposure to normal statistics in visual experience ( non-swapped condition ) typically builds size tolerance ( Li and DiCarlo , 2010 ) . To develop our unsupervised IT plasticity rule , we replicated the exact same experiment used in the monkeys on simulated IT neural sites . Figure 4A illustrates the exposure design for single IT sites , where the preferred object ( P ) and non-preferred object ( N ) of each neural site are defined by the magnitude of neuronal activity ( z-scored across all objects for each IT site ) . Selectivity of a neural site is measured by the difference of neuronal responses to its preferred and non-preferred objects ( P – N ) / ( p + N ) , the same as Li and DiCarlo , 2010 . We used a Hebbian-like ( associative ) plasticity rule ( Caporale and Dan , 2008; Hebb , 1949; Oja , 1982; Paulsen and Sejnowski , 2000 ) , which updates FR for each pair of images based on the difference of neuronal FR between the lagging and leading images ( see Materials and methods ) . Our plasticity rule states that the modification of FR of each IT unit to the leading image at time t is equal to the difference of FR between lagging and leading images multiplied by a plasticity rate α . This plasticity rule tends to reduce the difference in neuronal responses to consecutive images and implies a temporal association to images presented close in time . The plasticity rule is conceptually similar to previously proposed temporal continuity plasticity or a . k . a . slow feature analysis ( Berkes and Wiskott , 2005; Földiák , 1990; Földiák , 1991; Mitchison , 1991; Sprekeler et al . , 2007 ) . It is physiologically attractive because the findings on short-term synaptic plasticity revealed that synaptic efficacy changes over time in a way that reflects the history of presynaptic activity ( Markram et al . , 2012; Markram et al . , 1997 ) . Even though conceptually similar , our plasticity rule is a ‘descriptive’ rather than a ‘mechanistic’ rule of plasticity at all stages of the ventral stream . That is , the rule does not imply that all the underlying plasticity is in IT cortex itself – but only aims to quantitatively capture and predict the changes in IT responses resulting from unsupervised visual experience . It is general in a sense that it can make predictions for different objects or dimensions of variations , but it is ( currently ) limited in that it only applies to temporally paired image associations , ignores any correlation in the neural response patterns , and assumes that updates occur only in the responses to the exposed images ( i . e . , non-exposed object/size combinations are not affected ) . To show the effects of this unsupervised IT plasticity rule , we illustrate with an example simulated IT neural site . The simulated neural site in Figure 4B was initialized to be – like many adult monkey IT neurons – highly size tolerant: its response to a preferred object ( P ) is always greater than response to a non-preferred object ( N ) at each size . After applying the unsupervised exposure design in Figure 4A ( 200 exposure events for each arrow , 1600 exposure events in total ) , the responses to each of the six conditions ( 2 objects × 3 sizes ) evolved as shown in Figure 4B . We note two important consequences of this plasticity rule . First , because the rule was designed to decrease the difference in response across time , responses to images presented consecutively tend to become more similar to each other , which results in a reduction in the response difference between P and N at both the swapped and the non-swapped sizes . Second , once the neural site reached a state in which its response is no different over consecutively exposed images , the learning effect saturates . Notably , unlike adaptive changes in plasticity rate in the typical supervised optimization of deep neural networks ( Kingma and Ba , 2014 ) , our plasticity rate is kept constant over the ‘lifetime’ of the model . The gradual shrinkage of learning effect ( Δ ( P – N ) / ( p + N ) ) as more and more exposure events are provided was a consequence of the gradual reduction in the absolute difference between neuronal responses to the two consecutive images that makeup each exposure event . There is only one free parameter in our plasticity rule equation – the plasticity rate α . We determined this parameter using the single-electrode physiology data collected previously in the lab ( Li and DiCarlo , 2010 ) . Figure 4C shows the average IT plasticity effect that results from different settings of α ( here the plasticity effect is defined by the normalized changes in selectivity: Δ ( P – N ) / ( P – N ) , exactly as was done in Li and DiCarlo , 2010 ) . As expected , a higher plasticity rate ( α ) results in greater model IT plasticity effects ( Figure 4C ) . We chose the plasticity rate ( α ) that best matched the prior monkey IT neurophysiology results ( i . e . , the α that resulted in the minimal difference between the model IT plasticity effect [solid lines] and the experimentally reported IT plasticity effect [dashed lines] for swapped , non-swapped , and medium object sizes; see Figure 4C middle ) . The best α is 0 . 0016 nru per exposure event ( nru = normalized response units; see Materials and methods for intuition about approximate spike rate changes ) . Once we set the plasticity rate , we locked it down for the rest of this study ( otherwise noted later where we test rate impact ) . We next asked if our IT plasticity rule naturally captured the other IT plasticity effects reported in the monkey studies ( Li and DiCarlo , 2010 ) . Specifically , it was reported that , for each neural site , the selectivity that results from a fixed amount of unsupervised exposure depends on the initial selectivity of that site . Thus , the unsupervised ‘swapped’ experience manipulation causes a reduction of selectivity for neural sites that show a moderate level of initial P ( preferred object ) vs . N ( non-preferred object ) selectivity at the swapped size , and the same amount of unsupervised experience reverses the selectivity of neuronal sites that show a low level of initial selectivity at the swapped size ( i . e . , cause the site to , oxymoronically , prefer object N over object P ) . Li and DiCarlo , 2010 also reported that the more natural , ‘non-swapped’ experience manipulation caused a building of new selectivity ( for neuronal units that initially show a strong preference for P at some sizes , but happened to have low P vs . N selectivity at the non-swapped size ) . We tested for both of these effects in our model by selecting subsets of neural sites in the simulated IT population in exactly the same way as Li and DiCarlo , 2010 ( sampled from n = 1000 simulated IT units ) and then applied the plasticity rule to those units . We found a very similar dependency of the IT plasticity to those previously reported IT plasticity effects ( Figure 4D; cf . see Figures 6 and 7 of Li and DiCarlo , 2010 ) . Given that our IT plasticity rule tends to pull the response of temporally contiguous images toward each other ( Berkes and Wiskott , 2005; Földiák , 1990; Földiák , 1991; Mitchison , 1991; Sprekeler et al . , 2007 ) , it is not entirely obvious how this can build selectivity ( i . e . , pull response to P and N apart ) . The reason this occurs is that some IT neural sites have ( by chance draw from the generative model of IT , above ) initially high selectivity for P vs . N at the medium size and no selectivity at ( e . g . ) the big size . ( Indeed , such variation in the IT population exists as reported in Li and DiCarlo , 2010 . ) By design , the non-swapped ( ‘natural’ ) unsupervised exposure temporally links Pmed ( high response ) with Pbig , which – given the plasticity rule – tends to pull the Pbig response upward ( pull it up higher than Nbig ) . In addition , the non-swapped exposure links Nmed ( low response ) with Nbig , which can pull the Nbig response downward ( provided that the Nmed response is initially lower than the Nbig response ) . Both effects thus tend to increase the Pbig vs . Nbig response difference ( i . e . , both effects tend to ‘build’ selectivity for P vs . N at the big presentation size , which results in the neural site preferring object P over object N at both the medium and the big sizes – a property referred to as size ‘tolerance’ ) . This effect is observed in single IT neural site size-tuning curve for P and N before and after learning ( see Figure 3 in Li and DiCarlo , 2010 ) . Indeed , it is this effect that conceptually motivated temporal contiguity plasticity in the first place – natural-occurring statistics can be used to equalize the responses to the same object over nuisance variables ( such as size ) . In sum , our very simple IT plasticity rule quantitatively captures the average IT plasticity effects for which its only free parameter was tuned , and it also naturally captures the more subtle IT neural changes that have been previously described . To summarize , we have ( 1 ) built and tested a generative IT model that captured the object representation space and variability in the actual primate IT population; ( 2 ) locked down a set of parameters of a linear decoder rule that quantitatively links the current state of the simulated IT population to initial human performance on any discrimination task ( including the ones we plan to test ) ; and ( 3 ) defined an IT plasticity rule that describes how each individual IT neural site changes as a result of each unsupervised exposure event , and we locked down the only free parameter ( plasticity rate ) in that rule to match existing monkey IT plasticity data ( see Figure 1—figure supplement 1A ) . At this point , we could – without any further parameter tuning – combine each of these three model components into a single overall model that predicts the direction , magnitude , and time course of human unsupervised learning effects that should result from any unsupervised learning experiment using this exposure paradigm ( pairwise temporal image statistics ) . Specifically , to generate the predictions for each of unsupervised learning experiments ( u: u1 , u2 , u3; see Figure 2 ) , we ( 1 ) initialized a potential adult human IT ( from the generative IT model ) with a total of 260 simulated IT recording sites; ( 2 ) built linear decoders for the planned object discrimination tasks that read from all 260 sites , using 20 training examples for each and every task; ( 3 ) froze the parameters of all such decoders ( i . e . , froze the resulting weighting on each simulated IT neural site on the ‘subject’s’ object choice decision ) ; ( 4 ) ‘exposed’ the IT model population to the same unsupervised temporal exposure history as the human subjects , using the IT plasticity rule to update the model ‘IT’ after each temporally adjacent image exposure pair to update the responses of each simulated IT neural site ( note that the plasticity trajectory of each neural site is dependent on both its initial object/size response matrix [1] , and the sequence of images applied during unsupervised experience [u] ) ; ( 5 ) measured the changes in ‘behavioral’ performance of the overall model ( changes in object discrimination performance of the [frozen] decoders [2] ) ; and ( 6 ) took those changes as the model predictions of the changes in human performance that should result from that unsupervised experience ( u ) . Again we emphasize that , while the overall model relies heavily on data and parameters derived explicitly or implicitly from these prior studies ( Li and DiCarlo , 2010; Majaj et al . , 2015 ) , no components or parameters of this model nor its predictions depended on the behavioral data collected in this study . To give robust model estimates of the average predicted behavioral effects , we repeated this process ( 1–6 ) 100 times for each experiment ( u ) and averaged the results , which is analogous to running multiple subjects and averaging their results ( as we did with the human data; see Figure 2 ) . For clarity , we note that the prediction stochasticity is due to random sampling of the IT generative population , the clutter variability introduced in the generative IT model when generating the initial population response for each test image , the trial-by-trial variation in the simulated IT responses , the random unsupervised exposure event sequence ( see Materials and methods ) , and randomly drawn test images , all of which we expect to average out . Note that , in expecting that these overall model predictions might be accurate , we are implicitly making the following assumptions: ( 1 ) monkey IT and human IT are approximately identical ( Kriegeskorte et al . , 2008; Rajalingham et al . , 2015 ) , ( 2 ) the linkage of IT to behavioral performance is approximately identical ( as suggested by Majaj et al . , 2015 ) , ( 3 ) human IT unsupervised plasticity is the same as monkey IT unsupervised plasticity , and ( 4 ) humans do not re-learn or otherwise alter the assumed mechanistic linkage between IT and behavior during or after unsupervised visual experience ( at least not at the time scales of these experiments: 1 . 5–2 hr ) . Figure 5A , D , E show the model-predicted learning effects ( black solid line ) for each of the three unsupervised experiments ( u1 , u2 , u3 ) plotted on top of the observed measured human learning effects ( red line , reproduced from the learning effects shown in Figure 2 bottom ) . For each experiment , we found that the overall model did a very good job of predicting the direction , magnitude , and time course of the changes in human behavior . The directional predictions are not surprising given prior qualitative results , but the accurate predictions of the magnitude and time course are highly non-trivial ( see below ) . Despite these prediction successes , we also noticed that the predictions were not perfect , most notably after large numbers of unsupervised exposures ( e . g . Figure 5E , rightmost points ) , suggesting that one or more of our assumptions and corresponding model components are not entirely accurate ( see Discussion ) . Given the surprising overall quantitative accuracy of the model predictions straight ‘out of the box , ’ we wondered if those predictions might somehow occur even for models that we had not carefully tuned to the initial ( pre-exposure ) human performance and the previously reported IT plasticity . That is , which components of the model are critical to this predictive success ? We tested this in two ways ( focusing here on experiment u1 ) . First , we built model variants in which the IT plasticity rate ( α ) was either four times smaller or four times bigger than empirically observed in the prior monkey IT neurophysiology ( solid gray lines ) and re-ran the entire simulation procedure ( above ) . In both cases , the predictions of these ( non-biology-matched ) model variants were now clearly different in magnitude than the observations ( Figure 5A ) . This result is arguably the strongest evidence that the single-unit IT plasticity effects fully account for – and do not over-account for – the human unsupervised learning effects presented thus far . Second , we built model variants in which the two decoder hyperparameters ( number of neural sites and number of training images ) were no longer correctly aligned with the initial human performance levels . Figure 5B illustrates the two-dimensional hyperparameter space , and the dashed line represents potential choices of the two hyperparameters that match human initial performance ( the IT-to-COR-behavior-matching manifold; Figure 3F ) . Regions above ( or below ) that manifold indicate hyperparameter choices where the decoders are better ( or worse ) performing than initial human performance . We found that the unsupervised learning effects predicted by the overall model ( Figure 5C , two black lines on top of each other corresponding to two choices of hyperparameters , black dots in Figure 5B ) continued to well-approximate human learning effects . This was also true for other combinations of hyperparameters along the dashed black manifold in Figure 5B ( ~10 combinations tested; results were similar to those shown in Figure 5C–E , not shown ) . In other words , for model settings in which the model variant was in line with the biological initial state , the predictions of the unsupervised learning effects remained similarly accurate . This is a nice robustness check on the model simulations and predictions . ( However , as a side note orthogonal to our goals here , this result also means that , as in prior work [Majaj et al . , 2015] , we cannot use this analysis to determine which of these model variants is more matched to the biology . ) In contrast , when we built model variants in which the choices of the two hyperparameters did not match human initial performance , the unsupervised learning effect predicted by the overall model clearly differed from the observed human learning effect . Specifically , when an overall model starts off with ‘super-human’ performance , it overpredicted the learning effect; and when a different model starts off as ‘sub-human , ’ it underpredicted the learning effect . In sum , it is not the case that any model of the form we have built here will produce the correct predictions – proper ( biological ) setting of the unsupervised IT plasticity rate and proper ( biological ) setting of the IT-to-COR-behavior-linkage model are both critical . It is important to note that we did not tune these biologically constrained hyperparameters based on fitting the unsupervised behavioral learning effects in Figure 2 – they were derived in accordance with prior neurobiological work as outlined above . So far , we have established a quantitative overall model that quite accurately predicted the direction , magnitude , and time course of learning effects resulting from a range of unsupervised exposure manipulations . For each of those tests , we focused on object discrimination tasks that had an intermediate level of initial task difficulty ( face discrimination tasks with initial d’ around 2 . 0 ) , so that we had dynamic range to see both increases and decreases in performance ( e . g . , Figure 2 ) . However , we noticed that our IT plasticity rule seemed to imply that those learning effects would depend on the strength of the initial selectivity of individual IT neural sites for the exposed objects ( i . e . , the initial P vs . N response difference ) . The intuition is that this response difference is the driving force for IT plasticity updates ( e . g . , no difference leads to no update , large difference leads to large update ) . This in turn implied that the learning effect size should depend on the initial task performance ( d’ ) . To test for this dependence , we focused on the unsupervised size tolerance ‘breaking’ manipulation ( as in u1 , Figure 2A , but with 800 unsupervised ‘swapped’ exposures; see Materials and methods ) and tested new sets of human subjects using a wide range of initial task difficulties , ranging from subordinate object discriminations ( low d’ ) to basic-level object discriminations ( high d’ ) . We focused on 13 size-specific object discrimination sub-tasks with either small-medium-size swapping exposure or medium-big-size swapping exposure . Each subject received only one exposure variant ( see Materials and methods ) . For each exposure variant , 20–40 new human subjects were tested , and we quantified the unsupervised learning effect ( ‘breaking’ ) as the change ( from initial ) in performance ( relative to control objects , as in Figure 2A ) . Figure 6B shows that unsupervised learning effect plotted against pre-exposure task difficulty for all 13 object discrimination tasks . This result not only confirms that this unsupervised learning effect is observed for a range of object discriminations ( e . g . , not just face objects ) , but it also showed a relationship between task difficulty ( d’ ) and the magnitude of that learning effect . In particular , for initially easy tasks ( d’ > ~2 . 5 ) and initially difficult tasks ( d’ < ~0 . 5 ) , we observed a smaller learning effect than tasks with intermediate initial performance . We found that our overall model quite naturally – for the reasons outlined above – predicted the smaller learning effect for initially difficult tasks ( the left side of Figure 6B ) . Notably , the model as defined above did not naturally predict the lack of observed learning effects for the initially easy tasks ( high initial d’ ) – it tended to overpredict the magnitude of behavioral changes that will result in those high d’ task ( see Figure 6B , black dots ) . However , we realized that , if we assumed that the model also has a lapse rate similar to that of humans ( Prins , 2012 ) , then this discrepancy might go away . That is , we assume that there is some non-zero fraction of trials for which the subject guesses or makes motor errors , regardless of the quality of the sensory-driven information . The intuition here is that human subjects make task-independent mistakes ( ‘lapses’ ) , and even a low rate of random lapses puts a ceiling on the d’ value that can be experimentally measured ( Figure 6A ) . In the context of our learning experiments , this assumption would mean that the underlying neural representation might indeed be changing a great deal ( at least that is what our current model predicts ) , but those changes cannot be measured as changes in human performance in the face of a lapse-rate induced measurement ceiling ( e . g . , a sensory evidence d’ of 5 could change to a sensory evidence d’ of 3 . 5 [a large d’ change of 1 . 5] , but we would measure an observed behavioral d’ of ~3 in both cases and thus report a behavioral d’ change of ~0 ) . In contrast , the overall model that we described above had a zero lapse rate , which meant that we could measure changes in its behavioral performance for even very large initial d’ values . To explore this , we asked: what is the ( mean ) lapse rate of the human subjects in our experiments ? To estimate this , we used half our human data to rank the average initial human performance of each task which we take as an estimate of the ordering of those tasks in terms of available sensory evidence . We then used the remaining data to plot the observed human performance on each task ( Figure 6—figure supplement 1A ) . We found that the average performance tended to plateau around 90% , which we take as an indication of a performance ceiling that cannot be explained by a lack of sensory evidence , and thus we attribute to multiple types of downstream errors collectively referred to as lapses . We also directly quantified the distribution of task performance accuracy for basic-level tasks ( easy tasks: d’ > 2 . 5 ) in our experiments ( Figure 6 ) and found that the distribution has a maximum of 95 . 0% ( Figure 6—figure supplement 1B ) . These analyses suggest that the lapse rate of our subject pool is ~10% ( i . e . , 95% accuracy for two choice tasks with perfect sensory evidence ) , which is consistent with prior work on human lapse rates ( <20%; Manning et al . , 2018 ) . We simulated the effect of 9% lapse rate and 20% lapse rate ( i . e . , we told the model to make random guesses on 9% or 20% of trials , regardless of the strength of sensory evidence ) and found that these new overall models reasonably explained the dependence of the observed magnitude of human d’ changes as a function of initial human d’ ( Figure 6B ) . In sum , we interpret the lapse rate analyses not as a failure of the overall model , but instead as a limitation of our psychophysical experiments in this study . That is , the lapse rate estimate is consistent with the hypotheses that , in the high initial d’ range , the IT population is changing ( indeed , the exposure conditions are close to the conditions of the original monkey neural experiments; Li and DiCarlo , 2010 ) , but that , in the face of a lapse rate , the behavioral consequences of those changes are predicted to be small relative to the effects of downstream biological variability . That being said , it also means that the current study is simply not able to test the IT plasticity to behavioral-learning linkage in the initial high d’ range , and we take that as a target for future experimental work ( see Discussion ) . Human ( and monkey ) visual object recognition is highly tolerant to object viewpoint , even under short , but natural , viewing durations of ~200 ms referred to as ‘core object recognition’ ( COR ) ( DiCarlo et al . , 2012 ) . Much evidence suggests that this ability derives from neural non-linear processing ( and thus neural re-representation ) of the incoming image along the ventral visual stream , and some ANN models have become reasonably accurate emulators of that non-linear processing and of its supported COR behavior ( Cadieu et al . , 2014; Krizhevsky et al . , 2017; Kubilius et al . , 2018; Yamins et al . , 2014 ) . However , because the ‘learning’ of those models is highly non-biological ( in the sense that millions of labeled images are used to explicitly supervise the learning ) , a key question remains completely open: how does the ventral stream develop its non-linear processing strategy ? One proposed idea is that , during postnatal development and continuing into adulthood , naturally occurring temporally continuous visual experience can implicitly instruct plasticity mechanisms along the ventral stream that , working together , lead to the transform-invariant object representation ( Berkes and Wiskott , 2005; Einhäuser et al . , 2005; Földiák , 1991; Wallis et al . , 2009 ) . Intuitively , the physics of time and space in our natural world constrains the visual experience we gain in everyday life . Because identity-preserving retinal projections often occur closely in time , the spatiotemporal continuity of our viewing experience could thus be useful to instructing the non-linear processing that in turn supports highly view-tolerant object recognition behavior . Under this hypothesis , objects do not need to be labeled per se , they are simply the sources that statistically ‘travel together’ over time . We are not the first to propose this overarching hypothesis or variants of it as the theoretical idea dates back to at least ~1960 ( Attneave , 1954; Barlow , 1961 ) . Földiák suggested that the internal representation should mimic physical entities in real life , which are subject to continuous changes in time ( Földiák , 1990; Földiák , 1991 ) . This process is purely unsupervised and achieves transformation invariance by extracting slow features from quickly varying sensory inputs ( Berkes and Wiskott , 2005; Sprekeler et al . , 2007; Wiskott and Sejnowski , 2002 ) . A range of mathematical implementations of learning rules ( Berkes and Wiskott , 2005; Földiák , 1991; Isik et al . , 2012; Körding et al . , 2004; Wiskott and Sejnowski , 2002 ) all include variants of this same conceptual idea: to achieve response stability of each neuron over time ( while also maintaining response variance over the full population of neurons ) . Various synaptic plasticity mathematical rules and associated empirical observations support this form of unsupervised learning: Hebbian learning ( Hebb , 1949; Földiák , 1991; Löwel and Singer , 1992; Paulsen and Sejnowski , 2000 ) , anti-Hebbian learning ( Földiák , 1990; Mitchison , 1991; Pehlevan et al . , 2017 ) , BCM rule ( Bienenstock et al . , 1982; Toyoizumi et al . , 2005 ) , and spike-timing-dependent plasticity ( Caporale and Dan , 2008; Markram et al . , 1997; Rao and Sejnowski , 2001 ) . This prior work showed that unsupervised learning of neural representations of objects through temporal continuity was possible , at least in theory . Human psychophysics studies have provided empirical evidence supporting the role of unsupervised temporal contiguity plasticity in visual object recognition . Wallis and Bulthoff found that unsupervised exposure to temporal image sequences of different views of different faces led to performance deficits compared to sequences of the same face ( Wallis and Bülthoff , 2001; Wallis and Bülthoff , 1999 ) . They also pointed out that these results were only observed in similar face pairs ( i . e . , low d’ ) rather than very distinct faces ( i . e . , higher d’ ) . Cox et al . showed that ‘swapped’ unsupervised experience of pairs of images across saccades could reduce ( ‘break’ ) position tolerance of object discrimination ( Cox et al . , 2005 ) . Balas and Sinha showed that observing object motion can increase both generalization to nearby views and selectivity to exposed views ( Balas and Sinha , 2008 ) . These behavioral observations revealed that unsupervised temporal contiguity is constantly contributing to the tolerance of object recognition behavior , even in adults , and thus it must be inducing some kind of underlying neural changes somewhere in the brain . Our human psychophysical results reported here extend this prior work in three ways . First , we measured the learning effects over prolonged periods of time , which allowed us to test for accumulation and saturation . Second , we found that the behavioral learning effect is reversible ( Figure 2C ) . Third , we found that this unsupervised learning effect depended on initial task difficulty , which might explain why some studies report stronger effects than others . For example , Wallis and Bulthoff found that the learning effects on view tolerance were only observed in similar face pairs rather than very distinct faces ( Wallis and Bülthoff , 2001 ) , and those similar face pairs have initial d’ that happens to reside in the mid-range where we predict/observe the largest behavioral effects ( Figure 6B ) . Third , and most importantly , we designed our unsupervised visual statistical manipulations in the same way as previous monkey neurophysiology experiments , which allowed us to quantitatively compare our human behavioral results with prior monkey neuronal results . Because IT is , among other ventral stream areas , thought to most directly underlie object discrimination behavior ( DiCarlo et al . , 2012; Ito et al . , 1995; Rajalingham and DiCarlo , 2019 ) and IT plasticity has been found in many studies ( Baker et al . , 2002; Logothetis et al . , 1995; Messinger et al . , 2001 ) , reviewed by Op de Beeck and Baker , 2010 , it is natural to ask if temporally contiguous unsupervised experience also leads to plastic changes in IT neurons . Miyashita and colleagues showed that neurons in the temporal lobe shape their responses during learning of arbitrarily paired images such that each neuron’s response becomes more similar to images that were presented nearby in time ( Miyashita , 1988; Miyashita , 1993; Naya et al . , 2003; Sakai and Miyashita , 1991 ) . Li and DiCarlo directly tested the role of unsupervised visual experience in IT neuronal tolerance by manipulating the identities and properties of objects presented consecutively in time ( Li and DiCarlo , 2008; Li and DiCarlo , 2010 ) . They found that , over ~1 . 5 hr of unsupervised exposure of ‘swapped’ temporal statistics , the size and position tolerance of IT neuronal responses were significantly modified , and that these changes were not reward or task dependent ( Li and Dicarlo , 2012 ) . Qualitatively similar temporal continuity-dependent neuronal plasticity has also been observed in rodents during development ( Matteucci and Zoccolan , 2020 ) . Although prior experimental work seemed qualitatively consistent with the overarching theoretical idea , it did not demonstrate that the behavioral learning effects could be explained by the IT neural effects . The results of our study here show that those two effects are quantitatively consistent with each other – the behavioral effects can be largely accounted for by the IT neural effects . While this extends the work of others in the area ( see Introduction ) , some studies have reported null results or have proposed alternative mechanisms . Okamura et al . showed that continuous motion or view presentation is not necessary to form tolerance , rather , enough exposure to different views ( even in random sequence ) can support view-invariant object recognition ( Okamura et al . , 2014 ) . This evidence suggests an alternative , or additional , mechanism to form tolerant object recognition in addition to temporal continuity . Van Meel and Op de Beeck investigated whether temporal continuity experience can alter size-tolerance representation in human LOC using fMRI and reported no observable effects ( Van Meel and Op de Beeck , 2020 ) . However , because no behavioral learning effects are reported and fMRI signal has limited spatial and temporal resolution , this null result may not be inconsistent with the results presented here . Looking in rodents , Crijns et al . tested the temporal contiguity hypothesis in adult rat primary visual cortex and found that the tolerance in orientation selectivity across spatial frequency was not affected by temporal continuity manipulation ( Crijns et al . , 2019 ) , which may be caused by a different representation mechanism in lower levels of visual hierarchy . On the other hand , Matteucci and Zoccolan reported that reduced temporal continuity experience in early postnatal life led to a loss of complex cell functional properties in rat V1 ( Matteucci and Zoccolan , 2020 ) . In sum , the literature is still highly varied and future neurophysiological and behavioral experiments are necessary to test the boundary conditions of temporal contiguity induced effects . We believe that models similar to the one proposed here will be an important future direction in harmonizing results across spatial scales ( neurons to behavior ) and across species ( rodents to primates to humans ) , such as the studies outlined above . A second future direction is to extend our current overall model to other modalities , like view invariance or position invariance . This could be done by collecting further psychophysical data , adding proper tuning kernels to the current generative IT model , and using the same IT plasticity rule and decoding model . A third future direction is to extend our current model to other objects beyond those that have been tested in monkeys and humans . This could be achieved through testing new IT population responses to new and old objects and then embedding the new objects in the MDG model of the IT population representation space based on neuronal population response similarity . Alternatively , we can use image-computable deep ANN models that quite accurately predict ventral stream neuronal population responses ( Kubilius et al . , 2018; Yamins et al . , 2014 ) and use the ‘IT’ layer to build a much larger representation space of objects . A fourth future direction is to develop new unsupervised learning algorithms that implement some of the core ideas of temporal contiguity learning , but are scaled to produce high-performing visual systems that are competitive with state-of-the-art neural network systems trained by full supervision . Many computational efforts have touched on this direction ( Agrawal et al . , 2015; Bahroun and Soltoggio , 2017; Goroshin et al . , 2014; Higgins et al . , 2016; Kheradpisheh et al . , 2016; Lotter et al . , 2016; Srivastava et al . , 2015; Wang and Gupta , 2015; Whitney et al . , 2016 ) , and some are just beginning to make predictions about the responses along the ventral stream ( Zhuang et al . , 2021 ) . A key next step will be to put those full-scale models to experimental test at both the neurophysiological and behavioral levels . To build a quantitative linking model that predicts unsupervised learning effects in humans from neuronal response in IT , we used three experimental datasets: ( 1 ) human data: human psychophysics performance data collected with Amazon Mechanical Turk; ( 2 ) IT population data: simultaneous recordings of 168 sites with multi-electrode Utah array recordings implanted in monkey IT ( from a previous study; Majaj et al . , 2015 ) ; and ( 3 ) IT single-site learning data: multi-unit activity recorded with single electrodes in monkey IT ( from a previous study; Li and DiCarlo , 2010 ) . All processed data are available at https://github . com/jiaxx/temporal_learning_paper ( copy archived at swh:1:rev:bb355bb96286db2148c3abdc8f71b5880f657c5f ) , Jia , 2021 . We used the same 3D object models as previous published IT-behavior study ( Majaj et al . , 2015 ) and applied the same rendering mechanism ( ray-tracing software ) to each 3D object while parametrically varying its position , rotation , and size , and projected on a randomly chosen unique natural background ( out of a pool of 130 images ) to generate new test image examples . All images were achromatic . The ground truth of each image was the identity of the generating 3D model , and this was used to evaluate performance accuracy . This naturalistic image generation allows us to gain full control of all the object-related metadata in the images while preserving a relatively natural COR experience . For each object , we predefine a ‘baseline view’ ( i . e . , exact center of gaze , size of ~2° or 1/3 of the diameter of the image , and canonical pose; see Methods of Majaj et al . , 2015; Rajalingham et al . , 2018 ) . Variations in size , position , and rotation are transformations relative to baseline view of the object . Since our focus here was unsupervised learning of size-tolerant object selectivity , we intentionally introduced more images that only vary in size to measure size tolerance . Medium-sized objects were the ‘baseline’ size ( ~2° ) . Small-sized objects were 0 . 5× of baseline ( ~1° ) . Big-sized objects were 2× that of baseline ( ~4° ) . All test images for different sizes were generated with random naturalistic backgrounds . We thus created a set of 240 ‘size test’ images per object ( i . e . , 80 images per object at each of the three test sizes ) . The final test images were each 512 × 512 pixels and were always presented to the subject at a total extent of ~7° of visual angle at the center of gaze ( as in the prior neurophysiology studies above ) . To neutralize possible size-specific attentional effects and possible size-specific adaptation effects , we presented these ‘size test’ images intermixed with other ‘cover’ images of the same objects . These cover images were generated using mild variation in all of the object view parameters . Specifically , we sampled randomly and uniformly from the following ranges: [−1 . 2° , +1 . 2°] for object position in both azimuth ( h ) and elevation ( v ) ; [−2 . 4° , +2 . 4°] for rotation in all three axes; and [x0 . 7 , x1 . 3] for size . These cover images were mixed randomly with the ‘size test’ images ( above ) at a ratio of 1 cover image per ‘size test’ image to generate a set of psychophysical test images for each subject ( illustrated in Figure 1—figure supplement 1B ) . The behavioral results from the cover images were not part of the analyses . All human experiments were done in accordance with the MIT Committee on the Use of Humans as Experimental Subjects ( COUHES ) . We used Amazon Mechanical Turk ( MTurk ) , an online platform where subjects can participate in non-profit psychophysical experiments for payment based on the duration of the task . In the description of each task , it is clearly stated that participation is voluntary and subjects may quit at any time . Subjects can preview each task before agreeing to participate . Subjects will also be informed that anonymity is assured and the researchers will not receive any personal information . MTurk requires subjects to read task descriptions before agreeing to participate . If subjects successfully complete the task , they anonymously receive payment through the MTurk interface . Since it is easier and faster to recruit subjects through MTurk , we can collect a much larger dataset than traditional in-lab human psychophysics . A total of 505 ( 174 subjects in Figure 2 and 331 subjects in Figure 6 ) subjects successfully completed our tasks published through Amazon’s Mechanical Turk . Subjects who failed to complete the task or follow the instructions were rejected . Aspects of COR performance were measured based on the behavioral report following each test image presentation ( Rajalingham et al . , 2018 ) . Previous work compared the results of COR tasks measured in the laboratory setting with controlled viewing with results measured via Amazon MTurk and found virtually identical results ( Pearson correlation 0 . 94 ± 0 . 01; from Majaj et al . , 2015 ) . Each behavioral experiment contained two types of phases: a test phase in which specific aspects of object discrimination performance were measured ( see below ) and an exposure phase in which pairs of temporally contiguous images were experienced ( see Figure 1 ) . The main experiment consisted of five test phases ( 200 trials each; 6–8 min ) and four interleaved exposure phases ( 400 exposure events each; 12–20 min ) that together allowed us to measure exposure-induced changes in size-specific object discrimination over time ( total experiment time ranged from 90 min to 120 min ) . Our goal was to measure the discriminability of targeted ( exposed ) pairs of objects at targeted ( exposed ) sizes ( and , as references , we also measured discriminability for control object pairs and for target objects at a non-exposed size ) . Conceptually , each such discrimination sub-task ( size-specific object discrimination task ) is a generated set of images from object A at a specific size that must be discriminated from a generated set of images of object B at a specific size , and mapped to the same object at a medium size ( e . g . , see Figure 1B , choice images ) . For clarity , we note that , given this design , the only variation in each of these sub-task image test sets was the image background . These size-specific sub-tasks were randomly interleaved with cover trials to disguise this underlying fact from the subject ( see Figure 1—figure supplement 1B ) . To measure performance on each sub-task , we used a 2AFC design . Each 2AFC trial started with a central fixation point . Subjects were requested to fixate the black fixation point because the test image was always presented briefly at that location and they might miss it otherwise . After 500 ms , the fixation dot disappeared and a test image appeared centered at dot location ( center of the screen ) for 100 ms , followed by the presentation of two ‘choice’ images presented on the left and right of the screen ( Figure 1A ) . One of the choice images always matched the identity ( or category ) of the object that was used to generate the test image and was thus the correct choice , and its location was randomly assigned on each trial ( 50% on the right and 50% on the left ) . After mouse-clicking a choice image , the subject was given another fixation point ( i . e . , the next test phase trial began ) . No feedback on correctness of the choice was given . To measure size-specific discrimination performance , we created size-specific 2AFC sub-tasks . Specifically , each sub-task was a balanced ( i . e . , 50%/50% ) set of size-specific test images generated from objects A and B ( see above ) , and the two choices presented after each test image were ‘clean’ examples of objects A and B at a standard ( ‘medium’ ) size ( Figure 1A ) . For each subject , the test images were pseudorandomly drawn from a test image pool that contained the desired number of ‘size test’ images and cover images ( Figure 1—figure supplement 1B ) . Among the 200 trials ( 50 test images of each test object; four objects in total ) , 40% contained the ‘size test’ images ( 20 for each object; 10 for small and 10 for big ) , 10% contained baseline views ( medium size; five for each object ) , and the remaining 50% test images were ‘cover images’ ( see above ) that were not used in analyses ( see Figure 1—figure supplement 1B for example test images ) . The number of test images for target and control object pairs was thus balanced . The number of test images for small and big sizes was also balanced regardless of exposure type . As a result , for each subject , we created six size-specific 2AFC sub-tasks in total ( three different sizes for each object pair ) regardless of exposure type . The number of test images for target and control face pairs at different sizes in each test phase is specified in Figure 1—figure supplement 1B . To evaluate exposure-induced learning effects , we only calculated the discrimination performance of three exposure-relevant sub-tasks ( preplanned , Figure 1B ) : ( 1 ) the sub-task with exposed ( target ) objects at the exposure-manipulated size ( Figure 1B , red or blue d’ ) ; ( 2 ) the sub-task with non-exposed ( control ) objects at the exposure-manipulated size ( Figure 1B , black d’ ) ; and ( 3 ) the sub-task with exposed ( target ) objects at the non-manipulated size ( Figure 1B , dashed black d’ ) . For example , one subject might have been randomly assigned to exposure type = ( experiment u1 , swapped condition ) , target size = ( big size ) , target objects = ( face A , face B ) , control objects = ( face C , face D ) . In this example , each test phase aimed to measure performance ( d’ ) on three specific sub-tasks: [face A big vs . face B big] , [face C big vs . face D big] , and [face A small vs . face B small] . The sizes of the subject groups are provided in Results . The test trials for size and objects were always balanced in each subject group . In Figure 2 , the subject groups differ in the exposure type ( three subject groups ) . In each of these three groups , the target exposure size was the big size , and within each group , the specific face objects for target and control were randomly selected for each subject . In Figure 6 , the 13 subject groups correspond to the 13 sub-tasks that were targeted for exposure ( see below ) . Within each subject group , the targeted type of object ( i . e . , face or basic level ) and the targeted exposure size ( i . e . , medium-big or medium-small ) was the same for all subjects , and within each group , the specific objects for target and control were randomly selected within the targeted type . We computed the d’ for each exposure-relevant sub-task ( typically three d’ values for each subject group; see Figure 1B ) based on the population ( pooled ) confusion matrix of the entire subject group . For each sub-task , we constructed a 2 × 2 confusion matrix by directly filling the behavioral choices into hit , miss , false alarm , and correct rejection according to the stimuli and response of each trial ( Figure 1B ) . From the pooled confusion matrix , we computed the d’ for each sub-task . We used standard signal detection theory to compute d’s from the confusion matrix ( d’ = Z ( TPR ) – Z ( FPR ) , where Z is the inverse of the cumulative Gaussian distribution function , and TPR and FPR are true-positive and false-positive rates , respectively ) . The d’ value was bounded within −7 . 4 to 7 . 4 ( via an epsilon of 0 . 0001 ) . The mean d’ for each sub-task of each subject group was determined by averaging the d’ calculated from each bootstrapped subjects sample ( which converges to the d’ of the pooled confusion matrix ) . The error bar ( bootstrapped standard error ) of performance represents the standard deviation of population pooled d’ over all bootstrap samples ( 1000 samples in each case ) , which is performed by sampling with replacement across all trials ( aggregated for each subject group ) . p-value is directly estimated from the bootstrapped distribution of performance change by comparing to no change condition , which is by definition 0 . Each exposure trial ( a . k . a . exposure ‘event’ ) in the exposure phase was intended to deliver a pair of temporally contiguous images at the center of gaze . Each trial initiated with the presentation of a small , central black dot ( ~0 . 5° ) , and the subject was required to mouse-click on that dot ( this is intended to naturally bring the center of gaze to the dot ) . Immediately after a successful mouse-click ( within 0 . 5° of the dot ) , two images were presented sequentially at the location of the black dot . Each image was shown for 100 ms with no time lag between them . After the event , the black dot reappeared at a new , randomly chosen location ( out of nine possible locations ) on the screen ( i . e . , the next exposure trial began ) . The details of those images are described below in the context of the specific experiments carried out . Because we here focused on the effects of unsupervised exposure events on size tolerance , the size of object in each of the two sequential images was always different and always included the medium ( ‘baseline’ ) size: either big-sized objects paired with medium-sized objects or small-sized objects paired with medium-sized objects . In either variant , the order of those two images was counterbalanced , as in Li and DiCarlo , 2010 ( e . g . , approximately half of the events transitioned from medium to big objects and the other half from big to medium objects; signified by the double-headed arrows in Figure 1B ) . Following prior work ( Cox et al . , 2005; Li and DiCarlo , 2010; Wallis and Bülthoff , 2001 ) , there are two basic flavors of unsupervised exposure . The first flavor is referred to as the swapped exposure , in which the two images within each exposure event are generated from different objects ( here , at different sizes ) . Based on prior work ( Cox et al . , 2005; Wallis and Bülthoff , 2001 ) , this exposure flavor is expected to gradually ‘break’ ( disrupt ) size-tolerance discrimination of those two objects . The second flavor is non-swapped exposure , in which the two images are generated from the same object ( here , at different sizes ) . While this have been less studies in human psychophysics , based on prior IT neurophysiology results ( Li and DiCarlo , 2010 ) , this exposure flavor is expected to gradually build size-tolerant discrimination of those two objects . Our main experimental goal was to test the directions , magnitudes , and temporal profiles of changes In size-tolerant object discrimination ( assessed in the test phases , see above ) resulting from different types of unsupervised exposure conditions ( Figures 1 and 2 ) . To do that , we deployed the two flavors ( above ) in three types of unsupervised experience types ( u ) , and each subject was tested in only one of those three types . The first type ( u1 ) was a series of swapped exposure epochs ( intuitively , this aims for maximal ‘breaking’ ) . The second type ( u2 ) was a series of non-swapped exposure epochs ( intuitively , this aims for maximal ‘building’ ) . The third type ( u3 ) was two swapped exposure epochs followed by two non-swapped exposure epochs ( intuitively , this aims to test the reversibility of the unsupervised learning ) . Each type of experiment lasted for about 90 min , and each consisted of nine phases in total: five test phases ( 200 test images each ) and four exposure epochs ( 400 exposure events in each epoch; Figure 1A ) . This experiment was done with face objects only in a total of 174 subjects over all conditions ( u1 = 102 subjects , u2 = 36 subjects , u3 = 37 subjects ) . Our secondary experimental goal was to study how learning effect depends on the perceptual similarity of the exposed objects . To do this , we chose pairs of objects to cover a wide range of initial discrimination difficulties . Intuitively , it is easier to discriminate an elephant from a pear than it is to discriminate an apple from a pear . Specifically , we chose a total of 13 size-specific object pairs selected from a set of eight face objects ( n = 10 pairs ) and six basic-level objects ( n = 3 pairs ) . For subjects being exposed to faces , the control objects were also faces; for subjects exposed to basic-level objects , the control objects were other basic-level objects . These 13 pairs were selected based on pilot experiments that suggested that they would span a range of initial discrimination performance . Indeed , when tested in the full experiment ( below ) , we found that mean human initial discrimination difficulties ranged broadly ( d’ range: 0 . 4–2 . 6 , based on the first test phase ) . We thus ran 13 groups of subjects ( i . e . , one group per target object/size pair ) with ~20–40 subjects per group . Because the goal here was to test the magnitude of size-specific learning ( not the time course ) , we tested only the ‘swapped’ flavor of unsupervised experience using just one long exposure epoch ( consisting of 800 exposure events ) . Each subject was exposed with only one pair of objects and was exposed to one size variant of the exposure: small-medium-size swapping or medium-big-size swapping . We bracketed that unsupervised exposure with one pre-exposure test phase ( 200 trials ) and one post-exposure test phase ( also 200 trials ) . The learning effect was always measured at the exposed size ( e . g . , if exposed with small-medium swapping , the learning effect was measured as the performance change of small-size discrimination task of the exposed object pair ) , subtracting the performance change for control object pair at the exposed size ( all exactly analogous to Figure 2A ) . When conducting multiple tests of the same null hypothesis and considering any one of those tests to reject that null hypothesis , this results in an increase in the likelihood of incorrectly rejecting the null hypothesis by pure chance . To set an appropriate null rejection level , a correction for multiple comparison ( e . g . , Bonferroni correction or FDR ) needs to be conducted , which corrects the alpha level for each test to account for the number of tests of the same null hypothesis . In our testing of learning effects over exposure amount ( Figure 2 ) , we are not asking whether the learning effect for any exposure amount is different from 0 , which would require multiple-comparison correction for number of tests . Instead , each point is a single test of a different null hypothesis: ‘There is no learning effect at exposure amount x . ’ These results demonstrate how learning effect changes as a function of exposure time for different exposure types . Therefore , we do not believe that the multiple-comparison correction is applicable in this situation . In the statistical test for learning effect of different tasks ( Figure 6 ) , the dependent variables are the observed learning effects for tasks that differ in initial task difficulty . We are asking whether the learning effect that is measured at 800 exposures for a given task is significantly different from 0 ( the null hypothesis for all tasks ) . Thus , there is only one comparison for each dependent variable; therefore , we believe that a multiple-comparison correction is not necessary here . If we were asking whether there is any learning effect observed for any one of the measured tasks given the exposure , then a multiple-comparison correction would be necessary , but that is not the question being asked here . Instead , we are simply showing the trend of the effect size for each tested task , with bootstrapped standard deviations of the mean , to demonstrate the relationship between initial task difficulty and the learning effect size . We modeled the IT neuronal population response based on the IT population dataset collected from monkey IT cortex with a MDG model . This model assumes that the distribution of IT population response ( the distribution the mean responses of individual IT neurons to all images of an object ) to each object is Gaussian-like . We tested this hypothesis with a normality test and found that 81 . 25% ( 52 out of 64 distributions for 64 objects ) of the IT population response distributions were Gaussian ( reject when p<0 . 01 ) . This MDG model preserves the covariance matrix of neuronal responses to all 64 objects that have been tested in monkey IT cortex . A random draw ( of a 64 × 1 vector ) from the MDG is conceptualized as the average response ( over image repetitions ) of a simulated IT recording site to each of the 64 objects . To generate the simulated IT tuning over changes in object size , for each simulated IT site , we multiplied ( outer product ) that 64 × 1 vector with a randomly chosen size-tuning kernel ( 1 × 3 vector ) that was randomly sampled from a batch of size-tuning curves ( Figure 3—figure supplement 1B ) that we had obtained by fitting curves to real IT responses across changes in presented object size ( n = 168 recording sites; data from Majaj et al . , 2015 ) . This gives rise to perfectly size-tolerant simulated IT neurons ( i . e . , by construction , the tuning over object identity and over size are perfectly separable ) . Note that this produced a population of simulated IT neurons with a broad range of size tuning , but with that range approximating that observed across actual IT neurons . The distribution of the variance across size ( a . k . a . variance across size reflects the shape of size-tuning curve , e . g . , 0 variance corresponds to a flat tuning curve across sizes ) is shown in Figure 3—figure supplement 1C . To introduce more biological realism and to approximate the fact that each image is presented on a random background , we randomly jittered each value in the 64 × 3 matrix by a zero mean , iid shift of each matrix element ( randomly drawn from the distribution of variance across image exemplars for each object from IT neural data [Figure 3—figure supplement 1D]; σ2clutter ) . Given this procedure , we could generate a potentially infinite number of simulated IT neurons and their ( mean ) responses to each and every image condition of interest . We verified that , even with the simplifying assumptions imposed here , the population responses of simulated IT populations were quite similar to the actual IT neural population responses ( in the sense of image distances in the IT population space [Figure 3B] and variance level [Figure 3D] ) . To generate a hypothetical IT neural ( model ) population , we simply repeated the above process to obtain the requested number of model neurons in the simulated population ( note: the MDG and the size-tuning kernel pool was always fixed ) . In addition , when we ‘recorded’ from these neurons ( e . g . , in Figure 3A ) , we additionally added response ‘noise’ that was independently drawn on each repetition of the same image ( σ2repeats; mean zero , variance scaled with the mean to approximate known IT Poisson repetition ‘noise’; Figure 3—figure supplement 1E ) . To generate behavioral performance predictions from model IT population responses , we applied a previously defined IT-to-recognition-behavior-linking model ( Majaj et al . , 2015 ) . In that study , the authors used actual IT neural population responses to show that a set of possible IT-to-behavioral-linking models could each accurately describe and predict human performance on all tested recognition tasks within the reliability limits of the data . We here used one of the simplest , most biological plausible of those models – a linking model that seeks to infer the test image’s true label by computing the Pearson correlation between the mean IT population response to each possible object class ( computed on the IT response to the training images ) and the IT population response evoked by the current test image ( note that test images are never used in the training of decoders ) . In other words , the model’s ‘choice’ of object category for each test image was taken to be the choice object whose ( simulated ) IT population mean ( over the training images ) was closest to the population vector evoked by the current test image . The only difference from the prior work ( Majaj et al . , 2015 ) is that here we used simulated IT neurons ( see Generative IT model ) to drive the ‘behavior’ of the model . ( Note that the linking model has two key hyperparameters [see Results] and , for each simulation run , we held those constant . ) Since the model ( IT population + linking model ) could now be treated as a behaving ‘subject , ’ we analyzed the behavioral choices in exactly the same way as the actual human behavioral choices to arrive at d’ values that could be directly compared ( i . e . , generate a confusion matrix for each 2AFC sub-task , see above ) . Similarly , to test a new model ‘subject , ’ we simply generated an entirely new IT model population ( see above ) and then found the parameters of the IT-to-behavior-linking model for that subject . To simulate human lapses ( see Results ) , we introduced a ( fixed ) percentage of trials in each of the ‘behavioral’ confusion matrices where model responses were randomly chosen . Note that , when initial d’ is below ~2 , the lapse rate most consistent with the data ( 9% ) has little influence on measurable performance ( see Figure 6A ) and thus only a minor effect on the model in Figure 5 . Therefore , all predictions in Figure 5 were made with 0% lapse rate . We built a descriptive ( non-mechanistic ) learning rule with the same fundamental concept as previous computational models of temporal continuity learning ( Földiák , 1990; Földiák , 1991; Sprekeler et al . , 2007; Wiskott and Sejnowski , 2002 ) , except its mathematical implementation . In our setup , there are always only two images in each exposure event ( a leading image and a lagging image ) . Our plasticity rule states that , after each exposure event , the modification of the mean FRresponse to the leading image is updated as follows:ΔFRleading=α ( FRlagging−FRleading ) This plasticity rule tends to reduce the response difference between two exposed images ( i . e . , it tends to create response stability over time , assuming that the statistics of the future are similar ) . In our overall model , we apply this plasticity rule to each and every simulated IT neuron ( true ) after each and every exposure event . Note that , under repeated exposure events , the FR to all images will continue to change until there is no difference in responses to the leading and lagging images , which means that the responses will eventually reach a steady state . Compared with previous plasticity rules ( e . g . , Hebbian rule ) for temporal continuity learning , our plasticity rule is relatively simple . Our plasticity rule updates each IT unit’s output FR directly rather than its input weights ( Földiák , 1991 ) . Based on immediate activity history , our learning rule continuously changes each unit’s output by pulling its responses to consecutive images closer until reaching steady state . This learning rule has several features . First , it is temporally asymmetric , which means that the direction of rate change of the leading image depends on the sequence of leading and lagging image . In other words , the response to the lagging image is going to pull the response to the leading image toward it . However , since our experiments randomized the leading and lagging images on each exposure trial , this results in a change in the response to both images rather than an asymmetric change . Second , the effect of our plasticity rule is constrained to exposed image pairs and ignores any correlation in the neural representation space . Even though we do not yet have experimental data to accurately generalize the plasticity rule further than what has been presented in this paper , it is potentially generalizable to other types of tolerance ( position , pose ) and to other exposure paradigms . The neural plasticity data were collected by selecting preferred ( P ) and non-preferred ( N ) images for each unit , which are two different objects ( Li and DiCarlo , 2010 ) . We set the neural plasticity rate of the simulated neurons to match that observed in biological IT neurons . To do this , we focused on the same initial high d’ regime as the neural plasticity data were collected . Specifically , for each simulated IT site , objects P and N were chosen independently out of the 64 objects based on its mean response to each object ( most likely to be in the high d’ regime ) . The plasticity rule was applied to each simulated site as it underwent unsupervised exposure , with the neural response function updated based on its responses to images containing objects P and N at the exposed sizes ( see Materials and methods: Generative IT model for details ) . Because of the initial randomness in the size-tuning kernel selected for each neural site ( Figure 3—figure supplement 1B , C ) and the clutter variance introduced for the site’s responses to different image exemplars ( Figure 3—figure supplement 1D ) , the response profile of each simulated neuron is unique and thus the updated direction during simulated unsupervised learning is not always in the same direction across the population . The averaged learning effect across all simulated neurons was then computed ( as if these neurons had been observed in an experiment ) and that simulated learning effect was compared to the averaged learning effect observed in the biological IT neurons . The plasticity rate was optimized to minimize this difference . The ( fixed ) plasticity rate determined in this way could then be used with the plasticity rule to compute the expected individual IT neuronal response pattern changes to any pair images for which the image-driven responses are both known . In this study , that means we could apply it to any images in the space of the generative model of IT , but we note that this same plasticity rule could be applied to other models of IT responses ( e . g . , those from contemporary image-computable models; Kubilius et al . , 2018; Yamins et al . , 2014 ) . However , it is important to note that the learning rate value is determined by the IT population statistics and the plasticity rule chosen here and thus should not be taken as a universal value . Changes in the statistics of the simulated IT population ( i . e . , covariance matrix , variance across sizes or clutter variance , etc . ) can influence the initial state of the IT population , and as a consequence influence both plasticity rate value and the predicted changes for each simulated IT neuron . The plasticity rate that best matches neural data is 0 . 0016 nru per exposure event ( nru = normalized response units ) . The normalized response is calculated by Δ ( P – N ) / ( P – N ) , where P and N represent the z-scored FR ( across all objects ) to preferred and non-preferred objects . Z-score is measured in terms of standard deviations from the mean . Therefore , 1 normalized response unit is 1 std of the response ( FR ) distribution across all tested objects . Since the mean multi-unit FR is 90 ± 23 spk/s ( std across objects ) for the IT population across 64 objects , we estimate that 1 nru is ~23 spk/s . Therefore , 0 . 0016 nru corresponds to a FR change of ~0 . 035 spk/s per exposure event , which means that ~30 exposure events of this kind would give rise to 1 spk/s change in P vs . N selectivity .	A bear is a bear , regardless of how far away it is , or the angle at which we view it . And indeed , the ability to recognize objects in different contexts is an important part of our sense of vision . A brain region called the inferior temporal ( IT for short ) cortex plays a critical role in this feat . In primates , the activity of groups of IT cortical nerve cells correlates with recognition of different objects – and conversely , suppressing IT cortical activity impairs object recognition behavior . Because these cells remain selective to an item despite changes of size , position or orientation , the IT cortex is thought to underly the ability to recognise an object regardless of variations in its visual properties . How does this tolerance arise ? A property called ‘temporal continuity’ is thought to be involved – in other words , the fact that objects do not blink in and out of existence . Studies in nonhuman primates have shown that temporal continuity can indeed reshape the activity of nerve cells in the IT cortex , while behavioural experiments with humans suggest that it affects the ability to recognize objects . However , these two sets of studies used different visual tasks , so it is still unknown if the cellular processes observed in monkey IT actually underpin the behavioural effects shown in humans . Jia et al . therefore set out to examine the link between the two . In the initial experiments , human volunteers were given , in an unsupervised manner , a set of visual tasks designed similarly to the previous tests in nonhuman primates . The participants were presented with continuous views of the same or different objects at various sizes , and then given tests of object recognition . These manipulations resulted in volunteers showing altered size tolerance over time . Aiming to test which cellular mechanism underpinned this behavioural effect , Jia et al . built a model that simulated the plasticity of individual IT cells and the IT networks , to predict the changes of object recognition observed in the volunteers . A high predictability of the model revealed that the plasticity in IT cortex did indeed account for the behavioral changes in the volunteers . These results shed new light on the role that temporal continuity plays in vision , refining our understanding of the way the IT cortex helps to assess the world around us .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	Unsupervised changes in core object recognition behavior are predicted by neural plasticity in inferior temporal cortex
How positional information instructs adult tissue maintenance is poorly understood . Planarians undergo whole-body regeneration and tissue turnover , providing a model for adult positional information studies . Genes encoding secreted and transmembrane components of multiple developmental pathways are predominantly expressed in planarian muscle cells . Several of these genes regulate regional identity , consistent with muscle harboring positional information . Here , single-cell RNA-sequencing of 115 muscle cells from distinct anterior-posterior regions identified 44 regionally expressed genes , including multiple Wnt and ndk/FGF receptor-like ( ndl/FGFRL ) genes . Two distinct FGFRL-Wnt circuits , involving juxtaposed anterior FGFRL and posterior Wnt expression domains , controlled planarian head and trunk patterning . ndl-3 and wntP-2 inhibition expanded the trunk , forming ectopic mouths and secondary pharynges , which independently extended and ingested food . fz5/8-4 inhibition , like that of ndk and wntA , caused posterior brain expansion and ectopic eye formation . Our results suggest that FGFRL-Wnt circuits operate within a body-wide coordinate system to control adult axial positioning . Adult animals replace cells during tissue turnover and , in many cases , regeneration . How animals specify and maintain regional tissue identity during these processes is poorly understood . Planarians can regenerate any missing body part and replace aged tissues during homeostasis , presenting a powerful system for identifying adult positional information mechanisms ( Reddien and Sánchez Alvarado , 2004; Reddien , 2011 ) . Planarian regeneration requires an abundant population of dividing cells called neoblasts that includes pluripotent stem cells ( Wagner et al . , 2011; Rink , 2013; Reddien , 2013 ) . Accordingly , many genes required for regeneration are required for neoblast biology . However , some phenotypes associated with gene inhibition do not impact the capacity of animals to regenerate , but instead affect the outcome of regeneration , suggestive of a role for such genes in providing positional information . For example , inhibition of components of the Wnt signaling pathway causes regeneration of heads in place of tails , generating two-headed animals with heads facing opposing directions ( Petersen and Reddien , 2008; Gurley et al . , 2008; Iglesias et al . , 2008 ) . Neoblasts are also constantly utilized for the replacement of differentiated cells during natural tissue turnover . Several striking planarian phenotypes associated with altered regional tissue identity during tissue turnover have also been identified , including hypercephalized ( Wnt-signaling inhibition ) ( Petersen and Reddien , 2008; Gurley et al . , 2008; Iglesias et al . , 2008 ) and ventralized ( BMP-signaling inhibition ) ( Reddien et al . , 2007; Molina et al . , 2007; Orii and Watanabe , 2007 ) planarians . Reminiscent of the roles of Wnt and Bmp in planarian regeneration and tissue turnover , Wnt regulates anterior-posterior ( AP ) axis development ( Petersen and Reddien , 2009b; Niehrs , 2010 ) and Bmp regulates dorsal-ventral ( DV ) axis development ( De Robertis and Sasai , 1996 ) in many animal phyla . Many receptors , ligands , and secreted inhibitors belonging to key pathways that regulate development in many organisms , such as BMP and Wnt pathways are constitutively expressed in a regionalized manner across adult planarian body axes ( Reddien , 2011 ) . Interestingly , these genes are predominantly expressed together in the same planarian tissue , the body-wall muscle ( Witchley et al . , 2013 ) . Expression patterns of these genes can change dynamically following injury ( Petersen and Reddien , 2008; Petersen and Reddien , 2009a; Gurley et al . , 2010; Witchley et al . , 2013 ) , and some of these changes can occur in existing muscle cells in the absence of neoblasts ( Witchley et al . , 2013 ) . Body-wall muscle is distributed peripherally around the entire planarian body , and the known expression domains of candidate patterning molecules in muscle broadly span the AP , DV , and medial-lateral ( ML ) body axes , raising the possibility that muscle provides a body-wide coordinate system of positional information that controls regional tissue identity in tissue turnover and regeneration ( Witchley et al . , 2013 ) . However , the roles for many of these genes with regionally restricted expression in muscle are poorly understood , and it is likely that many genes with regionally restricted expression in muscle and roles in positional information await identification . Identification of muscle as a major site of expression of genes controlling regeneration and tissue turnover in adult planarians presented the opportunity for systematic characterization of positional information in an adult metazoan . To this end , we performed single-cell RNA sequencing on muscle cells isolated from 10 discrete regions along the planarian AP axis and found 44 genes for which expression within planarian muscle was restricted to specific AP domains . An RNA interference ( RNAi ) screen of many of these genes revealed two similar circuits each containing FGFRL and Wnt components that are required for the normal patterning of two distinct regions of the planarian body: the head and the trunk . The prior identification of a single , body-wide cell type ( body-wall muscle ) expressing genes implicated in patterning in restricted domains ( Witchley et al . , 2013 ) raised the possibility that RNA sequencing of muscle cells could systematically identify components of this candidate adult positional information system . We sought such genes with regional expression in muscle utilizing single-cell RNA sequencing of muscle cells isolated from different regions along the AP axis . Non-dividing single cells from 10 consecutive regions along the AP axis ( Figure 1A ) were isolated by fluorescence activated cell sorting ( FACS ) , and the resulting single-cell cDNA libraries were screened by qRT-PCR for expression of planarian muscle markers before sequencing ( Methods , Figure 1—figure supplement 1A–C , Supplementary file 1A ) . Cells expressed an average of 3 , 253 transcripts , within the range reported for planarian single-cell libraries ( Wurtzel et al . , 2015 ) . Principal component analysis ( PCA ) on the 177 single cells sequenced was performed using highly variable transcripts . Two significant principal components that separated cells by expression of muscle markers ( PC1<0 ) and expression of neoblast markers ( PC2<0 ) were identified ( details in Methods , Figure 1—figure supplement 1D–F , Supplementary file 1B ) . PCA and troponin expression confirmed the identity of 115 muscle cells , and these 115 cells were used in all subsequent analyses ( Figure 1—figure supplement 1D , Supplementary file 1A ) . 10 . 7554/eLife . 12845 . 003Figure 1 . Single-muscle-cell RNA sequencing identifies regionally expressed genes on the planarian AP axis . ( A ) Single cells from each colored AP region were isolated by FACS and resultant cDNA was screened by qRT-PCR for muscle marker expression . Positive cells were sequenced and analyzed for differential expression . ( B ) Whole-mount in situ hybridization ( ISH ) ( n=2 experiments ) shows expression of a subset of new and previously known ( # ) muscle regionally expressed genes ( mRGs ) . Images are representative of n>10 animals for new mRGs . Anterior , up . Scale bar , 100 μm . Right , violin plots show the expression distribution in muscle cells ( black dots ) within the 10 dissected regions . cpm , counts per million . ( C ) Double fluorescence ISH ( FISH ) show co-localization of several newly identified mRGs ( magenta ) and the muscle marker collagen ( green ) . DAPI was used to label nuclei DNA ( gray ) . Yellow arrows point to cells co-expressing both genes . Scale bar , 10 μm . ( D ) Heat map shows hierarchical clustering of the average expression per region of the 44 identified mRGs . Top color bar indicates dissected region . ( * ) marks genes that are named by best human BLASTx hits . ( E ) Pie chart shows the percentage , within the 44 genes shown in D , of Wnt-signaling genes , FGFRL , and Hox homologs . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 00310 . 7554/eLife . 12845 . 004Figure 1—figure supplement 1 . Single-muscle-cell sequencing and analysis . ( A ) Schematic of the 10 regions dissected and macerated to isolate single cells . ( B ) Representative FACS plot of Hoechst-stained cells from a single region indicating the gate used to isolate non-dividing cells . ( C ) Representative qRT-PCR plot for the muscle marker troponin used to screen single-cell cDNA libraries . The libraries from cells circled in red were sequenced . ( D ) Principal component ( PC ) analysis and troponin expression identified 115 muscle cells . Cells separated along two significant principal components: PC1 ( 29 . 9% of variance explained , p=1 . 4E-120 ) separated muscle from epidermal lineage and PC2 ( 8 . 3% of variance explained , p=4 . 4E-44 ) separated neoblasts from differentiated cells ( Supplementary file 1B ) . Cells to the left of the dashed line that expressed troponin were retained for further analysis as muscle cells . ( E ) Distribution of contigs with two or more reads in the 177 single-cell libraries used for PC analysis . Nearly all cells with a high number of expressed contigs that could signal a doublet event from FACS were categorized as non-muscle cells and excluded from differential expression analysis . ( F ) Muscle cells from all regions were evenly distributed throughout PC-space indicating that AP region of origin did not explain a significant proportion of the variance . Inset includes number of muscle cells analyzed per region . ( G ) Different differential expression analysis methods were tested for the ability to identify known mRGs . The rank order by p-value is shown on the y-axis in log10 scale for several canonical mRGs . Arrows mark the rank separating significant ( filled circle ) and not significant ( n . s . , unfilled circle ) genes at p<0 . 01 for each method . ( H ) Three differential expression analyses ( left ) using SCDE were performed between the indicated regions . Ranking of genes by a differential expression score was used to generate a receiver-operator curve ( right ) to evaluate whether the SCDE analysis correctly classified genes as mRGs compared to ISH validation ( Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 00410 . 7554/eLife . 12845 . 005Figure 1—figure supplement 2 . 44 mRGs are distributed along the AP axis . Colorimetric ISH of mRGs ( blue ) identified in SCDE analyses . Violin plots show the distribution of cells that express that gene in each of the 10 dissected regions; cpm , counts per million . # , previously known mRGs . ( A ) Genes with p<0 . 005 in any analysis; ( B ) Genes with p>0 . 005 . Anterior , up . Scale bar , 100 μm . Each image is representative of n>5 animals . At least 2 independent ISH experiments were performed for each new mRG . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 005 Single-cell differential expression ( SCDE , [Kharchenko et al . , 2014] ) analysis of the data was used to identify regionally expressed genes in the muscle single cell sequencing data , because of its ability to identify transcripts of genes with known anterior and posterior expression patterns ( details in Methods , Figure 1—figure supplement 1G ) . SCDE analyses of three different anterior-versus-posterior region comparisons ( Materials and methods , Figure 1A , Figure 1—figure supplement 1H , Supplementary file 1C–E ) identified transcripts of 99 genes as differentially expressed at p<0 . 005 . To further validate the regional expression of these candidate genes , RNA probes were generated for all statistically significant transcripts ( 88/99 successfully amplified ) and whole-mount in situ hybridization ( ISH ) was performed ( Figure 1B ) . 18 genes with regional expression in muscle have been previously identified ( Figure 1—figure supplement 2 , [Witchley et al . , 2013; Vogg et al . , 2014; Reuter et al . , 2015] ) . Although these three SCDE analyses correctly identified 13 of these 18 genes , those expressed in rare muscle cells ( wnt1 , foxD , zic-1 ) , in shallow gradients ( sFRP-2 ) , or broadly ( wntA ) , were below statistical significance ( Materials and methods , Figure 1—figure supplement 1H , Figure 1—figure supplement 2B ) . Therefore , an additional 168 genes , for which transcripts showed non-significant differential expression in the SCDE analysis , were tested by ISH . ISH reveals expression in all tissue types , which might obscure detection of regional expression within muscle cells for some genes by this method . Nonetheless , ISH analysis verified 44 of these genes as regionally expressed ( 35/44 with p<0 . 005 in any of anterior-versus-posterior SCDE analyses ) from the total 256 genes tested , including 26 previously not reported to be regionally expressed in muscle ( Figure 1B , Figure 1—figure supplement 2 , Supplementary file 1F ) . All newly identified regionally expressed genes tested were expressed at least in part in cells expressing the planarian muscle marker collagen ( Figure 1C ) . The term position control gene ( PCG ) has been used for genes with both regional adult expression , and patterning abnormal RNAi phenotypes or prediction by sequence to be in a pathway regulating planarian patterning ( Witchley et al . , 2013 ) . The function for many such PCGs awaits elucidation . Many of the genes identified here have as yet no known function and cannot be linked to known signaling pathways by sequence; we will therefore use in this manuscript the broadly inclusive term muscle regionally expressed gene ( mRG ) . Hierarchical clustering of the average expression per region of the 44 mRGs identified recapitulates the AP order of the regions ( Figure 1D ) . Interestingly , the 44 identified mRGs identified here were comprised mainly of genes encoding Wnt-signaling components , Hox-family transcription factors , and fibroblast growth factor receptor-like ( FGFRL ) proteins ( Figure 1E ) , suggesting that these gene families have prominent roles in providing positional information for maintaining and regenerating the planarian primary body axis . Combinatorial expression analysis using fluorescence ISH ( FISH ) of previously known mRGs and those newly described here generated a map depicting multiple , overlapping expression domains in planarian muscle along the planarian AP axis ( Figure 2A , Figure 2—figure supplement 1A ) . Few genes , like sFRP-2 and ptk7 ( Gurley et al . , 2010; Reuter et al . , 2015 ) , were expressed broadly in the trunk . The posterior involves multiple overlapping expression domains of genes encoding Wnt , Hox , and novel proteins ( Petersen and Reddien , 2008; Adell et al . , 2009; Iglesias et al . , 2008; Reuter et al . , 2015; Currie et al . , 2016 ) . The anterior region involves overlapping expression domains of several components of the Wnt pathway and genes of the FGFRL family ( Petersen and Reddien , 2008; Rink et al . , 2009 ) , some of which extended from the anterior head tip to varying posterior extents of the head and some were expressed in the pre-pharyngeal region ( Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 12845 . 006Figure 2 . Co-expression of mRGs along the AP axis . ( A ) FISH using mRGs maps discrete domains of mRG expression onto the planarian AP axis . Bars on left indicate the approximate extent of the expression domain for each of the genes analyzed . Images are representative of n≥5 animals . Anterior , up . Scale bar , 100 μm . ( B ) Heatmap shows co-expression of anterior FGFRL and Wnt pathway mRGs in the four anterior regions indicated in the cartoon ( 1–4 ) . Each column shows expression within a single cell with color bars above indicating the dissected region of origin for the cell . cpm , counts per million ( C ) FISH using different FGFRL/ndl probes and Wnt pathway mRGs show co-expression in the four regions depicted in the cartoon . Black boxes ( 1–4 ) in the cartoon in B show the region imaged for the FISH , as denoted by the number and colored rectangle next to the merged image . Scale bar , 10 μm . Images are representative of n≥5 animals . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 00610 . 7554/eLife . 12845 . 007Figure 2—figure supplement 1 . Axial mRG map and co-expression of multiple FGFRL genes and mRGs in the same muscle cell . ( A ) FISH using a combination of known and new mRG RNA probes show distributions of gradients along the AP axis . Anterior , left . Scale bar , 100 μm . Each image is representative of n>5 animals . ( B ) Heatmap shows hierarchical clustering of the identified 44 mRGs in each of the 115 muscle cells analyzed . Cartoon on top depicts the 10 regions dissected . Top color bar indicates region of origin for that cell . Expression values for each gene are scaled across each row as z-scores . ( * ) marks transcripts that are named by best human BLASTx hits . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 00710 . 7554/eLife . 12845 . 008Figure 2—figure supplement 2 . Phylogenetic analysis of SMED-FGFRL proteins . Top right: Domain diagram of FGFR and FGFRL proteins . IG , immunoglobulin domain; TM , transmembrane domain; TyrKc , Tyrosine kinase . Tree showing 54 FGFRL proteins from diverse organisms , which were aligned using MUSCLE with default settings and trimmed with Gblocks . Maximum likelihood analyses were run using PhyML with 100 bootstrap replicates , the WAG model of amino acid substitution , 4 substitution rate categories and the proportion of invariable sites estimated from the dataset . All ML bootstrap values are shown above or below respective branch . Hs , Homo sapiens; Mm , Mus musculus; Xt , Xenopus tropicalis; Sp , Strongylocentrotus purpuratus; Cs , Capitella sp . I; Lg , Lottia gigantean; Ci , Ciona intestinalis; Sm , Schistosoma mansoni; Smed , Schmidtea mediterranea; Dj , Dugesia japonica; Dl , Dendrocoelum lacteum; Ptor , Planaria torva; Pt , Polycelis tenuis; Pn , Polycelis nigra; Nv , Nematostella vectensis . Right , ISH of the 6 Schmidtea mediterranea FGFRL genes shown in tree . Images are representative of n>10 animals . Anterior , left . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 00810 . 7554/eLife . 12845 . 009Figure 2—figure supplement 3 . Pattern of FGFRL/ndl family expression in β-catenin-1 RNAi animals . FISH shows expression of ndl-5 , ndl-2 , ndl-3 , and sFRP-1 in control and β-catenin-1 RNAi animals after one , two , or four RNAi feedings ( 1F , 2F , 4F ) . Animals were fixed at different timepoints after initiation of RNAi , shown in brackets at the top . Yellow arrows show ectopic expression of anterior mRGs in posterior regions of the animal before ectopic eyes are visible . Red arrow indicates ectopic expression of the prepharyngeal mRG ndl-3 . opsin ( green ) marks eyes . Anterior , up . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 00910 . 7554/eLife . 12845 . 010Figure 2—figure supplement 4 . Inhibition of FGFRL genes does not significantly change expression of other members of the FGFRL family . ( A ) Heatmap shows efficiency of RNAi inhibition in each condition , and no significant effects in the expression level of other genes . Color scale represents mean log2 fold change in expression of each gene ( rows ) in the RNAi conditions ( columns ) compared to control RNAi in 6 dpa head fragments ( cartoon on left , screen RNAi feeding protocol was used , see Methods ) . At least three head fragments were analyzed by qRT-PCR in each condition . One-way ANOVA , * p<0 . 05 , p<0 . 01 , * p<0 . 001 , **** p<0 . 0001 . ( B ) FISH shows normal expression of ndk and wntP-2 in ndl-1; ndl-2; ndl-4; ndl-5 RNAi animals in prepharyngeal fragments 6 dpa ( cartoon on left , same RNAi feeding protocol as A ) . Anterior , up . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 010 FGFRL-family proteins , which lack the intracellular kinase domain present in FGFRs , have been little studied but are broadly conserved ( Figure 2—figure supplement 2 , [Cebrià et al . , 2002; Bertrand et al . , 2009] ) . The molecular mechanism of action of FGFRL proteins is not well understood . Planarians have six FGFRL-encoding genes , named nou darake ( ndk ) , the defining member of the FGFRL family ( Cebrià et al . , 2002 ) , and nou darake-like ( ndl ) -1 through ndl-5 ( Figure 2—figure supplement 2 ) . All of these FGFRL genes were identified in the SCDE analyses as AP mRGs ( Figure 1D ) . At least three of the five FGFRL genes with expression at the anterior-most region of the animal ( ndk , ndl-1 , ndl-2 , ndl-4 , ndl-5 ) were co-expressed in all 11 muscle cells isolated and sequenced from the anterior head tip ( region 1 ) ; 10/11 of these cells also co-expressed ndk and frizzled5/8–4 ( fz5/8–4 ) ( Figure 2B ) . In the single muscle cells sequenced from the pre-pharyngeal region ( region 4 ) , 3/6 cells expressing ndl-3 also expressed wntP-2 . FISH also demonstrated co-expression of FGFRL and Wnt-pathway genes together in the same cells in regions where their expression domains overlapped ( Figure 2C ) . Similarly , extensive co-expression of multiple mRGs in single muscle cells was observed in different regions along the entire AP axis ( Figure 2—figure supplement 1B ) . Wnt signaling is also required for maintenance of the AP axis , with β-catenin-1 RNAi animals developing ectopic heads around the entire body during tissue turnover ( Petersen and Reddien , 2008; Gurley et al . , 2008; Iglesias et al . , 2008 ) . At early timepoints following RNAi ( 7–9 days after first RNAi feeding ) , β-catenin-1 RNAi animals showed subtle posterior expansions of ndl-5 and ndl-2 expression domains ( Figure 2—figure supplement 3 ) . Later , ectopic expression of ndl-5 in posterior and lateral locations occurred and preceded ectopic expression of ndl-2 ( 9 days after first RNAi feeding ) even before the appearance of ectopic eyes . Fully formed ectopic heads ( 21 days after first RNAi feeding ) showed clear ectopic expression of the pre-pharyngeal mRG ndl-3 ( Figure 2—figure supplement 3 ) . Gross anatomical changes in the AP axis are therefore accompanied by corresponding changes in FGFRL expression domains . The axial expression map of genes in planarian adult muscle is reminiscent of regionalized gene expression patterns found during embryonic development in other species ( Pankratz and Jäckle , 1993; De Robertis et al . , 2000; Jaeger et al . , 2012 ) and provides a tool to dissect adult positional identity maintenance and regeneration . The regional expression of mRGs raises the possibility that many of these genes will have a role in controlling regional tissue identity . Therefore , we sought to determine with functional assays the roles of particular mRGs in the maintenance and/or regeneration of regional tissue identity . The prominence of a few gene families in the dataset of mRGs suggests that FGFRL/Wnt/Hox genes are major patterning determinants of the planarian AP axis . We , therefore , performed extensive single and multi-gene RNAi to identify the roles of these genes in adult positional identity ( Supplementary file 1G ) . Inhibition of single or combinations of Hox genes and a subset of FGFRL genes ( ndl-1 , ndl-2 , ndl-4 , and ndl-5 ) did not result in animals with a robust abnormal phenotype ( Supplementary file 1G ) . Additionally , expression of other members of the FGFRL family was not affected under these RNAi conditions at the time-point analyzed ( Figure 2—figure supplement 4 ) . However , we found striking AP patterning phenotypes when inhibiting a subset of Wnt pathway components and FGFRL genes as described below . The ndl-3 gene is expressed from below the eyes to the esophagus at the anterior end of the pharynx ( Figure 1B , [Rink et al . , 2009] ) , which is located centrally in the animal trunk ( Figure 3A ) . ndl-3 RNAi resulted in a striking phenotype: the formation of two or more mouths and two pharynges ( Figure 3A–C , Figure 3—figure supplement 1A–D ) . The ectopic mouths and pharynges of ndl-3 RNAi animals appeared within the trunk , posterior to the original mouth/pharynx location . This phenotype emerged both during tissue turnover in uninjured animals ( Figure 3—figure supplement 1C ) and following regeneration ( Figure 3A , Figure 3—figure supplement 1A ) . In the case of regeneration , animals initially regenerated a single mouth/pharynx , but as regenerating animals grew following feeding , ectopic mouths and pharynges emerged . Inhibition of the posterior mRG wntP-2/wnt11-5 also caused ectopic mouth and pharynx formation ( Figure 3A–C , Figure 3—figure supplement 1A–D ) , in agreement with a recent report ( Sureda-Gómez et al . , 2015 ) . Double RNAi of ndl-3 and wntP-2 was synergistic ( Figure 3A , Fisher’s exact test p<0 . 0001 for ndl-3 , p=0 . 0153 for wntP-2 , Figure 3—figure supplement 1C ) . Inhibition of ndl-3 and wntP-2 also resulted in pharyngeal cavity expansion ( Figure 3B , C , Figure 3—figure supplement 1B ) , and in increased numbers of para-pharyngeal cells ( Figure 3B ) , which express the matrix metalloproteinase mmp1 ( Newmark et al . , 2003 ) . In summary , when either ndl-3 or wntP-2 was inhibited , ectopic trunk structures were added sequentially as the animal grew and replaced tissues . 10 . 7554/eLife . 12845 . 011Figure 3 . ndl-3 and wntP-2 restrict trunk positional identity . ( A ) Live , ventral images of ectopic pharynges and mouths in 20–30 day post-amputation ( dpa ) RNAi animals . Right top , cartoon depicts esophagus , pharynx , and mouth . Left , pharynges ( yellow arrows ) and ectopic mouths without a protruding pharynx ( white arrows ) . Scale bar , 500 μm . Right bottom , mouths ( white arrows ) in 7 dpa RNAi animals . Anterior , left . Total number of animals were pooled from at least 2 independent RNAi experiments . ( B ) Increased numbers of para-pharyngeal mmp1+ cells in RNAi animals . NB . 22 . 1e labels mouths . Graph below shows mean ± SD ( n>8 animals/condition , 2 pooled experiments , One-way ANOVA ) . ( C ) Esophagus-gut connection in 20 dpa trunk fragments , region in dotted rectangle is shown at higher magnification below . FISH: mat ( gut ) , mhc-1 ( pharynx ) , and NB . 22 . 1e ( esophagus ) . Bracket , pharyngeal cavity length . ( D ) Time-lapse images of an ndl-3; wntP-2 RNAi animal eating liver through both pharynges ( yellow arrows ) , see Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 01110 . 7554/eLife . 12845 . 012Figure 3—figure supplement 1 . ndl-3 and wntP-2 restrict the number of mouths and pharynges in the trunk region . ( A ) Ectopic posterior mouths are observed in regenerating trunk pieces of ndl-3 , wntP-2 , and ndl-3; wntP-2 RNAi animals 7 dpa . Number of animals showing ectopic mouths are described in Figure 3A . Yellow arrows point to mouths . Anterior , left; ventral , up . Scale bar , 100 μm . ( B ) DAPI stainings of RNAi animals show pharynges in the different RNAi conditions . Bracket indicates pharyngeal cavity length . Anterior , up; Scale bar , 100 μm . Images are representative of n>10 animals per condition . ( C ) Graph shows the percentage of intact RNAi animals with a total of two or more mouths after 8 RNAi feedings . p-values , Fisher’s exact test . 3 independent RNAi experiments are pooled in this analysis . Number of animals with ectopic mouths out of total animals ( n ) are indicated . ( D ) FISH using RNA probes for ndl-3 and wntP-2 show a decrease in the expression of those genes following their RNAi demonstrating the efficiency of the inhibition ( total of 8 RNAi feedings ) . Graphs below ( mean ± SD ) show quantification of the mRNA levels by qRT-PCR . Student's t-test , * p<0 . 05 , ** p<0 . 01 . Cartoons show the region from where the mRNA was extracted . ( E ) Graph shows the percentage of intact RNAi animals with a total of two or more mouths after different number of RNAi feedings . The first five RNAi feedings were performed with only ndl-3; wntP-2 dsRNA . β-catenin-1 or control dsRNA was added in addition to ndl-3; wntP-2 starting on feeding six . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 012 The planarian pharynx is a long muscular organ that can extend through the mouth to ingest food ( Reddien and Sánchez Alvarado , 2004 ) and connects to the intestine through the esophagus at the medial anterior end of the pharyngeal cavity . In ndl-3 and wntP-2 RNAi animals , ectopic esophagi ( NB . 22 . 1e+ ) formed in variable locations , including from the side of the pharyngeal cavity wall and from a gut branch crossing the pharyngeal cavity ( Figure 3C ) . Despite variable positioning , ectopic pharynges always integrated through an esophagus into the intestine , demonstrating remarkable plasticity in the mechanisms underlying tissue organization . Ectopic pharynges in ndl-3; wntP-2 RNAi or wntP-2 RNAi animals were also functional – animals simultaneously projected both pharynges and each pharynx displayed independent food searching behavior , such as on opposite sides of the animal or in different directions ( Figure 3D , Video 1 ) . 10 . 7554/eLife . 12845 . 013Video 1 . Control , wntP-2 , and ndl-3; wntP-2 RNAi animals eating from one or two pharynges . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 013 Next , we examined whether trunk expansion in ndl-3 and wntP-2 RNAi animals with ectopic pharynges and mouths affected the mRG axial expression map described above . Anterior-most mRG expression domains ( sFRP-1 and ndl-5 ) were present and showed no overt changes following ndl-3 and wntP-2 RNAi ( Figure 4A–C , Figure 4—figure supplement 1A ) . By contrast , the ndl-3 expression domain was expanded to the ectopic posterior esophagus in wntP-2 RNAi animals ( Figure 4A , D ) . In both wntP-2 and ndl-3; wntP-2 RNAi animals , the expression domain of sFRP-2 ( Figure 4—figure supplement 1D ) was also extended towards the animal posterior . By contrast , expression of the pre-pharyngeal mRGs wnt2 and ndl-2 was changed only slightly or not at all ( Figure 4—figure supplement 1B , C ) . Conversely , the broad posterior expression domain of wntP-2 was significantly reduced in ndl-3 RNAi animals ( Figure 4B , E ) . Expression of other posterior mRGs such as fz4-1 and dd_13065 was still present in ndl-3 , wntP-2 , and ndl3; wntP-2 RNAi animals ( Figure 4—figure supplement 1E , F ) . Thus , both ndl-3 and wntP-2 are required for maintaining normal trunk tissue pattern including associated mRG expression domains , but not head or tail patterns of mRG expression . Altogether , these data suggest that the trunk patterning defects of ndl-3 and wntP-2 RNAi animals only affect local mRG expression within the axial map . 10 . 7554/eLife . 12845 . 014Figure 4 . Trunk mRG gradients are shifted in ndl-3 and wntP-2 RNAi animals with ectopic pharynges/mouths . mRG expression analyses by FISH: ( A ) expanded expression of trunk mRG ndl-3 , ( B ) reduction of the lateral expression of the posterior mRG wntP-2 . Left panel , ventral view . Right panel , dorsal view . Red arrows point to the mRG expression domain boundary shifted . White arrows point to mouths . Yellow arrows indicate esophagus . Anterior , up . Scale bar , 100 μm . All FISH images are representative of n>8 animals per condition , and at least 2 independent RNAi experiments have been performed . ( C–E ) Graphs show quantification of the shifts in expression domains for the mRGs shown in the FISH experiments ( mean ± SD , at least 3 independent experiments were pooled . One-way ANOVA for sFRP-1 , unpaired Student's t-tests for ndl-3 and wntP-2 ) . Cartoons on the left depict the expression domain in the wild-type animal and the distance that was measured in each case . Length of expression domain measured was normalized by total length of the animal . All measurements were scored blind . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 01410 . 7554/eLife . 12845 . 015Figure 4—figure supplement 1 . ndl-3 and wntP-2 restrict trunk but not head or tail mRG expression domains in animals with ectopic pharynges/mouths . ( A-D , F ) FISH using RNA probes for different mRGs and the esophagus and mouth marker NB . 22 . 1e ( yellow or green ) . White arrows indicate mouths . ( A–D ) Red arrows point to the posterior edge of the mRG expression gradient . ( A ) ndl-5 ( magenta ) and ( B ) wnt2 ( magenta ) expression domains do not obviously expand . ( C ) ndl-2 ( magenta ) and ( D ) sFRP-2 ( magenta ) expression domain slightly expands in some RNAi conditions . Graph on right ( mean ± SD ) shows quantification of gradient shifts . One-way ANOVA * p<0 . 05 , p<0 . 01 . Cartoons on the left depict the gradient in the wild-type animal and the distance that was measured in each case . All measurements were scored blindly . ( E ) Colorimetric whole-mount ISH using the RNA probe for the posterior mRG fz4-1 . Black arrows point to fz4-1 expression . ( F ) Expression of the posterior mRG dd_13065 ( magenta ) is still present . FISH images are representative of n>8 animals , ventral view . All FISH and ISH experiments have been repeated at least twice from independent RNAi experiments . All animals were fixed 20 dpa . The screen RNAi feeding protocol was used ( see Materials and methods ) . Anterior , up; scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 015 In addition to trunk patterning phenotypes , we found that fz5/8–4 RNAi caused ectopic eye formation and expansion of the brain posteriorly in both uninjured and regenerating animals ( Figure 5A , B , Figure 5—figure supplement 1A , C ) . fz5/8–4 showed graded anterior expression , strongest at the head tip , and brain expression ( Figure 1B ) . The fz5/8–4 RNAi phenotype was similar to that previously described for ndk and wntA RNAi ( Figure 5—figure supplement 1B , [Cebrià et al . , 2002; Kobayashi et al . , 2007; Adell et al . , 2009; Hill and Petersen , 2015] ) . ndk is expressed in head muscle and the brain ( Figure 1B , [Cebrià et al . , 2002; Witchley et al . , 2013] ) and restricts brain tissues to the animal head ( Cebrià et al . , 2002 ) . wntA is expressed broadly , with strong expression at the posterior base of the brain ( Figure 1—figure supplement 2B , [Kobayashi et al . , 2007; Adell et al . , 2009; Hill and Petersen , 2015] ) . wntA; ndk double RNAi animals showed a stronger phenotype in homeostasis ( Figure 5A , B ) and regeneration ( Kobayashi et al . , 2007 ) , than did single gene RNAi animals . Double RNAi of fz5/8–4 and either ndk or wntA also showed a synergistic effect during tissue turnover ( Figure 5A , B , Figure 5—figure supplement 1B , Fisher’s exact test p<0 . 0001 for both ndk and wntA ) . Additionally , RNAi of four out of five other ndl ( FGFRL ) -family members further enhanced the fz5/8–4; ndk double RNAi phenotype ( Figure 5—figure supplement 1D , Supplementary file 1G ) , suggesting that multiple FGFRL genes synergize to control head pattern with ndk . 10 . 7554/eLife . 12845 . 016Figure 5 . fz5/8–4 , wntA , and ndk restrict head positional identity . ( A ) Posterior ectopic eyes seen in uninjured RNAi animals . Black arrows , ectopic eyes . Total number of animals have been pooled from 3 independent RNAi experiments . Cartoon on left shows area imaged . Graph below shows the percentage of intact animals with ectopic posterior eyes in each RNAi condition . ( B , C ) Posterior expansion of neuronal markers ( B ) ChAT and notum and eyes ( anti-ARRESTIN/VC-1 antibody , images representative of n>5 ) and ( C ) glutamic acid decarboxylase ( gd , red arrows mark posterior-most cell ) and photoreceptor marker opsin . Cartoon on left shows area imaged . Below , graph shows increased gd+ cell numbers , mean ± SD ( n>5 animals/condition , 2 independent RNAi experiments , One-way ANOVA ) normalized by the length from head tip to the esophagus . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 01610 . 7554/eLife . 12845 . 017Figure 5—figure supplement 1 . fz5/8–4 , wntA , and ndk restrict the brain tissue to the head region . ( A ) Ectopic eyes are shown in 7 dpa trunk fragments of fz5/8–4 RNAi animals . Black arrows point to ectopic eyes . Total numbers of RNAi animals are indicated . ( B ) Ectopic eyes ( black arrows ) are shown in an intact wntA RNAi and in a fz5/8–4; wntA double RNAi intact animal after 6 RNAi feedings . RNAi experiments were performed three times . Total numbers of RNAi animals are indicated . ( C ) FISH using the RNA probe fz5/8–4 shows decreased expression of this gene following its RNAi demonstrating the efficiency of the inhibition ( 6 RNAi feedings ) . Graph below ( mean ± SD ) shows quantification of the mRNA levels by qRT-PCR in different RNAi conditions . mRNA was extracted from 6 dpa head fragments . One-way ANOVA , ** p<0 . 0001 . ( D ) Synergistic RNAi effect of several members of the FGFRL family on the fz5/8–4; ndk RNAi phenotype . RNAi experiments have been performed twice . Total number of RNAi animals is indicated . Black arrows point to ectopic eyes . ( E , F ) FISH using the neuronal markers: ( E ) cintillo , and ( F ) notum , and the photoreceptor marker opsin ( E ) . Graph shows the percentage of cintillo+ cells ( mean ± SD ) in intact RNAi animals normalized by the length of the animal ( from the tip of the head to the esophagus ) , ( n>8 animals per RNAi condition , two independent RNAi experiments , One-way ANOVA ) . Yellow dotted line shows the esophagus location , * shows pharynx . White arrows show the posterior-most cell expressing the neuronal marker analyzed . Anterior , up; dorsal view unless indicated . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 017 The ndk RNAi phenotype is poorly understood . For instance , are mRG expression domains expanded along with the brain in ndk RNAi animals , and what diversity of cell types expands posteriorly ? Following eight RNAi feedings , fz5/8–4; ndk , wntA; ndk , and fz5/8–4; wntA double RNAi intact animals showed posterior expansion of multiple neuron classes from different head regions suggesting that the entire head-restricted nervous system expanded posteriorly ( Figure 5B , C , Figure 5—figure supplement 1E , F ) . After eight RNAi feedings , a time point at which brain expansion was visible , non-neural mag-1+ adhesive gland cells ( Zayas et al . , 2010 ) showed normal distribution ( Figure 6—figure supplement 1A ) . By 12 RNAi feedings , however , mag-1+ cells were disorganized ( Figure 6A , Figure 6—figure supplement 1D ) , indicating that non-neural head cell types were eventually affected , but not visibly posteriorized , by these RNAi conditions . 10 . 7554/eLife . 12845 . 018Figure 6 . Anterior and prepharyngeal mRG gradients are shifted in fz5/8–4 , ndk , and wntA RNAi animals with expanded brain tissue and ectopic eyes . ( A , B ) Expansion of mRG expression domains towards the animal posterior . White bracket marks distance between mRG posterior boundary and esophagus; red arrows mark expression domain shifts . White dotted lines outline pharynx . ( A ) ndl-5 , and ( B ) ndl-2 . ( A ) Disorganization of mag-1 expression ( yellow arrows ) . White arrows and opsin expression mark eyes . ( C ) Retraction of the pre-pharyngeal mRG ndl-3 . Red arrows points to the shift towards the posterior of the anterior gradient boundary . White bracket indicates distance from the tip of the head to the anterior edge of the ndl-3 gradient . In all panels , anterior is up . Scale bar , 100 μm . All FISH images are representative of n>10 animals and at least 2 independent RNAi experiments were performed . ( D–F ) Graphs show quantification of the expression domain shifts for the mRGs shown in the FISH experiments ( mean ± SD , at least 3 independent experiments were pooled , One-way ANOVA ) . Cartoons on the left depict the expression domain in the wild-type animal and the distance ( normalized to total length ) that was measured in each case . All measurements were scored blind . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 01810 . 7554/eLife . 12845 . 019Figure 6—figure supplement 1 . fz5/8–4 , wntA , and ndk locally restrict mRG expression in animals with expanded brain tissue and ectopic eyes . FISH and ISH in intact RNAi animals after 8 or 12 RNAi feedings . ( A , D ) FISH experiments show normal ( A , after only 8 RNAi feedings ) and abnormal ( D , after 12 RNAi feedings ) organization in the expression of the secretory cell marker mag-1 ( green ) . ( B , C ) FISH experiments show expression domain shifts of prepharyngeal mRGs ndl-2 ( magenta , B ) and wnt2 ( C ) . ( B–D ) Posterior mRG expression domains ( wntP-2 , wnt11-1 , wnt11-2 , both wnt11 genes pooled in C ) did not change . opsin ( green or magenta ) and NB . 22 . 1e ( yellow ) . For all FISH images , red arrows point to the posterior edge of the mRG expression domain . White arrows point to eyes . Anterior , up . Scale bar , 100 μm . FISH images are representative of n>8 , FISH experiments . DAPI shows ectopic eyes in the fz5/8–4; ndk RNAi animal imaged ( yellow arrows ) in B . FISH was performed twice from independent RNAi experiments . ( E ) ISH using ndk RNA probe shows posterior expansion of the ndk expression domain in a fz5/8–4; wntA RNAi animal after 8 RNAi feedings . Images are representative of n≥5 animals per condition . Anterior , left . Right graph shows quantification of increased ndk expression by qRT-PCR . Cartoons on top indicate the region from which mRNA was extracted . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 019 We next examined whether the axial mRG map changed in RNAi animals with posterior ectopic eyes . Anterior-most ( sFRP-1 and ndl-4 ) and posterior ( wntP-2 , wnt11-1 , and wnt11-2 ) mRG expression domains did not visibly change in fz5/8–4; ndk RNAi or wntA; ndk double RNAi animals , even under strong RNAi conditions ( 12 RNAi feedings , multiple ectopic eyes ) ( Figure 6B , C , Figure 6—figure supplement 1B–D ) . By contrast , ndl-2 expression , which is normally restricted to a domain immediately posterior to the eyes , was expanded into the pre-pharyngeal region after eight RNAi feedings ( Figure 6B , E ) , and showed a more severe posterior expansion after 12 RNAi feedings ( Figure 6—figure supplement 1B ) . Similarly , ndl-5 , ndk , and wnt2 expression domains extended posteriorly into the pre-pharyngeal region in RNAi animals with strong phenotypes ( Figure 6A , D , Figure 6—figure supplement 1C , E ) . By contrast , the anterior end of the pre-pharyngeal ndl-3 expression domain was significantly posterior-shifted ( Figure 6C , F ) . Our results indicate that ndk , wntA , and fz5/8–4 are required to restrict anterior tissues and associated expression domains of mRGs to the head region , while leaving the anterior tip , trunk , and tail mRG domains unaffected . Single-cell sequencing has recently been used to identify transcriptomes for multiple planarian cell types ( Wurtzel et al . , 2015 ) . We utilized single-cell sequencing to map axial gene expression within the planarian muscle . Planarian muscle was previously found to express several genes with known roles in adult tissue patterning in planarians , raising the possibility that muscle functions to produce a body-wide coordinate system of positional information ( Witchley et al . , 2013 ) . Traditional RNA sequencing approaches to identifying candidate adult positional information in planarians is limited by the diversity of gene expression in heterogeneous tissue . The identification of a particular cell type expressing genes associated with regional tissue identity allowed application of regional single-cell sequencing to surmount this challenge . We applied this approach to the AP axis , and identified mRGs that constitute an expression map of muscle cells of the planarian primary axis ( Figure 7A ) . Coordinate systems of positional information , such as those proposed to control embryonic development ( De Robertis et al . , 2000; Petersen and Reddien , 2009b; Niehrs , 2010 ) , might exist within adult tissues of many animals , including humans ( Rinn et al . , 2006 ) , however there is little functional data regarding positional information and maintenance of the adult body plan . Here , we described several mRGs that work together to pattern and maintain two distinct body regions , the head and the trunk . 10 . 7554/eLife . 12845 . 020Figure 7 . Two FGFRL-Wnt circuits control AP patterning in planarians . ( A ) Expression domains of all identified mRGs along the planarian AP axis . Wnt pathway ( purple ) , FGFRL ( orange ) , and Hox genes ( green ) . In bold , genes shown here to be involved in maintaining regional identity . ( B ) Cartoons summarize the characterized RNAi phenotypes . ndl-3 and wntP-2 restrict the number of pharynges and mouths in the trunk region . wntP-2 RNAi animals with ectopic pharynges/mouths have an expanded ndl-3 domain whereas ndl-3 RNAi animals with ectopic pharynges/mouths have a reduced wntP-2 expression domain . fz5/8–4 , ndk , and wntA restrict the brain tissue to the head . Inhibition of these genes results in ectopic posterior eyes , brain expansion , and expanded domains of head mRGs . ( C ) Expression domains of the two FGFRL-Wnt circuits are shown . Black brackets indicate the region controlled by the FGFRL-Wnt circuits . DOI: http://dx . doi . org/10 . 7554/eLife . 12845 . 020 Constitutive regional expression of planarian orthologs to genes with key developmental roles in metazoans has been hypothesized to be important for multiple aspects of planarian body plan maintenance ( Reddien , 2011 ) . Here we show that both the planarian head and trunk require an FGFRL-Wnt circuit to maintain adult regional tissue identity . Different FGFRL and Wnt genes are used in the two body locations; however , in both cases , an FGFRL expression domain is juxtaposed by a posterior Wnt expression domain . Strikingly , inhibition of either gene ( FGFRL or Wnt ) caused posterior expansion and sequential duplications of structures normally found within the head and trunk regions , resulting in expanded brain and ectopic eyes in one case , and ectopic pharynges and mouths in the other ( Figure 7B ) . Following inhibition of any of the components of these FGFRL-Wnt circuits , the axial expression map shows local shifts in expression domains coincident with the expansion of specific regionally restricted tissues such as brain and pharynx ( Figure 7B ) . These results suggest that muscle , a tissue found uniformly throughout the animal , marks different AP regions through combinatorial expression of mRGs . This implies that communication exists between muscle cells and the underlying region-specific tissues . Understanding the coordination between muscle cells and tissues within AP regions is necessary for determining how planarians are able to robustly maintain and regenerate their entire body plan . Positional information must be integrated into the decision to generate and pattern new tissue during planarian growth and regeneration . mRGs might therefore influence the regional behavior of neoblasts and/or their division progeny . The two FGFRL-Wnt circuits described in this work are striking examples of body plan plasticity during homeostatic tissue turnover . In both cases , inhibition by RNAi of FGFRL genes ( i . e . , reduction or absence of the FGFRL expression domain ) and inhibition of Wnt pathway components ( i . e . , expansion of the FGFRL expression domain ) are coincident with the same phenotype of expanded regional identity . Future characterization of the biochemical properties of FGFRLs and elucidation of the mechanisms of interaction between Wnt/Fz pathways and FGFRLs might help in understanding this property . We propose that FGFRL proteins confine the regions where specific tissues in both the head and trunk can normally form and that the Wnt gene of each circuit acts by restricting tissues at the anterior end of its expression domain ( Figure 7B , C ) . Given the similarities of the distinct FGFRL-Wnt circuits for patterning two different body regions , FGFRL-Wnt circuits might be broadly utilized , but presently underappreciated , patterning modules of animal body plans . Asexual Schmidtea mediterranea strain ( CIW4 ) animals starved 7–14 days prior experiments were used . Animals were dissected into 10 adjacent regions along the AP axis , and only the midline region ( i . e . , in between the ventral nerve cords ) of each segment was utilized , to minimize heterogeneity caused by gradients expressed along the medio-lateral axes . 10 regions were chosen to balance consistency of amputation and AP resolution . The pharynx was dissected out and discarded for the regions 5 and 6 . Fragments were dissociated into single cells . Single cell suspensions for each region were stained labeled with Hoechst , and non-dividing single cells were sorted by flow cytometry into 96 well plates containing 5 ul of total cell lysis buffer ( Qiagen , Germany ) with 1% β-mercaptoethanol . Subsequently , amplified cDNA libraries were made from each single cell using the SmartSeq2 method ( Picelli et al . , 2013; 2014; Wurtzel et al . , 2015 ) , and tested by qRT-PCR for the expression of the muscle markers collagen and troponin ( collagen Fw: GGTGTACTTGGAGACGTTGGTTTA , collagen Rv: GGTCTACCTTCTCTTCCTGGAAC; troponin Fw: ACAGGGCCTTGCAACTATTTTCATC , troponin Rv: GAAGCTCGACGTCGACAGGA ) . Cells expressing either or both of these muscle-specific genes ( ~5 in 96 cells ) were used to make libraries using the Nextera XT kit ( Illumina , Inc ) . Libraries were sequenced ( Illumina Hi-seq ) and fastq files generated by Illumina 1 . 5 and examined by fastqc . Sequencing data was submitted to the GEO database as GSE74360 . Each cell was sequenced twice , once with 80 bp reads and once with 40 bp reads , and reads from both sequencing runs were concatenated . Reads were trimmed using cutadapt to remove Nextera transposon sequences CTGTCTCTTATA and TATAAGAGACAG ( overlap 11 bp ) and low quality 3′ base pairs ( quality score less than 30 ) before mapping to the dd_Smed_v4 assembly ( http://planmine . mpi-cbg . de; [Liu et al . , 2013] ) using bowtie 1 ( Langmead et al . , 2009 ) with -best alignment parameter . Bowtie 1 was used because of its better sensitivity mapping <50 bp reads . Read counts from prominent mitochondrial and ribosomal RNAs ( dd_smedV4_0_0_1 , dd_Smed_v4_7_0_1 , and dd_Smed_v4_4_1_1 ) were discarded . Reads from the same isotig were summed to generate raw read counts for each transcript ( Wurtzel et al . , 2015 ) . Libraries with fewer than 1000 expressed ( >2 reads ) transcripts were discarded , leaving 177 cells , expressing an average of 3 , 253 unique transcripts with an average of 430 , 114 reads mapped ( Supplementary file 1A , Figure 1—figure supplement 1E ) . Counts per million reads ( cpm ) were log transformed after addition of a pseudocount and used as expression values for violin plots and heatmap in Figure 2B . Principal component ( PC ) analysis on a set of highly expressed transcripts ( 4 < mean expression < 8 ) with high variance ( dispersion > 1 . 2 ) was extended to the entire set of transcripts to identify two significant PCs ( Seurat [Satija et al . , 2015] , Supplementary file 1B ) . The transcripts defining PC1<0 are all found within the top 45 of the published set of muscle-enriched transcripts ( Wurtzel et al . , 2015 ) . Two clear populations were separated by PC analysis , one of which included 115 cells that expressed troponin ( >4 cpm ) ( Figure 1—figure supplement 1D , F ) . These 115 muscle cells were used for all subsequent analysis ( Supplementary file 1A ) . Average expression per region ( Seurat ) for each transcript ( Figure 1D ) or expression per cell ( Figure 2—figure supplement 1B ) was centered and scaled to generate expression z-scores used for heatmap visualization . Dendrograms show complete hierarchical clustering using Euclidean distance ( Figure 1D , Figure 2—figure supplement 1B ) . Rv3 . 2 . 2 was used for all subsequent data analysis and visualization , relying on the following packages: Seurat ( Satija et al . , 2015 ) , SCDE ( Kharchenko et al . , 2014 ) , matrixStats , ROCR ( Sing et al . , 2005 ) , ggplot2 , RColorBrewer ( http://colorbrewer2 . org ) . To determine the best differential expression analysis method to identify putative mRGs , we tested three statistical methods: SCDE ( Kharchenko et al . , 2014 ) , bimod ( McDavid et al . , 2013 ) , and t-test ( Student's t-test ) for their ability to identify known mRGs . For each of the statistical tests , we compared cells from the head tip ( region 1 ) versus those from the tail tip ( region 10 ) and determined the rank and statistical significance of 10 canonical mRGs ( Figure 1—figure supplement 1G ) . Based on its ability to identify mRGs present only in a subset of cells within a region ( e . g . , wnt11-1 , sFRP-1 , notum ) , we chose SCDE for all further differential expression analysis . Note that SCDE explicitly accounts for drop-out rates due to single-cell sequencing by calculating a probability distribution for each transcript in each cell before calculating differential expression between groups . To identify putative mRGs , we performed three differential expression analyses: anterior ( regions 1 , 2 , 3; n=23 cells ) versus posterior ( regions 8 , 9 , 10; n=38 cells ) ; head ( region 1; 11 cells ) versus post-pharyngeal ( regions 7 , 8 , 9; n=35 cells ) ; pre-pharyngeal ( regions 2 , 3 , 4; n=22 cells ) versus tail ( region 10; n=12 cells ) ( Figure 1A ) . All transcripts with a \|Z\| score greater than 2 . 58 ( p<0 . 005 ) in any of the SCDE analyses were screened by ISH ( Supplementary file 1C–E ) . Z-scores corrected for multiple hypothesis testing are reported in Supplementary file 1C–E , however we used uncorrected Z-scores due to their ability to rank many more transcripts . In addition , 168 genes below our statistical cutoff were successfully amplified from cDNA and screened by ISH ( Supplementary file 1C–E ) . To determine if our method correctly classified transcripts as mRGs , we used a combined score from all three SCDE analyses ( Figure 1—figure supplement 1H , [Wan and Sun , 2012] ) . If there is no differential expression of a gene along the AP axis , then the minimum p-value from any of the analyses , p[1] = min ( p1 , … , pk ) where pi is the p-value from ith analysis , is expected to follow a beta distribution with parameters 1 and k ( Tippett , 1931 ) . A differential expression score was calculated based on beta distribution , Smin= -log ( P ( beta ( 1 , k ) < p[1] ) ) , and used to rank all transcripts . Note that this scoring system only ranks transcripts , and that statistical significance of differential expression is only interpretable within the analysis performed . We then quantified how well our analyses , as scored by Smin , classified transcripts as mRGs , as determined by ISH validation . The receiver-operator curve plots the false positive rate versus the true positive rate for each value of Smin based on ISH validation . The area under the curve ( 0 . 88 , perfect classification = 1 , random classification = 0 . 5 ) indicates that Smin and therefore our SCDE analyses are able to correctly classify transcripts as mRGs visible by ISH . Primers used to PCR amplify all planarian transcripts are listed in Supplementary file 1C–E . 44 mRGs were cloned from cDNA into the pGEM vector ( Promega , Madison , WI ) . RNA probes were synthesized and nitroblue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate ( NBT/BCIP ) colorimetric whole-mount in situ hybridizations ( ISH ) were performed as described ( Pearson et al . , 2009 ) . Fluorescence in situ hybridizations ( FISH ) were performed as described ( King and Newmark , 2013 ) with minor modifications . Briefly , animals were killed in 5% NAC before fixation in 4% formaldehyde . Following treatment with proteinase K ( 2 μg/ml ) and overnight hybridizations , samples were washed twice in pre-hyb buffer , 1:1 pre-hyb:2X SSC , 2X SSC , 0 . 2X SSC , PBST . Subsequently , blocking was performed in 5% casein ( 10X solution , Sigma , St . Louis , MO ) and 5% inactivated horse serum PBST solution when anti-DIG or anti-DNP antibodies were used , and in 10% casein PBST solution when an anti-FITC antibody was used . Post-antibody binding washes and tyramide development were performed as described ( King and Newmark , 2013 ) . Peroxidase inactivation with 1% sodium azide was done for 90 min at RT . Live animal images were taken with a Zeiss Discovery Microscope . Fluorescent images were taken with a Zeiss LSM700 Confocal Microscope . Co-localization analysis of FISH signals was performed using Fiji/ImageJ . For each channel , histograms of fluorescence intensity were used to determine the cut-off between signal and background . All FISH images shown are maximal intensity projections . A median filter was applied using the ImageJ Despeckle function . Images are representative of results seen in >5 animals per panel . dsRNA was prepared from in vitro transcription reactions ( Promega ) using PCR-generated templates with flanking T7 promoters , followed by ethanol precipitation , and annealed after resuspension in water . dsRNA was then mixed with planarian food ( liver ) ( Rouhana et al . , 2013 ) and 2 ul per animal of the liver containing dsRNA was used in feedings . For the RNAi screen ( Supplementary file 1G ) , the following feeding protocol was used: animals were fed six times in three weeks , cut in four pieces ( head , pre-pharyngeal , trunk and tail pieces ) , allowed to regenerate for 10 days , fed all together six times in another three weeks , and cut again into the four pieces described above . Seven days following amputation ( 7 dpa ) , trunk pieces were scored ( Supplementary file 1G , Figure 3A right panels , Figure 3—figure supplement 1A , and Figure 5—figure supplement 1A ) , and fixed at day 20 following amputation for further analysis ( Figure 3 and Figure 4 ) . Five to 10 trunk pieces were kept after the first regeneration cycle , and were fed once a week for another 12 weeks and were scored after that period ( homeostasis , Supplementary file 1G ) . Animals for homeostasis RNAi experiments for trunk patterning studies ( Figure 3—figure supplement 1C ) were fed eight times in four weeks and scored a week after the last feeding . In RNAi experiments for head patterning analysis , animals were fed eight times in four weeks , scored after the first six feedings ( Figure 5A , Figure 5—figure supplement 1B , D ) , and fixed seven days after the last feeding ( without amputation ) . For longer time point experiments , animals were fed twelve times in six weeks and fixed seven days after the last feeding . For all RNAi conditions tested , the total amount of dsRNA per feeding per animal was kept constant . Therefore , for example , when RNAi of two or more genes was performed , dsRNA for each gene was diluted in half . For combinations of dsRNAs , synergistic effects of double RNAi were calculated using Fisher’s exact test ( Figure 3A , 5A , Figure 3—figure supplement 1C ) . Minimum sample sizes were estimated using difference of proportion power calculation with h=0 . 4 ( ectopic mouths ) or h=0 . 8 ( ectopic eyes ) , sig . level=0 . 05 , and power=0 . 8 ( n=98 . 1 or n=24 . 5 ) . RNAi animals with ectopic pharynges/mouths were treated with 0 . 2% chlorotone , which results in muscle relaxation and pharynx protrusion through the mouth ( Figure 3A , left panels ) . For RNAi enhancement experiments of ndl-3; wntP-2 RNAi with β-catenin-1 , animals were fed five times with a combination of ndl-3 and wntP-2 , and starting in the sixth feeding , β-catenin-1 or control dsRNA was added to the mix of ndl-3 and wntP-2 for another three , four , or five feedings ( being a total of eight , nine or 10 feedings , Figure 3—figure supplement 1E ) . For the β-catenin-1 RNAi experiment shown in Figure 2—figure supplement 3 , animals were fed once , twice , or four times with β-catenin-1 or control dsRNA . Animals were fixed at different days after the first RNAi feeding . Samples were processed and analyzed as described ( Owen et al . , 2015 ) . Briefly , total RNA was isolated from fragments from individual intact worms or from individual regenerated fragments , as indicated by cartoons in figures , in 0 . 75mL Trizol ( Life Technologies , Carlsbad , CA ) following manufacturer's instructions . Samples were homogenized for 30s using TissueLyser II ( Qiagen ) . Following RNA purification and resuspension in dH20 , concentrations for each sample were measured by Qubit using RNA HS Assay Kit ( Life Technologies ) . 5 ng of RNA were treated with 1U amplification-grade DNAse I ( Life Technologies ) for 15 min at room temperature before DNAse heat-inactivation for 10 min at 65°C in the presence of 2 . 5 mM EDTA . Multiplex reverse-transcription and 15 cycles of PCR amplification were performed on DNAse-treated RNA using pooled outer primers at 50 nM each and Superscript III/Platinum Taq enzyme mix ( Supplementary file 1H ) . Following outer primer digestion with ExoI ( 15U , New England Biolabs , Ipswich , MA ) , samples were diluted to 500pg/ul and checked for presence of g6pd by qRT-PCR ( 7500 Fast PCR System , Applied Biosystems ) . Samples and inner primers were loaded onto a 96 . 96 Dynamic Array Fast IFC chip ( Fluidigm BioMark ) and analyzed as described ( Supplementary file 1H , [Owen et al . , 2015] ) . Ct values from two technical replicates were averaged and normalized by the average Ct value of three housekeeping genes ( g6pd , clathrin , and ubiquilin , [van Wolfswinkel et al . , 2014] ) to generate ΔCt values . Log2 fold-changes were determined by the ΔΔCt method by calculating the difference from the average ΔCt value of control RNAi replicates . Heatmap of average ΔΔCt values was generated by pheatmap in R . Bar graphs show mean ΔΔCt +/- standard deviation with individual ΔΔCt values . Statistical tests ( unpaired Student’s t-test or one-way ANOVA followed by Dunnett’s multiple comparisons test ) were performed between individual ΔΔCt values . FISH of the mRG of interest was performed in control RNAi animals and RNAi animals showing phenotypes ( ectopic pharynx or ectopic eyes ) in at least three independent experiments , and images taken with same intensity settings within an experiment . The extent of an mRG expression domain was measured in ImageJ as depicted in cartoons by blind scoring maximal intensity projections . For ndl-2 , the extent of the domain with strong expression and not total expression was measured . The length of the mRG expression domain was normalized by the length of the animal . Statistical analysis of expression domain shifts were determined by one-way ANOVA followed by Dunnett’s multiple comparisons test . Animals were fixed as for in situ hybridizations and then treated as described ( Newmark and Sánchez Alvarado , 2000 ) . A mouse anti-ARRESTIN antibody ( kindly provided by Kiyokazu Agata ) was used in a 1:5000 dilution , and an anti-mouse-Alexa conjugated antibody was used in a 1:500 dilution . Numbers of cintillo+and gd+cells were counted and normalized by the length between the anterior tip of the animal and the esophagus in control , fz5/8–4; ndk , and wntA; ndk RNAi animals after eight RNAi feedings ( Figure 5C , Figure 5—figure supplement 1E ) . One-way ANOVA and Dunnet’s post-test were used to determine significant differences between the different conditions and the control . Similarly , cells expressing the metalloproteinase mmp1 were counted in control , wntP-2 , ndl-3 , and ndl-3; wntP-2 RNAi trunk pieces after 12 RNAi feedings and two rounds of regeneration ( 20 dpa , screen RNAi protocol , Figure 3B ) . One-way ANOVA and Dunnet post-test were used to determine significant differences between the different conditions and the control . Minimum sample-size estimations were calculated using balanced one-way analysis of variance power calculation with k=4 ( mmp1 ) or k=3 ( cintillo and gd ) , f=0 . 8 , sig . level=0 . 05 , and power=0 . 8 ( n=5 . 3 or n=6 . 1 ) . Genbank: Homo sapiens: NP_068742 . 2 ( FGFRL1 ) . Mus musculus: NP_473412 . 1 ( FGFRL1 ) . Xenopus tropicalis NP_001011189 . 1 ( FGFRL1 ) . Strongylocentrotus purpuratus NP_001165523 . 1 ( FGFRL1 ) . Dj , Dugesia japonica: BAC20953 . 1 ( Ndk ) , BAP15931 . 1 ( Ndl-2 ) , BAQ21471 . 1 ( Ndl-3 ) , BAQ21471 . 1 ( Ndl-1 ) . Nematostella vectensis XP_001635234 . 1 ( FGFRL-1 ) . Uniprot: Ciona intestinalis F7BEX9 ( FGFRL1 ) . Genomic database: Capitella sp . I: CAPC1_170033 ( FGFRL1 ) . Lottia gigantean: LOTGI_167118 ( FGFRL1 ) . Schistosoma mansoni: Smp_052290 ( Ndk ) , Smp_036020 ( Ndl-5 ) Planmine/Genbank: Schmidtea mediterranea: dd_11285/ADD84674 . 1 ( Ndk ) , dd_12674/ADD84675 . 1 ( Ndl-4 ) , dd_5102/AFJ24803 . 1 ( Ndl-5 ) , dd_6604/ADD84676 . 1 ( Ndl-3 ) , dd_8310 ( Ndl-1 ) , dd_8340 ( Ndl-2 ) . Dendrocoelum lacteum: Dlac_193209/JAA92597 . 1 ( Ndk ) , Dlac_194186/JAA92596 . 1 ( Ndl-4 ) , Dlac_184398 ( Ndl-5 ) , Dlac_178408 ( Ndl-3-2 ) , Dlac_182339 ( Ndl-3-1 ) , Dlac_189993 ( Ndl-1 ) , Dlac_170672 ( Ndl-2-1 ) , Dlac_181923/JAA92595 . 1 ( Ndl-2-2 ) . Planaria torva: Ptor_24279 ( Ndk-1 ) , Ptor_18635 ( Ndk-2 ) , Ptor_24521 ( Ndl-4-1 ) , Ptor_34251 ( Ndl-4-2 ) , Ptor_36400 ( Ndl-5-1 ) , Ptor_68870 ( Ndl-5-2 ) , Ptor_24828 ( Ndl-3-1 ) , Ptor_27132 ( Ndl-3-2 ) , Ptor_29905 ( Ndl-1 ) , Ptor_23702 ( Ndl-2 ) . Polycelis tenuis: Pten_63627 ( Ndk-1 ) , Pten_14428 ( Ndk-2 ) , Pten_6171 ( Ndl-4 ) , Pten_43037 ( Ndl-5 ) , Pten_46975 ( Ndl-3 ) , Pten_39799 ( Ndl-1-1 ) , Pten_47685 ( Ndl-1-2 ) , Pten_41107 ( Ndl-2 ) . Polycelis nigra: Pnig_15421 ( Ndk-1 ) , Pnig_6593 ( Ndk-2 ) , Pnig_29523 ( Ndl-4-2 ) , Pnig_3947 ( Ndl-4-1 ) , Pnig_25001 ( Ndl-5 ) , Pnig_3933 ( Ndl-3 ) , Pnig_25308 ( Ndl-1 ) , Pnig_22111 ( Ndl-2 ) .	Some animals can regrow tissues that have been amputated . A group of flatworms called planarians are often used as a model to study the regeneration process because they are able to restore any lost tissue or even an entire animal from tiny pieces of the body . For regeneration to be successful , it is important to ensure that the new tissues form in the correct locations in the body . The planarian body is divided into three main parts: head , trunk and tail . Several gene products involved in specifying what tissues regenerate are made by muscle cells along the planarian body . Some of the genes are involved in mechanisms that allow cells to communicate with each other , such as the Wnt signaling pathway . These genes could form a coordinated system to control regeneration , but their precise roles remain poorly understood . Two groups of researchers have now independently identified genes that provide cells with information about their location in the flatworm body . Scimone , Cote et al . used a technique called RNA sequencing in individual muscle cells to identify 44 genes that have different levels of expression across the head , trunk and tail regions . These genes included multiple components of the Wnt signaling pathway and others that encode members of the FGFRL family of signaling proteins . Further experiments revealed two distinct sets of genes , or “gene circuits” , that provide information to correctly position tissues in the head and trunk regions of the worm . For example , inhibiting the activity of the wntP-2 or ndl-3 genes increased the size of the trunk of the worms and caused extra mouths and pharynges ( muscular organ used for eating ) to form . On the other hand , blocking the activity of genes in the other gene circuit caused the brain to expand and extra eyes to form . Another study by Lander and Petersen found that wntP-2 and ndl-3 act with another gene called ptk7 , which encodes another component of the Wnt signaling pathway . Together these findings suggest that the Wnt-FGFRL circuits act in a body-wide system that co-ordinates where and which new tissues form during regeneration . A future challenge is to find out how the genes identified in these studies interact and how the cells of the animal interpret this information to properly regenerate missing tissues .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2016	Two FGFRL-Wnt circuits organize the planarian anteroposterior axis

Subsets and Splits

No community queries yet

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